An energy storage battery residual life prediction method based on an INFO-SVR algorithm

Abstract

The application discloses a kind of based on INFO-SVR algorithm's energy storage battery residual life prediction method, comprising: obtaining energy storage battery residual life cycle dataset;The cycle dataset is analyzed to obtain health factor;The health factor is normalized, and the health factor after normalization is divided into test data and training data;Initial model is trained based on the training data, and the initial model is verified based on the test data, to generate INFO-SVR battery residual life prediction model, wherein the INFO-SVR battery residual life prediction model is used to predict the residual life of energy storage battery.Utilize the stronger exploration of INFO algorithm, escape local optimal capacity and faster development convergence speed, effectively solve the optimal super parameter value problem in SVR algorithm, with better prediction effect and precision to the state of energy storage battery.

Description

Technical Field

[0001] This invention belongs to the field of energy storage battery state monitoring technology, specifically relating to a method for predicting the remaining lifespan of energy storage batteries based on the INFO-SVR algorithm. Background Technology

[0002] With the introduction of dual-carbon goals, energy storage batteries have become an indispensable part of the entire industrial production process, and their operational status directly affects the stability of the entire system. However, as energy storage batteries are repeatedly used, their lifespan decreases. When the battery capacity decays to 80% or the internal resistance reaches 100%, it signifies that the battery's lifespan has ended. Failure to replace the battery in a timely manner when its lifespan has expired will significantly reduce the reliability of the entire system, potentially leading to equipment or system failure. Furthermore, the increased internal resistance also increases the battery's inherent safety hazards, making it prone to thermal runaway, internal short circuits, and other problems, which could potentially lead to catastrophic accidents. Therefore, accurately predicting the remaining lifespan of energy storage batteries is of great significance for extending battery life and reducing equipment maintenance costs.

[0003] Currently, data-driven methods for predicting the remaining life (SOH) of energy storage batteries are widely used. While domestic and international scholars have conducted extensive research to improve prediction accuracy, significant issues remain. Most studies use data obtained under constant temperature and humidity conditions, failing to consider the impact of discharge current variations on prediction accuracy. Chinese patent application CN115684972A proposes a lithium-ion battery SOH estimation method based on the SSA-SVR model. This method uses the SSA-SVR model to estimate the SOH of lithium batteries, but significant errors are still evident in the prediction accuracy, which cannot be guaranteed. Furthermore, the paper does not consider the influence of temperature and humidity on the battery during data collection. The literature [Wang Yingzhou, Ni Yulong, Zheng Yuqing. Prediction of remaining lifespan of lithium-ion batteries based on ALO-SVR [J]. Journal of Electrical Engineering, 2021, 41(04)] used the Antlion optimized SVR algorithm to predict the remaining lifespan of the battery, but the prediction error after optimization was still very large. The article still did not take into account the influence of temperature and humidity on the battery lifespan in the data acquisition. Summary of the Invention

[0004] In view of the above, the present invention provides a method for predicting the remaining life of energy storage batteries based on the INFO-SVR algorithm, which can realize the prediction of the remaining life of energy storage batteries and effectively improve the prediction accuracy of the model.

[0005] A method for predicting the remaining lifetime of energy storage batteries based on the INFO-SVR algorithm includes the following steps:

[0006] Obtain the remaining life cycle dataset of the energy storage battery;

[0007] The cyclic dataset is analyzed to obtain health factors;

[0008] The health factors are normalized, and the normalized health factors are divided into test data and training data.

[0009] The initial model is trained based on the training data and validated based on the test data to generate an INFO-SVR battery remaining life prediction model, wherein the INFO-SVR battery remaining life prediction model is used to predict the remaining life of the energy storage battery.

[0010] Optionally, obtaining the remaining life cycle data set of the energy storage battery includes:

[0011] Based on the experimental chamber, daily humidity and temperature change data around the location of the placed device are collected. The placed device contains at least three batteries and is placed inside the experimental chamber.

[0012] The battery was charged and discharged using DAQ equipment and MATLAB experimental control device.

[0013] When the battery cycle is divided into charging modes, constant current charging is performed at 2.6A, and when the voltage reaches 4.2V, constant voltage charging is performed at 4.2V until the current drops to 0.08A, and then charging is stopped.

[0014] When the battery cycle is divided into discharge modes, the discharge is oscillating with alternating currents of 2A and 3A per minute, and the discharge stops when the detected voltage reaches the set cutoff voltage of 2.5V.

[0015] The battery is subjected to multiple charge-discharge cycle tests until its capacity reaches 80% of its rated capacity. The test is then terminated, and the remaining life cycle data of the energy storage battery generated during the test is obtained.

[0016] Optionally, the initial model is trained based on the training data, and the initial model is validated based on the test data to generate an INFO-SVR battery remaining life prediction model, including:

[0017] Establish an SVR model as the initial model;

[0018] The INFO algorithm is used to optimize the SVR model to obtain the penalty factor and kernel parameters;

[0019] The optimized SVR model is trained based on the training data to obtain the trained INFO-SVR battery remaining life prediction model.

[0020] Optionally, before optimizing the SVR model using the INFO algorithm to obtain the penalty factor and kernel parameters, the method further includes:

[0021] Set the parameters for the INFO algorithm, including: population size, maximum number of iterations, number of variables, upper bound of parameters, lower bound of parameters, and fitness function.

[0022] Optionally, the step of using the INFO algorithm to optimize the SVR model to obtain the penalty factor and kernel parameters includes:

[0023] The INFO algorithm parameters are initialized, and the search domain is defined according to the set INFO algorithm parameters. A vector group is randomly generated in the search domain, and the objective function value of each vector in the vector group is calculated to determine the optimal vector x in the vector group. bs ;

[0024] Within the search domain, three sets of vectors with different directions are randomly selected, and the weights corresponding to each vector are calculated using wavelet functions.

[0025] Based on the preset first formula and the weights, calculate the vector corresponding to each iteration. and

[0026] Based on the preset second formula, the vector The new vector obtained by merging in, This is the vector generated under the condition rand < 0.5, where rand is a randomly generated number within [0, 1].

[0027] When rand < 0.5, based on the preset third formula, the result after the g-th iteration is determined. The nearby new solution vector x rng And perform position determination and update;

[0028] The generated optimal vector x bs Return the result as the optimal solution.

[0029] Optionally, the preset first formula is:

[0030]

[0031]

[0032]

[0033] σ=2α×rand-α

[0034] in: These represent three sets of vectors randomly selected in the g-th iteration, where g is the current iteration number, Maxg is the maximum iteration number, and x represents the direction vector. bs x bt x ws Let be the optimal solution vector, the second-best solution vector, and the third-best solution vector in the population during the g-th iteration.

[0035] Optionally, the preset second formula is:

[0036]

[0037] in, For the generated new vector, For generating vectors with the condition rand < 0.5, This is the new vector obtained by merging vectors in the g-th generation, where rand1 and rand2 are uniformly distributed random real numbers in [0,1].

[0038] Optionally, the preset third formula is:

[0039] x rnd =φ×x avg +(1-φ)×(φ×xbt+(1-φ)×x bs )

[0040]

[0041] Where: x rng It is made by x avg x bt and x bs The new solution vector is formed to increase the randomness of the algorithm and achieve better search in the solution space. φ represents a random number in (0,1); where v1 and v2 are two random numbers that increase the influence of the optimal position on the vector.

[0042] The formulas for calculating v1 and v2 are as follows:

[0043]

[0044]

[0045] Where p represents a uniformly distributed random real number in the range [0,1].

[0046] Based on the above technical solution, the present invention has the following beneficial technical effects:

[0047] In conducting energy storage battery charging and discharging experiments, this invention takes into account the influence of temperature and humidity at the device placement location on the data. It collects real daily humidity and temperature change data around the device placement location for environmental changes and conducts experiments, providing support for the study of battery life under special environments.

[0048] This invention addresses the difficulty of hyperparameter selection in SVR algorithms by using the INFO algorithm for SVR hyperparameter optimization. Compared to traditional SVR algorithms, the INFO algorithm, with its stronger ability to explore and avoid local optima and faster convergence speed, provides better prediction results and accuracy for the state of energy storage batteries. Furthermore, this invention has better extensibility for other research. Attached Figure Description

[0049] Figure 1 The flowchart shows a method for predicting the remaining life of energy storage batteries based on the INFO-SVR algorithm.

[0050] Figure 2 This is a flowchart of the remaining life prediction process for energy storage batteries based on the INFO-SVR algorithm of this invention.

[0051] Figure 3 This is a schematic diagram of the INFO algorithm flow of the present invention;

[0052] Figure 4 This is a schematic diagram showing the prediction results of the remaining lifespan of the energy storage battery under the algorithm model of this invention. Detailed Implementation

[0053] To describe the present invention in more detail, the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] like Figure 1 As shown, the present invention provides a method for predicting the remaining life of energy storage batteries based on INFO-SVR, which includes the following steps:

[0055] Step 101: Obtain the remaining life cycle dataset of the energy storage battery.

[0056] Based on the experimental chamber, daily humidity and temperature change data around the location of the placed device are collected. The placed device contains at least three batteries and is placed inside the experimental chamber.

[0057] The battery was charged and discharged using DAQ equipment and MATLAB experimental control device.

[0058] When the battery cycle is divided into charging modes, constant current charging is performed at 2.6A, and when the voltage reaches 4.2V, constant voltage charging is performed at 4.2V until the current drops to 0.08A, and then charging is stopped.

[0059] When the battery cycle is divided into discharge modes, the discharge is oscillating with alternating currents of 2A and 3A per minute, and the discharge stops when the detected voltage reaches the set cutoff voltage of 2.5V.

[0060] The battery is subjected to multiple charge-discharge cycle tests until its capacity reaches 80% of its rated capacity. The test is then terminated, and the remaining life cycle data of the energy storage battery generated during the test is obtained.

[0061] Step 102: Analyze the cyclic dataset to obtain health factors;

[0062] This implementation example uses the NASA battery cycling dataset to predict remaining battery life.

[0063] Step 103: Normalize the health factors and divide the normalized health factors into test data and training data.

[0064] The NASA battery cycling dataset was divided into data samples, health factors were extracted, and the health factors were preprocessed using normalization. The normalization method was as follows:

[0065]

[0066] Where: x represents the data before normalization, x' represents the data after normalization, xmax represents the maximum value of the sample data, and xmin represents the minimum value of the sample data. 80% of the sample data is selected as the training set, and the remainder is the test set.

[0067] Step 104: Train the initial model based on the training data and validate the initial model based on the test data to generate the INFO-SVR battery remaining life prediction model, wherein the INFO-SVR battery remaining life prediction model is used to predict the remaining life of the energy storage battery.

[0068] An SVR model was established, and then the INFO algorithm was used to optimize the SVR model to obtain the optimal penalty factor and kernel parameters. The optimized SVR algorithm was then trained using the training set to obtain the INFO-SVR battery remaining life prediction model. Figure 2 This is a flowchart illustrating the remaining life prediction process for energy storage batteries based on the INFO-SVR algorithm of this invention. Given a battery cycle data training set:

[0069] S={(x1,y1),(x2,y2),…,(x i ,y i )},y i ∈R

[0070] Where: x i Let y be the input vector. i For the corresponding output vector, the nonlinear mapping function is defined as follows:

[0071] f(x) = wφ(x) + b

[0072] Where: w is the direction vector, and b is the distance between the hyperplane and the far point; the goal of SVR is to make f(x) and y as close as possible, but there will always be a deviation between them. Therefore, a maximum tolerance deviation value ε (ε>0) is introduced. When the deviation between f(x) and y is less than ε, it is considered that there is no loss; otherwise, the loss is included in the calculation. The model is then:

[0073]

[0074] Since the degree of relaxation differs in the two instances on the hyperplane, a relaxation variable ξ is introduced. i and ξ i After correcting for irregular factors, the objective function transforms into the following constrained form:

[0075]

[0076]

[0077] In the formula: ξ i and ξ i Using the relaxation factor, the regression problem is transformed into a problem of minimizing the objective function. At this point, the Lagrange penalty operator is introduced, transforming the objective function into a dual problem:

[0078]

[0079]

[0080] a in the objective function i and a i 'Using the Lagrange multiplier, the expression for the nonlinear mapping SVR can be obtained by minimizing the Lagrange function:

[0081]

[0082] In the formula: K(x) is the kernel function. i The inner product operation can transform a low-dimensional linearly inseparable problem into a high-dimensional linearly separable problem. The kernel function chosen is the radial basis function, defined as:

[0083]

[0084] Where σ represents the kernel parameter, and the values ​​of the penalty factor C and the kernel parameter σ directly affect the prediction performance of the final SVR model. Inaccuracy of the penalty factor will reduce the model's generalization ability. The kernel parameter implicitly determines the distribution of data after mapping to the new feature space. However, there is currently a lack of guiding methods for selecting the penalty factor and kernel parameter. The vector weighted mean algorithm has strong capabilities in developing, exploring, escaping local optima, and convergence, and can achieve good numerical optimization results for the penalty factor and kernel parameter. It can effectively improve the prediction accuracy of the remaining life of energy storage batteries in the SVR algorithm. The specific process is as follows:

[0085] First, the vector weighted mean algorithm is used for optimization. Figure 3 This is a schematic diagram of the INFO algorithm flow of the present invention. In the initialization phase of the INFO algorithm, a vector group is randomly generated in the defined search domain D. There are N vectors in the defined search domain D. p Given a vector, the search domain is:

[0086] D = (X l,j ={X l,j ,X l,j ,…,X l,j}, l=1,2,…,Np)

[0087] Two main control parameters are defined: the weighted average factor δ and the vector scaling factor σ. They are defined as follows:

[0088] δ=2β×rand-β

[0089] σ=2α×randn-α

[0090] Where rand represents a random number that follows a standard normal distribution, and α and β are defined as follows:

[0091]

[0092]

[0093] Here, g represents the current iteration number, and Maxg represents the maximum iteration number.

[0094] (1) Rule update phase

[0095] The INFO algorithm increases population diversity during the search process through an update rule phase. First, it extracts the weighted mean of vectors from a set of random vectors. Then, it moves this weighted mean information from a randomly selected set of vectors to the next solution, increasing population diversity through optimal, second-best, and worst solutions. The MeanRule formula is defined as follows:

[0096]

[0097] Where r is a random number between [0, 0.5]; and The weighted average formula is defined as follows:

[0098]

[0099] In the formula, ε represents a constant value. In this study, the weight of each vector is calculated based on a wavelet function, where w1, w2, and w3 are wavelet functions. Wavelet functions can generate effective fluctuation information during the optimization process, thereby achieving global search in the solution space. The wavelet function is defined as follows:

[0100]

[0101]

[0102]

[0103] ω=max(f(x a1 ),f(x a2 ),f(x a3 ))

[0104] The definition is as follows:

[0105]

[0106] in:

[0107]

[0108]

[0109]

[0110] ω=f(x ws )

[0111] Where f(x) is the objective function value, x bs x bt x ws These represent the optimal, second-best, and third-best solution vectors in the population at the g-th iteration, respectively. After each iteration, all population vectors are sorted to obtain x. bs x bt x ws .

[0112] Simultaneously, a convergence acceleration component is added during the rule update phase. The optimal vector is used to move the current vector within the search domain; here, the optimal solution represents the solution vector closest to the global optimum. Furthermore, a random number within the range [0,1] is multiplied to ensure that each vector has a different step size in each iteration, as shown in the following formula:

[0113]

[0114] Where Randn is a normally distributed random number. The new vector can be calculated using the standard formula CA:

[0115]

[0116] Update rules are based on x bs x bt , and Define the position of the new vector after the g-th iteration. and The formula is as follows:

[0117]

[0118]

[0119] (2) Vector merging stage

[0120] The vector merging stage can improve the local search capability of the INFO algorithm, thus bringing better vectors closer together. The main idea is to combine the two new vectors generated in the previous stage... and Vectors with condition rand < 0.5 Combine to generate a new vector

[0121]

[0122] in, This is the new vector obtained by merging vectors in the g-th generation; where μ = 0.05 × randn.

[0123] (3) Local search phase

[0124] The local search phase effectively prevents the INFO algorithm from getting stuck in local optima. In this process, A new vector can be found nearby, where rand is a random value in [0,1].

[0125]

[0126] Where, x rng It is made by xavg x bt and x bs The new solution vector is formed to increase the randomness of the algorithm and achieve better search in the solution space. Its formula is defined as follows:

[0127] x rnd =φ×x avg +(1-φ)×(φ×xbt+(1-φ)×x bs )

[0128]

[0129] Where φ represents a random number in (0,1); and v1 and v2 are two random numbers that increase the influence of the optimal position on the vector, defined as follows:

[0130]

[0131]

[0132] p represents a random number in the range [0,1]. The overall flowchart of the INFO algorithm is as follows: Figure 1 As shown;

[0133] Through repeated iterations until the requirements are met, the optimal solutions for the two hyperparameters, the penalty factor and the kernel parameter, are output. The INFO algorithm then terminates, and the obtained optimal hyperparameters are returned to the SVR model to construct the INFO-SVR model. The root mean square error (RMSE) is used as the error evaluation metric.

[0134]

[0135] In the formula: and y i These are the predicted capacity and the actual capacity for the i-th iteration, respectively, where n represents the number of samples.

[0136] The prediction evaluation indicators selected are the mean absolute error, mean absolute percentage error, and root mean square error. The mean absolute error (MAE) is the average of the absolute differences between the actual and estimated values ​​of the SOH, used to describe the overall performance of the method; the mean absolute percentage error (MAPE) measures the relative percentage error between the estimated and actual SOH values; and the root mean square error (RMSE) is the square root of the mean square error, representing the deviation between the estimated and actual values.

[0137]

[0138]

[0139]

[0140] in SOH i These represent the predicted and actual remaining lifespan of the energy storage battery, respectively.

[0141] like Figure 4 As shown, based on the simulation results of the remaining life of the B0005 battery, RMSE = 0.0018963, which is a good result.

[0142] In this embodiment, a cyclic dataset of remaining battery life is first obtained. This dataset is then analyzed to obtain health factors. These health factors are then normalized and divided into test data and training data. Finally, an initial model is trained based on the training data and validated based on the test data to generate an INFO-SVR battery remaining life prediction model. This model is used to predict the remaining life of the energy storage battery. Therefore, by utilizing the stronger exploration and local optima avoidance capabilities and faster convergence speed of the INFO algorithm, the optimal hyperparameter value problem in the SVR algorithm is effectively solved, resulting in better prediction performance and accuracy for the energy storage battery state. Furthermore, during energy storage battery charge / discharge experiments, considering the influence of temperature and humidity on the data, experiments are conducted using real daily humidity and temperature variation data of the energy storage battery's placement location, providing support for research on battery life under special environments.

[0143] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be made to the above embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining lifespan of energy storage batteries based on the INFO-SVR algorithm, characterized in that, include: Obtaining the remaining life cycle dataset of the energy storage battery, wherein obtaining the remaining life cycle dataset of the energy storage battery includes: Based on the experimental chamber, daily humidity and temperature change data around the location of the placed device are collected. The placed device contains at least three batteries and is placed inside the experimental chamber. The battery was charged and discharged using DAQ equipment and MATLAB experimental control device. When the battery cycle is divided into charging modes, constant current charging is performed at 2.6A, and when the voltage reaches 4.2V, constant voltage charging is performed at 4.2V until the current drops to 0.08A, and then charging is stopped. When the battery cycle is divided into discharge modes, the discharge is oscillating with alternating currents of 2A and 3A per minute, and the discharge stops when the detected voltage reaches the set cutoff voltage of 2.5V. The battery is subjected to multiple charge-discharge cycle tests until the battery capacity reaches 80% of the rated capacity. The test is then terminated, and the remaining life cycle data of the energy storage battery generated during the test is obtained. The cyclic dataset is analyzed to obtain health factors; The health factors are normalized, and the normalized health factors are divided into test data and training data. The initial model is trained based on the training data and validated based on the test data to generate an INFO-SVR battery remaining life prediction model, wherein the INFO-SVR battery remaining life prediction model is used to predict the remaining life of the energy storage battery. The step of training an initial model based on the training data and validating the initial model based on the test data to generate an INFO-SVR battery remaining life prediction model includes: Establish an SVR model as the initial model; The INFO algorithm is used to optimize the SVR model to obtain the penalty factor and kernel parameters; The optimized SVR model is trained based on the training data to obtain the trained INFO-SVR battery remaining life prediction model. Before using the INFO algorithm to optimize the SVR model and obtain the penalty factor and kernel parameters, the process further includes: Set the parameters for the INFO algorithm, including: population size, maximum number of iterations, number of variables, upper bound of parameters, lower bound of parameters, and fitness function; The process of optimizing the SVR model using the INFO algorithm to obtain the penalty factor and kernel parameters includes: The INFO algorithm parameters are initialized, and the search domain is defined based on the set INFO algorithm parameters. A vector group is randomly generated within the search domain, and the objective function value of each vector in the vector group is calculated to determine the optimal vector in the vector group. ; Within the search domain, three sets of vectors with different directions are randomly selected, and the weights corresponding to each vector are calculated using wavelet functions. Based on the preset first formula and the weights, calculate the vector corresponding to each iteration. and ; Based on the preset second formula, the vector , , The new vector obtained by merging ,in, As a condition The generation vector, rand is a randomly generated number in [0,1]; when At that time, based on the preset third formula, it is determined that after the g-th iteration... New solution vectors in the vicinity And perform position determination and update; The generated optimal vector Return the result as the optimal solution.

2. The method according to claim 1, characterized in that, The preset first formula is: in: , , These represent three sets of randomly selected direction vectors in the g-th iteration, where g is the current iteration number and Maxg is the maximum iteration number. , , Let be the vectors of the optimal solution, the second-best solution, and the third-best solution in the population during the g-th iteration. This is an adjustment factor that varies with the number of iterations. This is the weighting adjustment coefficient; The average rule term is calculated based on three randomly selected sets of different direction vectors. It is used to characterize the comprehensive guidance information of multiple direction vectors and participate in the generation calculation of new vectors. rand is a number randomly generated in [0,1] to enhance its randomness and search capability in the iterative optimization process.

3. The method according to claim 1, characterized in that, The preset second formula is: in, , For the generated new vector, As a condition The generation vector, This is the new vector obtained by merging vectors in the g-th generation, where rand1 and rand2 are uniformly distributed random real numbers in [0,1].

4. The method according to claim 1, characterized in that, The preset third formula is: in: It is by , and The new solution vector is formed to increase the randomness of the algorithm, thereby achieving a better search in the solution space, where ϕ represents a random number in (0,1); where and These are two random numbers that increase the influence of the optimal position on the vector. and The calculation formula is as follows: Where p represents a uniformly distributed random real number in the range [0,1].

Images