阅读说明：本技术 一种基于长短时神经网络的居民负荷曲线分解方法 (Long-time neural network-based resident load curve decomposition method ) 是由 林顺富 黄佳凌 李东东 詹银枫 于 2021-07-30 设计创作，主要内容包括：本发明公开了一种基于长短时神经网络的居民负荷曲线分解方法,包括利用数据采集设备采集负荷的电压与电流数据,并计算出负荷的总有功功率P,根据开关事件合集和所述负荷总有功功率数据P选择数据样本X,将所述数据样本X划分训练集和验证集；对所述训练集和所述验证集进行归一化处理；对归一化处理后的训练集和验证集进行神经网络训练,得到预测的负荷数据；基于所述预测的负荷数据和所述开关事件集合进行负荷曲线分解,得到单一负荷曲线。本发明提出的基于长短时神经网络的居民负荷曲线分解方法,既能实现组合负荷曲线中单一负荷的曲线分解,得到单一负荷详细运行曲线图,又能分解出每一种负荷具体的运行情况和消耗电能。(The invention discloses a resident load curve decomposition method based on a long-time and short-time neural network, which comprises the steps of collecting voltage and current data of a load by using data collection equipment, calculating the total active power P of the load, selecting a data sample X according to a switching event set and the total active power data P of the load, and dividing the data sample X into a training set and a verification set; carrying out normalization processing on the training set and the verification set; carrying out neural network training on the training set and the verification set after the normalization processing to obtain predicted load data; and decomposing a load curve based on the predicted load data and the switching event set to obtain a single load curve. The resident load curve decomposition method based on the long and short time neural network can realize the curve decomposition of a single load in a combined load curve to obtain a detailed operation curve chart of the single load, and can decompose the specific operation condition and the consumed electric energy of each load.)

1. A method for decomposing a resident load curve based on a long-time and short-time neural network is characterized by comprising the following steps:

acquiring voltage and current data of a load by using data acquisition equipment, calculating the total active power P of the load, selecting a data sample X according to a switching event set and the total active power data P of the load, and dividing the data sample X into a training set and a verification set;

carrying out normalization processing on the training set and the verification set;

carrying out neural network training on the training set and the verification set after the normalization processing to obtain predicted load data;

and decomposing a load curve based on the predicted load data and the switching event set to obtain a single load curve.

2. The decomposition method of load curve of residents based on long and short term neural network as claimed in claim 1, wherein: performing on and off event determination on the set of switching events comprises,

if the event is an open event, recording the open event load type;

and if the event is an off event, returning to the continuous judgment.

3. The long-and-short-term neural network-based decomposition method of load curve of residents according to claim 1 or 2, wherein: matching the corresponding event includes, according to the switch event load type,

if the load is a constant-power load, matching a first load switch event of the same type after the switch event in the switch event set;

if the load is a nonlinear variable power load, matching the last off event in the operation period of the switch event set;

and a new set of switching events is formed by discharging in the matched order.

4. The decomposition method of load curve of residents based on long and short term neural network as claimed in claim 1, wherein: the neural network comprises a neural network having a plurality of neural networks,

a forgetting layer, an input layer, an update layer, and an output layer.

5. The long-term and short-term based neural network of claim 4The decomposition method of the load curve of the residents in the network is characterized in that: the forgetting layer outputs information f_tComprises the steps of (a) preparing a mixture of a plurality of raw materials,

the forgetting layer screens current input information and previous output information through a sigmoid function and judges whether the part of information needs to be reserved or not;

if 0 is finally output after screening, the part of information can be forgotten, and if 1 is finally output after screening, the part of information needs to be reserved;

passing said retained part of information to the next unit, said forgetting layer outputting information f_tComprises the steps of (a) preparing a mixture of a plurality of raw materials,

f_t＝σ(W_f[h_t-1,x_t]+b_f)

where σ denotes a sigmoid function, W_fWeight vector representing forgetting layer, [ h ]_t-1，x_t]Representing the combination of the previous output information and the current input information into an information vector, b_fA bias term representing a forgetting layer.

6. The decomposition method of load curve of residents based on long and short term neural network as claimed in claim 5, wherein: the obtaining of the candidate information may include,

the current input information x of the input layer is input_tAnd the previous output information h_t-1Information confirmation is carried out through a sigmoid function to obtain updated information i_tMeanwhile, candidate information is obtained through tanh functionComprises the steps of (a) preparing a mixture of a plurality of raw materials,

i_t＝σ(W_i[h_t-1,x_t]+b_i)

wherein, W_i、W_cDifferent weight vectors representing two functions of the input layer, b_i、b_cRepresenting the bias terms of the two functions of the input layer.

7. The decomposition method for load curve of residents based on long and short term neural network as claimed in any one of claims 4 to 6, wherein: based on previous cell information C_t-1And said forgetting layer output information f_tConstructing old information and utilizing the updated information i_tAnd the candidate informationObtaining new information;

the updating layer combines old information and new information to generate current information, and inputs the current information into the output layer, the current information C_tComprises the steps of (a) preparing a mixture of a plurality of raw materials,

wherein, C_t-1Representing previous unit information, f_tIndicating forgetting layer output information, i_tThe information is represented as an update of the information,representing candidate information.

8. The long-and-short-term neural network-based decomposition method of load curve of residents according to claim 7, wherein: the output layer outputs intermediate information and information input to a next unit, the intermediate information including,

o_t＝σ(W_o[h_t-1，x_t]+b₀)

wherein o is_tRepresenting intermediate information, W_oWeight vector representing output layer, b₀A bias term representing an output layer;

the information input to the next unit includes,

h_t＝o_ttanh(C_t)

wherein h is_tInformation indicating that the next unit is input.

9. The decomposition method of load curve of residents based on long and short term neural network as claimed in claim 1, wherein: the obtaining of the predicted load data may include,

and when the error value of the training neural network is smaller than the set error value or the training times are larger than the set training times, stopping training to obtain the predicted load data.

Technical Field

The invention relates to the technical field of non-invasive load monitoring, in particular to a resident load curve decomposition method based on a long-time and short-time neural network.

Background

The proportion of the domestic electricity consumption of urban and rural residents in the electricity consumption of the whole society is gradually improved, and with the popularization and application of the electricity internet of things technology and the like, the perception and acquisition of the use information of the internal electric appliances of the family users become a research hotspot in recent years; the use information of the household electrical appliances can enable an electric power company to further know the load composition condition of residents, and is beneficial to implementation of energy-saving measures and economic operation of a power grid; the traditional intrusive load monitoring needs to install a separate sensing device such as an intelligent socket and the like for each resident electrical appliance, so that the cost is high and the maintenance is inconvenient; the non-intrusive load monitoring (NILM) technology proposed by Hart professor of massachusetts university of america has the potential to obtain the power consumption data of each electrical appliance in a household through data such as total household power, total current and the like, and as residential household monitoring equipment based on the NILM technology has the advantages of low cost and easy maintenance, the residential household monitoring equipment has been widely paid attention and researched in recent years.

With the improvement of the intelligent technology level, load decomposition also becomes a research hotspot in the non-invasive monitoring technology, but the research of load curve decomposition is rarely carried out; anyilon and xu have just proposed a non-invasive load decomposition method based on a deep sequence translation model, but the method needs to establish a huge state code library, if the data in the state code library is less, the decomposition accuracy is affected, and it is difficult to establish state code libraries of various accurate loads; the technical scheme is that the non-invasive load decomposition method considering the state behaviors of time segments is proposed by the people of red mercy, grandfather, Liu flare and the like, but the two methods can only decompose the energy of the load and cannot decompose the specific operating condition curve of each load.

Disclosure of Invention

This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.

The present invention has been made in view of the above-mentioned conventional problems.

Therefore, the technical problem solved by the invention is as follows: the prior art method can only decompose the energy of the load, and cannot decompose the specific operating condition curve of each load.

In order to solve the technical problems, the invention provides the following technical scheme: acquiring voltage and current data of a load by using data acquisition equipment, calculating the total active power P of the load, selecting a data sample X according to a switching event set and the total active power data P of the load, and dividing the data sample X into a training set and a verification set; carrying out normalization processing on the training set and the verification set; carrying out neural network training on the training set and the verification set after the normalization processing to obtain predicted load data; and decomposing a load curve based on the predicted load data and the switching event set to obtain a single load curve.

As a preferable aspect of the method for decomposing the load curve of the residents based on the long and short term neural network according to the present invention, wherein: performing on and off event determination on the set of switching events comprises,

if the event is an open event, recording the open event load type;

and if the event is an off event, returning to the continuous judgment.

As a preferable aspect of the method for decomposing the load curve of the residents based on the long and short term neural network according to the present invention, wherein: matching the corresponding event includes, according to the switch event load type,

if the load is a constant-power load, matching a first load switch event of the same type after the switch event in the switch event set;

if the load is a nonlinear variable power load, matching the last off event in the operation period of the switch event set;

and a new set of switching events is formed by discharging in the matched order.

As a preferable aspect of the method for decomposing the load curve of the residents based on the long and short term neural network according to the present invention, wherein: the neural network is a long-time and short-time neural network, and the long-time and short-time neural network comprises a forgetting layer, an input layer, an updating layer and an output layer.

As a preferable aspect of the method for decomposing the load curve of the residents based on the long and short term neural network according to the present invention, wherein: the forgetting layer screens the current input information and the previous output information through a sigmoid function to judge whether the part of information needs to be reserved and included at present,

if the commander is selected and 0 is finally output, the part of information can be forgotten, and if the commander is selected and 1 is finally output, the part of information needs to be reserved;

passing said retained part of information to the next unit, said forgetting layer outputting information f_tComprises the steps of (a) preparing a mixture of a plurality of raw materials,

f_t＝σ(W_f[h_t-1,x_t]+b_f)

where σ denotes a sigmoid function, W_fWeight vector representing forgetting layer, [ h ]_t-1，x_t]Representing the combination of the previous output information and the current input information into an information vector, b_fA bias term representing a forgetting layer.

As a preferable aspect of the method for decomposing the load curve of the residents based on the long and short term neural network according to the present invention, wherein: the obtaining of the candidate information may include,

the current input information x of the input layer is input_tAnd the previous output information h_t-1Information confirmation is carried out through a sigmoid function to obtain updated information i_tMeanwhile, candidate information is obtained through tanh functionComprises the steps of (a) preparing a mixture of a plurality of raw materials,

i_t＝σ(W_i[h_t-1,x_t]+b_i)

wherein, W_i、W_cDifferent weight vectors representing two functions of the input layer, b_i、b_cRepresenting the bias terms of the two functions of the input layer.

As a preferable aspect of the method for decomposing the load curve of the residents based on the long and short term neural network according to the present invention, wherein: based on previous cell information C_t-1And said forgetting layer output information f_tConstructing old information and utilizing the updated information i_tAnd the candidate informationObtaining new information;

the updating layer combines old information and new information to generate current information, and inputs the current information into the output layer, the current information C_tComprises the steps of (a) preparing a mixture of a plurality of raw materials,

wherein, C_t-1Representing previous unit information, f_tIndicating forgetting layer output information, i_tThe information is represented as an update of the information,representing candidate information.

The output layer outputs intermediate information and inputs information of a next unit,

the intermediate information may include, for example,

o_t＝σ(W_o[h_t-1，x_t]+b₀)

wherein o is_tRepresenting intermediate information, W_oWeight vector representing output layer, b₀A bias term representing an output layer;

the information input to the next unit includes,

h_t＝o_ttanh(C_t)

wherein h is_tInformation indicating that the next unit is input.

As a preferable aspect of the method for decomposing the load curve of the residents based on the long and short term neural network according to the present invention, wherein: the obtaining of the predicted load data may include,

and when the error value of the training neural network is smaller than the set error value or the training times are larger than the set training times, stopping training to obtain the predicted load data.

The invention has the beneficial effects that: the invention overcomes the defects that the traditional method can only decompose the energy of the load and can not decompose the specific operating condition curve of each load, provides a dramatic load curve decomposition method based on a long-time and short-time neural network, realizes the curve decomposition of a single load in a combined load curve, and obtains a detailed operating curve graph, operating conditions and electric energy consumption of the single load.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise. Wherein:

fig. 1 is a flow chart of switching event matching of a method for decomposing a load curve of a resident based on a long-and-short-term neural network according to an embodiment of the present invention;

fig. 2 is a schematic diagram of an operating principle of a long-term neural network of a method for decomposing a load curve of a resident based on the long-term neural network according to an embodiment of the present invention;

fig. 3 is a group scene power curve diagram of a decomposition method of a residential load curve based on a long-and-short-term neural network according to an embodiment of the present invention;

fig. 4 is a combined data power data graph of a decomposition method of a load curve of a resident based on a long-and-short-term neural network according to an embodiment of the present invention;

fig. 5 is a graph after data prediction and filling of a method for decomposing a load curve of a resident based on a long-and-short-term neural network according to an embodiment of the present invention;

FIG. 6 is a graph of a long-term neural network-based decomposition method of load curves of residents according to an embodiment of the present invention after the operation of the long-term neural network-based decomposition method of load curves of residents is performed;

fig. 7 is a combined scene load curve exploded view of a decomposition method of a residential load curve based on a long-and-short-term neural network according to an embodiment of the present invention;

fig. 8 is an exploded view of a linear interpolation load curve of a decomposition method of a residential load curve based on a long-and-short-term neural network according to an embodiment of the present invention;

fig. 9 is a REDD data power curve diagram of a method for decomposing a load curve of a resident based on a long-and-short-term neural network according to an embodiment of the present invention.

Detailed Description

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, specific embodiments accompanied with figures are described in detail below, and it is apparent that the described embodiments are a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present invention, shall fall within the protection scope of the present invention.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.

Furthermore, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.

The present invention will be described in detail with reference to the drawings, wherein the cross-sectional views illustrating the structure of the device are not enlarged partially in general scale for convenience of illustration, and the drawings are only exemplary and should not be construed as limiting the scope of the present invention. In addition, the three-dimensional dimensions of length, width and depth should be included in the actual fabrication.

Meanwhile, in the description of the present invention, it should be noted that the terms "upper, lower, inner and outer" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of describing the present invention and simplifying the description, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation and operate, and thus, cannot be construed as limiting the present invention. Furthermore, the terms first, second, or third are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.

The terms "mounted, connected and connected" in the present invention are to be understood broadly, unless otherwise explicitly specified or limited, for example: can be fixedly connected, detachably connected or integrally connected; they may be mechanically, electrically, or directly connected, or indirectly connected through intervening media, or may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.

Example 1

Referring to fig. 1 to 2, in an embodiment of the present invention, a method for decomposing a load curve of a resident based on a long-time and short-time neural network is provided, including:

s1: the method comprises the steps of collecting voltage and current data of a load by using data collection equipment, calculating total active power P of the load, selecting a data sample X according to a switching event collection and the total active power data P of the load, and dividing the data sample X into a training set and a verification set. It should be noted that:

(1) judging the on and off events of the switch event set; if the event is an open event, recording the open event load type; if the event is a closing event, returning to continue judging;

(2) matching the corresponding event includes, based on the switching event load type,

if the load is a constant-power load, matching a first load-off event of the same type after the on event in the switch event set;

if the load is a nonlinear variable power load, matching the last off event in the operation period of the switch event set;

and a new set of switching events is formed by discharging in the matched order.

S2: and carrying out normalization processing on the training set and the verification set.

S3: and carrying out neural network training on the training set and the verification set after the normalization processing to obtain predicted load data. It should be noted that:

the neural network is a long-time and short-time neural network, and the long-time and short-time neural network comprises a forgetting layer, an input layer, an updating layer and an output layer;

forgetting layer output information f_tComprises the steps of (a) preparing a mixture of a plurality of raw materials,

the forgetting layer screens the current input information and the previous output information through a sigmoid function and judges whether the part of information needs to be reserved or not;

if 0 is finally output, the partial information can be forgotten, if 1 is finally output, the partial information needs to be reserved,

passing the retained part of the information to the next unit, forgetting the output information f of the layer_tThe method comprises the following steps:

f_t＝σ(W_f[h_t-1,x_t]+b_f)

where σ denotes a sigmoid function, W_fWeight vector representing forgetting layer, [ h ]_t-1，x_t]Representing the combination of the previous output information and the current input information into an information vector, b_fA bias term representing a forgetting layer;

further, the obtaining of the candidate information includes,

inputting the current input information x of the input layer_tAnd the previous output information h_t-1Information confirmation is carried out through a sigmoid function to obtain updated information i_tMeanwhile, candidate information is obtained through tanh functionComprises the steps of (a) preparing a mixture of a plurality of raw materials,

i_t＝σ(W_i[h_t-1,x_t]+b_i)

wherein, W_i、W_cDifferent weight vectors representing two functions of the input layer, b_i、b_cBias terms representing two functions of the input layer;

still further, based on the previous cell information C_t-1And forgetting layer output information f_tBuild old information, utilize updated information i_tAnd candidate informationObtaining new information; the updating layer combines the old information and the new information to generate current information, and inputs the current information into the output layer, the current information C_tComprises the steps of (a) preparing a mixture of a plurality of raw materials,

wherein, C_t-1Representing previous unit information, f_tIndicating forgetting layer output information, i_tThe information is represented as an update of the information,representing candidate information;

further, the output layer outputs the intermediate information and inputs the information of the next unit,

the intermediate information includes the information of the location of the mobile terminal,

o_t＝σ(W_o[h_t-1，x_t]+b₀)

wherein o is_tRepresenting intermediate information, W_oWeight vector representing output layer, b₀A bias term representing an output layer;

the information input to the next cell includes,

h_t＝o_ttanh(C_t)At)

wherein h is_tInformation indicating that the next unit is input.

The obtaining of the predicted load data may include,

and when the error value of the training neural network is smaller than the set error value or the training times are larger than the set training times, stopping training to obtain the predicted load data.

S4: and decomposing the load curve based on the predicted load data and the switching event set to obtain a single load curve.

The resident load curve decomposition method based on the long and short time neural network can realize the curve decomposition of a single load in a combined load curve to obtain a detailed operation curve chart of the single load, and can decompose the specific operation condition and the consumed electric energy of each load.

Example 2

Referring to fig. 3 to 9, a second embodiment of the present invention is different from the first embodiment in that a verification test of a long-and-short-term neural network-based resident load curve decomposition method is provided, and in order to verify the technical effects adopted in the method, the embodiment adopts a conventional technical scheme and the method of the present invention to perform a comparison test, and compares the test results by means of scientific demonstration to verify the real effects of the method.

Taking the combined data as an example, fig. 3 is a combined data power graph, and the operation sequence of the load switch event is shown in table 1.

Table 1: a combined scene switch event operational sequence.

Taking a Microwave Oven (MO) as an example, fig. 4 is an example of a process for extracting a power curve of a microwave oven, where 20 data points before an on event and 20 data points after an off event of the microwave oven are captured to predict blank data, and then a total power curve is subtracted from a filled curve, so that an operation curve of a single load can be extracted from the two curves; FIGS. 4-6 are graphs of a microwave oven curve extraction process clearly showing a single load curve under combined data extracted using a switching event and a long-and-short neural network; fig. 7 is a diagram showing the superposition comparison between the single load curve after the combined power curve decomposition and the original power curve, and in order to verify the accuracy of the load curve decomposition algorithm provided by the present invention, a Linear Interpolation (LI) method is used to extract the single load curve, and the decomposition result is shown in fig. 8; comparing fig. 7 and fig. 8, it can be seen that the power curve of each single load decomposed by the long and short time neural network is more stable and closer to the real load operation curve.

In the invention, the power decomposition accuracy and the energy decomposition accuracy are used as evaluation indexes of load decomposition. The formula for the power resolution accuracy is given by:

the formula for energy decomposition accuracy is given as:

table 2 shows the results of the decomposition of the long-and-short-term neural network and the linear interpolation method provided by the present invention, and the comparison results show that in the combined scene, the load curve decomposition method provided by the present invention also has obvious advantages, the accuracy of each load is above 90%, and the accuracy of the present invention is higher than that of the Linear Interpolation (LI) method.

Table 2: power accuracy and energy decomposition accuracy under the combined data.

As shown in fig. 9, the low-frequency power data in the REDD database is selected for verification, and the power decomposition accuracy and the energy decomposition accuracy under the TEDD data are shown in table 3.

Table 3: power resolution accuracy and energy resolution accuracy under REDD data.

From the results, compared with a Linear Interpolation (LI) method, the method for decomposing the load curve of the residents based on the long and short-term neural network has a better decomposition effect on the complex mixed power curve, and the effectiveness of the method is proved again.

It should be noted that the above-mentioned embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.