阅读说明：本技术 解耦建模方法和系统 (Decoupling modeling method and system ) 是由 吕健 王继禹 陆宁 曾温特 于 2017-10-09 设计创作，主要内容包括：一种解耦ETP模型处理器配置为：存储从电力系统获得的电力消耗数据；将所述电力消耗数据转换成电力激活时间周期和电力不激活时间周期；推导室外温度中的每个下对于预定热流(Q)的热阻(R)参数和热容(C)参数；将转换后的电力激活时间周期与实际电力激活时间周期比较；将转换后的电力不激活时间周期与实际电力不激活时间周期比较；针对与实际电力激活时间周期比较的转换后的电力激活时间周期,计算第一改善的热阻-热容-热流(RCQ)参数组和相应的第一室外温度；计算在所述电力激活时间周期期间,在每个室外温度下的Q参数；以及,计算在所述电力不激活时间周期期间,在每个室外温度下的R参数和C参数。(A decoupled ETP model processor configured to: storing power consumption data obtained from the power system; converting the power consumption data into a power active time period and a power inactive time period; deriving a thermal resistance (R) parameter and a thermal capacity (C) parameter for a predetermined heat flow (Q) at each of the outdoor temperatures; comparing the converted power activation time period with an actual power activation time period; comparing the converted power inactivity time period with an actual power inactivity time period; calculating a first improved set of thermal resistance-thermal capacitance-thermal flow (RCQ) parameters and a corresponding first outdoor temperature for the converted power activation time period compared to the actual power activation time period; calculating a Q parameter at each outdoor temperature during the power activation time period; and calculating an R parameter and a C parameter at each outdoor temperature during the power inactive time period.)

1. A method of improving energy parameter estimation, comprising:

storing power consumption data obtained from a plurality of power systems in a power consumption database;

converting, by a processing circuit, the power consumption data into a power active time period and a power inactive time period;

calculating, by the processing circuit, median time values for the power active time period and the power inactive time period at respective outdoor temperatures;

deriving, for the plurality of power systems, a thermal resistance (R) parameter and a thermal capacitance (C) parameter for a predetermined heat flow (Q) parameter at each of the respective outdoor temperatures;

comparing, by the processing circuit, the converted power activation time period to an actual power activation time period for the plurality of power systems;

comparing, by the processing circuit, the converted power inactivity time period to an actual power inactivity time period for the plurality of power systems;

calculating a first improved set of thermal resistance-thermal capacitance-thermal flow (RCQ) parameters and a corresponding first outdoor temperature for the converted power activation time period compared to the actual power activation time period;

calculating a second improved set of RCQ parameters and a corresponding second outdoor temperature for the converted power inactivity time period compared to the actual power inactivity time period; and

improving an energy parameter estimate by performing the first and second improved sets of RCQ parameters at the respective first and second outdoor temperatures for each of the plurality of power systems, wherein the improved energy parameter estimate identifies an energy efficiency to reduce an overall energy load within each of the plurality of power systems.

2. The method of claim 1, wherein the improved set of RCQ parameters is determined by minimizing an error between a converted power-on time period and an actual power-on time period, and between a converted power-off time period and an actual power-off time period.

3. The method of claim 1, wherein the power consumption data is limited to data within a temperature range between a predetermined upper temperature limit and a predetermined lower temperature limit.

4. The method of claim 1, wherein the method comprises an Equivalent Thermal Parameter (ETP) model for improving energy parameter estimation.

5. The method of claim 1, further comprising:

calculating, for the plurality of power systems, a Q parameter at each of the respective outdoor temperatures during the power activation time period; and

calculating, for the plurality of power systems, an R parameter and a C parameter at each of the respective outdoor temperatures during the power inactivity time period.

6. The method of claim 5, further comprising:

calculating, by the processing circuit, an estimated duration of each of the power activation time periods from the Q parameter calculated at each of the respective outdoor temperatures;

calculating, by the processing circuit, an estimated duration of each of the power inactivity time periods from the R and C parameters calculated at each of the respective outdoor temperatures; and the number of the first and second groups,

improving energy parameter estimation by decoupling daytime parameters from nighttime parameters by means of an estimated duration of each of the power active time periods estimated from the Q parameter and by means of an estimated duration of each of the power inactive time periods estimated from the R and C parameters to reduce errors corresponding to variations in each of the respective outdoor temperatures.

7. The method of claim 1, further comprising:

calculating an R adjustment coefficient, wherein the R adjustment coefficient is the ratio of the R parameter in the whole day to the R parameter at night;

calculating a C adjustment coefficient, wherein the C adjustment coefficient is the ratio of the C parameter in the whole day to the C parameter at night;

calculating a Q adjustment factor at each of the respective outdoor temperatures, the Q adjustment factor being a ratio of a full day Q parameter to a night Q parameter at each of the respective outdoor temperatures; and

calculating an RC adjustment factor at each of the respective outdoor temperatures that is a ratio of a full-day RC parameter to a night RC parameter at each of the respective outdoor temperatures, wherein the full-day R parameter, the full-day C parameter, the full-day Q parameter, and the full-day RC parameter are calculated from data taken over a 24 hour period, and

wherein the night R parameter, the night C parameter, the night Q parameter and the night RC parameter are calculated from data acquired during a period of no sunshine.

8. The method of claim 7, further comprising:

calculating a daytime R parameter, which is the product of the R adjustment coefficient and the nighttime optimal R parameter;

calculating a daytime C parameter which is a product of the C adjustment coefficient and the nighttime optimal C parameter;

calculating a daytime Q parameter that is a product of the Q adjustment coefficient at each of the respective outdoor temperatures and a nighttime optimal Q parameter;

calculating a daytime RC parameter that is a product of the RC adjustment coefficient at each of the respective outdoor temperatures and a nighttime optimal RC parameter; and

the energy parameter estimation is improved by adjusting a night parameter to be used as a daytime parameter by means of the daytime R parameter, the daytime C parameter, the daytime Q parameter, and the daytime RC parameter, to reduce an error corresponding to a change in sunshine.

9. The method of claim 1, wherein the plurality of power systems comprises a plurality of thermostatic control devices (TCAs).

10. The method of claim 9, wherein the plurality of TCAs includes a plurality of heating, ventilation, and air conditioning (HVAC) systems, and the power consumption database includes an HVAC consumption database.

11. A decoupled Equivalent Thermal Parameter (ETP) model processor, comprising:

circuitry configured to:

storing power consumption data obtained from a plurality of power systems in a power consumption database;

converting the power consumption data into a power active time period and a power inactive time period;

calculating median time values of the power activation time period and the power deactivation time period at the corresponding outdoor temperature;

deriving, for the plurality of power systems, parameters of a thermal resistance (R) parameter and a thermal capacity (C) parameter for a predetermined heat flow (Q) parameter at each of the respective outdoor temperatures;

comparing the converted power activation time period to an actual power activation time period for the plurality of power systems;

comparing, for the plurality of power systems, the converted power inactivity time period to an actual power inactivity time period;

calculating a first improved set of thermal resistance-thermal capacitance-thermal flow (RCQ) parameters and a corresponding first outdoor temperature for the converted power activation time period compared to the actual power activation time period;

calculating a second improved set of RCQ parameters and a corresponding second outdoor temperature for the converted power inactivity time period compared to the actual power inactivity time period;

calculating, for the plurality of power systems, a Q parameter at each of the respective outdoor temperatures during the power activation time period; and

calculating, for the plurality of power systems, an R parameter and a C parameter at each of the respective outdoor temperatures during the power inactivity time period.

12. A decoupled ETP model processor in accordance with claim 11, wherein said circuitry is further configured to:

calculating an estimated time length of each of the power activation time periods from the Q parameter calculated at each of the respective outdoor temperatures; and

calculating an estimated time length of each of the power non-activation time periods from the R parameter and the C parameter calculated at each of the respective outdoor temperatures,

wherein the calculated Q parameter, the calculated R parameter, and the calculated C parameter improve energy parameter estimation by decoupling daytime parameters from nighttime parameters by way of an estimated duration of each of the power active time periods estimated from the Q parameter and by way of an estimated duration of each of the power inactive time periods estimated from the R parameter and C parameter to reduce errors corresponding to variations in each of the respective outdoor temperatures.

13. A decoupled ETP model processor in accordance with claim 11, wherein the improved set of RCQ parameters is determined by minimizing an error between the converted power-on time period and the actual power-on time period, and between the converted power-off time period and the actual power-off time period.

14. A decoupled ETP model processor in accordance with claim 11, wherein the plurality of power systems comprises a plurality of thermostatic control devices (TCAs).

15. A post-tune decoupled Equivalent Thermal Parameter (ETP) model processor, comprising:

circuitry configured to:

storing power consumption data obtained from a plurality of power systems in a power consumption database;

converting the power consumption data into a power active time period and a power inactive time period;

calculating median time values of the power activation time period and the power deactivation time period at the corresponding outdoor temperature;

deriving, for the plurality of power systems, parameters of a thermal resistance (R) parameter and a thermal capacity (C) parameter for a predetermined heat flow (Q) parameter at each of the respective outdoor temperatures;

comparing the converted power activation time period to an actual power activation time period for the plurality of power systems;

comparing, for the plurality of power systems, the converted power inactivity time period to an actual power inactivity time period;

calculating a first improved set of thermal resistance-thermal capacitance-thermal flow (RCQ) parameters and a corresponding first outdoor temperature for the converted power activation time period compared to the actual power activation time period;

calculating a second improved set of RCQ parameters and a corresponding second outdoor temperature for the converted power inactivity time period compared to the actual power inactivity time period;

calculating an R adjustment coefficient, wherein the R adjustment coefficient is the ratio of the R parameter in the whole day to the R parameter at night;

calculating a C adjustment coefficient, wherein the C adjustment coefficient is the ratio of the C parameter in the whole day to the C parameter at night;

calculating a Q adjustment factor at the respective outdoor temperature, the Q adjustment factor being a ratio of a whole day Q parameter at the respective outdoor temperature to a night Q parameter at the respective outdoor temperature; and

calculating an RC adjustment factor at the respective outdoor temperature, the RC adjustment factor being a ratio of an entire day RC parameter at the respective outdoor temperature to a night RC parameter at the respective outdoor temperature.

16. The retrospectively decoupled ETP model processor of claim 15, wherein the circuitry is further configured to:

calculating a daytime R parameter, which is the product of the R adjustment coefficient and the nighttime optimal R parameter;

calculating a daytime C parameter which is a product of the C adjustment coefficient and the nighttime optimal C parameter;

calculating a daytime Q parameter, which is a product of the Q adjustment coefficient at the corresponding outdoor temperature and a nighttime optimal Q parameter; and

calculating a daytime RC parameter that is a product of the RC adjustment coefficient at each of the respective outdoor temperatures and a nighttime optimal RC parameter;

wherein the calculated daytime R parameter, the calculated daytime C parameter, the calculated daytime Q parameter, and the calculated daytime RC parameter improve the energy parameter estimation by adjusting a nighttime parameter to be used as the daytime parameter to reduce an error corresponding to a change in sunshine.

17. The post-tune decoupled ETP model processor of claim 15, wherein the full-day R, C, Q, and RC parameters are calculated from data taken over a 24-hour period, and

wherein the night R parameter, the night C parameter, the night Q parameter, and the night RC parameter are calculated from data acquired during a period in which there is no sunshine.

18. The post-tune decoupled ETP model processor of claim 15, wherein the improved set of RCQ parameters is determined by minimizing an error between the converted power-on time period and the actual power-on time period, and between the converted power-off time period and the actual power-off time period.

19. The retrospectively decoupled ETP model processor of claim 15, wherein the plurality of power systems includes a plurality of thermostatic control devices (TCAs).

20. The retrospective decoupled ETP model processor of claim 19, wherein the plurality of TCAs includes a plurality of heating, ventilation, and air conditioning (HVAC) systems and the power consumption database includes an HVAC consumption database.

Background

Thermostatic control equipment (TCAs), such as heating, ventilation and air conditioning (HVAC) units and water heaters, are commonly used as load-side resources for demand response projects and Home Energy Management (HEM) systems. In order to control the power consumption of a TCA unit without disturbing the user comfort, an Equivalent Thermal Parameter (ETP) model of the TCA unit is normally required to predict the ON/OFF (ON/OFF) state of the unit for hours. The ETP model uses the constant temperature set point, the outdoor temperature, and the initial room temperature as inputs to predict the subsequent on/off cycle.

In Ruzzelli et al, a load detection method based on signal processing methods (e.g., signal extraction, neural networks, spectral analysis, V-I trajectories, and wavelet transforms) is proposed. However, these methods are used to detect the load rather than deriving a model that can predict the load behavior over a later time interval. See A.G.Ruzzelli, C.Nicolas, A.Schoofsand G.M.P.O' Hare, "Real-Time registration and Profiling of applications through interface electric Sensor," 20107 th annular IEEE Communications society conference Sensor, Mesh and Ad Hoc Communications and Networks (SECON), Boston, MA,2010, pp.1-9, S.Makonin, F.Popowich, L.Bartram, B.Gill and I.V."AMPds:A public dataset for load disaggregation and eco-feedback research,"2013 IEEE Electrical Power&Energy Conference, halimax, NS,2013, pp.1-6, m.weiss, a.hellfenstein, f.matern and t.staake, "Leveraging meter Data storage hierarchy home applications," 2012 IEEE International Conference on innovative computing and Communications, Lugano,2012, pp.190-197, m.dong, p.c.m.meira, w.xuand c.y.chung, "Non-intuitive signal acquisition for Major reactive loads," in IEEE Transactions on Smart Grid, vol.4, No.3, 1421-1430, sept.2013, a.i.color Data and "interaction Data and" expression ", n.12. runtime on + g, m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.t.m.m.m.m.p.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.m.p., w.xuand c.m.m.n.t.m.n.1-2, m.m.m.m.m.m.m.m.2, m.m.m.m.m.ep.m.1-2, m.m.m.m.m.m.m.m.m.p.2, 2, m.m.m.m.m.p.1, m.m.m.m.m.m.2, 2, m.m.m.p.1, 2, m.m.m.m.m.p.p.m.m.m.m.m.2, 2.

For residential applications, researchers typically apply first or second order differential equations to represent the thermodynamics (thermal dynamics) of a single family when outdoor temperature changes. See, s.katipamola and n.lu, "Evaluation of residual HVAC control sequences for controlled response programs," ashrae trans, "vol.1, No.12, pp.1-12,2006, which is fully incorporated herein by reference. ETP model parameters are derived by physics-based methods or measurement-based methods.

Physics-based methods typically model houses in detail. See s.shao, m.pitpatanaspornand s.rahman, "Development of physical-based controlled response-enabled driver load models," in IEEE Transactions on Power Systems, vol.28, No.2, pp.607-614, May 2013; S.Ihara and F.C. Schwell, "physical based modification of color load pickup up," IEEE Trans.Power application. Syst., vol.PAS-100, No.9, pp.4142-4150, Sep.1981, E.Agneholm and J.Dalder, "Cold load pick-up of identification load," Proc.Inst.Electron.Eng., Gen., Transm, Distrib, vol.147, No.1, pp.44-50, Jan.2000, J.Yan, Q.G, Y.Liang, L.201and Z.Li, "Transmission and multiplication of electric impedance device for software," Sound and noise, sound and noise, and "sound and noise, sound and noise, sound and noise, sound and noise, sound and noise, sound, noise, sound, noise, sound, noise, sound, noise, sound, noise, sound, noise, sound, noise, sound, noise, sound, noise. Inputs include the material and thickness of the walls, the size and number of windows, the thermal mass in the house, the orientation of the house, etc. This type of approach is commonly used to model typical residential houses. When modeling a real house for HEM control purposes, it is impractical because many unpredictable factors (e.g., building material variations, tree shadowing, and vent locations) can have a significant impact on model accuracy.

Measurement-based methods have been used to overcome some of the deficiencies of physics-based methods for deriving ETP parameters. Lu, n.lu, proposes a data-driven HVAC model in which parameters are derived from curve fitting HVAC consumption curves. See n.lu, "An evaluation of the HVAC load potential for providing a compressive balancing service," IEEE trans.smart Grid, vol.3, No.3, pp.1263-1270, sep.2012, which is fully incorporated by reference in this specification. To account for the effects of outdoor temperature, adjustments may be made using a look-up table containing a set of HVAC parameters at each temperature range. However, because high-resolution HVAC consumption data for fractional metering is not available, HVAC consumption data generated from physics-based higher-order ETP models has been used primarily to derive and test this data-driven model. During validation using metered one minute HVAC consumption, the accuracy of the model was unsatisfactory.

The description of "background" provided herein is for the purpose of generally illustrating the background of the disclosure. Work of the presently named inventors, to the extent it is described in this background section, as well as aspects of the description that may not otherwise qualify as conventional at the time of filing, are neither expressly nor impliedly admitted as conventional for the present disclosure.

Disclosure of Invention

The embodiments described herein include the following aspects:

(1) a method of improving energy parameter estimation, comprising: storing power consumption data obtained from a plurality of power systems in a power consumption database; converting, by a processing circuit, the power consumption data into a power active time period and a power inactive time period; calculating, by the processing circuit, median time values for the power active time period and the power inactive time period at respective outdoor temperatures; deriving, for the plurality of power systems, a thermal resistance (R) parameter and a thermal capacitance (C) parameter for a predetermined heat flow (Q) parameter at each of the respective outdoor temperatures; comparing, by the processing circuit, the converted power activation time period to an actual power activation time period for the plurality of power systems; comparing, by the processing circuit, the converted power inactivity time period to an actual power inactivity time period for the plurality of power systems; calculating a first improved set of thermal resistance-thermal capacitance-thermal flow (RCQ) parameters and a corresponding first outdoor temperature for the converted power activation time period compared to the actual power activation time period; calculating a second improved set of RCQ parameters and a corresponding second outdoor temperature for the converted power inactivity time period compared to the actual power inactivity time period; and improving an energy parameter estimate by performing the first and second improved sets of RCQ parameters at the respective first and second outdoor temperatures for each of the plurality of power systems, wherein the improved energy parameter estimate identifies an energy efficiency to reduce an overall energy load within each of the plurality of power systems.

(2) The method of improving energy parameter estimation of (1), wherein the improved set of RCQ parameters is determined by minimizing an error between a converted power-on time period and an actual power-on time period, and between a converted power-off time period and an actual power-off time period.

(3) The method of improving energy parameter estimation of any of (1) or (2), wherein the power consumption data is limited to data within a temperature range between a predetermined upper limit temperature and a predetermined lower limit temperature.

(4) The method of improving energy parameter estimation of any of (1) to (3), wherein the method includes an Equivalent Thermal Parameter (ETP) model for improving energy parameter estimation.

(5) The method of improving energy parameter estimation of any of (1) to (4), the method further comprising: calculating, for the plurality of power systems, a Q parameter at each of the respective outdoor temperatures during the power activation time period; and calculating, for the plurality of power systems, an R parameter and a C parameter at each of the respective outdoor temperatures during the power inactivity time period.

(6) The method of improving energy parameter estimation of any of (1) to (5), the method further comprising: calculating, by the processing circuit, an estimated duration of each of the power activation time periods from the Q parameter calculated at each of the respective outdoor temperatures; calculating, by the processing circuit, an estimated duration of each of the power inactivity time periods from the R and C parameters calculated at each of the respective outdoor temperatures; and improving the energy parameter estimation by decoupling the daytime parameter from the nighttime parameter by means of the estimated duration of each of the power active time periods estimated from the Q parameter and by means of the estimated duration of each of the power inactive time periods estimated from the R and C parameters to reduce the error corresponding to the variation of each of the respective outdoor temperatures.

(7) The method of improving energy parameter estimation of any of (1) to (6), the method further comprising: calculating an R adjustment coefficient, wherein the R adjustment coefficient is the ratio of the R parameter in the whole day to the R parameter at night; calculating a C adjustment coefficient, wherein the C adjustment coefficient is the ratio of the C parameter in the whole day to the C parameter at night; calculating a Q adjustment factor at each of the respective outdoor temperatures, the Q adjustment factor being a ratio of a full day Q parameter to a night Q parameter at each of the respective outdoor temperatures; and calculating an RC adjustment coefficient at each of the respective outdoor temperatures, the RC adjustment coefficient being a ratio of a whole-day RC parameter to a night RC parameter at each of the respective outdoor temperatures, wherein the whole-day R parameter, the whole-day C parameter, the whole-day Q parameter, and the whole-day RC parameter are calculated from data taken during a 24-hour period, and wherein the night R parameter, the night C parameter, the night Q parameter, and the night RC parameter are calculated from data taken during a no-sun period.

(8) The method of improving energy parameter estimation of any of (1) to (7), the method further comprising: calculating a daytime R parameter, which is the product of the R adjustment coefficient and the nighttime optimal R parameter; calculating a daytime C parameter which is a product of the C adjustment coefficient and the nighttime optimal C parameter; calculating a daytime Q parameter that is a product of the Q adjustment coefficient at each of the respective outdoor temperatures and a nighttime optimal Q parameter; calculating a daytime RC parameter that is a product of the RC adjustment coefficient at each of the respective outdoor temperatures and a nighttime optimal RC parameter; and improving the energy parameter estimation by adjusting a nighttime parameter to be used as a daytime parameter by means of the daytime R parameter, the daytime C parameter, the daytime Q parameter, and the daytime RC parameter, to reduce an error corresponding to a change in sunshine.

(9) The method of improving energy parameter estimation of any of (1) to (8), wherein the plurality of power systems includes a plurality of thermostatic control devices (TCAs).

(10) The method of improving energy parameter estimation of any of (1) through (9), wherein the plurality of TCAs includes a plurality of heating, ventilation, and air conditioning (HVAC) systems, and the power consumption database includes an HVAC consumption database.

(11) A decoupled Equivalent Thermal Parameter (ETP) model processor includes a circuit. The circuit is configured to: storing power consumption data obtained from a plurality of power systems in a power consumption database; converting the power consumption data into a power active time period and a power inactive time period; calculating median time values of the power activation time period and the power deactivation time period at the corresponding outdoor temperature; deriving, for the plurality of power systems, parameters of a thermal resistance (R) parameter and a thermal capacity (C) parameter for a predetermined heat flow (Q) parameter at each of the respective outdoor temperatures; comparing the converted power activation time period to an actual power activation time period for the plurality of power systems; comparing, for the plurality of power systems, the converted power inactivity time period to an actual power inactivity time period; calculating a first improved set of thermal resistance-thermal capacitance-thermal flow (RCQ) parameters and a corresponding first outdoor temperature for the converted power activation time period compared to the actual power activation time period; calculating a second improved set of RCQ parameters and a corresponding second outdoor temperature for the converted power inactivity time period compared to the actual power inactivity time period; calculating, for the plurality of power systems, a Q parameter at each of the respective outdoor temperatures during the power activation time period; and calculating, for the plurality of power systems, an R parameter and a C parameter at each of the respective outdoor temperatures during the power inactivity time period.

(12) The decoupled ETP model processor of (11), wherein the circuitry is further configured to: calculating an estimated time length of each of the power activation time periods from the Q parameter calculated at each of the respective outdoor temperatures; and calculating an estimated duration of each of the power inactive time periods from the R and C parameters calculated at each of the respective outdoor temperatures, wherein the calculated Q parameter, the calculated R parameter, and the calculated C parameter improve energy parameter estimation by decoupling a daytime parameter from a nighttime parameter by the estimated duration of each of the power active time periods estimated from the Q parameter and by the estimated duration of each of the power inactive time periods estimated from the R and C parameters to reduce an error corresponding to a change in each of the respective outdoor temperatures.

(13) The decoupled ETP model processor of any of (11) or (12), wherein the improved set of RCQ parameters is determined by minimizing an error between the converted power-on time period and the actual power-on time period, and between the converted power-off time period and the actual power-off time period.

(14) The decoupled ETP model processor of any of (11) through (13), wherein the plurality of power systems includes a plurality of thermostatic control devices (TCAs).

(15) A post-tune decoupled ETP model processor includes circuitry. The circuit is configured to: storing power consumption data obtained from a plurality of power systems in a power consumption database; converting the power consumption data into a power active time period and a power inactive time period; calculating median time values of the power activation time period and the power deactivation time period at the corresponding outdoor temperature; deriving, for the plurality of power systems, parameters of a thermal resistance (R) parameter and a thermal capacity (C) parameter for a predetermined heat flow (Q) parameter at each of the respective outdoor temperatures; comparing the converted power activation time period to an actual power activation time period for the plurality of power systems; comparing, for the plurality of power systems, the converted power inactivity time period to an actual power inactivity time period; calculating a first improved set of thermal resistance-thermal capacitance-thermal flow (RCQ) parameters and a corresponding first outdoor temperature for the converted power activation time period compared to the actual power activation time period; calculating a second improved set of RCQ parameters and a corresponding second outdoor temperature for the converted power inactivity time period compared to the actual power inactivity time period; calculating an R adjustment coefficient, wherein the R adjustment coefficient is the ratio of the R parameter in the whole day to the R parameter at night; calculating a C adjustment coefficient, wherein the C adjustment coefficient is the ratio of the C parameter in the whole day to the C parameter at night; calculating a Q adjustment factor at the respective outdoor temperature, the Q adjustment factor being a ratio of a whole day Q parameter at the respective outdoor temperature to a night Q parameter at the respective outdoor temperature; and calculating an RC adjustment factor at the corresponding outdoor temperature, the RC adjustment factor being a ratio of an entire day RC parameter at the corresponding outdoor temperature to a night RC parameter at the corresponding outdoor temperature.

(16) The post-tune decoupled ETP model processor of (15), wherein the circuitry is further configured to: calculating a daytime R parameter which is a product of the R adjustment coefficient and the nighttime optimized R parameter; calculating a daytime R parameter, which is the product of the R adjustment coefficient and the nighttime optimal R parameter; calculating a daytime C parameter which is a product of the C adjustment coefficient and the nighttime optimal C parameter; calculating a daytime Q parameter, which is a product of the Q adjustment coefficient at the corresponding outdoor temperature and a nighttime optimal Q parameter; and calculating a daytime RC parameter that is a product of the RC adjustment coefficient at each of the respective outdoor temperatures and a nighttime optimal RC parameter; wherein the calculated daytime R parameter, the calculated daytime C parameter, the calculated daytime Q parameter, and the calculated daytime RC parameter improve the energy parameter estimation by adjusting a nighttime parameter to be used as the daytime parameter to reduce an error corresponding to a change in sunshine.

(17) The post-tune decoupled ETP model processor of any of (15) or (16), wherein the full-day R, C, Q, and RC parameters are calculated from data taken during a 24 hour period, and wherein the night R, C, Q, and RC parameters are calculated from data taken during periods of no sun exposure.

(18) The post-tune decoupled ETP model processor of any one of (15) to (17), wherein the improved set of RCQ parameters is determined by minimizing an error between the converted power-on time period and the actual power-on time period, and between the converted power-off time period and the actual power-off time period.

(19) The post-tune decoupling ETP model processor of any one of (15) through (18), wherein the plurality of power systems includes a plurality of thermostatic control devices (TCAs).

(20) The post-tune decoupled ETP model processor of any of (15) to (19), wherein the plurality of TCAs includes a plurality of heating, ventilation, and air conditioning (HVAC) systems and the power consumption database includes an HVAC consumption database.

The above paragraphs are provided as a general summary and are not intended to limit the scope of the claims. The described embodiments, together with further advantages, will be best understood by reference to the following detailed description taken in conjunction with the accompanying drawings.

Drawings

A more complete appreciation of the invention and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings, wherein:

FIG. 1A is a diagram illustrating an exemplary modeling system according to one embodiment;

FIG. 1B illustrates a second exemplary modeling system for an HVAC system according to one embodiment;

FIG. 2 is a diagram illustrating an output based on a first order HVAC ETP model according to one embodiment;

FIG. 3A is a diagram illustrating a first type of data for midnight data, according to one embodiment;

FIG. 3B is a diagram illustrating a second type of data for 24-hour whole day data, according to one embodiment;

FIG. 3C is a diagram illustrating a third type of data, which is discarded data, according to one embodiment;

FIG. 4 is a flow diagram of algorithm 1 showing six steps in the RCQ parameter estimation process for a first order ETP model, according to one embodiment;

FIG. 5 is a diagram illustrating an example of a circular eigenvector S matrix according to an embodiment;

FIG. 6A is a graph of different outdoor temperatures T according to one embodiment_outBox plot of down-time (ON) duration;

FIG. 6B is a boxplot of OFF time durations at different outdoor temperatures according to one embodiment;

FIG. 6C is a graph illustrating median values at different outdoor temperatures according to one embodimentAndwherein the complete working cycle t_ONAnd a complete off period t_OFFIs namedAnd

FIG. 7A is a diagram illustrating a set of RCQ values that are not capable of satisfying criteria A, where R represents thermal resistance, C represents thermal capacity, and Q represents heat flow, according to one embodiment.

FIG. 7B is a diagram illustrating criteria for ensuring t when modeling HVAC setpoint variations, according to one embodiment_ONAnd t_OFFWill change;

FIG. 8 is a median value comparing measured on/off durations according to one embodimentAndand the estimated operation/non-operation duration using the optimal RCQ groupAndbar graph of (2);

FIG. 9A is a diagram illustrating R × C and T according to one embodiment_outA linear correlation between;

FIG. 9B is a diagram illustrating Q and T according to one embodiment_outA linear correlation between;

FIG. 10 is a flow diagram of algorithm 2 decoupling the ETP model according to one embodiment;

FIG. 11 is a bar graph illustrating a comparison between median values of estimated on/off time periods and actual on/off time periods versus outdoor temperature according to one embodiment;

FIG. 12A is a diagram illustrating K according to one embodiment_Q(heat flow adjustment coefficient) and when i is 1 to 10(at T)_outWhole day heat flow model parameters of (at) c_outNight heat flow model parameters) with T_outThe relationship between;

FIG. 12B is a diagram illustrating a linear fit model according to one embodimentAnd T_outThe relationship between;

FIG. 13A is a diagram illustrating K according to one embodiment_QAnd when i is 1 to 10(at T)_outWhole day thermal resistance and thermal capacity model parameters at/_outNight time thermal resistance and thermal capacity model parameters) and T)_outThe relationship between;

FIG. 13B is a diagram illustrating a second order polynomial model according to one embodimentAnd T_outThe relationship between;

FIG. 14 is a flowchart of algorithm 3 for post-tune decoupling of the ETP model (Adjusted Decoupled-ETPmodel), according to one embodiment;

15A and 15B are bar graphs illustrating the mean Absolute Percent Error (APE) and Standard Deviation (SD) of total operating time based on nighttime data according to one embodiment;

15C and 15D are bar graphs illustrating APE and SD for total switch times according to one embodiment;

16A and 16B are bar graphs illustrating the on-time APE and switching APE, respectively, of the decoupled ETP1 model, according to one embodiment;

16C and 16D are bar graphs illustrating the on-time APE and switching APE, respectively, of the decoupled ETP2 model, according to one embodiment;

16E and 16F are bar graphs illustrating the on-time APE and the switching APE, respectively, of the adjusted decoupled ETP model, according to one embodiment; and

FIG. 17 is a block diagram showing a hardware description of a computer, according to one embodiment.

Detailed Description

The following description is intended to further explain the present disclosure by providing specific examples and embodiments thereof. These embodiments are illustrative and not exhaustive. The full scope of the disclosure is not to be limited to any specific embodiment disclosed in the specification, but only by the claims.

It will be appreciated that in the development of any such actual embodiment, numerous implementation-specific decisions must be made to achieve the developers' specific goals, such as compliance with application-and business-related constraints, which will vary from one implementation to another and from one developer to another.

A data-driven decoupled modeling approach for deriving model parameters for a thermostatic control device (TCA) is described herein using a heating, ventilation, and air conditioning (HVAC) unit as an example. The method estimates parameters of an HVAC Equivalent Thermal Parameter (ETP) model using outdoor temperature and HVAC power consumption as inputs.

The model parameter estimation process and adjustment algorithm are illustrated with HVAC as an example. Residential houses are used herein as the primary focus. Therefore, no model of an HVAC unit with a variable frequency drive is considered. However, adjustments may be made to integrate the embodiments described herein in an HVAC unit having a variable frequency drive. Moreover, the methods and systems described herein can be readily extended to parameter estimation for numerous other thermostatic control devices (TCAs), such as water heaters and refrigerators.

FIG. 1A is a diagram illustrating an exemplary modeling system 100 used in accordance with embodiments described herein. The method is tested and validated on a number of power systems 110 (e.g., TCA or HVAC systems of associated residential houses). For example, an HVAC system includes a power system 110 that provides heating and/or air conditioning to a premises, and also includes associated meters for controlling the power system 110. The power system may also include other TCAs, such as water heaters and refrigerators. Consumption data (e.g., actual one minute HVAC consumption data) is stored in the power consumption database 120.

The power ETP model processor 130 estimates parameters of the power ETP model using power consumption data (e.g., HVAC power consumption data) and outdoor temperature data from the power consumption database 120 as inputs. Algorithm 1 for estimating parameters of the power ETP model is described in detail with reference to fig. 4.

The power decoupling ETP model processor 140 is configured with circuitry for decoupling modeling of "on" and "off" periods of the power unit to improve modeling accuracy. In one example, the power decoupled ETP model processor 140 is an HVAC decoupled ETP model processor. The algorithm 2 for decoupling the modeling of the "on" and "off" periods is described in detail with reference to fig. 10.

The power adjusted decoupled ETP model processor 150 is configured with circuitry for deriving model parameters using only midnight data. In one example, the power post-tune decoupling ETP model processor 150 is an HVAC post-tune decoupling ETP model processor. This method is suitable for a case where daytime power consumption is highly distorted due to household activity. The algorithm 3 for decoupling and adjusting the model parameters is described in detail with reference to fig. 14.

The prediction processor 160 is configured with circuitry for processing and outputting models for predicting power unit behavior for energy management applications based on one or more of the power ETP model processor 130, the power decoupling ETP model processor 140, and the power adjustment ETP model processor 150. For example, without knowing the detailed parameters of the particular house in which the HVAC unit is located, the model output uses the outdoor temperature and HVAC power consumption data.

FIG. 1A shows the modeling system 100 as a single unit. However, the processors described herein may be combined into one or more individual processing units, or there may be other processing units not shown in fig. 1. The processing units shown in fig. 1 contain processing circuitry configured to perform the processing steps described herein, either individually or as a combined whole.

FIG. 1B illustrates a second exemplary modeling system 200 for an HVAC system. The data collection module 210 shows the devices in which data is collected, such as a thermometer 211 for each house and a thermostat 212 for each house.

The data processing module 215 stores the data collected from the data collection module 210 in the data processing database 220. The acquired data is processed by processor 225. Processor 225 may be a single processor or multiple processors operating in parallel. As shown in algorithm 1 in fig. 4, processor 225 derives an ETP model 230. Processor 225 also derives decoupled ETP model 240 as shown in algorithm 2 in fig. 10. Processor 225 also derives a post-tune decoupled ETP model 250 as shown in algorithm 3 in fig. 14.

FIG. 2 is a diagram illustrating a first order HVAC ETP model. For the current time step at T k, the indoor room temperature T at T k +1_roomThe first order ETP model can be expressed as:

wherein, T_outAnd T_roomRespectively outdoor temperature and room temperature, u_acIs the HVAC on/off state, Δ t is the duration of each time step, R represents the thermal resistance, C represents the thermal capacity, and Q represents the heat flow provided by the HVAC unit. Thus, the HVAC state u at the next time step k +1 can be determined by the following equation_ac：

Wherein, T^-And T⁺Respectively, a lower limit and an upper limit of the HVAC temperature deadband. May depend on specific processing objectives, equipment limitations, etcA predetermined upper temperature limit and a predetermined lower temperature limit are determined. For example, the predetermined upper temperature limit may be a higher temperature of the indoor comfort zone, such as 72 ° F, and the predetermined lower temperature limit may be a lower temperature of the indoor comfort zone, such as 68 ° F.

If T is_outHeld constant, (1) can be rewritten as:

wherein, t_ONAnd t_OFFRefers to the complete on and off cycle as shown in fig. 2.

One major advantage of the first-order ETP model for data-driven modeling is its simplicity. RCQ values can be easily derived from the group of five variables: t is⁺，T^-，T_out，t_ONAnd t_OFF. One disadvantage of the ETP model is that t_ONAnd t_OFFAt T_outAnd will change when changed. Thus, to meet the modeling accuracy requirement, each T_outA range requires a set of RCQ values. Furthermore, due to solar radiation, the daytime and nighttime RCQ values may not be the same, even though they are at the same T_outThe derived is pushed down. For example, one set of RCQ values may be for T of only 5-10 ° F during the nighttime portion of the day_outAnd (3) a range. A decoupled ETP model with tuning methods is described herein to address these modeling issues.

HVAC consumption P_acCan be converted into an active/inactive state u based on the following formula_ac(t)：

HVAC consumption P_acAnd outdoor temperature T_outCollected by researchers on the PECAN street program. See pecan street inc., "Dataport,",2017, which is incorporated by referenceThe literature is incorporated by reference in its entirety in this specification. One hundred houses with HVAC power consumption and outdoor temperature data measured in minutes were selected. Summer time periods are used to model the HVAC unit in its cooling mode.

To derive the RCQ parameters, data is filtered to exclude periods when HVAC power consumption is significantly distorted by household activity (e.g., manually turning the HVAC unit on and off, frequently changing thermostat set points, opening doors and/or windows for extended periods of time, etc.). Thus, energy usage data is divided into three types. FIG. 3A is a graph showing a 48 hour HVAC power consumption curve. The first type of data is midnight data from zero to 5 a.m., where minimal occupant activity may affect HVAC operation. FIG. 3B is a graph showing a 48 hour HVAC power consumption curve for a second type of data, which is 24 hour full day data, in which the occupant causes very little disruption to the HVAC operation. Figure 3C is a graph showing a third type of data, which is discarded data in which the HVAC unit is either not operating stably or is changing on/off modes frequently. When a home has an undersized HVAC unit or units, the HVAC operation cannot be properly modeled by an ETP model based on RCQ parameters.

The following exemplary algorithm may be used to select a data set according to embodiments described herein.

Exemplary Algorithm 0-select data set

Step 1 inputs the data set of house i.

Step 2 based on equation (5), P_ac(t) transition to the active/inactive state u_ac(t)。

Step 3 creating a work vector (ON vector) S_ON(i)＝{t_ON(i),T_out,avg(i),t_START(i)},i＝1,2…,N_ON。

Calculating the duration t of each working cycle_onCorresponding average T_out(rounded to 1 ℃ F.) and a start time t_start。

If workerMaximum and minimum T during a cycle_outThe difference between is larger than delta t_ON(e.g., 4 ℃ F.), discarding S_ON(i)。

If t_ON>Δt_ON,max(e.g., 40min) or t_ON<Δt_ON,min(e.g., 2min), then S is discarded_ON(i)。

Step 4 creates a null vector (OFF vector) S_OFF(i)＝{t_OFF(i),T_out,avg(i),t_START(i)},i＝1,2…,N_OFF。

Calculating the duration t of each off-cycle_OFFCorresponding average T_out(rounded to 1 ℃ F.) and a start time t_start。

Maximum and minimum T if during the off period_outThe difference between is larger than delta t_OFF(e.g., 4 ℃ F.), discarding S_OFF(i)。

If t_OFF>Δt_OFF,max(e.g., 60min) or t_OFF<Δt_OFF,min(e.g., 5min), then S is discarded_OFF(i)。

Step 5 derives the HVAC night database.

Calculate when t_START(i) From 6 pm to 6 am (variable in different seasons), at S_ON(i),i＝1,2…,N_ONNumber of all duty cycles N in_ON,NIGHT. Calculate when t_START(i) From 6 pm to 6 am (variable in different seasons), at S_OFF(i),i＝1,2…,N_OFFNumber of all inactive periods N in_OFF,NIGHT。

If N_ON,NIGHT<N_ON,NIGHT,min(e.g., 5min) or N_OFF,NIGHT<N_{OFF,NIGHT,min}(e.g., 5min), the data set is discarded.

Step 6 derives the HVAC whole day database.

Calculate when t_START(i) From 6 am to 6 pm (variable in different seasons), at S_ON(i),i＝1,2…,N_ONNumber of all duty cycles N in_ON,DAY. Calculate when t_START(i) From 6 am to 6 pm (variable in different seasons), at S_OFF(i),i＝1,2…,N_OFFNumber of all inactive periods N in_OFF,DAY。

If N_ON,NIGHT<N_ON,NIGHT,min(e.g., 2min) or N_OFF,NIGHT<N_{OFF,NIGHT,min}(e.g., 5min) or N_ON,DAY<N_ON,DAY,min(e.g., 2min) or N_OFF,DAY<N_OFF,DAY,min(e.g., 5min), the data set is discarded.

Step 7 steps 1 to 6 are repeated for all data sets in each house.

FIG. 4 is a flow chart of algorithm 1 showing the process steps in the RCQ parameter estimation process for a first order ETP model of an HVAC unit. The process steps of algorithm 1 may be implemented on a power ETP model processor 130 or other similar processor having circuitry configured to implement the process steps of algorithm 1. The HVAC unit is described for illustrative purposes only. Other power units, such as hot water heaters and refrigerators, may also be integrated with the process steps of algorithm 1.

In step S410, HVAC power consumption data and ambient temperature data of the next house are input. Algorithm 1 is described using residential houses as the structure in which the power units are located. However, other physical structures or buildings (e.g., commercial structures, commercial buildings, and warehouses) may also be included in algorithm 400.

In step S420, it is determined whether the entire day data can be used. If the data for the entire day can be used (yes decision in step S420), 24-hour data is selected in step S421. If the entire day data may not be used (NO decision in step S420), it is determined whether midnight data may be used in step S422. If midnight data can be used (yes decision in step S422), data from zero to 5 am is selected in step S423. If the midnight data cannot be used (no decision in step S422), the data set is discarded in step S424.

Obtaining HVAC operating/non-operating curves

In step (b)In step S430, the HVAC power consumption data is converted to an HVAC state using equation (5), where P_ac,maxIs the maximum power consumption during t ∈ τ, τ being the set of total time ranges of the input data, u_acIs the HVAC state.

Establishing a cyclic feature vector S

In step S440, a circulation vector S is calculated. Calculating the duration t of each duty and off cycle_ONAnd t_OFFAnd a corresponding average outdoor temperature T during each on-or off-cycle_{out_avg}. Each T is_{out_avg}Are rounded to their nearest integer values. Will t_ON(i)，t_OFF(i) And T_{out_avg}Stored in the cyclic feature vector s (i) corresponding to period i. Consistent with embodiments described herein, outdoor temperature may refer to a temperature outside of a physical structure in which an HVAC unit or other electrical power unit is located.

When the difference between the maximum and minimum outdoor temperatures in an active or inactive cycle is greater than a threshold Δ T_outThen the cycle is discarded to remove outliers. Use of Δ T_outA value of 4F because RCQ may not be representative when the temperature varies greatly over a period. The diagram of fig. 5 shows an example of a circular eigenvector S matrix (also shown in table I).

Table i. an example of a vector S

S(i)

S(1)

S(2)

S(3)

S(4)

S(5)

S(6)

…

tON(i)

NA

14min

NA

17min

NA

13min

…

tOFF(i)

10min

NA

11min

NA

15min

NA

…

Tout_avg(i)

85°F

85°F

89°F

89°F

84°F

81°F

…

So that t is_ONAnd t_OFFAnd T_{out_avg}Association

In step S450 of fig. 4, the median value of S is calculated to obtain vector M. For each T in S_{out_avg}Selecting t_ONAnd t_OFFTo obtain a new vector M. M is used when for a certain T_{out_avg}T of_ONOr t_OFFIn the absence, deducing (t)_ON,t_OFF) And T_{out_avg}The correlation between them. T is_{out_avg}Not included in M. An example of M is shown in Table II. The outdoor temperature group in M is renamed to T_o。t_ONAnd t_OFFIs named as the group of median valuesAndfig. 6A is a box plot of operating time at different outdoor temperatures. FIG. 6B is T_outBox plot of off-time durations at different values of (1), where T_outIs the outdoor temperature. FIG. 6C is a graph showing different outdoor temperaturesAndthe median value of (c).

Table ii an example of a vector M

M(i)

M(1)

M(2)

M(3)

M(4)

M(5)

M(6)

…

tM ON(i)

10min

10min

11min

11min

12min

13min

…

tM OFF(i)

35min

30min

27min

27min

18min

19min

…

To(i)

75°F

76°F

77°F

78°F

88°F

89°F

…

Derivation of RCQ parameters

In step S460 of fig. 4, R, C and Q parameters are calculated. At each outdoor temperature T_out∈T_oUnder, there is one groupAndand (4) parameters. Since Q represents the heat flow contributed by the HVAC unit, the value of Q only affects the duty cycle. During the curve fitting process, the RCQ values no longer accurately represent their physical characteristics, as compared to physics-based model methods. Therefore, a value of-500 is selected for all households. Only the values of R and C are adjusted. Thus, from equations (3) and (4), when the value of Q is fixed, R and C can be calculated as:

for the embodiments described herein, it is assumed that for each HVAC unit, T⁺And T-are selected to be 72 ° F and 68 ° F, respectively. When selecting a value for Q, two points are noted. First, Q is only a variable that affects the length of the HVAC duty cycle because during the off-cycle, the room temperature decay is independent of HVAC power consumption. Secondly, from equations (6) and (7), one can deduce: if the problem is treated as a curve fitting problem without regard to the physical meaning of the RCQ parameters, then any value can be selected for Q to calculate a corresponding set of values for R and C, while the measured duty/duty cycle characteristics can still be reproduced by a corresponding decoupled ETP model. Thus, while-500 is selected as the value for Q, others may select another value, such as-1000 or-1500. They will obtain another set of values for R and C, which can also be generated at a given T_outThe same curve fitting results below. However, in the next step, let T_outFrom min(T_o) Change to max (T)_o). RCQ values are excluded in subsequent steps when they do not yield satisfactory results at other outdoor temperatures.

Discard bad RCQ group

In step S470, the unacceptable R, C, and Q groups are discarded. Poor sets of RCQ values do not reproduce T_outFrom min (T)_o) Change to max (T)_o) The HVAC cycle characteristics. The following two criteria were used to exclude the poor set of RCQ values obtained.

In the A-standard for meeting HVAC capability requirements, equation (1) is rewritten as:

the inverse logarithm of the logarithmic function in equations (8) and (9) is required at T_outIn min (T)_o) To max (T)_o) Is positive, which results in:

Q×R＞T⁺-T_out(10)

the graph of FIG. 7A shows a set of RCQ values when the A criteria cannot be met. The physical meaning means that the HVAC unit is undersized and is no longer able to bring the room temperature down to T when the outdoor temperature is too high^-. Therefore, any set of RCQ values that violate the A standard needs to be discarded.

In the a standard, the modeling setpoint change requirement is satisfied. FIG. 7B is a graph illustrating the A criteria used to ensure when the HVAC setpoint changes (e.g., [ T ] T [)^-,T⁺]Change to) When t is_ONAnd t_OFFWill change accordingly. Mathematically, ifAnd isThe following equation holds true:

wherein f is_ON(ΔT,T_out) And f_OFF(ΔT,T_out) Is for determining Δ T and T for HVAC setpoint changes_outT of_ONAnd t_OFFIs a function of the minimum change in. It is to be noted that by usingAndsubstitution of T^-And T⁺Calculated from equations (8) and (9)Andany set of RCQ values that violate the a standard is discarded.

Selecting the best RCQ group

In step S480 of fig. 4, the best set R, C, and Q is selected. The best RCQ group (defined as being at the outdoor temperature T) of the remaining N groups may be selected by minimizing the error between the estimated on/off time period and the actual on/off time period using the following formula_fR obtained below_f、C_fAnd Q_fWherein, T_fIs to generate R_f、C_fAnd Q_fSpecific T of_out)：

Wherein the content of the first and second substances,andis the outdoor temperature T calculated using equations (8) and (9)_out(ii) an estimated operating duration and an off duration;andis at T from the actual measurement that can be found in the vector M_outThe median value of the down on/off duration. The optimum R can be determined_fC_fQ_fParameter set and corresponding R_fC_fQ_fParameter outdoor temperature T_fWherein the converted power active time period and the converted power inactive time period match the actual power active time period and the actual power inactive time period, respectively, within a predetermined error range. A set R, C and Q is obtained for one HVAC unit or other TCA by calculating the best RCQ set in step S480; the group is marked R_f、C_fAnd Q_f. These R_f、C_fAnd Q_fThe value is then used as an input to the decoupled ETP model. FIG. 8 is a bar graph comparing median measured on/off durationsAnd(ON, m and OFF, m in FIG. 8) and using the best R_fC_fQ_fGroup estimated on/off durationAnd(ON, e and OFF, e in FIG. 8). In this case, R_fC_fQ_fValues were derived at an outdoor temperature of 84 ° F.

From this comparison, the following observations can be made:

when 80 ℃ F. is less than or equal to T_outThe on/off duration of the model calculations shown in FIG. 8 fit the measurements well at 93F.

When T is, however_out<At 80 ° F, the estimated off-cycle duration is significantly longer than the measurement duration. When T is_out>At 93 ° F, the estimated duty cycle duration is significantly longer than the measurement duration.

This occurs when a first order is used to approximate higher order thermodynamic processes because the non-linearity of the actual thermodynamic process results in an estimated thermodynamic processAndincreasingly deviating from the measurementAndthis is at T_outSignificantly above or below the temperature T at which the RCQ parameter is derived_fOccurs when. To address this modeling problem, a decoupled ETP model is described herein.

In step S485, it is determined whether the current house under consideration is the last house on the list of houses to be processed. If the current house is the last house (YES determination in step S485), the method is terminated. If the current house is not the last house (NO decision in step S485), a counter is used to go to the (i +1) th house in step S490. The method restarts for the (i +1) th house in step S410.

The following exemplary algorithm may be used to select parameters of an ETP HVAC model according to embodiments described herein.

Exemplary Algorithm 1: selecting ETP Parameters of HVAC model

Step 1 input S_ONAccording to T_outTo S_ONAnd (6) sorting.

Step 2 calculationAnd

·is S_ONOf (2) is the same as T_out,avg.All of t_ON(i) Average value of (1), T₀(i) Is T_out,avg.。

·Finger S_ONOf (2) is the same as T_out,avg.All of t_OFF(i) Average value of (1), T₀(i) Is T_out,avg.。

Step 3 calculationIf M is_ON(i) And M_OFF(i) T in (1)₀(i) Are the same.

Step 4 calculation for at T using equations (6) to (7)₀(i) Each active/inactive pair of M (i) belowRCQ parameter of (1).

Suppose Q is-500, T⁺72 ° F, and T-68 ° F. Based on equations (6) to (7), N is calculated_MR and C values of group, wherein

T_out＝T₀(i)And is

Step 5 for each set R, C and Q obtained in step 4, a calculation is made for T⁺If 72 ° F, equation (10), T is satisfied_outIn the range of M (i), i-1, 2_MFrom min { T }₀(i) To max (T)₀(i) And (4) dividing. If not, the set R, C and Q will be discarded.

Q×R＞T⁺-T_out(10)

Step 6 calculates whether equation (11) is satisfied for each set R, C and Q remaining after step 5. If not, the set R, C and Q will be discarded.

Step 7 for each set R, C and Q remaining after step 6, select the set R, C and Q that minimizes equation (13), wherein,andis a result of the calculation of the time-domain,can be found in M (i). Selected R, C and Q are designated R_f，C_fAnd Q_f。

And 8: and (5) repeating the steps 1 to 7 for all the houses.

As shown in figures 7A and 7B,andhave different trends. If one R is used_fC_fQ_fBy grouping to model both duty cycles and off-cycles, simple parameter adjustment is not feasible because when one models both duty cycles and off-cycles, it is not possible to do soWhen the error of (2) is reduced,the error of (2) increases. From equations (8) and (9), t_ONAnd R_f、C_f、Q_fIs related, but t_OFFRelating only to R × C therefore, a decoupled ETP model is used to separate the modeling of the duty cycle and the off-duty cycle_outTo reduce the function corresponding to T_outVarying modeling errors. In contrast to equations (1) and (2), in the decoupled ETP model, T is k +1_roomAnd u_acThe calculation is as follows:

when T is_outConstant, equation (14) can be rewritten as:

wherein R can be obtained as described above_f、C_f. Modifications will be discussed below.

For the off period, R_f×C_f＝f_RC(T_out) Is defined as such that R_fC_fThe value being T_outAs a function of (c). From the equation (17),

wherein the content of the first and second substances,is the estimated dead time by using a decoupled ETP model. Find f_RCAnd T_outThe relationship between them. Calculating at each T by using the method set forth above_outR of_f×C_fNumerical value to calculate R derived from data of one hundred houses_f×C_fAnd T_outLinear dependence between them. The average correlation coefficient was 0.973, and the standard deviation was 0.0253. FIG. 9A is a graph showing actual numerical values and R_f×C_fAnd T_outLinear dependence between them. Therefore, for calculating f_RC(T_out) The linear regression model of (a) is:

f_RC(T_out)＝k_RC×T_out+b_RC(19)

to avoid modeling temperature sensitivity for all three RCQ parameters, let R ═ R_fAnd C ═ C_f(obtained above) to derive the duty cycle RCQ parameter. When R and C are fixed, Q is changed into T_outResults in:

calculating Q and T derived from one hundred houses_outThe correlation between them. The correlation coefficient had an average value of 0.941 and a standard deviation of 0.048. FIG. 9B is a diagram showing Q and T_out.Linear dependence between them. For each T_outThe linear regression model for Q was calculated as:

f_Q(T_out)＝k_Q×T_out+b_Q(21)

for each T using equation (21)_outAfter Q is calculated, the estimated duty cycle duration may be calculatedThe calculation is as follows:

fig. 10 is a flow chart of algorithm 2 for decoupling the ETP model. The process steps of algorithm 2 may be implemented on a power decoupled ETP model processor 140 or other similar processor having circuitry configured to implement the process steps of algorithm 2.

In step S1010, R_f、C_f、Q_fAnd HVAC power consumption and ambient temperature T of the ith premises_outIs an input. Calculating the input parameter R as described herein with reference to step S480 in FIG. 4_f、C_f、Q_f. HVAC Power consumption and ambient temperature T_outObtained by the HVAC consumption database 120 for the ith house.

In step S1020, f is calculated_RC(T_out)。

Calculate each T_outIs as follows

Wherein, T_out、T_offAnd M is calculated by step S450 in fig. 4.

FIG. 9A is a graph showing for each T_outF of (a)_RC(T_out) The actual calculation result of (2).

In step S1030, an implementation is made for each T_outF of (a)_RC(T_out) An example of which is shown in fig. 9A.

In step S1040, for each T, based on the following equation_outCalculating Q (T)_out)：

Wherein, T_out、And M is obtained in vector M in step S450 of fig. 4.

FIG. 9B is a graph showing for each T_outQ (T) of_out) The actual calculation result of (2).

In step S1050, an implementation is performed for each T_outQ (T) of_out) An example of which is shown in fig. 9B.

In step S1060, a decoupled ETP model is established for the ith house.

Wherein

In step S1065, it is determined whether the house list end point has been reached. If the house list end point has been reached ("YES" decision in step S1065), the process is stopped. If the house list end point has not been reached ("NO" decision in step S1065), the process proceeds to step S1070.

In step S1070, the counter is incremented to the (i +1) th house and algorithm 2 returns to step S1010 to repeat the process for the next house.

The following exemplary algorithm may be used to derive a decoupled ETP model according to embodiments described herein.

Exemplary algorithm 2: derivation decoupling ETP model

Step 1 inputs the ith (i ═ 1,2, …, N_house) Data set of individual house and repeat steps 1 to 7 of algorithm 1

Step 2 vectors are established for each group R, C and Q remaining after step 6 in Algorithm 1Wherein the content of the first and second substances,T₀(i) is used to obtain the corresponding data sets in M (i) of R (i), C (i), Q (i).

Step 3 calculation

Step 4 Using Linear regression f_RC(T_out)＝k_RC×T_out+b_RCTo fit withAnd T₀(i) The relation between i 1,2_D。

Step 5 calculation

Step 6 Using Linear regression f_Q(T_out)＝k_Q×T_out+b_QTo fit withAnd T₀(i) The relation between i 1,2_D。

Step 7 establishes a decoupled ETP model based on equations (14) to (15).

Step 8 steps 1 to 7 are repeated for all data sets for each house.

Fig. 11 is a bar graph showing a comparison between the median values of the estimated operation and non-operation periods and the actual operation/non-operation periods with respect to the outdoor temperature. The accuracy of the prediction of the decoupled ETP model is significantly improved compared to fig. 8 for the ETP model. A validation test was performed for one hundred houses using the PECAN street data set. Calculate t_ONAnd t_OFFOf the actual median value and estimatedAndaverage error between. One hundred houses at T_outThe average error over the entire range of values is 4.02 and 2.05 for the ETP model and the decoupled ETP model, respectively. If only low outdoor temperature (T) is considered_out<76 ° F) or high outdoor temperature (T)_out>95 ° F), the modeling error of the decoupled ETP model is typically 10 to 20 minutes less than that of the ETP model.

The decoupled ETP model can be applied to model all TCAs as long as consumption is accommodated by consistent thermostat settings that are not distorted by household activity (e.g., long-term opening of doors or windows, frequent changing of thermostat set points, or manual closing of TCA devices, etc.). However, in practice, TCA operation during the daytime may be severely distorted by human activity, as shown in fig. 3A and 3C. For most premises, the most useful data for deriving the TCA model is data measured between midnight and early morning when the resident is asleep (so that the impact of their activities is minimal). However, for HVAC units, using only midnight data may lead to inaccuracies because it does not take into account the thermodynamic effect of solar radiation on the house.

To address this problem, the parameters of the decoupled ETP model are adjusted using a post-tune decoupled ETP model that is derived using only midnight data. By decoupling the ETP model after tuning, the daytime operation can also be modeled accurately.

When the HVAC unit is working, T can be T of k +1_roomThe calculation is as follows:

when the HVAC unit is not in operation, T can be T at k +1_roomThe calculation is as follows:

wherein:

in equations (28) and (29), R_f,N，C_f,N，f_Q,N(T_out) And f and_RC,N(T_out) Are parameters of the decoupled ETP model calculated using the midnight dataset.K_Q(T_out) And K_RC(T_out) Are adjustment coefficients used to model daytime HVAC cycle behavior.

To derive these coefficients, n houses with such a 24 hour HVAC consumption of at least five days are selected: during this period, there was no significant distortion caused by household activity. The night model and the whole day model represent decoupled ETP models derived using midnight data and whole day data, respectively. The night model parameters for the ith HVAC unit includeAndthe parameters of the model all day includeAnd

the adjustment coefficient may be calculated as follows:

wherein, T_out∈T_o。

To account for the saturation effects of the HVAC unit in the low or high temperature range, an upper limit linear regression model is used.

FIG. 12A is a diagram showing K_QAnd when i ranges from 1 to 10An example of (1) and T_outThe relationship between them. FIG. 12B is a graph showing a linear fitting modelRelative to T_outThe figure (a).

Will K_RC(T_out) The calculation is as follows:

wherein, T_out∈T_o. Using an upper-bound second-order polynomial regression modelTo fit K_RCAnd T_outThe relationship between, such that:

FIG. 13A is a diagram showing K_QAnd when i ranges from 1 to 10An example of (1) and T_outThe relationship between them. FIG. 13B is a diagram illustrating a second order polynomial modelRelative to T_outThe figure (a).

When deriving the adjustment coefficients, only midnight data is used to derive decoupled ETP model parameters. The night model parameters are adjusted using equations (28) and (29) so that the effects of solar radiation can be properly reflected when modeling daytime HVAC cycle behavior.

Fig. 14 is a flow chart of algorithm 3 for post-tuning decoupling the ETP model. The process steps of algorithm 3 may be implemented on a power-adjusted decoupled ETP model processor 150 or other similar processor having circuitry configured to implement the process steps of algorithm 3.

In step S1410, n pieces of non-warped data are input. For example, the data may cover a 5 day, 24 hour period of home HVAC power consumption and ambient temperature data from the HVAC consumption database 120. FIG. 3B is a graph showing acceptable HVAC consumption curves associated with step 1.

In step S1420, the decoupled ETP model parameters for each home HVAC are calculated using midnight data from the estimates of algorithm 2 (see fig. 10):

in step S1430, the decoupled ETP model parameters for each home HVAC are calculated using the entire day data from the estimates of algorithm 2 (see fig. 10):

in step S1440, the adjustment coefficient is calculated based on the following formulaK_Q(T_out) And K_RC(T_out)：

In step S1450, a linear fit K is performed_Q(T_out) Fitting K with a polynomial_RC(T_out)。

In step S1460, a post-tune decoupled ETP model is established for each HVAC unit.

Wherein

The following exemplary algorithm may be used to derive a post-tune decoupled ETP model according to embodiments described herein.

Exemplary algorithm 3: derivation post-modulation decoupling ETP model

Step 1 selecting at least N of the same season_adjIndividual house data (P)_ac(t),T_out(t)), and repeating algorithm 1 to select at least N that enables derivation of both nighttime and full-day HVAC models_SSingle family house (single family house), N_AApartment (accommodation) and N_TA townhouse (townhouse).

Step 2 for each selected single family house i, steps 1 to 7 of algorithm 2 are repeated to obtain the whole day decoupled ETP model parametersAnd decoupling ETP model parameters at night

Step 3 calculating the adjustment gain for a single family data s modification

Step 4 fitting using the upper linear regression and upper quadratic regression in equations (38) and (40)And T_outThe relationship between them.

Step 5 for selected N_ARepeating the steps 2 to 4 for each apartment and obtaining the adjustment gainK_Q，A(T_out)，K_RC，A(T_out). For selected N_TRepeating the steps 2 to 4 for each villa and obtaining the adjustment gainK_Q，T(T_out)，K_RC，T(T_out)。

Step 6 for each house to be modeled, steps 1 to 7 of algorithm 2 are repeated, with parametersAnd establishing a night decoupling ETP model. Selecting corresponding adjustment gains obtained in the steps 4 to 5 according to the type of the houseK_Q(T_out)，K_RC(T_out)。

Step 7, establishing a post-tuning decoupling ETP model by using the following equation:

if u is_ac(k)＝1 (59)

If u is_ac(k)＝0(60)

The simulation results compare the performance of the ETP model with the decoupled ETP model and the post-tuning decoupled ETP model. Two performance criteria were used. The first criterion is the error in the measured and estimated total operating duration and the second criterion is the total number of handovers. The total operating time reflects the model accuracy in estimating the total power consumption. The total number of handovers indicates the model accuracy in estimating the handover behavior.

The accuracy of the ETP model and the decoupled ETP model was verified using the data for house # 3192, and the verification used 100 HVAC units.

Verification result using data of house # 3192

The results shown in table III compare the error in the total operating time and the total number of switches when using the ETP model and the decoupled ETP model, respectively. APE is defined as the absolute percent error.

Where x is the actual measured value,is an estimate using an ETP model or a decoupled ETP model. As shown in table III below, the results,the decoupled ETP model performs much better than the ETP model.

TABLE III Performance comparison

Verification results using 100 HVAC units

The bar graphs of fig. 15A and 15B show the Mean Absolute Percent Error (MAPE) and Standard Deviation (SD) of total operating time based on nighttime data for one hundred HVAC units. The error distribution is for an ETP or decoupled ETP model based on midnight data. The bar graphs of fig. 15C and 15D show MAPE and SD for the total number of switches for one hundred HVAC units. Table IV also shows these results. These results show that MAPE and SD, which decoupled the total working time and total number of switches of the ETP model, are much smaller than those of the ETP model. The error distribution of the ETP model for the work time APE shown in fig. 15A also confirms that the decoupled ETP model for the work time APE shown in fig. 15B performs significantly better than the ETP model and that its performance is consistent from household to household.

TABLE IV Performance comparison Using 100 House data

The performance of the ETP model, decoupled ETP model, and post-tune decoupled ETP model were verified using 24-hour whole day data as shown below. The performance of the post-tune decoupled ETP model was verified using data for house 3456, and the verification results used 50 HVAC units.

Verification result using 3456 house data

Table V shows the error comparison of the total operating time and the number of switching times for the different models.

Table v. comparison of results, premises 3456, austin, texas, from 7 months, 12 days, 15:18, 2015 to 21:57 at 8 months and 2 days in 2015

The "data input" column refers to the type of data used to build the model. The day-wide data uses 24 hour consumption of the HVAC unit in room number 3456 as input, while the night data uses the HVAC consumption from zero to 5 a.m. as input. The decoupled ETP model using the whole day data as input has the best performance with the least error. When the midnight model is used without being adjusted, the modeling error increases significantly. When the post-tune decoupled ETP model was used, the error was reduced by nearly 50%. The error is defined as the cumulative error over 21 days, the average daily modeling error for the total operating time is within 18 minutes, and the average error for the number of switches is within two. This shows improved performance when modeling HVAC switching characteristics.

Verification of 100 HVAC units

Table VI and fig. 16A-16F validate the results of case 1 using data collected from fifty houses during the summer months. The bar graphs of fig. 16A and 16B show the on-time APE and the switching APE, respectively, of the decoupling ETP1 model. The bar graphs of fig. 16C and 16D show the on-time APE and the switching APE, respectively, of the decoupled ETP2 model. The bar graphs of fig. 16E and 16F show the operating time APE and the switching APE, respectively, of the adjusted decoupled ETP model. These results show that the performance of all three decoupled ETP models is consistent from household to household. If only midnight data is available, a post-tune decoupled ETP model can be used to model daytime HVAC behavior.

Table vi performance comparison using data for 50 houses

Previous systems and methods have attempted to reduce peak power by controlling generators, thermostats, etc. based on a centralized system model, architecture, or configuration. Some systems and methods have used a state queuing model to model uncertainty in the aggregated HVAC load, where no control is involved. Other methods reshape the energy demand curve of a polymeric building, which is also based on a centralized system design and model. An HVAC aggregation model is based on an ETP model and a centralized control strategy. Other systems and methods use HVAC load control strategies to follow target load curves, which are based on a centralized multi-objective system model, architecture, or configuration. Still other systems and methods use a state space based HVAC simulation model based on an ETP model, where control is not involved.

In contrast, embodiments herein describe data-driven methods and models for estimating ETP model parameters of a TCA using power consumption and ambient temperature data of the TCA as inputs. Modeling of the HVAC unit is used to decouple the modeling of the on and off cycles, which simplifies the derivation of modeling parameters and significantly improves modeling accuracy. To adjust the distortion of the HVAC cycle behavior by human activity, the midnight data is used for estimation of ETP model parameters. The night model is adjusted using an adjustment procedure due to the influence of solar radiation.

In one embodiment, the functions and processes of modeling system 100 may be implemented in part by computer 1700 shown in FIG. 17. In particular, the power consumption database 120, the power ETP model processor 130, the power decoupling ETP model processor 140, the power post-tune decoupling ETP model processor 150, and/or the prediction processor 160 may be implemented by the computer 1700, either individually or together as one or more units.

A hardware description of a computer 1700 according to an exemplary embodiment is described with reference to fig. 17. In fig. 17, computer 1700 includes a CPU 1701 that implements the processes described herein. Process data and instructions may be stored in the memory 1702. These processes and instructions may also be stored on a storage media disk 1704, such as a hard disk (HDD) or portable storage media, or may be stored remotely. Moreover, the claimed improvements are not limited by the form of computer-readable media storing the instructions of the inventive processes. For example, the instructions may be stored on a CD, DVD, flash memory, RAM, ROM, PROM, EPROM, EEPROM, hard disk, or any other information processing device (e.g., a server or computer) in communication with computer 1700.

Moreover, the claimed improvements may be implemented as a combination of the CPU 1701 and an operating system (e.g.:Solaris，Appleand other systems known to those skilled in the art) executing utility applications, daemons, or components of an operating system, or combinations thereof to provide a utility function

To implement the computer 1700, the hardware elements may be implemented by numerous circuit elements known to those skilled in the art. For example, the CPU 1701 may be from Intel Corporation of the United statesOrProcessors, or AMD from the United statesA processor, or may be other processor types as will be appreciated by those skilled in the art. Alternatively, the CPU 1701 may be implemented on an FPGA, ASIC, PLD, or using discrete logic circuitry, as will be appreciated by those skilled in the art. Also, the CPU 1701 may be implemented as multiple processors working in concert in parallel to execute the instructions of the inventive process described above.

The computer 1700 in FIG. 17 also includes a network controller 1706, such as an Intel Ethernet PRO network interface card from Intel Corporation of the United states, for interacting with the network 1724. As can be appreciated, the network 1724 can be a public network, such as the Internet, or a private networkThe network, such as a LAN or WAN network, or any combination thereof, may also include PSTN or ISDN subnets. The network 1724 may also be wired, such as an ethernet network, or may be wireless, such as a cellular network, including EDGE, 3G, and 4G wireless cellular systems. The wireless network may also beOr any other known form of wireless communication.

The computer 1700 also includes a display 1710 (e.g., Hewlett packard)HPL2445w liquid crystal display) interactive display controller 1708, such as from NVIDIA Corporation of the united statesGTX orA graphics adapter. The general purpose I/O interface 1712 interfaces with a keyboard and/or mouse 1714 and an optional touch screen panel 1716 on or separate from the display 1710. The general I/O interface 1712 also connects to a variety of peripheral devices 1718, including printers and scanners, such as from Hewlett PackardOr

The general purpose storage controller 1720 interfaces the storage media disk 1704 with a communication bus 1722, which may be an ISA, EISA, VESA, PCI, or similar device, for interconnecting all of the components of the computer 1700. Since features of the display 1710, keyboard and/or mouse 1714, as well as the display controller 1708, storage controller 1720, network controller 1706, and general purpose I/O interface 1712 are known, a description of their general features and functionality is omitted here for the sake of brevity.

A number of embodiments have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the disclosure. For example, preferred results may be achieved if the steps of the disclosed techniques were performed in a different sequence, if components in the disclosed systems were combined in a different manner, or if components were replaced or supplemented by other components. The functions, processes, and algorithms described herein may be implemented in hardware or software executed by hardware, including a computer processor and/or programmable circuitry configured to execute program code and/or computer instructions to perform the functions, processes, and algorithms described herein. Furthermore, embodiments may be implemented on modules or hardware different from those described. Accordingly, other embodiments are within the scope of what may be claimed.

The foregoing discussion describes merely exemplary embodiments of the present disclosure. As will be understood by those skilled in the art, the present disclosure may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. Accordingly, the disclosure is intended to be illustrative, but not limiting, of the scope of the disclosure and claims. The present disclosure, including any readily discernible variants of the teachings herein, define, in part, the scope of the foregoing claim terminology such that no inventive subject matter is dedicated to the public.