A method, device and electronic equipment for predicting the remaining life of an energy storage battery

Abstract

The application discloses a kind of energy storage battery remaining life prediction method, device and electronic equipment, the energy storage battery remaining life prediction method uses the improved phase space reconstruction method based on chaos theory, overcome the severe fluctuation that exists when energy storage battery is in life prediction, can effectively extract energy storage battery data feature information.The battery capacity data after reconstruction is divided into training set and test set in the present application, training set is input into MSSA-informer neural network model, long time series prediction with lower complexity can be realized, finally the accuracy of life prediction model is verified using test set, the accuracy of energy storage battery remaining life prediction can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of lithium-ion battery health status detection technology, and in particular, to a method, apparatus and electronic device for predicting the remaining life of an energy storage battery. Background Technology

[0002] Predicting the lifespan of energy storage batteries is currently a hot research topic in the battery field. With the rapid development of batteries, research on predicting their health status is also deepening. Accurately assessing their health status helps predict the overall lifespan of the battery and improve charging and discharging strategies to avoid battery abuse. Currently, most studies use characteristic parameters of battery aging to define the health status, such as capacity and internal resistance. Due to the nonlinear, time-varying, and uncertain characteristics of batteries, their temporal features are difficult to extract. Furthermore, most machine learning methods for time-series prediction are based on recurrent neural networks, such as RNNs and LSTMs. While these models can better solve sequence prediction problems compared to other deterministic and traditional statistical models, they are prone to gradient vanishing and gradient exploding problems because they extract time-series information sequentially and propagate it backwards. This makes the model difficult or impossible to train, severely impacting prediction accuracy. While Transformer models can achieve long-term sequence prediction, their high computational complexity makes lifespan prediction challenging and also affects prediction accuracy. Summary of the Invention

[0003] To address the aforementioned technical problems, this application provides a method for predicting the remaining lifespan of energy storage batteries, thereby solving the technical problem of low accuracy in predicting the remaining lifespan of existing energy storage batteries.

[0004] The technical solution adopted in this application is as follows:

[0005] A method for predicting the remaining lifespan of an energy storage battery, specifically including the following steps:

[0006] S1: Preprocess the battery capacity data of the energy storage battery that has missing, abnormal and duplicate data, and divide the extracted data into training set and test set; use the improved phase space reconstruction method based on chaos theory to transform the one-dimensional time series of battery capacity data into a two-dimensional chaotic sequence, and learn and extract the effective nonlinear information in the two-dimensional chaotic sequence.

[0007] S2: Chaotic mapping is introduced to initialize and adjust the relevant parameters of the SSA algorithm. Secondly, in order to improve the diversity of the population and enhance the search of the surrounding space, a logarithmic spiral strategy is introduced to improve the SSA algorithm. Finally, an adaptive step size strategy is introduced to improve the SSA algorithm, better control the development and exploration capabilities of the algorithm, and enhance the convergence speed of the algorithm.

[0008] S3: Based on the MSSA algorithm, the key parameters in the informer network model are optimized. The training set of reconstructed battery capacity data is input into the informer network model to train the prediction model. The MSSA algorithm finds the optimal solution for encoder sequence dimension, decoder sequence dimension, number of heads, number of encoder layers, number of decoder layers, learning rate, and batch size in the informer model, initializes the informer model parameters, constructs the informer network model, and outputs the predicted battery capacity value.

[0009] S4: Input the test set of the reconstructed battery capacity data into the trained prediction model to obtain the predicted value of the battery capacity.

[0010] Further, in step S1, an improved phase space reconstruction method based on chaos theory is used to transform the one-dimensional time series of battery capacity data into a two-dimensional chaotic sequence, and to learn and extract the effective nonlinear information in the two-dimensional chaotic sequence. The specific steps are as follows:

[0011] S101. Setting a univariate chaotic time series:

[0012] s={s i :i=1,2,…,N}

[0013] In the formula: N is the time length; s i Let be the sequence value at time i;

[0014] Assuming a univariate chaotic time series, the coordinate reconstruction method requires two important parameters: the delay time τ and the embedding dimension d. The sequence values ​​from every τ time steps after time i are inserted into the current vector until a d-dimensional vector is formed. This vector represents the state vector x corresponding to the i-th time step in the reconstructed and improved phase space. i :

[0015] x i =(s i ,s i+τ ,s i+2τ ,…,s i+(d-1)τ ) T i = 1, 2, ..., m

[0016] In the formula: m = N - (d-1)τ, m is the total number of states in the reconstructed and improved phase space, and finally the reconstructed and improved phase space is obtained:

[0017] X = (x1, x2, ..., x m );

[0018] S102. Select the delay time τ using the mutual information method:

[0019]

[0020] Wherein, P(x n ), P(x n+τ Let I(τ) be the probability, and τ be the delay time interval, where I(τ) is the first local minimum. Based on the characteristic that mutual information can effectively solve nonlinear systems but cannot easily and efficiently solve linear time series, an improved phase space reconstruction method is proposed:

[0021]

[0022] When the system has a linear time relationship, choose R. m The first zero of (τ) is the time delay τ; when the system has a nonlinear time relationship, choose... and The minimum value is the time delay τ, where The factor enables the improved phase space method to be extended to higher dimensions, and the method can solve linear and nonlinear time-series systems.

[0023] For chaotic time series, the existing autocorrelation method defines the autocorrelation function as follows:

[0024]

[0025] When the value of the autocorrelation function R(τ) decreases to (1-1 / e) times the initial value R(0), the corresponding τ is a suitable delay time value. However, the choice of a decrease coefficient of (1-1 / e) times in the autocorrelation method is somewhat subjective and needs to be adjusted according to specific circumstances. Furthermore, it cannot be extended to higher dimensions for research. Additionally, the autocorrelation method can only extract the linear correlation of time series. After summarizing the shortcomings of the autocorrelation method, this embodiment proposes a method for judging the nonlinear relationship of a system: the mutual information method. The mutual information method can effectively solve systems with nonlinear relationships, but it cannot easily and efficiently solve time series with linear relationships. Therefore, this embodiment proposes the improved phase space reconstruction method mentioned above, thereby easily and efficiently solving the problem of time series with linear relationships.

[0026] S103. The improved Cao method is used to select the embedding dimension d. The Cao method is an improvement over the spurious nearest neighbor method. Its advantage is that it only requires the parameter delay time τ. In the reconstructed improved phase space, the Euclidean distance between every two pairs of state points is calculated, i.e.:

[0027] R d,i =|x i -x j |

[0028] x i With x j After extending to one dimension, the Euclidean distance is calculated again:

[0029] R' d,i =|x i '-x' j |

[0030] However, the Cao method is easily affected by the dimensions of the sequence, and the correlation between variables can affect the calculation results. Therefore, this embodiment improves the Cao method and proposes an improved Cao method:

[0031] The distance between every two pairs of state points, i.e.:

[0032]

[0033] Where, μ i Let be the expected value of the i-th element, and S be the covariance matrix;

[0034] x i and After extending it to one dimension, calculate the distance again:

[0035]

[0036] In the improved Cao's method, D is used directly. M (x), D M The ratio of (x)' to i defines the index:

[0037]

[0038] The distance metric chooses the infinite norm, defined as follows:

[0039]

[0040] In the formula: E(d) represents the overall spurious nearest neighbor situation, and how E(d) changes as the embedding dimension changes from d to d+1. Defined as follows:

[0041]

[0042] If the time series is chaotic, then E1(d) will tend to stabilize after d is greater than a certain value, so it is more appropriate to use this certain value as the embedding dimension d.

[0043] Furthermore, step S2 specifically includes the following steps:

[0044] S201: A chaotic mapping is introduced to initialize and adjust the relevant parameters of the SSA algorithm. By leveraging the characteristics of chaos, variables are mapped to the value range of a chaotic variable space. Finally, the solution is linearly transformed into the optimization variable space. Specifically, the Circle chaotic mapping is used when generating the initial population, and its definition is:

[0045]

[0046] Where: mod is the modulo function;

[0047] S202: The SSA algorithm is improved by introducing a logarithmic spiral strategy. A logarithmic spiral search model is introduced to solve the problem. The mathematical model is described as follows:

[0048]

[0049] l = 2(1 - 1 / iter) max )-1

[0050] In the formula: a is a constant that determines the shape of the helix, with a value of 1; l is a parameter that decreases linearly from 1 to -1; X pbest To determine the optimal position for the current iteration, each generation of individuals gradually approaches the position in a spiral motion when updating their position, and the position update formula is adjusted as follows:

[0051]

[0052] Among them, X i，j It represents the position information of the i-th sparrow in the j-th dimension; t represents the current iteration number; iter max R is a constant representing the maximum number of iterations; R2 (R2∈(0,1]) and ST (ST∈[0.5,1]) represent the warning value and the safety value, respectively; Q is a random number that follows a normal distribution; L represents a matrix in which each element is 1; R3 is a random number that is uniformly distributed from 0 to 1; and p is a constant with a value of 0.5.

[0053] S203: An adaptive step-size strategy is introduced to improve the SSA algorithm, and an adaptive step-size update formula is proposed:

[0054]

[0055] K = (2rand-1) × (1-t / iter) max ) 1 / 2

[0056]

[0057] In the formula, X represents the average fitness of the dominant population, SEL represents the proportion of the dominant population, and X is 0.35. best Represents the current global optimal position; β represents the step size control parameter, which follows a normal distribution with a mean of 0 and a variance of 1; K∈[-1,1] is a random number; represents a matrix, where each element is randomly assigned a value of 1 or -1; f i f is the fitness value of the current individual sparrow;g and f w This represents the current best and worst fitness values ​​globally; ε is a constant to avoid zero in the denominator.

[0058] Furthermore, the Informer network model comprises an encoder and a decoder, and predicts the remaining lifespan of the energy storage battery by analyzing the trend of battery cycle capacity. Specifically, it includes the following steps:

[0059] S301: The reconstructed battery capacity data is used as input to the Informer network model. Location encoding (local timestamp and global timestamp) is used to mark the temporal relationship between the local and global data, so as to fully explore the temporal correlation of the data.

[0060] S302: The multi-head attention mechanism of the Informer network model is used to focus attention on data features with more obvious degradation trends and obtain long-term dependencies of the data. The decoder input consists of two parts: one part is the implicit intermediate feature data about the capacity data output by the encoder, and the other part requires the capacity data to be predicted to be placed with 0 when inputting, and a masking mechanism is added to prevent each position from focusing on the capacity information of future time points.

[0061] S303: Connects the data to a multi-head attention mechanism, which in turn connects to a fully connected layer to output the energy storage battery capacity data, thereby enabling prediction of the remaining lifespan of the energy storage battery.

[0062] Furthermore, step S4 also includes the step of:

[0063] The performance of the prediction model was validated using mean absolute percentage error (MAPE), root mean square error (RMSE), and mean absolute error (MAE).

[0064] Furthermore, the mean absolute percentage error (MAPE) is calculated as follows:

[0065]

[0066] The root mean square error (RMSE) is calculated as follows:

[0067]

[0068] The mean absolute error (MAE) is calculated as follows:

[0069]

[0070] Where n is the number of samples, y pre This represents the predicted battery capacity; y ture This represents the actual battery capacity. This represents the average battery capacity.

[0071] Furthermore, step S4 also includes the step of:

[0072] The model's universality was verified by comparing it with NASA and Maryland datasets.

[0073] This application also provides a device for predicting the remaining lifespan of an energy storage battery, specifically including:

[0074] The information extraction module is used to preprocess the battery capacity data of the energy storage battery that has missing, abnormal and duplicate data, and divide the extracted data into training set and test set; the one-dimensional time series of battery capacity data is transformed into two-dimensional chaotic sequence by using an improved phase space reconstruction method based on chaos theory, and the effective nonlinear information in the two-dimensional chaotic sequence is learned and extracted.

[0075] The parameter adjustment module is used to introduce chaotic mapping to initialize and adjust the relevant parameters of the SSA algorithm. Secondly, in order to improve the diversity of the population and enhance the search of the surrounding space, a logarithmic spiral strategy is introduced to improve the SSA algorithm. Finally, an adaptive step size strategy is introduced to improve the SSA algorithm, better control the algorithm's development and exploration capabilities, and enhance the algorithm's convergence speed.

[0076] The model training module is used to optimize key parameters in the informer network model based on the MSSA algorithm. The training set of reconstructed battery capacity data is input into the informer network model to train the prediction model. The MSSA algorithm finds the optimal solution for encoder sequence dimension, decoder sequence dimension, number of heads, number of encoder layers, number of decoder layers, learning rate, and batch size in the informer network model, initializes the informer network model parameters, constructs the informer network model, and outputs the predicted battery capacity value.

[0077] The model testing module is used to input the test set of reconstructed battery capacity data into the trained prediction model to obtain the predicted value of battery capacity.

[0078] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the energy storage battery remaining life prediction method.

[0079] This application also provides a storage medium including a stored program that, when the program is executed, controls the device where the storage medium is located to perform the steps of the energy storage battery remaining life prediction method.

[0080] Compared with the prior art, this application has the following advantages:

[0081] This application employs an improved phase space reconstruction method based on chaos theory to overcome the severe fluctuations in energy storage battery lifetime prediction, effectively extracting characteristic information from the battery data. The reconstructed battery capacity data is divided into training and testing sets. The training set is input into the MSSA-informer neural network model, enabling long-term series prediction with low complexity. Finally, the testing set is used to verify the accuracy of the lifetime prediction model, effectively improving the accuracy of remaining lifetime prediction for energy storage batteries.

[0082] In addition to the purposes, features, and advantages described above, this application has other purposes, features, and advantages. The application will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0083] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0084] Figure 1 A schematic flowchart of a method for predicting the remaining life of an energy storage battery, provided as a preferred embodiment of this application;

[0085] Figure 2 This is a schematic diagram of the Informer network model structure of the present invention;

[0086] Figure 3 This is a schematic diagram of the NASA charge-discharge cycle capacity of the present invention;

[0087] Figure 4 This is a schematic diagram of a battery remaining life prediction device module according to a preferred embodiment of this application;

[0088] Figure 5 This is a schematic block diagram of an electronic device according to a preferred embodiment of this application;

[0089] Figure 6 This is an internal structural diagram of a computer device according to a preferred embodiment of this application. Detailed Implementation

[0090] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0091] Reference Figure 1 A preferred embodiment of this application provides a method for predicting the remaining lifespan of an energy storage battery, specifically including the following steps:

[0092] S1: Preprocess the battery capacity data of the energy storage battery that has missing, abnormal and duplicate data, and divide the extracted data into training set and test set; use the improved phase space reconstruction method based on chaos theory to transform the one-dimensional time series of battery capacity data into a two-dimensional chaotic sequence, and learn and extract the effective nonlinear information in the two-dimensional chaotic sequence.

[0093] S2: Chaotic mapping is introduced to initialize and adjust the relevant parameters of the SSA algorithm. Secondly, in order to improve the diversity of the population and enhance the search of the surrounding space, a logarithmic spiral strategy is introduced to improve the SSA algorithm. Finally, an adaptive step size strategy is introduced to improve the SSA algorithm, better control the development and exploration capabilities of the algorithm, and enhance the convergence speed of the algorithm.

[0094] S3: Based on the MSSA algorithm, the key parameters in the informer network model are optimized. The training set of reconstructed battery capacity data is input into the informer network model to train the prediction model. The MSSA algorithm finds the optimal solution for encoder sequence dimension, decoder sequence dimension, number of heads, number of encoder layers, number of decoder layers, learning rate, and batch size in the informer model, initializes the informer model parameters, constructs the informer network model, and outputs the predicted battery capacity value.

[0095] S4: Input the test set of the reconstructed battery capacity data into the trained prediction model to obtain the predicted value of the battery capacity.

[0096] Preferably, in step S1, an improved phase space reconstruction method based on chaos theory is used to transform the one-dimensional time series of battery capacity data into a two-dimensional chaotic sequence, and to learn and extract the effective nonlinear information in the two-dimensional chaotic sequence. The specific steps are as follows:

[0097] S101. Setting a univariate chaotic time series:

[0098] s={s i :i=1,2,…,N}

[0099] In the formula: N is the time length; s i Let be the sequence value at time i;

[0100] Assuming a univariate chaotic time series, the coordinate reconstruction method requires two important parameters: the delay time τ and the embedding dimension d. The sequence values ​​from every τ time steps after time i are inserted into the current vector until a d-dimensional vector is formed. This vector represents the state vector x corresponding to the i-th time step in the reconstructed and improved phase space. i :

[0101] x i =(s i ,s i+τ ,s i+2τ ,…,s i+(d-1)τ ) T i = 1, 2, ..., m

[0102] In the formula: m = N - (d-1)τ, m is the total number of states in the reconstructed and improved phase space, and finally the reconstructed and improved phase space is obtained:

[0103] X = (x1, x2, ..., x m )

[0104] S102. Select the delay time τ using the mutual information method:

[0105]

[0106] Wherein, P(x n ), P(x n+τ Let I(τ) be the probability, and τ be the delay time interval, where I(τ) is the first local minimum. Based on the characteristic that mutual information can effectively solve nonlinear systems but cannot easily and efficiently solve linear time series, an improved phase space reconstruction method is proposed:

[0107]

[0108] When the system has a linear time relationship, choose R. m The first zero of (τ) is the time delay τ; when the system has a nonlinear time relationship, choose... and The minimum value is the time delay τ, where The factor enables the improved phase space method to be extended to higher dimensions, and the method can solve linear and nonlinear time-series systems.

[0109] For chaotic time series, the existing autocorrelation method defines the autocorrelation function as follows:

[0110]

[0111] When the value of the autocorrelation function R(τ) decreases to (1-1 / e) times the initial value R(0), the corresponding τ is a suitable delay time value. However, the choice of a decrease coefficient of (1-1 / e) times in the autocorrelation method is somewhat subjective and needs to be adjusted according to specific circumstances. Furthermore, it cannot be extended to higher dimensions for research. Additionally, the autocorrelation method can only extract the linear correlation of time series. After summarizing the shortcomings of the autocorrelation method, this embodiment proposes a method for judging the nonlinear relationship of a system: the mutual information method. The mutual information method can effectively solve systems with nonlinear relationships, but it cannot easily and efficiently solve time series with linear relationships. Therefore, this embodiment proposes the improved phase space reconstruction method mentioned above, thereby easily and efficiently solving the problem of time series with linear relationships.

[0112] S103. The improved Cao method is used to select the embedding dimension d. The Cao method is an improvement over the spurious nearest neighbor method. Its advantage is that it only requires the parameter delay time τ. In the reconstructed improved phase space, the Euclidean distance between every two pairs of state points is calculated, i.e.:

[0113] R d,i =|x i -x j |

[0114] x i With x j After extending to one dimension, the Euclidean distance is calculated again:

[0115] R' d,i =|x i '-x' j |

[0116] However, the Cao method is easily affected by the dimensions of the sequence, and the correlation between variables can affect the calculation results. Therefore, this embodiment improves the Cao method and proposes an improved Cao method:

[0117] The distance between every two pairs of state points, i.e.:

[0118]

[0119] Where, μ i Let be the expected value of the i-th element, and S be the covariance matrix;

[0120] x i and After extending it to one dimension, calculate the distance again:

[0121]

[0122] In the improved Cao's method, D is used directly. M (x), D M The ratio of (x)' to i defines the index:

[0123]

[0124] The distance metric chooses the infinite norm, defined as follows:

[0125]

[0126] In the formula: E(d) represents the overall spurious nearest neighbor situation, and how E(d) changes as the embedding dimension changes from d to d+1. Defined as follows:

[0127]

[0128] If the time series is chaotic, then E1(d) will tend to stabilize after d is greater than a certain value, so it is more appropriate to use this certain value as the embedding dimension d.

[0129] In this application, the SSA algorithm (Sparrow Search Algorithm) is described as follows:

[0130] The foraging process of sparrows follows a discoverer-follower model, with a superimposed detection and early warning mechanism. Highly adapted individuals act as producers, while others are followers. Simultaneously, a certain proportion of individuals in the population are selected for detection and early warning. If danger is detected, these individuals will fly away to find a new location.

[0131] Producers, followers, and guardians update their positions according to their respective rules, which are as follows:

[0132]

[0133] Among them, X ij It represents the position information of the i-th sparrow in the j-th dimension; t represents the current iteration number; iter max α is a constant representing the maximum number of iterations; α∈(0,1] is a random number; R2(R2∈(0,1]) and ST(ST∈[0.5,1]) represent the warning value and the safety value, respectively; Q is a random number that follows a normal distribution; L represents a matrix in which every element is 1.

[0134] The remaining sparrows act as followers, and their positions are updated using the following formula:

[0135]

[0136] Among them, X P Indicates the current optimal position occupied by the discoverer; X worstA represents the worst position in the global current order; A represents a matrix, with each element randomly assigned a value of 1 or -1; when i ≥ n / 2, the i-th participant with the lowest fitness value has not obtained food and is in a state of extreme hunger. At this time, it needs to fly to other places to find food in order to obtain more energy.

[0137] Randomly select a portion of sparrows from the population as watchdogs, and update the watchdog positions using the following formula:

[0138]

[0139] Among them, X best represents the current global optimal position; β represents the step size control parameter, which follows a normal distribution with a mean of 0 and a variance of 1; K∈[-1,1] is a random number; represents a matrix, with each element randomly assigned a value of 1 or -1; when i≥n / 2, the i-th joiner with a lower fitness value has not obtained food and is in a state of extreme hunger, at which point it needs to fly to other places to forage for food in order to obtain more energy. i f is the fitness value of the current individual sparrow; g and f w This represents the current best and worst fitness values ​​globally; ε is a constant to avoid zero in the denominator.

[0140] Preferably, step S2 specifically includes the following steps:

[0141] S201: A chaotic mapping is introduced to initialize and adjust the relevant parameters of the SSA algorithm. By leveraging the characteristics of chaos, variables are mapped to the value range of a chaotic variable space. Finally, the solution is linearly transformed into the optimization variable space. Specifically, the Circle chaotic mapping is used when generating the initial population, and its definition is:

[0142]

[0143] Where: mod is the modulo function;

[0144] S202: An improvement to the SSA algorithm is introduced using a logarithmic spiral strategy. Experiments show that the original SSA algorithm is prone to getting trapped in local optima, leading to premature convergence. While the discoverer approaches the individual optimal solution line with each iteration, demonstrating strong discovery capabilities, it loses exploration of the surrounding search space in the process of approaching the optimal individual, resulting in reduced population diversity and a tendency to get trapped in local optima. Therefore, we introduce a logarithmic spiral search model to address this problem. The mathematical model for this solution is described below:

[0145]

[0146] l = 2(1 - 1 / iter) max )-1

[0147] In the formula: a is a constant that determines the shape of the helix, with a value of 1; l is a parameter that decreases linearly from 1 to -1; X pbest To determine the optimal position for the current iteration, each generation of individuals gradually approaches the position in a spiral motion when updating their position, and the position update formula is adjusted as follows:

[0148]

[0149] Among them, X i，j It represents the position information of the i-th sparrow in the j-th dimension; t represents the current iteration number; iter max R is a constant representing the maximum number of iterations; R2 (R2∈(0,1]) and ST (ST∈[0.5,1]) represent the warning value and the safety value, respectively; Q is a random number that follows a normal distribution; L represents a matrix in which each element is 1; R3 is a random number that is uniformly distributed from 0 to 1; p is a constant with a value of 0.5. In this embodiment, when updating the position, each generation of individuals gradually approaches in a spiral shape, which increases the search of the surrounding space, maintains the diversity of the population, and enhances the exploration ability of the algorithm.

[0150] S203: In SSA, two strategies are used for updating the guardian's position. For individuals with poor fitness, a Gaussian distribution is used to generate the step size. The Gaussian distribution has a higher probability of generating smaller step sizes, which is beneficial for the algorithm's global search. For individuals with good fitness, a random step size strategy is used, but there is still a high probability of large step sizes in later iterations, which is not conducive to algorithm convergence. Based on the above analysis, in order to balance the algorithm's development and exploration capabilities and enhance the algorithm's convergence speed, this embodiment introduces an adaptive step size strategy to improve the SSA algorithm, proposing the following adaptive step size update formula:

[0151]

[0152] K = (2rand-1) × (1-t / iter) max ) 1 / 2

[0153]

[0154] In the formula, X represents the average fitness of the dominant population, SEL represents the proportion of the dominant population, and X is 0.35. best Represents the current global optimal position; β represents the step size control parameter, which follows a normal distribution with a mean of 0 and a variance of 1; K∈[-1,1] is a random number; represents a matrix, where each element is randomly assigned a value of 1 or -1; f i f is the fitness value of the current individual sparrow; gand f w This represents the current best and worst fitness values ​​globally; ε is a constant to avoid zero in the denominator.

[0155] Preferably, such as Figure 2 As shown, the Informer network model consists of an encoder and a decoder, and predicts the remaining lifespan of the energy storage battery by analyzing the trend of battery cycle capacity. Specifically, it includes the following steps:

[0156] S301: The reconstructed battery capacity data is used as input to the Informer network model. Location encoding (local timestamp and global timestamp) is used to mark the temporal relationship between the local and global data, so as to fully explore the temporal correlation of the data.

[0157] S302: The multi-head attention mechanism of the Informer network model is used to focus attention on data features with more obvious degradation trends and obtain long-term dependencies of the data. The decoder input consists of two parts: one part is the implicit intermediate feature data about the capacity data output by the encoder, and the other part requires the capacity data to be predicted to be placed with 0 when inputting, and a masking mechanism is added to prevent each position from focusing on the capacity information of future time points.

[0158] S303: Connects the data to a multi-head attention mechanism, which in turn connects to a fully connected layer to output the energy storage battery capacity data, thereby enabling prediction of the remaining lifespan of the energy storage battery.

[0159] Preferably, step S4 further includes the step:

[0160] The performance of the prediction model was validated using mean absolute percentage error (MAPE), root mean square error (RMSE), and mean absolute error (MAE).

[0161] Furthermore, the mean absolute percentage error (MAPE) is calculated as follows:

[0162]

[0163] The root mean square error (RMSE) is calculated as follows:

[0164]

[0165] The mean absolute error (MAE) is calculated as follows:

[0166]

[0167] Where n is the number of samples, y pre This represents the predicted battery capacity; y ture This represents the actual battery capacity. This represents the average battery capacity.

[0168] Preferably, step S4 further includes the step:

[0169] The model's universality was verified by comparing it with NASA and Maryland datasets.

[0170] The following explanation uses the capacity of a No. 5 battery as an example to further illustrate this application:

[0171] This embodiment is based on energy storage battery data provided by NASA (see...). Figure 3 This data was obtained through charge-discharge experiments on the batteries. The charging experiment involved charging at a constant current (CC) of 1.5A until the battery voltage reached 4.2V, then continuing charging at a constant voltage (CV) until the charging current dropped to 20mA. The discharging experiment involved discharging at a constant current (CC) of 2A until the voltages of batteries 5, 6, 7, and 18 dropped to 2.7V, 2.5V, 2.2V, and 2.5V, respectively. The experiment was terminated when the battery reached the end-of-life (EOL) standard—a 20% decrease in rated capacity (from 2Ahr to 1.4Ahr). The experiment recorded data on battery voltage, power, impedance, and temperature, which can be used for battery life prediction.

[0172] Taking the capacity of a No. 5 battery as an example, the energy storage battery capacity data is extracted and divided into training and testing sets. The training and testing sets are divided into training and testing sets in a 7:3 ratio. An improved phase space reconstruction method based on chaos theory is used to transform the one-dimensional time series of battery capacity data into a two-dimensional chaotic sequence. The effective nonlinear information in the chaotic sequence is learned and extracted to reconstruct the battery capacity data.

[0173] Chaotic mapping is introduced to initialize the main parameters of SSA. The initial sparrow population size is 10, the number of iterations is 10, the proportion of sparrows among producers is 0.2, and the optimization dimension is 4. A logarithmic spiral strategy is introduced to improve the sparrow search mechanism. MSSA is used to optimize key parameters in the Informer network model. The training set of reconstructed battery capacity data is input into the neural network to train the prediction model. After initializing the parameters using the MSSA algorithm, the optimal solutions for encoder sequence dimension, decoder sequence dimension, number of heads, number of encoder layers, number of decoder layers, learning rate, and batch size in the Informer network model are found to be 30, 20, 8, 1, 1, 0.001, and 10, respectively.

[0174] Initialize the Informer model algorithm parameters, using the features of the battery capacity at the previous time step and the loop count features of the prediction period as input. The input sequence length (seq_len) and the prior sequence length (label_len) are 24 and 12, respectively, and the prediction sequence length is 6. Output the predicted battery capacity value.

[0175] The reconstructed battery capacity data test set was input into the trained prediction model to obtain the model's performance evaluation metrics: Mean Absolute Percentage Error (MAPE) of 0.167865, Root Mean Square Error (RMSE) of 0.129616, and Mean Absolute Error (MAE) of 0.120595. The above steps were repeated to predict remaining battery life using the Maryland dataset, and the differences between the two datasets were compared to verify the algorithm's generalizability.

[0176] This invention addresses the problem of predicting the lifespan of energy storage batteries. First, it performs an improved phase space reconstruction of the battery capacity. Then, it employs a sparrow search algorithm that integrates multiple schemes to optimize the parameters of the Informer network model. The reconstructed data is then input into the optimized Informer network model. The resulting prediction model is simple to implement, easy to implement, and has high accuracy.

[0177] like Figure 4 As shown, this application also provides a device for predicting the remaining lifespan of an energy storage battery, specifically including:

[0178] The information extraction module is used to preprocess the battery capacity data of the energy storage battery that has missing, abnormal and duplicate data, and divide the extracted data into training set and test set; the one-dimensional time series of battery capacity data is transformed into two-dimensional chaotic sequence by using an improved phase space reconstruction method based on chaos theory, and the effective nonlinear information in the two-dimensional chaotic sequence is learned and extracted.

[0179] The parameter adjustment module is used to introduce chaotic mapping to initialize and adjust the relevant parameters of the SSA algorithm. Secondly, in order to improve the diversity of the population and enhance the search of the surrounding space, a logarithmic spiral strategy is introduced to improve the SSA algorithm. Finally, an adaptive step size strategy is introduced to improve the SSA algorithm, better control the algorithm's development and exploration capabilities, and enhance the algorithm's convergence speed.

[0180] The model training module is used to optimize key parameters in the informer network model based on the MSSA algorithm. The training set of reconstructed battery capacity data is input into the informer network model to train the prediction model. The MSSA algorithm finds the optimal solution for encoder sequence dimension, decoder sequence dimension, number of heads, number of encoder layers, number of decoder layers, learning rate, and batch size in the informer network model, initializes the informer network model parameters, constructs the informer network model, and outputs the predicted battery capacity value.

[0181] The model testing module is used to input the test set of reconstructed battery capacity data into the trained prediction model to obtain the predicted value of battery capacity.

[0182] Each module in the aforementioned energy storage battery remaining life prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0183] like Figure 5 As shown, a preferred embodiment of this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the energy storage battery remaining life prediction method in the above embodiments.

[0184] like Figure 6 As shown, a preferred embodiment of this application also provides a computer device, which may be a terminal or a liveness detection server, and its internal structure diagram may be as follows. Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with other external computer devices via a network connection. When the computer program is executed by the processor, it implements the aforementioned method for predicting the remaining lifespan of the energy storage battery.

[0185] Those skilled in the art will understand that Figure 6The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0186] A preferred embodiment of this application also provides a storage medium, the storage medium including a stored program, which, when the program is executed, controls the device where the storage medium is located to execute the energy storage battery remaining life prediction method in the above embodiments.

[0187] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0188] If the functions described in this embodiment are implemented as software functional units and sold or used as independent products, they can be stored in one or more computing device-readable storage media. Based on this understanding, the parts of this application's embodiments that contribute to the prior art or the technical solutions can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computing device (which may be a personal computer, server, mobile computing device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0189] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting the remaining lifespan of an energy storage battery, characterized in that: The specific steps include: S1: Preprocess the battery capacity data containing missing, abnormal, and duplicate data in the energy storage battery, and divide the extracted data into training and testing sets; use an improved phase space reconstruction method based on chaos theory to transform the one-dimensional time series of battery capacity data into a two-dimensional chaotic sequence, learn and extract the effective nonlinear information in the two-dimensional chaotic sequence, specifically including: The improved phase space reconstruction method is as follows: ； in, x i To reconstruct the improved phase space of the first i The state vector corresponding to the given time; N The duration; τ To delay time, d for Embedding dimension; m=N - (d - 1)τ; , For probability; The value of the autocorrelation function; when the system has a linear time relationship, select... The first zero point is the time delay. When the system has a nonlinear time relationship, select and The minimum value is the time delay. ; Calculate the distance between every two pairs of state points: ； in, μ i It is the first i The expected value of each element. S It is the covariance matrix; Will x i and After extending it to one dimension, calculate the distance again: ； Use directly , The ratio is defined as an indicator: ； The distance metric chooses the infinite norm, defined as follows: ； In the formula: E ( d The value of ) represents the overall spurious nearest neighbor situation when the embedding dimension increases from d Change to d When +1, E ( d The changes in ) are defined as follows: ； If the time series is chaotic, then E 1( d ) will d If the value tends to stabilize after exceeding a certain specific value, then that specific value can be used as the embedding dimension. d ; S2: Chaotic mapping is introduced to initialize and adjust the relevant parameters of the SSA algorithm. Secondly, in order to improve the diversity of the population and enhance the search of the surrounding space, a logarithmic spiral strategy is introduced to improve the SSA algorithm. Finally, an adaptive step size strategy is introduced to improve the SSA algorithm, better control the development and exploration capabilities of the algorithm, and enhance the convergence speed of the algorithm. S3: Based on the MSSA algorithm, the key parameters in the informer network model are optimized. The training set of reconstructed battery capacity data is input into the informer network model to train the prediction model. The MSSA algorithm finds the optimal solution for encoder sequence dimension, decoder sequence dimension, number of heads, number of encoder layers, number of decoder layers, learning rate, and batch size in the informer model, initializes the informer model parameters, constructs the informer network model, and outputs the predicted battery capacity value. S4: Input the test set of the reconstructed battery capacity data into the trained prediction model to obtain the predicted value of the battery capacity.

2. The method for predicting the remaining lifespan of an energy storage battery according to claim 1, characterized in that: Step S2 specifically includes the following steps: S201: A chaotic mapping is introduced to initialize and adjust the relevant parameters of the SSA algorithm. By leveraging the characteristics of chaos, variables are mapped to the value range of a chaotic variable space. Finally, the solution is linearly transformed into the optimization variable space. Specifically, the Circle chaotic mapping is used when generating the initial population, and its definition is: ； Where: mod is the modulo function; S202: The SSA algorithm is improved by introducing a logarithmic spiral strategy. A logarithmic spiral search model is introduced to solve the problem. The mathematical model is described as follows: ； In the formula: a The constant that determines the shape of the spiral is set to 1. l The parameters decrease linearly from 1 to -1. X pbest To determine the optimal position for the current iteration, each generation of individuals gradually approaches the position in a spiral motion when updating their position, and the position update formula is adjusted as follows: ； Among them, among them, X i，j Indicates the first i The sparrow in the first j Location information within the dimension; t Represents the current iteration number; iter max It is a constant representing the maximum number of iterations; R 2 and ST These represent the warning value and the safety value, respectively. R 2∈(0,1], ST ∈[0.5,1]; Q It is a random number that follows a normal distribution; L This represents a matrix where every element is 1. R 3 is a random number uniformly distributed from 0 to 1. p It is a constant, with a value of 0.5; S203: An adaptive step-size strategy is introduced to improve the SSA algorithm, and an adaptive step-size update formula is proposed: ； ； ； In the formula, The average fitness of the dominant population. SEL This represents the proportion of the dominant species, with a value of 0.

35. X best This represents the current globally optimal position; This represents the step size control parameter, which follows a normal distribution with a mean of 0 and a variance of 1. A random number; represents a matrix, where each element is randomly assigned the value 1 or -1. f i This represents the current fitness value of the individual sparrow. f g and f w This represents the current best and worst fitness values ​​globally. It is a constant to avoid zero in the denominator.

3. The method for predicting the remaining lifespan of an energy storage battery according to claim 1, characterized in that, The Informer network model consists of an encoder and a decoder, and predicts the remaining life of the energy storage battery by analyzing the trend of battery cycle capacity. Specifically, it includes the following steps: S301: The reconstructed battery capacity data is used as input to the Informer network model. Location encoding is used to mark the local and global temporal positional relationships of the data, so as to fully explore the temporal correlation of the data. S302: The multi-head attention mechanism of the Informer network model is used to focus attention on data features with more obvious degradation trends and obtain long-term dependencies of the data. The decoder input consists of two parts: one part is the implicit intermediate feature data about the capacity data output by the encoder, and the other part requires the capacity data to be predicted to be placed with 0 when inputting, and a masking mechanism is added to prevent each position from focusing on the capacity information of future time points. S303: Connects the data to a multi-head attention mechanism, which in turn connects to a fully connected layer to output the energy storage battery capacity data, thereby enabling prediction of the remaining lifespan of the energy storage battery.

4. The method for predicting the remaining lifespan of an energy storage battery according to claim 1, characterized in that: Step S4 further includes the following steps: The performance of the prediction model was verified using mean absolute percentage error, root mean square error, and mean absolute error.

5. The method for predicting the remaining lifespan of an energy storage battery according to claim 4, characterized in that: The mean absolute percentage error is calculated as follows: ； The root mean square error is calculated as follows: ； The mean absolute error is calculated as follows: ； in, n For the sample size, y pre This is a predicted value for battery capacity. y ture This represents the actual battery capacity. This represents the average battery capacity.

6. The method for predicting the remaining lifespan of an energy storage battery according to claim 5, characterized in that, Step S4 further includes the following steps: The model's universality was verified by comparing it with NASA and Maryland datasets.

7. A device for predicting the remaining life of an energy storage battery, used to implement the method as described in any one of claims 1 to 6, characterized in that, Specifically, it includes: The information extraction module is used to preprocess the battery capacity data of the energy storage battery that is missing, abnormal and duplicate, and divide the extracted data into training set and test set; An improved phase space reconstruction method based on chaos theory is used to transform the one-dimensional time series of battery capacity data into a two-dimensional chaotic sequence, and to learn and extract the effective nonlinear information in the two-dimensional chaotic sequence. The parameter adjustment module is used to introduce chaotic mapping to initialize and adjust the relevant parameters of the SSA algorithm. Secondly, in order to improve the diversity of the population and enhance the search of the surrounding space, a logarithmic spiral strategy is introduced to improve the SSA algorithm. Finally, an adaptive step size strategy is introduced to improve the SSA algorithm, better control the algorithm's development and exploration capabilities, and enhance the algorithm's convergence speed. The model training module is used to optimize key parameters in the informer network model based on the MSSA algorithm. The training set of reconstructed battery capacity data is input into the informer network model to train the prediction model. The MSSA algorithm finds the optimal solution for encoder sequence dimension, decoder sequence dimension, number of heads, number of encoder layers, number of decoder layers, learning rate, and batch size in the informer network model, initializes the informer network model parameters, constructs the informer network model, and outputs the predicted battery capacity value. The model testing module is used to input the test set of reconstructed battery capacity data into the trained prediction model to obtain the predicted value of battery capacity.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the energy storage battery remaining life prediction method as described in any one of claims 1 to 6.

9. A storage medium comprising a stored program, characterized in that, When the program is running, it controls the device containing the storage medium to perform the steps of the energy storage battery remaining life prediction method as described in any one of claims 1 to 6.

Images