A method and system for predicting remaining service life of energy storage battery

Abstract

The present invention discloses a method and system for predicting the remaining service life of an energy storage battery. The method obtains battery capacity time series data; uses TVF-EMD to decompose the battery capacity time series data to obtain intrinsic mode components; uses an improved collaborative group optimization algorithm to optimize the number and size of convolution kernels of a time convolution neural network model; normalizes the complete battery capacity time series data, and inputs the normalized data into the time convolution neural network model optimized by the improved collaborative group optimization algorithm to predict the remaining service life. The present invention uses the time convolution neural network model optimized by the improved collaborative group optimization algorithm to make its prediction more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage battery life prediction, and in particular to a method and system for predicting the remaining service life of an energy storage battery. Background Art

[0002] Energy combustion is the main source of carbon dioxide emissions. Emissions from the power industry account for a high proportion of energy industry emissions, and the task of reducing emissions is arduous. The key to promoting clean and low-carbon energy transformation is to accelerate the development of new energy. However, new energy has problems such as instability. The large-scale development and utilization of new energy brings challenges to the safe operation of the power grid. It is urgent to improve the system's flexible adjustment capabilities through measures such as the development of energy storage;

[0003] As a dynamic and nonlinear electrochemical system, the capacity of energy storage batteries will gradually decay with the increase of charge and discharge times and the extension of continuous use time. Accurately estimating the state of health (SOH) of the battery is crucial for battery management and maintenance. According to IEEE Standard 1188-1996, when the SOH of the energy storage battery is less than 80%, the battery needs to be replaced.

[0004] Currently, accurately predicting the life of energy storage batteries still faces great challenges. Traditional methods mainly rely on simple statistical models or empirical formulas, which are difficult to fully consider the complex factors in the battery use process, such as ambient temperature, charging and discharging mode, maintenance, etc. These factors will have a significant impact on the battery life. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the above-mentioned prior art and propose a method and system for predicting the remaining service life of energy storage batteries. The battery capacity time series data is decomposed by TVF-EMD. Then, the number and size of convolution kernels of the temporal convolutional neural network (TCN) model are optimized by using the improved collaborative group optimization algorithm, thereby improving the prediction accuracy of the energy storage battery life.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is: a method for predicting the remaining service life of an energy storage battery, comprising the following steps:

[0007] Step 1: Obtain battery capacity time series data;

[0008] Step 2: Use TVF-EMD to decompose the battery capacity time series data to obtain the intrinsic mode component (IMF);

[0009] Step 3: Build a temporal convolutional neural network model;

[0010] Step 4: Use the improved collaborative group optimization algorithm to optimize the number and size of convolution kernels of the temporal convolutional neural network model; the improved collaborative group optimization algorithm improves the speed update formula:

[0011] ;

[0012] In the formula, is the updated speed of the candidate solution of the jth dimension of the i-th individual, is the inertia weight value, is the individual optimal coefficient, is the global optimal coefficient, is the dynamic attraction coefficient, is the adaptive neighborhood interaction coefficient, is the diversity preservation factor;

[0013] Step 5: Normalize the complete battery capacity time series data, and input the normalized data into the time convolutional neural network (TCN) model optimized by the improved collaborative group optimization algorithm to predict the remaining service life.

[0014] Further preferably, the process of decomposing the battery capacity time series data using TVF-EMD to obtain the intrinsic mode component (IMF) is as follows:

[0015] Step 2.1: First, use Hilbert transform to find the instantaneous frequency and instantaneous amplitude of the battery capacity time series data, then calculate the local maximum and local minimum of the instantaneous frequency and instantaneous amplitude respectively, perform difference budget on the local maximum and local minimum of the instantaneous frequency and instantaneous amplitude to get the difference between the local maximum and local minimum of the instantaneous frequency and the difference between the local maximum and local minimum of the instantaneous frequency, and finally use time-varying filter to process the battery capacity time series data, get the instantaneous mean and instantaneous envelope, and then calculate the local cutoff frequency:

[0016] Step 2.2: Reconstruct the battery capacity time series data according to the local cutoff frequency:

[0017] Step 2.3: Determine whether the cutoff frequency meets the stopping criteria:

[0018] Step 2.4: When the cutoff frequency is less than or equal to the Loughlin instantaneous bandwidth threshold, the input data of the current iteration number corresponding to the cutoff frequency is an eigenmode component.

[0019] Further preferably, the improved collaborative swarm optimization algorithm introduces Logistic-Tent chaotic mapping to initialize the population.

[0020] Further preferably, during the iteration process of the improved collaborative group optimization algorithm, an escape formula is introduced to enable it to jump out of the local optimal value;

[0021] ;

[0022] ;

[0023] In the formula, represents the candidate solution after the update of the jth dimension of the i-th individual, represents the current candidate solution of the jth dimension of the i-th individual, represents the solution at the last iteration, represents a random value of (0, 1), represents a constant that changes with the number of iterations, represents the current number of iterations, T represents the maximum number of iterations, and e is a natural constant.

[0024] More preferably, in step 1, outlier detection and interpolation are performed on the battery capacity time series data using the quartile method.

[0025] The present invention also provides a system for predicting the remaining service life of an energy storage battery. The system includes a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, the various steps of the above-mentioned method for predicting the remaining service life of the energy storage battery are implemented.

[0026] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the various steps of the above-mentioned method for predicting the remaining service life of an energy storage battery are implemented.

[0027] The beneficial effects of the present invention are as follows: TVF-EMD is used to decompose the battery capacity time series data, reducing the stationarity of the data, thereby reducing the impact on the prediction. The improved collaborative group optimization algorithm is used to optimize the number and size of the convolution kernels of the time convolution neural network (TCN) model, improving the prediction accuracy of wind power, reducing the randomness of the prediction of a single time convolution neural network (TCN) model, narrowing the prediction error range, and making the prediction of the time convolution neural network (TCN) model more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flow chart of a method for predicting the remaining service life of an energy storage battery;

[0029] Figure 2 Flowchart of the optimization algorithm for the collaborative group;

[0030] Figure 3Convergence curve of collaborative group optimization algorithm before and after improvement. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0032] Reference Figure 1 , a method for predicting the remaining service life of an energy storage battery, comprising the following steps:

[0033] Step 1: Obtain battery capacity time series data;

[0034] Step 2: Use TVF-EMD to decompose the battery capacity time series data to obtain the intrinsic mode component (IMF);

[0035] Step 3: Build a temporal convolutional neural network model;

[0036] Step 4: Use the improved collaborative group optimization algorithm to optimize the number and size of convolution kernels of the temporal convolutional neural network model;

[0037] Step 5: Normalize the complete battery capacity time series data, and input the normalized data into the time convolutional neural network (TCN) model optimized by the improved collaborative group optimization algorithm to predict the remaining service life.

[0038] In step 1 of this embodiment, the battery capacity time series data is also subjected to outlier detection and interpolation using the quartile method, and the process is as follows:

[0039] Sort the values ​​in each divided interval of battery capacity time series data from small to large, and calculate the 25% quantile of the sorted battery capacity time series data and the 75% quantile ; By quantile and quantiles Calculate the interquartile range :

[0040] ;

[0041] Define the normal value interval by calculating the lower boundary of the normal value interval and the upper boundary of the normal value interval , get the normal value interval , for those beyond the normal range The inner limit value is the abnormal value identified by the quartile method; Values ​​within the limit are retained.

[0042] In this embodiment, the process of decomposing the battery capacity time series data using TVF-EMD to obtain the intrinsic mode component (IMF) is as follows:

[0043] Step 2.1: First, use Hilbert transform to find the instantaneous frequency of the battery capacity time series data x(t) and instantaneous amplitude a(t), and then calculate the instantaneous frequency and the local maximum and local minimum of the instantaneous amplitude a(t), for the instantaneous frequency The instantaneous frequency is obtained by performing a difference budget with the local maximum and local minimum of the instantaneous amplitude a(t) The difference between the local maximum and the local minimum The difference between the local maximum and local minimum of the instantaneous frequency a(t) Finally, the time-varying filter is used to process the battery capacity time series data to obtain the instantaneous mean and the instantaneous envelope ,according to , , and Calculate the local cutoff frequency :

[0044] ;

[0045] In the formula, Indicates time;

[0046] Step 2.2: Based on the local cutoff frequency Reconstruct battery capacity time series data:

[0047] ;

[0048] In the formula, Represents the reconstructed battery capacity time series data; represents integral;

[0049] Step 2.3: Determine the cutoff frequency Whether the stopping criteria are met:

[0050] ;

[0051] ;

[0052] ;

[0053] ;

[0054] ;

[0055] In the formula, represents the weighted mean instantaneous frequency; Indicates The instantaneous frequency of the order component; represents the Loughlin instantaneous bandwidth; represents the instantaneous amplitude of the first-order component; represents the instantaneous amplitude of the second-order component; represents the instantaneous frequency of the first-order component; The instantaneous frequency of the second-order component;

[0056] Step 2.4: Let Loughlin instantaneous bandwidth threshold be ξ, when ≤ξ, then judge The input data corresponding to the current iteration number is an intrinsic mode component (IMF), otherwise, B-spline interpolation is used to Corresponding Approximate and get the approximation result m(t), that is, , The approximated data is used as the input data for the next iteration, and steps 2.1 to 2.3 are repeated.

[0057] In this embodiment, the temporal convolutional neural network model is constructed according to the following process:

[0058] Step 3.1, set the number of residual blocks to 10, the number of convolutional layers in each residual block to 3, and the basic parameters of each convolution layer;

[0059] Step 3.2, set the convolution layer parameters, step size 1, expansion factor [2 4 8 16 32 64 128];

[0060] Step 3.3: Construct a residual block with 3 convolutional layers, sigmoid activation function, dropout layer and batch normalization.

[0061] In this embodiment, Figure 2 As shown in the figure, the process of optimizing the number and size of convolution kernels of the temporal convolutional neural network model using the improved collaborative group optimization algorithm is as follows:

[0062] Step 4.1, set the maximum number of iterations to 100 and the population size to 30;

[0063] Step 4.2: Introduce Logistic-Tent chaotic mapping to initialize the population to solve the uneven distribution problem caused by random initialization;

[0064] ;

[0065] ;

[0066] ;

[0067] In the formula, X represents the initialized population, represents the candidate solution of the jth dimension of the i-th individual before the Logistic-Tent chaotic mapping, Represents the current candidate solution of the jth dimension of the i-th individual, i∈1,2…n; j∈1,2,…,d; n is the population size, d is the dimension of the problem; each individual represents a set of convolution kernel number and convolution kernel size parameters of the temporal convolutional neural network model, and are the upper and lower bounds of the problem, Rand represents a random number between 0 and 1, mod1 represents the remainder operator, r is a random number, .

[0068] Step 4.3, calculate the fitness and iterate, and update the fitness after iteration;

[0069] ;

[0070] In the formula, represents the candidate solution after the update of the jth dimension of the i-th individual, Represents the current speed of the candidate solution of the jth dimension of the i-th individual.

[0071] In view of the fact that the original algorithm is easily affected by bad individuals, the speed update formula is improved and the convergence speed is improved by using the average method;

[0072] ;

[0073] In the formula, is the updated speed of the candidate solution of the jth dimension of the i-th individual, is the inertia weight value, is the individual optimal coefficient, is the global optimal coefficient, is the dynamic attraction coefficient, is the adaptive neighborhood interaction coefficient, is the diversity preservation factor.

[0074] ;

[0075] In the formula, Indicates the current iteration number, For the Inertia weight parameter at iteration.

[0076] ;

[0077] In the formula, For the The inertia weight parameter at the iteration, k is the constant that determines the inertia reduction rate.

[0078] ;

[0079] In the formula, is the random value parameter of the individual optimal coefficient, is the reduction factor, is the solution of a random individual;

[0080] ;

[0081] In the formula, is the random value parameter of the global optimal coefficient, is the best global solution.

[0082] ;

[0083] In the formula, is the random value parameter of the dynamic attraction coefficient, is the position with the highest local attraction value near the i-th individual, is the additional acceleration coefficient of the dynamic attraction coefficient.

[0084] ;

[0085] In the formula, is the random value parameter of the adaptive neighborhood interaction coefficient, is the fitness value of a random individual, is the fitness value of the global optimal individual, is a random value between 0 and 1. is the fitness value of the worst individual.

[0086] ;

[0087] In the formula, is the random value parameter of the diversity preservation coefficient, is the position with the maximum diversity near the i-th individual in the population, is the additional acceleration factor for the diversity preservation factor.

[0088] 4.4: During the iteration process, in order to prevent the algorithm from falling into the local optimal value, an escape formula is introduced to make it jump out of the local optimal value;

[0089] ;

[0090] ;

[0091] In the formula, represents the solution at the last iteration, represents a random value of (0, 1), represents a constant that changes with the number of iterations, T represents the maximum number of iterations, and e is a natural constant.

[0092] 4.5: Determine whether the current iteration has reached the maximum number of iterations. If not, continue to iterate; otherwise, stop iterating and output the optimal solution, that is, the number of convolution kernels of the temporal convolutional neural network (TCN) model is 7 and the convolution kernel size is 3, which is the optimal value.

[0093] Depend on Figure 3 It can be seen that the collaborative group optimization algorithm before and after the improvement is far superior to that before the improvement in terms of convergence speed and convergence accuracy, which makes the method of the present invention more feasible.

[0094] Table 1 Comparison of prediction performance of different models

[0095]

[0096] As shown in Table 1, the method of the present invention is compared with the commonly used method. As shown in Table 1, MAE is the mean absolute error, MAPE is the mean absolute percentage error, and RMSE is the root mean square error. The method of the present invention is the smallest under the three evaluation indicators where the smaller the better, and the prediction accuracy is higher.

[0097] Another embodiment of the present invention provides a system for predicting the remaining useful life of an energy storage battery, the system comprising a memory, a processor, and a computer program stored in the memory, wherein the computer program implements the various steps of the above-mentioned method for predicting the remaining useful life of an energy storage battery when executed by the processor.

[0098] Another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the various steps of the above-mentioned method for predicting the remaining service life of an energy storage battery are implemented.

[0099] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the remaining service life of an energy storage battery, characterized in that: The following steps are involved: Step 1: Obtain battery capacity time series data; Step 2: Use TVF-EMD to decompose the battery capacity time series data to obtain the intrinsic mode components; Step 3: Build a temporal convolutional neural network model; Step 4: Use the improved collaborative group optimization algorithm to optimize the number and size of convolution kernels of the temporal convolutional neural network model; the improved collaborative group optimization algorithm improves the speed update formula: ； In the formula, is the updated speed of the candidate solution of the jth dimension of the i-th individual, is the inertia weight value, is the individual optimal coefficient, is the global optimal coefficient, is the dynamic attraction coefficient, is the adaptive neighborhood interaction coefficient, is the diversity preservation factor; ； In the formula, Indicates the current iteration number, For the Inertia weight parameter at the iteration; ； In the formula, For the The inertia weight parameter at the iteration, k is the constant that determines the inertia weight reduction rate; ； In the formula, is the random value parameter of the individual optimal coefficient, is the reduction factor, is the solution of a random individual, Represents the current candidate solution of the jth dimension of the i-th individual; ； In the formula, is the random value parameter of the global optimal coefficient, is the best global solution; ； In the formula, is the random value parameter of the dynamic attraction coefficient, is the position with the highest local attraction value near the i-th individual, is the additional acceleration coefficient of the dynamic attraction coefficient; ； In the formula, is the random value parameter of the adaptive neighborhood interaction coefficient, is the fitness value of a random individual, is the fitness value of the global optimal individual, is a random value between 0 and 1. is the fitness value of the worst individual; ； In the formula, is the random value parameter of the diversity preservation coefficient, is the position with the maximum diversity near the i-th individual in the population, is the additional acceleration factor of the diversity preservation factor; During the iteration process of the improved collaborative group optimization algorithm, an escape formula is introduced to enable it to jump out of the local optimal value; ； ； In the formula, represents the candidate solution after the update of the jth dimension of the i-th individual, represents the solution at the last iteration, represents a random value of (0, 1), represents a constant that changes with the number of iterations, represents the current number of iterations, T represents the maximum number of iterations, and e is a natural constant; Step 5: Normalize the complete battery capacity time series data, and input the normalized data into the time convolutional neural network model optimized by the improved collaborative group optimization algorithm to predict the remaining service life.

2. The method for predicting the remaining service life of an energy storage battery according to claim 1, characterized in that: The process of using TVF-EMD to decompose the battery capacity time series data to obtain the intrinsic mode components is as follows: Step 2.1: First, use Hilbert transform to find the instantaneous frequency and instantaneous amplitude of the battery capacity time series data, then calculate the local maximum and local minimum of the instantaneous frequency and instantaneous amplitude respectively, perform difference budget on the local maximum and local minimum of the instantaneous frequency and instantaneous amplitude to get the difference between the local maximum and local minimum of the instantaneous frequency and the difference between the local maximum and local minimum of the instantaneous frequency, and finally use time-varying filter to process the battery capacity time series data, get the instantaneous mean and instantaneous envelope, and then calculate the local cutoff frequency: Step 2.2: Reconstruct the battery capacity time series data according to the local cutoff frequency: Step 2.3: Determine whether the cutoff frequency meets the stopping criteria: Step 2.4: When the cutoff frequency is less than or equal to the Loughlin instantaneous bandwidth threshold, the input data of the current iteration number corresponding to the cutoff frequency is an eigenmode component.

3. The method for predicting the remaining service life of an energy storage battery according to claim 1, characterized in that: The improved collaborative swarm optimization algorithm introduces Logistic-Tent chaotic mapping to initialize the population.

4. The method for predicting the remaining service life of an energy storage battery according to claim 3, characterized in that: The process of optimizing the number and size of convolution kernels of the temporal convolutional neural network model using the improved collaborative group optimization algorithm is as follows: Step 4.1, set the maximum number of iterations and population size; Step 4.2, introduce Logistic-Tent chaotic mapping to initialize the population; Step 4.3, calculate the fitness and iterate, and update the fitness after iteration; Step 4.4: During the iteration process, the escape formula is introduced to make it jump out of the local optimal value; Step 4.5: Determine whether the current iteration has reached the maximum number of iterations. If not, continue iterating; otherwise, stop iterating and output the optimal solution.

5. The method for predicting the remaining service life of an energy storage battery according to claim 1, characterized in that: In the step 1, the battery capacity time series data is also subjected to outlier detection and interpolation using the quartile method.

6. A system for implementing the method for predicting the remaining service life of an energy storage battery according to any one of claims 1 to 5, characterized in that: The system includes a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, each step of the above-mentioned method for predicting the remaining service life of the energy storage battery is implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, each step of the method for predicting the remaining service life of an energy storage battery according to any one of claims 1 to 5 is implemented.

Images