阅读说明：本技术 蓄电池组健康预测的方法 (Method for health prediction of storage battery pack ) 是由 王延涛 刘松 金欣 左长华 潘志鹏 王兆敏 韩其东 赵利 于 2021-05-31 设计创作，主要内容包括：本发明公开了一种蓄电池组健康预测的方法,包括以下步骤：利用前馈神经网络对蓄电池组剩余容量进行预测,并建立放电模型,通过放电模型对蓄电池组剩余容量进行验算；在线获取电池参数,将参数输入放电模型,放电模型输出蓄电池SOH；在蓄电池放电时,获取放电电压和蓄电池SOH,利用放电电压和蓄电池SOH进行模糊分类以及自适应求解,在核容20％蓄电池组标称容量的前提下实时准确预测后备蓄电池在断电情况下的供电能力。可见采用本发明的方法,可实现电池故障的提前预警,并提供中长期失效预测,以便能在事故前发现蓄电池组的故障问题。(The invention discloses a method for predicting the health of a storage battery pack, which comprises the following steps: predicting the residual capacity of the storage battery pack by utilizing a feedforward neural network, establishing a discharge model, and checking the residual capacity of the storage battery pack through the discharge model; acquiring battery parameters on line, inputting the parameters into a discharge model, and outputting a storage battery SOH by the discharge model; when the storage battery is discharged, the discharging voltage and the storage battery SOH are obtained, fuzzy classification and self-adaptive solution are carried out by using the discharging voltage and the storage battery SOH, and the power supply capacity of the backup storage battery under the condition of power failure is accurately predicted in real time on the premise of the nominal capacity of the storage battery pack with the nuclear capacity of 20%. Therefore, the method can realize early warning of battery faults and provide medium-long term failure prediction so as to find the fault problem of the storage battery pack before an accident.)

1. A method for battery pack health prediction, comprising the steps of:

predicting the residual capacity of the storage battery pack by utilizing a feedforward neural network, establishing a discharge model, and checking the residual capacity of the storage battery pack through the discharge model;

secondly, battery parameters are acquired on line, the parameters are input into a discharge model, the discharge model outputs storage battery SOH, and the health state of the storage battery pack is predicted according to the storage battery SOH;

and step three, acquiring a discharging voltage and a storage battery SOH when the storage battery is discharged, carrying out fuzzy classification and self-adaptive solution by using the discharging voltage and the storage battery SOH, and accurately predicting the power supply capacity of the backup storage battery under the condition of power failure in real time on the premise of checking the nominal capacity of the storage battery pack by 20%.

2. The battery pack health prediction method of claim 1, wherein the discharge model comprises a feedforward neural network model and a capacity accumulation algorithm model;

then, in the first step, checking the remaining capacity of the storage battery pack through a discharge model includes:

s1, obtaining the specification and the brand of the storage battery;

s2, judging whether the specifications and brands of the storage batteries are the same

S3, if the residual capacity of the storage battery is the same, obtaining the residual capacity of the storage battery by using a feedforward neural network model;

if the residual capacity of the storage battery is different from the residual capacity of the storage battery, obtaining the residual capacity of the storage battery by using a capacity accumulation algorithm model;

and S4, comparing and checking the residual capacity of the storage battery with the predicted residual capacity.

3. The battery pack health prediction method according to claim 2, wherein the capacity accumulation algorithm model is specifically:

calculating the residual capacity Q residual by using a stepped load and a piecewise calculation method, wherein the Q residual is Q total-Q discharge, Q1+ Q2+ Q3+ Q4, Q1 is I1T 1/eta [1+ a (T-25) ], Q2 is I2T 2/eta [1+ a (T-25) ], Q3 is I3T 3/eta [1+ a (T-25) ], and Q4 is I4T 4/eta [1+ a (T-25) ];

i1, I2, I3 and I4 are load currents in each stage; t1, T2, T3 and T4 are the number of discharge hours in each stage; eta is a discharge capacity coefficient; t is the lowest environment temperature value of the actual battery location, and a is the battery temperature coefficient.

4. The battery pack health prediction method according to claim 2, wherein the feedforward neural network model is a BP neural network, and the optimization of calculation of the BP neural network based on LM comprises:

setting initial values of each weight and threshold value of the network;

calculating the actual output of the network and the state of the hidden layer unit, wherein the excitation function is a Sigmoid function;

judging an error, and if the error is larger than a set expected error, correcting the weight and the threshold;

the correction method after LM optimization is to correct the weight and the threshold, and the specific correction method is as follows:

W_i+1＝W_i–(H+λdiag[H])^-1din the formula, H is Hessian matrix, diag [ H ]]Is a diagonal matrix of the Hessian matrix, lambda is a variable parameter which changes according to the training error, and d is the descending gradient of the correction parameter;

and calculating the corrected weight value threshold until the error reaches a set expected value, finishing the circular calculation, and outputting a neural network fitting value.

5. The method for battery pack health prediction according to claim 4, characterized in that the weights and threshold matrices of each layer are obtained by calculation through training and learning of learning samples based on the LM optimized BP neural network, wherein the learning samples are extracted from results obtained by analyzing existing battery pack operating data; and in the field, after the system runs for a period of time, automatically extracting learning samples and teacher samples from the field storage battery running data, and performing re-learning training to obtain new network weights and thresholds.

6. The battery pack health prediction method of claim 1, wherein the internal resistance measurement method is a phase discrimination process, wherein a constant ac excitation signal is applied to the battery, the ac component is filtered out by a filter by multiplying the response signal by a sinusoidal signal generating a constant current source, and COS Φ is mapped into the internal resistance.

7. The method for battery pack health prediction according to claim 1, wherein in the third step, fuzzy classification and adaptive solution are performed by using the discharging voltage and the storage battery SOH, and the capacity and the remaining discharging time of the storage battery are calculated by combining the previous discharging capacity of the storage battery or performing fuzzy classification by using the storage battery SOH and the current discharging voltage, and then performing adaptive solution with various standard discharging curves; the fuzzy classification includes SOH classification and discharge voltage classification:

the SOH classification is carried out according to the obtained prediction performance of the float model and a corresponding membership function;

the discharge voltage classification is based on the discharge voltage V at the discharge time t in the discharge process and the analog output voltage V with the corresponding capacity at the time_i(t) establishing a membership function, wherein i is 1, 2, 3 or 4, and the membership function is:

f₁(t)＝min

(|V₁(t)-V|，|V₂(t)-V|，|V₃(t)-V|，|V₄(t)-V|)

in the formula f₁(t) is the minimum of 4 parameters, f₂(V) is the class number of the classification;

and constructing a fuzzy controller, establishing a reasonable fuzzy rule controller based on the fuzzy rule by the fuzzy controller based on the performance classification and the discharge voltage classification of the floating charge model, and obtaining a final classification number through the fuzzy controller.

8. The method of claim 2, wherein when the discharge model is a feedforward neural network model, the parameters in step two include the dispersion of the floating voltage, the internal resistance, the tank bottom voltage, and the recovery voltage, and the dispersion of the floating voltage, the internal resistance, the tank bottom voltage, and the recovery voltage.

Technical Field

The invention relates to the technical field of storage batteries, in particular to a storage battery pack health prediction method.

Background

At present, the capacity of a storage battery pack of a transformer substation is detected mainly by manually using a discharge instrument, so that the testing time is long, a voltage loss fault occurs in the testing process, and the storage battery pack does not have enough capacity to guarantee the normal operation of a load.

The daily inspection is to test the internal resistance, voltage and temperature of each storage battery once a month, so as to ensure the normal daily work of the battery pack. The manual operation and maintenance mode is extremely low in efficiency, and when a battery fails, problems cannot be found in time, so that safety accidents occur.

The daily operation and maintenance work of the storage battery of the direct current system of the transformer substation mainly depends on manual work to carry out one-by-one inspection, charging and discharging operation and maintenance of the storage battery of each transformer substation. According to the national grid storage battery operation and maintenance specification, a deep capacity check test is carried out on a new battery pack every 2 years, and the battery pack with the capacity more than 4 years needs to be subjected to the least deep capacity check every year. The problems of long operation and maintenance time, large equipment investment, low efficiency and the like exist in manual operation and maintenance.

Disclosure of Invention

Aiming at the defects, the technical problems to be solved by the invention are as follows: the method for predicting the health of the storage battery pack is provided, so that early warning of battery faults is realized, medium-long term failure prediction is provided, the fault problem of the storage battery pack can be found before an accident, and maintenance suggestions and maintenance bases are provided for the batteries needing to be replaced in time.

In order to solve the technical problems, the technical scheme of the invention is as follows:

a method of battery pack health prediction, comprising the steps of:

predicting the residual capacity of the storage battery pack by utilizing a feedforward neural network, establishing a discharge model, and checking the residual capacity of the storage battery pack through the discharge model;

secondly, battery parameters are acquired on line, the parameters are input into a discharge model, the discharge model outputs storage battery SOH, and the health state of the storage battery pack is predicted according to the storage battery SOH;

and step three, acquiring a discharging voltage and a storage battery SOH when the storage battery is discharged, carrying out fuzzy classification and self-adaptive solution by using the discharging voltage and the storage battery SOH, and accurately predicting the power supply capacity of the backup storage battery under the condition of power failure in real time on the premise of checking the nominal capacity of the storage battery pack by 20%.

2. The battery pack health prediction method of claim 1, wherein the discharge model comprises a feedforward neural network model and a capacity accumulation algorithm model;

then, in the first step, checking the remaining capacity of the storage battery pack through a discharge model includes:

s1, obtaining the specification and the brand of the storage battery;

s2, judging whether the specifications and brands of the storage batteries are the same

S3, if the residual capacity of the storage battery is the same, obtaining the residual capacity of the storage battery by using a feedforward neural network model;

if the residual capacity of the storage battery is different from the residual capacity of the storage battery, obtaining the residual capacity of the storage battery by using a capacity accumulation algorithm model;

and S4, comparing and checking the residual capacity of the storage battery with the predicted residual capacity.

Preferably, the capacity accumulation algorithm model specifically includes:

calculating the residual capacity Q residual by using a stepped load and a piecewise calculation method, wherein the Q residual is Q total-Q discharge, Q1+ Q2+ Q3+ Q4, Q1 is I1T 1/eta [1+ a (T-25) ], Q2 is I2T 2/eta [1+ a (T-25) ], Q3 is I3T 3/eta [1+ a (T-25) ], and Q4 is I4T 4/eta [1+ a (T-25) ];

i1, I2, I3 and I4 are load currents in each stage; t1, T2, T3 and T4 are the number of discharge hours in each stage; eta is a discharge capacity coefficient; t is the lowest environment temperature value of the actual battery location, and a is the battery temperature coefficient.

Preferably, the feedforward neural network model is a BP neural network, and the calculation based on the LM optimized BP neural network includes:

setting initial values of each weight and threshold value of the network;

calculating the actual output of the network and the state of the hidden layer unit, wherein the excitation function is a Sigmoid function;

judging an error, and if the error is larger than a set expected error, correcting the weight and the threshold;

the correction method after LM optimization is to correct the weight and the threshold, and the specific correction method is as follows:

W_i+1＝W_i–(H+λdiag[H])^-1din the formula, H is Hessian matrix, diag [ H ]]Is a diagonal matrix of the Hessian matrix, lambda is a variable parameter which changes according to the training error, and d is the descending gradient of the correction parameter;

and calculating the corrected weight value threshold until the error reaches a set expected value, finishing the circular calculation, and outputting a neural network fitting value.

The optimization method comprises the following steps of training and learning a learning sample based on an LM optimization BP neural network, and obtaining a weight and a threshold matrix of each layer through calculation, wherein the learning sample is extracted from a result obtained by analyzing the operation data of the existing storage battery pack; and on the spot, after the system runs for a period of time, the learning sample and the teacher sample are automatically extracted from the running data of the storage battery on the spot, and the learning training is carried out again to obtain a new network weight and a new threshold value.

The preferred mode is that the measuring method of the internal resistance is a phase discrimination processing method, a constant alternating current excitation signal is applied to the storage battery, the response signal is multiplied by a sine signal which generates a constant current source, an alternating current component is filtered by a filter, and then COS phi is mapped into the internal resistance.

In the third step, fuzzy classification and self-adaptive solution are carried out by utilizing the discharging voltage and the storage battery SOH, the previous discharging capacity of the storage battery is combined, or the storage battery SOH and the current discharging voltage are used for carrying out fuzzy classification, and then the fuzzy classification and the self-adaptive solution are carried out with various standard discharging curves to calculate the capacity and the residual discharging time of the storage battery; the fuzzy classification includes SOH classification and discharge voltage classification:

the SOH classification is carried out according to the obtained prediction performance of the float model and a corresponding membership function;

the discharge voltage classification is to establish a membership function according to a discharge voltage V at a discharge time t in a discharge process and an analog output voltage Vi (t) with a capacity corresponding to the discharge time, wherein i is 1, 2, 3 or 4, and the membership function is as follows:

f₁(t)＝min

(|V₁(t)-V|，V₂(t)-V|，|V₃(t)-V|，|V₄(t)-V|)

in the formula f₁(t) is the minimum of 4 parameters, f₂(V) is the class number of the classification;

and constructing a fuzzy controller, establishing a reasonable fuzzy rule controller based on the fuzzy rule by the fuzzy controller based on the performance classification and the discharge voltage classification of the floating charge model, and obtaining a final classification number through the fuzzy controller.

Preferably, when the discharge model is a feedforward neural network model, the parameters in the second step include the dispersion of the floating charge voltage, the internal resistance, the bottom voltage of the cell, and the recovery voltage.

After the technical scheme is adopted, the invention has the beneficial effects that:

the method for predicting the health of the storage battery pack comprises the following steps: predicting the residual capacity of the storage battery pack by utilizing a feedforward neural network, establishing a discharge model, and checking the residual capacity of the storage battery pack through the discharge model; secondly, acquiring battery parameters on line, inputting the parameters into a discharge model, and outputting a storage battery SOH by the discharge model; and step three, acquiring a discharging voltage and a storage battery SOH when the storage battery is discharged, carrying out fuzzy classification and self-adaptive solution by using the discharging voltage and the storage battery SOH, and accurately predicting the power supply capacity of the backup storage battery under the condition of power failure in real time on the premise of checking the nominal capacity of the storage battery pack by 20%. Therefore, the method can realize early warning of battery faults and provide medium and long-term failure prediction so as to find the fault problem of the storage battery pack before an accident and provide maintenance suggestions and maintenance bases for the batteries needing to be replaced in time.

Drawings

Fig. 1 is a BP neural network.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the following embodiments further describe the present invention in detail. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

A method of battery pack health prediction, comprising the steps of:

predicting the residual capacity of the storage battery pack by utilizing a feedforward neural network, establishing a discharge model, and checking the residual capacity of the storage battery pack through the discharge model;

secondly, battery parameters are acquired on line, the parameters are input into a discharge model, the discharge model outputs storage battery SOH (abbreviation of English state of health, Chinese is battery health degree, and is the percentage of the current capacity and the factory capacity of the battery), and the health state of the storage battery pack is predicted according to the storage battery SOH; and step three, acquiring a discharging voltage and a storage battery SOH when the storage battery is discharged, carrying out fuzzy classification and self-adaptive solution by using the discharging voltage and the storage battery SOH, and accurately predicting the power supply capacity of the backup storage battery under the condition of power failure in real time on the premise of checking the nominal capacity of the storage battery pack by 20%.

Therefore, the method can realize early warning of battery faults and provide medium and long-term failure prediction so as to find the fault problem of the storage battery pack before an accident and provide maintenance suggestions and maintenance bases for the batteries needing to be replaced in time.

In the embodiment, the discharge model comprises a feedforward neural network model and a capacity accumulation algorithm model, when the discharge model is the feedforward neural network model, the parameters in the second step comprise the dispersion of the floating charge voltage, the internal resistance, the bottom voltage and the recovery voltage, and the dispersion of the floating charge voltage, the internal resistance, the bottom voltage and the recovery voltage are measured;

in the first step, checking the remaining capacity of the storage battery pack through a discharge model includes:

s1, obtaining the specification and the brand of the storage battery;

s2, judging whether the specifications and brands of the storage batteries are the same

S3, if the residual capacity of the storage battery is the same, obtaining the residual capacity of the storage battery by using a feedforward neural network model;

if the residual capacity of the storage battery is different from the residual capacity of the storage battery, obtaining the residual capacity of the storage battery by using a capacity accumulation algorithm model;

and S4, comparing and checking the residual capacity of the storage battery with the predicted residual capacity.

The capacity accumulation algorithm model specifically comprises the following steps:

calculating the residual capacity Q by using a stepped load and a piecewise calculation method, wherein Q is Q1+ Q2+ Q3+ Q4, Q1 is I1T 1/eta [1+ a (T-25) ], Q2 is I2T 2/eta [1+ a (T-25) ], Q3 is I3T 3/eta [1+ a (T-25) ], Q4 is I4T 4/eta [1+ a (T-25) ], and I1, I2, I3 and I4 are load currents of each stage; t1, T2, T3 and T4 are the number of discharge hours in each stage; eta is a discharge capacity coefficient; t is the lowest environment temperature value of the actual battery location, and a is the battery temperature coefficient. Currently, the minimum value of m is 6 seconds, and the precision can be customized according to requirements. And calculating the residual capacity in real time through the formula. When the discharge hour rate is more than or equal to 10, taking alpha as 0.006; when the discharge hour rate is more than 10 and is more than or equal to 1, taking alpha as 0.008; when the discharge hour rate is less than 1, taking alpha as 0.01; discharge capacity coefficient (η) table:

hours of battery discharge (H)	0.5	1	2	3	4	6	8	10	≥20
										Discharge end voltage (V)	1.70	1.75	1.75	1.80	1.80	1.80	1.80	1.80	≥1.85
Coefficient of discharge capacity (eta)	0.45	0.40	0.55	0.61	0.75	0.79	0.88	1.00	1.00

Wherein, the feedforward formula neural network model is BP neural network, optimizes BP neural network's calculation based on LM, includes:

setting initial values of each weight and threshold value of the network;

calculating the actual output of the network and the state of the hidden layer unit, wherein the excitation function is a Sigmoid function;

judging an error, and if the error is larger than a set expected error, correcting the weight and the threshold;

the correction method after LM (Levenberg-Marquardt) optimization is to correct the weight and the threshold, and the specific correction method is as follows:

W_i+1＝W_i–(H+λdiag[H])^-1din the formula, H is Hessian matrix, diag [ H ]]Is Hessian moment

And calculating the corrected weight value threshold until the error reaches a set expected value, finishing the circular calculation, and outputting a neural network fitting value.

Training and learning a learning sample based on the LM optimization BP neural network, and calculating to obtain a weight and a threshold matrix of each layer, wherein the learning sample is extracted from a result obtained by analyzing the operation data of the existing storage battery pack; on site, after the system runs for a period of time, automatically extracting learning samples and teacher samples from the running data of the storage battery on site, and performing re-learning training to obtain new network weights and threshold values; the BP neural network is shown in fig. 1.

In the third step of the embodiment, the fuzzy classification and the self-adaptive solution are performed by using the discharging voltage and the storage battery SOH, and the method comprises the following steps: the capacity and the residual discharge time of the storage battery are calculated by combining the previous discharge capacity of the storage battery or performing fuzzy classification by using the estimated SOH and the current discharge voltage, and then performing self-adaptive solution on the capacity and the residual discharge time of the storage battery and various standard discharge curves; the fuzzy classification includes:

and (3) SOH classification, wherein according to the obtained prediction performance of the float model, the SOH classification can be carried out according to a corresponding membership function:

in the formula Cap_iNo more than-performance value predicted by the ith battery;

f(Cap_i) -the class number of the classification.

The discharge voltage classification is to establish a membership function according to a discharge voltage V at a discharge time t in a discharge process and an analog output voltage Vi (t) of a capacity corresponding to the discharge time, wherein i is 1, 2, 3 or 4, and specifically comprises the following steps: according to the discharge voltage V at the discharge time t in the discharge process, the analog output voltage V1(t) of the first type capacity, the analog output voltage V2(t) of the second type capacity, the analog output voltage V3(t) of the third type capacity and the analog output voltage V4(t) of the fourth type capacity, the 5 parameters establish a membership function, namely the membership function:

f₁(t)＝min

(|V₁(t)-V|，V₂(t)-V|，V₃(t)-V|，|V₄(t)-V|)

in the formula f₁(t) -the minimum of the 4 parameters;

f₂(V) -class number of class.

Constructing a fuzzy controller, establishing a reasonable fuzzy rule controller based on the performance classification and the discharge voltage classification of the float model, establishing a fuzzy rule controller according to a fuzzy rule, obtaining a final classification number through the fuzzy controller;

TABLE 1 fuzzy rules

The measuring method of the internal resistance is a phase discrimination processing method, which applies a constant alternating current excitation signal to the storage battery, multiplies a response signal by a sinusoidal signal generating a constant current source, filters an alternating current component by a filter, and maps COS phi into the internal resistance.

In order to solve the problem of traditional storage battery system maintenance, a dummy load is used for discharging, operation and maintenance equipment is transported to a site, a large amount of time is spent on connecting a monitoring unit, a battery pack is discharged after being separated from the system, and the system is connected back to charge the storage battery after the discharging is finished. When the storage battery is discharged, the battery pack of the conventional storage battery maintenance equipment must be separated from the original system, and the combined storage battery pack is in an off-line state. In addition, the existing storage battery operation and maintenance system has no effective monitoring means, cannot find the problems of the storage battery pack in the operation process in time, and lacks scientific and effective data information to master the accurate information of the storage battery pack. However, with the rapid development of power electronic technology, the maturity of technologies such as ac to dc, dc to dc, communication protocol and module design provides an effective means for solving various problems of the series-connected storage batteries, and the technical conditions for realizing the core-capacitance grid connection of the storage batteries are met.

1) And (3) information acquisition and realization principles. The method comprises the steps that a sensor is added on physical hardware, a single acquisition module is added on each battery, the voltage, the internal resistance and the pole temperature of each battery are sampled, and sampling information is collected through a group end convergence module. And the group terminal convergence module samples the group terminal voltage, the charger voltage and the current. The hardware part of the storage battery comprehensive management system consisting of the single acquisition module and the group end convergence module is matched with background management software and algorithm models of all functions to realize the operation of the whole system.

2) And (3) realizing a principle of core capacity current. And feeding back the voltage of the storage battery pack to the power grid through an inverter.

3) And (4) realizing the principle of unmanned operation and maintenance. The voltage, the internal resistance and the temperature of a single storage battery are monitored, the single operation and maintenance module is combined to boost and discharge in a constant current mode, the external discharge data of the batteries can be recorded in real time, and the capacity of each battery can be calculated according to a capacity accounting formula. By adding a remote control software platform, the storage battery can be monitored in real time and charge and discharge can be controlled remotely.

The above-described preferred embodiments of the present invention are not intended to limit the present invention, and any modifications, equivalent improvements in the method of battery pack health prediction, which are within the spirit and principle of the present invention, are intended to be included within the scope of the present invention.