WO2023130776A1

WO2023130776A1 - Method and system for predicting working condition health status of battery in energy storage power station

Info

Publication number: WO2023130776A1
Application number: PCT/CN2022/121940
Authority: WO
Inventors: 张雪松; 林达; 钱平; 赵波; 杨帆; 章雷其; 马瑜涵; 葛晓慧; 刘敏; 陈凌宇; 陈哲
Original assignee: 国网浙江省电力有限公司电力科学研究院
Priority date: 2022-01-07
Filing date: 2022-09-28
Publication date: 2023-07-13
Also published as: CN114547849A

Abstract

Disclosed are a method and system for predicting the working condition health status of a battery in an energy storage power station. The prediction method of the present invention comprises: acquiring historical operation data of a battery in an energy storage power station, and pre-processing the original data; intercepting a platform voltage interval of the battery according to the characteristics of the battery, using the platform voltage interval as a reference to acquire, from a historical operation data charging section, charging platform interval voltage and current data of the battery in a period of time, acquiring a charging capacity value for each instance by means of an ampere-hour integral, and acquiring an original battery health status change sequence; separating a periodic change section and a linear degradation section of the battery from the original battery health status change sequence by using an empirical mode decomposition algorithm; establishing data-driven models for the periodic change section and the linear degradation section, respectively, and predicting the change trend of the linear degradation section and a seasonal periodic section of the battery. In the present invention, the demand during battery health status evaluation on battery data is reduced while precision is ensured to a certain extent, and the precision of battery health status prediction and evaluation is increased.

Description

Method and system for predicting the health status of batteries in energy storage power stations

technical field

The invention relates to the technical field of energy storage batteries, in particular to a method and system for predicting the health state of batteries in energy storage power stations.

Background technique

With the development of electrochemical energy storage technology, the cost of lithium-ion batteries has been decreasing year by year. In addition to power batteries, a large number of energy storage batteries represented by lithium iron phosphate batteries have been put into the field of electrochemical energy storage. At the same time, the development of the electric vehicle industry has also brought about a wave of decommissioning of power batteries. On the one hand, a large number of decommissioned power batteries are directly dismantled and recycled, and some decommissioned batteries with better performance are sorted and enter the cascade utilization stage. These cascade batteries are often used It is used for energy storage on the user side, especially in peak shaving scenarios. The installed capacity of energy storage batteries is generally large, and evaluating and predicting the health status of batteries is crucial to the healthy and stable operation of the system. Accurate status assessment and trend prediction help to find problems with batteries, facilitate system maintenance, and ensure energy storage The health and stability of the system.

Conventional data-driven SOH prediction method: The current SOH prediction of batteries generally uses the historical data of battery decline to fit the empirical decline model or some commonly used data-driven models. This method can easily capture the law of battery decline, and some better results in the scene. This method requires a large amount of relatively complete battery decline historical data to fit the model, which is generally a linear or exponential decline process, and is not suitable for the superimposition of periodic factors (such as temperature, etc.). Higher, generally it is relatively complete charge and discharge data, otherwise it is necessary to estimate the SOH of the battery first and then use the estimated value to realize the SOH prediction, resulting in error superposition.

Contents of the invention

Aiming at the problem that it is difficult to directly obtain the true value of battery capacity (energy) due to incomplete battery charging and discharging data, the present invention provides a method and system for predicting the health status of batteries in energy storage power stations, which approximates by intercepting the capacity (energy) of the platform area The full interval capacity (energy) of the battery, while ensuring a certain accuracy, reduces the demand for battery data for battery health status evaluation. It is suitable for SOH evaluation of energy storage systems working under peak load conditions, and solves the health status of energy storage batteries in complex environments. For the problem of multi-modal superposition, the corresponding signal decomposition algorithm is used to realize the decoupling of the complex factors of the battery SOH sequence, so as to pay attention to the essential law of battery decline and improve the accuracy of battery health state prediction and evaluation.

For this reason, a technical solution adopted by the present invention is as follows: a method for predicting the health status of batteries in energy storage power stations, which includes:

Step 1: Obtain the historical operation data of the battery of the energy storage power station, and preprocess the original data;

Step 2: According to the characteristics of the battery, intercept the platform voltage range of the battery, and use this as a reference to obtain the voltage and current data of the battery charging platform range for a period of time from the charging period of historical operating data, and obtain the charging capacity value each time through the integration of ampere hours , to obtain the original battery health state change sequence;

Step 3: Use the empirical mode decomposition algorithm to separate the battery cycle change segment and the linear decay segment from the original battery health state change sequence;

Step 4: Establish the data-driven models for the periodical change segment and the linear decline segment respectively, and realize the forecast of the change trend of the battery linear decline segment and the seasonal cycle segment.

Aiming at the problem that it is difficult to directly obtain the true value of battery capacity (energy) due to incomplete battery charge and discharge data, it is proposed to use the capacity (energy) of the battery platform area as the reference value of the battery's health status. The battery platform area contains about 80% of the capacity (energy) of the battery. For the energy storage lithium iron phosphate battery curve, the platform area and the two fast polarization areas of the battery can be easily distinguished.

Due to the low frequency of on-site battery voltage and current sampling, the commonly used methods based on capacity increment curve characteristics and differential voltage characteristics are basically unavailable due to accuracy problems. In order to minimize the impact of sampling accuracy on characteristic measurement values, current integration is used. The method obtains the accumulative ampere-hours in the battery platform area, which is used as a reference for the capacity and health status of the battery.

In view of the fact that the health status of the battery under the condition of no temperature control or weak temperature control is greatly affected by the temperature, the annual power station operation data is processed to obtain the daily charging amount of the battery cell in the platform area, and then the signal decomposition method is used to obtain the linear decline period and season of the battery In order to realize the decoupling of battery health status and temperature influence, obtain the approximate linear decline path of the battery, so as to realize the prediction of battery SOH based on the field data of the power station.

Further, in step 1, the historical battery operation data of the energy storage power station includes: battery pack voltage data, battery pack current data, battery cell voltage data, battery cell current data, battery cell temperature data, battery module temperature data and the accumulated running time of the energy storage power station; among them, battery cell voltage data, battery cell current data and battery cell temperature data are necessary data;

The data preprocessing: using data interpolation to unify the sampling interval of the data; removing outliers, interpolating and smoothing the data whose sampling interval has been unified.

Further, in step 2, the plateau voltage range of the battery is defined by the battery capacity incremental curve, and the voltage platform is required to include all peaks of the battery capacity incremental curve.

Furthermore, in step 2, the battery capacity increment curve is obtained by the following method:

Obtain the voltage and current data of the complete charging section of the battery at a certain charging rate;

Obtain the cumulative charging ampere-hours of each sampling point during the charging process by the ampere-hour integration method to obtain the battery voltage-capacity curve;

Set a reasonable voltage interval dV, obtain the voltage-capacity curve increasing at a fixed voltage interval through the linear interpolation method, and then obtain the battery IC curve data through the differential method, as follows:

In the above formula, dV is the voltage interval, dQ is the capacity corresponding to the equal voltage interval of the battery, and IC is the battery capacity increment.

Furthermore, in step 2, based on the battery IC curve, the voltage interval of the battery platform area is quantitatively fixed to obtain the reference value of the platform area capacity; The capacity sequence of the platform area of the charging section of the battery is obtained by using the method, which is the original battery health state change sequence.

Further, the specific steps of step 3 are as follows:

a. According to all the local maximum and minimum points found from the original signal, the upper envelope and the lower envelope of the original signal are constructed by means of the curve fitting method; the original signal is taken from the original The battery health state change sequence;

b. Find the mean value of the upper and lower envelopes and construct the mean value curve, and then subtract the curve from the original signal to obtain the intermediate component of the signal;

c. Determine whether the intermediate component of the obtained signal satisfies the constraints of the intrinsic mode function IMF: if it is satisfied, the component is an IMF; if it is not satisfied, repeat steps a to b with the component as the signal to be decomposed until it is satisfied The constraint condition of IMF, or the screening threshold value is less than the preset threshold value;

d. After the first IMF is obtained through the above steps, subtract the original signal from the first IMF, repeat steps a to c as a new original signal, iterate until the residual component meets the preset condition, and the remaining signal component is is the residual component, and the iteration ends;

After the above decomposition process, the original battery health state change sequence is decomposed into a number of empirical modal components and residual sequences, in which the residual sequence is mostly the noise of the signal, which is ignored; some empirical modal component sequences include some periodic signal sequences and Approximate linearly decaying signal sequence.

Furthermore, the specific content of step 4 is as follows:

Based on the obtained quasi-periodic signal sequence and approximate linear decay signal sequence, the modeling methods include linear regression, support vector machine and Gaussian process regression, and the square sine kernel function is used for the quasi-periodic signal sequence The Gaussian process regression characterizes the periodic variation characteristics of periodic signals;

The steps for modeling with Gaussian process regression are as follows:

a. According to the decomposition results of the empirical mode decomposition algorithm, the data of the linear decay process and the periodic process are respectively selected as the training set and cross-validation is turned on;

b. According to different training set data characteristics, select different kernel functions and initialize hyperparameters;

c. Use the training set data to train the regression model, and use the maximum likelihood estimation to obtain the optimal estimate of the hyperparameters to obtain the posterior model;

d. Use the trained model to predict the input test set data, and compare the predicted value with the verification set data to analyze the predictive ability of the model.

Another technical solution adopted by the present invention is: a system for predicting the health status of batteries in energy storage power stations, which includes:

Data acquisition and preprocessing unit: acquire the historical operation data of the battery of the energy storage power station, and preprocess the original data;

Original battery health state change sequence acquisition unit: According to the characteristics of the battery, intercept the battery platform voltage range, use this as a reference to obtain a period of time from the historical operating data charging section voltage and current data of the battery charging platform interval, and obtain it through ampere-hour integration Get the original battery health state change sequence for each charging capacity value;

Separation unit: Use the empirical mode decomposition algorithm to separate the battery cycle change segment and linear decay segment from the original battery health state change sequence;

Battery change trend prediction unit: establish data-driven models for the cycle change segment and the linear decline segment respectively, and realize the change trend prediction of the battery linear decline segment and the seasonal cycle segment.

The beneficial effects that the present invention has are:

The present invention approximates the capacity (energy) of the entire battery interval by intercepting the capacity (energy) of the platform area, which greatly reduces the demand for battery data for battery health status evaluation while ensuring a certain accuracy, and is suitable for storage batteries working under peak-shaving conditions. Energy system SOH evaluation; solve the problem of multi-modal superposition of energy storage battery health status in a complex environment, and use the corresponding signal decomposition algorithm to realize the decoupling of complex factors in the battery SOH sequence, so as to pay attention to the essential law of battery decline and improve battery health Accuracy of state prediction and assessment.

Description of drawings

Fig. 1 is a schematic flow chart of the method for predicting the health state of the battery working condition of the energy storage power station according to the present invention;

Fig. 2 is the battery IC graph in the application example of the present invention;

Fig. 3-4 is the comparison chart before and after working condition data processing in the application example of the present invention (Fig. 3 is the schematic diagram before working condition data processing, and Fig. 4 is the schematic diagram after working condition data processing);

Fig. 5 is the original sequence diagram of the platform capacity of all monomers in the battery cluster in the application example of the present invention;

Fig. 6 is a comparison chart between the predicted value and the actual value of the battery cell capacity in the application example of the present invention;

Fig. 7 is a structural diagram of the system for predicting the health state of the battery in the energy storage power station according to the present invention.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific implementation methods.

Example 1

Referring to FIG. 1 , this embodiment provides a method for predicting the health state of a battery in an energy storage power station, the steps of which are as follows:

Step 1: Obtain the historical operation data of the battery of the energy storage power station, specific to the battery cell, with a sampling accuracy of 1min to 5min, and preprocess the original data, including data interpolation and data cleaning.

Step 3: Use the empirical mode decomposition algorithm to separate the battery cycle change segment and linear decay segment from the original battery health state sequence;

On the basis of the above technical solution, the historical operation data of the energy storage power station, and the data preprocessing method:

Battery pack voltage data, battery pack current data, battery cell voltage data, battery cell current data, battery cell temperature data, battery module temperature data and the accumulated time of system operation. Among them, the voltage data, current data and temperature data of the battery cells are necessary data.

Data preprocessing: due to poor sampling accuracy or short-term failure of the monitoring system, some sampled data may appear outliers or missing data to varying degrees, resulting in data distortion, thereby affecting subsequent data extraction. Therefore, it is necessary to delete outlier data and interpolate missing data in this step to obtain usable data that can be further processed. At the same time, the power station does not participate in peak shaving every day, so the battery of the power station does not work or only works in a small DOD interval at certain times. This part of the data is not representative and it is difficult to extract new information from it, so this part of the data will be Clean up and make empty.

Specifically, the platform voltage range of the battery described in step 2 is obtained through the battery capacity increment curve, and the voltage platform is required to include all peaks of the capacity increment curve.

Specifically, the capacity increment curve of the battery is obtained by the following method:

Obtain the voltage and current data of the complete charging section of the battery at a certain charging rate.

The accumulative charging ampere-hours of each sampling point in the charging process of the battery is obtained by the ampere-hour integration method, and the battery voltage-capacity curve is obtained.

Set a reasonable voltage interval (dV), obtain the voltage-capacity curve increasing at a fixed voltage interval by the linear interpolation method, and then obtain the battery IC curve data (IC(V)) by the differential method, as follows:

In the above formula, dV is the voltage interval, dQ is the capacity corresponding to the equal voltage interval of the battery, and IC is the battery capacity increment sequence.

Based on the obtained specific IC curve, the voltage interval of the plateau area of the battery can be fixed quantitatively, and the reference value for obtaining the capacity of the plateau area can be obtained. Based on the obtained voltage reference of the platform area, the data of each battery charging section is intercepted, and the capacity sequence of the platform area of the charging section of the battery is obtained by the ampere-hour integration method, which is the reference sequence of the battery SOH.

Specifically, in order to separate the seasonal periodic signal and the approximate linear decay signal of the battery in the acquisition sequence, the empirical mode decomposition (EMD) algorithm is used to decompose the original sequence, and the steps of the EMD algorithm are as follows:

1. Construct the upper envelope and lower envelope of the original signal according to all the local maximum and minimum points found from the original signal, with the help of curve fitting method;

2. Find the mean value of the upper and lower envelopes and construct the mean value curve, and then subtract the curve from the original signal to obtain the intermediate component of the signal;

3. Judge whether the intermediate component of the obtained signal satisfies the constraints of the intrinsic mode function (IMF): if it is satisfied, then the component is an IMF; if not, repeat steps 1 to 3 with the component as the signal to be decomposed until Meet the constraints of the IMF, or the screening threshold is less than the preset threshold (usually 0.2 to 0.3);

4. After the first IMF is obtained through the above steps, subtract the original signal from the IMF, repeat steps 1 to 4 as a new original signal, iterate until the residual component meets the preset conditions, and the remaining signal component is the residual component, the iteration ends.

After the above decomposition process, the original sequence of battery decline is decomposed into several empirical mode components and residuals, among which the residual sequence is mostly the noise of the signal and can be ignored. Several empirical mode components include local characteristic signals of different time scales, mainly including some quasi-periodic signals and approximate linear decay signals.

Specifically, based on the obtained periodic signal and linear signal, there are many ways to model the approximate linear sequence, such as linear regression, support vector machine, Gaussian process regression and other regression algorithms. For the periodic sequence, the square sine kernel function is used Gaussian process regression can well describe the periodic variation characteristics of periodic signals. In this example, in order to maintain unity, Gaussian process regression is used to model the two sequences:

The steps to train a model using Gaussian process regression are as follows:

1. According to the decomposition results of the EMD algorithm, select the linear decay process and periodic process data as the training set and enable cross-validation;

2. According to the data characteristics of different training sets, select different kernel functions and initialize hyperparameters;

3. Use the training set data to train the regression model, and use the maximum likelihood estimation to obtain the optimal estimate of the hyperparameters to obtain the posterior model;

4. Use the trained model to predict the input test set data, and compare the predicted value with the verification set data to analyze the predictive ability of the model.

Above, the battery health status trend prediction with seasonal cycle signals is completed.

Example 2

Referring to Fig. 7, this embodiment provides a system for predicting the state of health of a battery in an energy storage power station, which includes:

Specifically, in the data acquisition and preprocessing unit, the historical battery operation data of the energy storage power station includes: battery pack voltage data, battery pack current data, battery cell voltage data, battery cell current data, battery cell temperature data, Battery module temperature data and the accumulated time of energy storage power station operation;

Specifically, in the acquisition unit of the original battery health state change sequence, the plateau voltage range of the battery is defined by the battery capacity increment curve, and the voltage plateau is required to include all peaks of the battery capacity increment curve.

The battery capacity increment curve is obtained by the following method:

Quantitatively fix the voltage interval of the battery platform area based on the battery IC curve, and obtain the benchmark value of the platform area capacity; based on the benchmark value of the platform area capacity, intercept the data of each battery charging section, and obtain the platform area of the battery charging section through the ampere-hour integral method The capacity sequence is the original battery health state change sequence.

Specifically, the specific steps of the separation unit are as follows:

Specifically, the specific content of the battery change trend prediction unit is as follows:

The steps for modeling with Gaussian process regression are as follows:

Refer to Example 1 for the parts not described in detail in Example 2.

Application example

The data of the application example comes from the operation data of a demonstration project using battery energy storage in an echelon. The data recording period is 1 year. The specific real-time steps are as follows:

Step 1. Obtain the historical operation data of the energy storage power station, specific to the battery cell, with a sampling accuracy of 1min to 5min, and perform preprocessing on the original data, including data interpolation and data cleaning.

Since the data sampling intervals at different times are not uniform, in order to facilitate analysis and calculation, the data sampling points are first unified, and the measure taken is to interpolate the sparsely sampled data and unify the data sampling interval to 1min. Due to the poor quality of the original voltage data due to insufficient sampling accuracy, outlier points, interpolation, and smoothing are performed on the data that has been uniformly sampled to obtain a smoother voltage curve of the battery. The battery voltage curve before and after processing is shown in Figure 2 .

Step 2, according to the characteristics of the battery, intercept the platform voltage range of the battery, and use this as a reference to obtain the voltage and current data of the battery charging platform range for a period of time from the charging period of historical operating data, and obtain the charging capacity value each time through the ampere-hour integration , to obtain the battery health state change sequence.

Among them, the voltage interval of the battery platform area is defined by the capacity increment curve of the battery. In this embodiment, a lithium iron phosphate battery of the same batch as the power station battery is taken, and the constant current charging data of the battery at a certain rate is obtained to draw a battery capacity increment curve. In this embodiment, dV takes a fixed value of 0.002V, dQ corresponding to the voltage interval can be obtained by using the difference method, and the capacity increment curve of the battery is obtained according to the definition of the capacity increment curve, as shown in Figure 3-4.

The range of the plateau area can be easily determined through the capacity increment curve, and the voltage range of the plateau area is limited to 3.25-3.4V in this embodiment. Based on the obtained voltage interval, intercept the daily charging section data and integrate the current to obtain the battery platform area capacity sequence. In this embodiment, the normalized battery capacity sequence operation is not performed to make the result more intuitive, and a certain battery cluster is obtained. The original sequences of all the monomers in are shown in Figure 5.

Through the original sequence, it can be found that the capacity change trend of the battery platform area is multi-modal aliasing, and the capacity of some battery platform areas is significantly affected by temperature, so the signal decomposition of the original data must be performed.

Step 3. In this application example, the empirical mode decomposition method is used to decompose the historical data of the battery, and the mode decomposition can be realized by using the EMD algorithm toolkit in the signal processing toolbox of Matlab software. In this embodiment, the maximum IMF The number is set to 3. At this point, the post-signal decomposition and decoupling work is completed, and the battery linear segment and cycle segment sequence are obtained.

Step 4. Establish data-driven models for the periodical variation segment and the linear recession segment respectively, so as to realize the prediction of the change trend of the battery linear recession segment and the seasonal cycle segment.

For the decomposed linear decay process, when using Gaussian process regression to model the process, the selected kernel function is a square exponential kernel function that can characterize the battery's long-term decline trend. The expression of the kernel function is:

Among them, σ, l and α are hyperparameters. Since the analytical expression of the kernel function contains several hyperparameters, selecting appropriate parameter values is crucial for establishing an accurate Gaussian process regression model. The hyperparameter estimation method generally adopts the maximum likelihood estimation method, which is a commonly used model parameter estimation method: set the hyperparameter set θ={θ ₁ ,...,θ _n }, through Bayesian Theorem shows that the posterior distribution of hyperparameters is:

Among them, x represents the input of training data, and y represents the output of training data. If there is no prior knowledge about the hyperparameter set θ={θ ₁ ,...,θ _n }, then the maximum a posteriori estimate of θ is the maximum likelihood estimate of p(y|x,θ), namely:

In this way, the posterior estimated value of the hyperparameters of the kernel function can be obtained according to the training data.

For seasonal periodic signals, in this application example, the square sine kernel function is used to describe the relationship between sample variables, and its form is shown in the following formula:

After the battery decay model is built, in order to verify the accuracy of the model, two battery cells in the battery cluster are selected, and the field data from June 2020 to January 2021 are used as training data to predict the battery life from February to June 2021. For the decline situation, the predicted value and actual value of the obtained battery cell capacity are shown in Figure 6. The linear decline trend of the battery is well fitted, and the maximum prediction error is less than 5%.

In the present invention, specific examples are used to illustrate the principle and implementation of the present invention, and the descriptions of the above embodiments are only used to help understand the method and core idea of the present invention. It should be pointed out that for those skilled in the art, without departing from the principle of the present invention, some improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

The method for predicting the state of health of a battery in an energy storage power station is characterized in that it includes:

Step 1: Obtain the historical operation data of the battery of the energy storage power station, and preprocess the original data;

Step 2: According to the characteristics of the battery, intercept the platform voltage interval of the battery, and use this as a reference to obtain the voltage and current data of the battery charging platform interval for a period of time from the historical operation data charging section, and obtain the charging capacity value each time through the integration of ampere hours , to obtain the original battery health state change sequence;

Step 3: Use the empirical mode decomposition algorithm to separate the battery cycle change segment and the linear decay segment from the original battery health state change sequence;

Step 4: Establish the data-driven models for the periodical change segment and the linear decline segment respectively, and realize the forecast of the change trend of the battery linear decline segment and the seasonal cycle segment.
The method for predicting the state of health of a battery in an energy storage power station according to claim 1, wherein in step 1, the historical battery operation data of the energy storage power station includes: battery pack voltage data, battery pack current data, battery cell Voltage data, battery cell current data, battery cell temperature data, battery module temperature data and the accumulated time of energy storage power station operation;

The data preprocessing: using data interpolation to unify the sampling interval of the data; removing outliers, interpolating and smoothing the data whose sampling interval has been unified.
The method for predicting the state of health of a battery in an energy storage power station according to claim 1, wherein in step 2, the platform voltage range of the battery is defined by the battery capacity incremental curve, and the voltage platform is required to include the battery capacity incremental curve all peaks.
The method for predicting the state of health of a battery in an energy storage power station according to claim 3, wherein in step 2, the battery capacity increment curve is obtained by the following method:

Obtain the voltage and current data of the complete charging section of the battery at a certain charging rate;

Obtain the cumulative charging ampere-hours of each sampling point during the charging process by the ampere-hour integration method to obtain the battery voltage-capacity curve;

Set a reasonable voltage interval dV, obtain the voltage-capacity curve increasing at a fixed voltage interval through the linear interpolation method, and then obtain the battery IC curve data through the differential method, as follows:

In the above formula, dV is the voltage interval, dQ is the capacity corresponding to the equal voltage interval of the battery, and IC is the battery capacity increment.
The method for predicting the state of health of a battery in an energy storage power station according to claim 4, wherein in step 2, the voltage interval of the battery platform area is quantitatively fixed based on the battery IC curve, and the reference value for obtaining the capacity of the platform area is obtained; based on the platform Intercept the data of each charging section of the battery, and obtain the platform area capacity sequence of the charging section of the battery through the ampere-hour integration method, which is the original battery health state change sequence.
The method for predicting the state of health of a battery in an energy storage power station according to claim 1, wherein the specific steps of step 3 are as follows:

a. According to all the local maximum and minimum points found from the original signal, the upper envelope and the lower envelope of the original signal are constructed by means of the curve fitting method; the original signal is taken from the original The battery health state change sequence;

b. Find the mean value of the upper and lower envelopes and construct the mean value curve, and then subtract the curve from the original signal to obtain the intermediate component of the signal;

c. Judging whether the intermediate component of the obtained signal satisfies the constraints of the intrinsic mode function IMF: if it is satisfied, the component is an IMF; if it is not satisfied, then repeat steps a to b with this component as the signal to be decomposed until it is satisfied The constraint condition of IMF, or the screening threshold value is less than the preset threshold value;

d. After the first IMF is obtained through the above steps, subtract the original signal from the first IMF, repeat steps a to c as a new original signal, iterate until the residual component meets the preset condition, and the remaining signal component is is the residual component, and the iteration ends;

After the above decomposition process, the original battery health state change sequence is decomposed into a number of empirical modal components and residual sequences, in which the residual sequence is mostly the noise of the signal, which is ignored; some empirical modal component sequences include some periodic signal sequences and Approximate linearly decaying signal sequence.
The method for predicting the state of health of a battery in an energy storage power station according to claim 6, wherein the specific content of step 4 is as follows:

Based on the obtained quasi-periodic signal sequence and approximate linear decay signal sequence, the modeling methods include linear regression, support vector machine and Gaussian process regression, and the square sine kernel function is used for the quasi-periodic signal sequence The Gaussian process regression characterizes the periodic variation characteristics of periodic signals;

The steps for modeling with Gaussian process regression are as follows:

a. According to the decomposition results of the empirical mode decomposition algorithm, the data of the linear decay process and the periodic process are selected as the training set and cross-validation is turned on;

b. According to different training set data characteristics, select different kernel functions and initialize hyperparameters;

c. Use the training set data to train the regression model, and use the maximum likelihood estimation to obtain the optimal estimate of the hyperparameters to obtain the posterior model;

d. Use the trained model to predict the input test set data, and compare the predicted value with the verification set data to analyze the predictive ability of the model.
The system for predicting the state of health of a battery in an energy storage power station is characterized in that it includes:

Data acquisition and preprocessing unit: acquire the historical operation data of the battery of the energy storage power station, and preprocess the original data;

Original battery health state change sequence acquisition unit: According to the characteristics of the battery, intercept the battery platform voltage range, use this as a reference to obtain a period of time from the historical operating data charging section voltage and current data of the battery charging platform interval, and obtain it through ampere-hour integration Get the original battery health state change sequence for each charging capacity value;

Separation unit: Use the empirical mode decomposition algorithm to separate the battery cycle change segment and linear decay segment from the original battery health state change sequence;

Battery change trend prediction unit: establish the data-driven models of the cycle change segment and the linear decline segment respectively, and realize the change trend prediction of the battery linear decline segment and the seasonal cycle segment.
The system for predicting the state of health of the battery working condition of the energy storage power station according to claim 8, wherein the specific steps of the separation unit are as follows:

a. According to all the local maximum and minimum points found from the original signal, the upper envelope and the lower envelope of the original signal are constructed by means of the curve fitting method; the original signal is taken from the original The battery health state change sequence;

b. Find the mean value of the upper and lower envelopes and construct the mean value curve, and then subtract the curve from the original signal to obtain the intermediate component of the signal;

c. Determine whether the intermediate component of the obtained signal satisfies the constraints of the intrinsic mode function IMF: if it is satisfied, the component is an IMF; if it is not satisfied, repeat steps a to b with the component as the signal to be decomposed until it is satisfied The constraint condition of IMF, or the screening threshold value is less than the preset threshold value;

d. After the first IMF is obtained through the above steps, subtract the original signal from the first IMF, repeat steps a to c as a new original signal, iterate until the residual component meets the preset condition, and the remaining signal component is is the residual component, and the iteration ends;

After the above decomposition process, the original battery health state change sequence is decomposed into a number of empirical modal components and residual sequences, in which the residual sequence is mostly the noise of the signal, which is ignored; some empirical modal component sequences include some periodic signal sequences and Approximate linearly decaying signal sequence.
The system for predicting the state of health of a battery in an energy storage power station according to claim 9, wherein the specific content of the battery change trend prediction unit is as follows:

Based on the obtained quasi-periodic signal sequence and approximate linear decay signal sequence, the modeling methods include linear regression, support vector machine and Gaussian process regression, and the square sine kernel function is used for the quasi-periodic signal sequence The Gaussian process regression characterizes the periodic variation characteristics of periodic signals;

The steps for modeling with Gaussian process regression are as follows:

a. According to the decomposition results of the empirical mode decomposition algorithm, the data of the linear decay process and the periodic process are respectively selected as the training set and cross-validation is turned on;

b. According to different training set data characteristics, select different kernel functions and initialize hyperparameters;

c. Use the training set data to train the regression model, and use the maximum likelihood estimation to obtain the optimal estimate of the hyperparameters to obtain the posterior model;

d. Use the trained model to predict the input test set data, and compare the predicted value with the verification set data to analyze the predictive ability of the model.