Method for monitoring and predicting service life of intelligent battery

Abstract

The invention discloses a method for monitoring and predicting the service life of an intelligent battery. The method comprises the following steps: (a) collecting historical data of the battery and carrying out data preprocessing on the historical data; (b) selecting the data preprocessed in the step (a) by using a random forest to obtain new characteristic data of the battery charging and discharging process; (c) initializing parameters of the BI-LSTM network model, and performing PSO algorithm optimization on the parameters of the BI-LSTM network model; (d) obtaining an optimal BI-LSTM network model parameter combination and taking the optimal BI-LSTM network model parameter combination as an initial value of the BI-LSTM network model after the maximum number of iterations is reached or a precision condition is met; and (e) training and predicting the BI-LSTM network model by using the parameter combination in the step (d), and predicting by using the trained BI-LSTM network model to obtain the remaining life of the battery. And the accuracy and robustness of the battery remaining service life prediction model are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery life prediction, and in particular relates to a method for intelligent battery life monitoring and prediction. Background Art

[0002] There are two main types of existing intelligent battery life monitoring and prediction methods: model-based methods and data-based methods. Model-based battery life prediction methods mainly include: electrode model, electrolyte model, thermal model and mechanical model, etc.; data-based battery life prediction methods mainly include: statistical model, machine learning model, etc.

[0003] Model-based battery life prediction methods usually use electrochemical models or physical models to describe the battery degradation process to predict the battery life. Such models have high computational costs, uncertain parameters, and are more dependent on experimental data, lacking dynamic response. Data-based battery life prediction methods usually use statistical models such as Bayesian reasoning, Kalman filter (KF) model, or autoregressive moving average (ARMA) model, specifically using past measured capacity values ​​and environmental inputs as input to estimate the current capacity. However, the battery degradation mechanism is complex, real-time prediction is difficult, and interpretation and debugging are difficult. Alternatively, machine learning models such as BP neural network, support vector machine, decision tree, and random forest are used. The generalizability of machine learning models is limited, and BP neural network is prone to fall into local optimality, which limits its accuracy and effectiveness in practical applications.

[0004] The Chinese invention patent with application number 202111165151.9 discloses a control method for optimizing the use and life monitoring of cascade energy storage batteries, including: establishing a physical model of grid battery compensation, obtaining the initial maximum power and initial longest usage time of each battery module, and the server calculates the compensation rate through the compensation rate optimization algorithm based on the estimated load power, actual load power, and power supply power, and sets the power threshold of the battery module to protect and stop discharging the battery module whose remaining power is lower than the power threshold, and judges the life of the battery module by comparing the maximum usage time deviation ratio with the size of the second threshold. This application needs to use the compensation rate optimization algorithm to estimate the load power, the real-time collected power supply end and load end power combined with the historical compensation rate to obtain the compensation rate and then process the compensation operation. The operation steps are complicated and the accuracy needs to be improved. Summary of the invention

[0005] In view of the above-mentioned shortcomings of the prior art, the present invention provides a method for intelligent battery life monitoring and prediction.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for intelligent battery life monitoring and prediction, comprising the following steps:

[0007] (a) Collect historical data of batteries and perform data preprocessing;

[0008] (b) using random forest to select the data preprocessed in step (a) to obtain new feature data of the battery charging and discharging process;

[0009] (c) Initialize the BI-LSTM network model parameters and optimize them using the PSO algorithm;

[0010] (d) After reaching the maximum number of iterations or satisfying the accuracy condition, the optimal BI-LSTM network model parameter combination is obtained and used as the initial value of the BI-LSTM network model;

[0011] (e) Use the parameter combination in step (d) to train and predict the BI-LSTM network model, and use the trained BI-LSTM network model to predict the remaining battery life.

[0012] Optimally, in step (a), the data preprocessing is to perform outlier processing and data standardization on the historical data.

[0013] Optimally, in step (a), the data standardization process is performed by using a normalization formula for calculation:

[0014]

[0015] In the formula, x′ is the original value, x′ max is the maximum value of the data, x′ min is the minimum value of the data.

[0016] Optimally, in step (b),

[0017] First, characteristic data with high correlation with battery charging and discharging are screened out, wherein the characteristic data at least includes voltage and temperature;

[0018] Then, new characteristic data that can intuitively reflect the battery charging and discharging process is reconstructed, and the new characteristic data at least includes pressure difference and temperature difference.

[0019] Optimally, in step (c), the BI-LSTM network model parameters include at least the learning rate and the number of hidden layer units.

[0020] Further, in step (c), the PSO algorithm optimization includes the following steps:

[0021] (c1) population parameter initialization;

[0022] (c2) Particle fitness evaluation and comparison;

[0023] (c3) Calculate and compare the individual optimum and the group optimum;

[0024] (c4) Update the particle's velocity and position;

[0025] (c5) Determine whether the optimization conditions are met: when the optimization conditions are met, proceed to step (d); when the optimization conditions are not met, proceed to step (c2).

[0026] Furthermore, the PSO algorithm optimization is specifically as follows:

[0027] (c1) Define the particle population and initialize the particle speed and position to random values; each particle represents a BI-LSTM network model parameter combination;

[0028] (c2) Train a BI-LSTM model using the current parameters of each particle and evaluate its performance according to the fitness function;

[0029] (c3) Record the best parameter pbest of each particle in history, and select the particle with the highest fitness among all particles as the best parameter gbest of the group;

[0030] (c4) Calculate the average optimal position mbest, and then update the speed and position of the particle; then calculate the fitness value of each particle according to the selected fitness function, compare the current individual optimal value and the current global optimal value with their previous values, and obtain a smaller fitness value, thereby adjusting the individual optimal position and the global optimal position of the particle;

[0031] (c5) judging whether the optimization result of the PSO algorithm reaches the set target or satisfies the iterative loop termination condition according to the fitness value; when the optimization condition is met, proceed to step (d); when the optimization condition is not met, proceed to step (c2)

[0032] Compared with the prior art, the present invention has the following beneficial effects: the method for intelligent battery life monitoring and prediction of the present invention uses a bidirectional long-short term memory neural network (Bi-LSTM) as the basic model for battery life prediction, which can simultaneously consider the context information of the sequence, and enhances the model's expression ability and prediction accuracy compared to the traditional LSTM; the Bi-LSTM network is optimized by combining the particle swarm (PSO) algorithm to improve the global search ability, enhance the generalization ability and improve the accuracy of the prediction results. The proposed battery life prediction model based on PSO-BI-LSTM avoids the complex electrochemical reaction modeling of the battery, and uses the battery external characteristic data to derive the upcoming state trend of the battery system, overcoming the problems of poor dynamic accuracy and universality of the model-based method, and also improves the accuracy and robustness of the battery remaining service life prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 The figure is a flow chart of the method for monitoring and predicting the life of an intelligent battery of the present invention. DETAILED DESCRIPTION

[0035] Explanation of technical terms:

[0036] Remaining Useful Life (RUL): Mainly refers to the remaining service life of a system after it has been running for a period of time. This concept has been applied in many fields to accurately predict the remaining service life of the system, thereby greatly reducing the losses caused by system downtime and improving the operational reliability of the system.

[0037] Battery State of Health (SOH): Describes key indicators such as battery capacity, life and performance, usually expressed as a percentage. The SOH of a battery is a ratio determined by comparing the current performance of the battery with the performance when it left the factory. A new battery has an SOH of 100%, while a completely scrapped battery has an SOH of 0%. As the battery is used and the number of charging cycles increases, the SOH will gradually decrease, resulting in a decrease in battery performance.

[0038] Random Decision Forests (RF): Random forests are composed of many independent decision trees (ranging from dozens to hundreds), similar to a dense forest. It forms a final prediction by aggregating the predictions of all decision trees. The final result is obtained by voting or weighted average calculation of the predictions of all trees.

[0039] Bidirectional Long Short-Term Memory (BI-LSTM): LSTM is a variant of a recurrent neural network (RNN) with a gating mechanism that can better capture dependencies in time series data. Compared with traditional RNNs, LSTM uses forget gates, input gates, and output gates to control the flow of information, thereby avoiding the problems of gradient vanishing and gradient explosion. Bidirectional LSTM (BI-LSTM) combines LSTM in two directions and is able to consider both past and future contextual information.

[0040] Particle Swarm Optimization (PSO): PSO is a random search algorithm based on group optimization. It first randomly assigns different speeds and positions to each particle in the group, then evaluates the pros and cons of the current position through the fitness function, and finally updates the position and speed, and determines the optimal result through several iterations.

[0041] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] like Figure 1 The method for intelligent battery life monitoring and prediction shown includes the following steps:

[0043] (a) collecting historical data of the battery and performing data preprocessing on the data; the data preprocessing is to perform outlier processing and data standardization on the historical data;

[0044] Among them, outlier processing specifically refers to filling missing values ​​and processing extreme values ​​using the quartile method; among them, processing extreme values ​​using the quartile method includes the following steps:

[0045] (1) Calculate quartiles:

[0046] Q1=25th percentile, Q3=75th percentile;

[0047] (2) Calculate the interquartile range (IQR):

[0048] IQR = Q3-Q1;

[0049] (3) Determine the range of outliers

[0050] Lower limit: Q1-1.5×IQR;

[0051] Upper limit: Q3+1.5×IQR;

[0052] If a data point is less than the lower limit or greater than the upper limit, it is considered an outlier;

[0053] (4) Handling extreme values:

[0054] Replace these extreme values ​​with the mean.

[0055] The data standardization process is calculated using the normalization formula:

[0056]

[0057] In the formula, x′ is the original value, x′ max is the maximum value of the data, x′ min is the minimum value of the data.

[0058] (b) using random forest to select the data preprocessed in step (a) to obtain new feature data of the battery charging and discharging process;

[0059] Specifically, the characteristic data with high correlation with battery charging and discharging are first screened out, and the characteristic data at least includes voltage and temperature; and then new characteristic data that can intuitively reflect the battery charging and discharging process are reconstructed, and the new characteristic data at least includes pressure difference and temperature difference.

[0060] (c) Initialize the BI-LSTM network model parameters and optimize them using the PSO algorithm; the BI-LSTM network model parameters at least include the learning rate and the number of hidden layer units.

[0061] The PSO algorithm optimization includes the following steps:

[0062] (c1) population parameter initialization;

[0063] (c2) Particle fitness evaluation and comparison;

[0064] (c3) Calculate and compare the individual optimum and the group optimum;

[0065] (c4) Update the particle's velocity and position;

[0066] (c5) Determine whether the optimization conditions are met: when the optimization conditions are met, proceed to step (d); when the optimization conditions are not met, proceed to step (c2).

[0067] Specifically:

[0068] (c1) Define the particle population, initialize the particle speed and position to random values; each particle represents a BI-LSTM network model parameter combination (i.e., the aforementioned BI-LSTM network model parameter combination);

[0069] (c2) Train a BI-LSTM model using the current parameters of each particle and evaluate its performance according to the fitness function;

[0070] (c3) Record the best parameter pbest of each particle in history, and select the particle with the highest fitness among all particles as the best parameter gbest of the group;

[0071] (c4) Calculate the average optimal position mbest, and then update the speed and position of the particle; then calculate the fitness value of each particle according to the selected fitness function, compare the current individual optimal value and the current global optimal value with their previous values, and obtain a smaller fitness value, thereby adjusting the individual optimal position and the global optimal position of the particle;

[0072] (c5) judging whether the optimization result of the PSO algorithm reaches the set target or satisfies the iterative loop termination condition according to the fitness value; when the optimization condition is met, proceed to step (d); when the optimization condition is not met, proceed to step (c2)

[0073] (d) After reaching the maximum number of iterations or satisfying the accuracy condition, the optimal BI-LSTM network model parameter combination is obtained and used as the initial value of the BI-LSTM network model;

[0074] (e) Use the parameter combination in step (d) to train and predict the BI-LSTM network model, and use the trained BI-LSTM network model to predict the remaining battery life.

[0075] The above method is used to predict the remaining battery life, and the results are shown in Table 1.

[0076] Table 1 Prediction table of different network models

[0077] Prediction Model Training time / s Accuracy / % Accuracy / % False positive rate / % SVR 75.59 93.46 94.17 7.1 LSTM 58.72 96.56 95.32 5.8 BI-LSTM 45.95 97.92 96.87 3.6 This embodiment 40.45 98.83 97.56 2.9

[0078] The method of intelligent battery life monitoring and prediction of the present invention uses a bidirectional long-short term memory neural network (Bi-LSTM) as the basic model for battery life prediction, which can simultaneously consider the context information of the sequence, and enhances the model's expressiveness and prediction accuracy compared to the traditional LSTM; the Bi-LSTM network is optimized in combination with the particle swarm (PSO) algorithm to enhance global search capabilities, enhance generalization capabilities, and improve the accuracy of prediction results. The proposed battery life prediction model based on PSO-BI-LSTM avoids the complex electrochemical reaction modeling of the battery, and uses the battery's external characteristic data to derive the upcoming state trend of the battery system, overcoming the problems of poor dynamic accuracy and universality of the model-based method, while also improving the accuracy and robustness of the battery remaining service life prediction model. Specifically:

[0079] (1) Using the random forest model to analyze the correlation between different factors and the remaining battery life can more accurately describe the behavior of the battery, thereby improving the model performance;

[0080] (2) BI-LSTM is used as the basic model for battery remaining life prediction, which can simultaneously consider the context information of the sequence. Compared with the traditional LSTM, it enhances the model's expressiveness and prediction accuracy.

[0081] (3) The particle swarm optimization (PSO) algorithm is used to further optimize the initial parameters of the BI-LSTM model, so that the algorithm can iterate globally and reduce the probability of falling into the local optimum; it can also effectively reduce the number of iterations of the algorithm and speed up the convergence of the algorithm.

[0082] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent battery life monitoring and prediction, characterized in that: The following steps are involved: (a) Collect historical data of batteries and perform data preprocessing; (b) using random forest to select the data preprocessed in step (a) to obtain new feature data of the battery charging and discharging process; (c) Initialize the BI-LSTM network model parameters and optimize them using the PSO algorithm; (d) After reaching the maximum number of iterations or satisfying the accuracy condition, the optimal BI-LSTM network model parameter combination is obtained and used as the initial value of the BI-LSTM network model; (e) Use the parameter combination in step (d) to train and predict the BI-LSTM network model, and use the trained BI-LSTM network model to predict the remaining battery life.

2. The method for monitoring and predicting the life of an intelligent battery according to claim 1, characterized in that: In step (a), the data preprocessing is to perform outlier processing and data standardization on the historical data.

3. The method for monitoring and predicting the life of an intelligent battery according to claim 1, characterized in that: In step (a), The data standardization process is to use the normalization formula for calculation: In the formula, x′ is the original value, x′ max is the maximum value of the data, x′ min is the minimum value of the data.

4. The method for monitoring and predicting the life of an intelligent battery according to claim 1, characterized in that: In step (b), First, characteristic data with high correlation with battery charging and discharging are screened out, wherein the characteristic data at least includes voltage and temperature; Then, new characteristic data that can intuitively reflect the battery charging and discharging process are reconstructed, and the new characteristic data at least includes pressure difference and temperature difference.

5. The method for monitoring and predicting the life of an intelligent battery according to claim 1, characterized in that: In step (c), the BI-LSTM network model parameters include at least the learning rate and the number of hidden layer units.

6. The method for monitoring and predicting the life of an intelligent battery according to claim 1 or 5, characterized in that: In step (c), the PSO algorithm optimization includes the following steps: (c1) population parameter initialization; (c2) Particle fitness evaluation and comparison; (c3) Calculate and compare the individual optimum and the group optimum; (c4) Update the particle's velocity and position; (c5) Determine whether the optimization conditions are met: when the optimization conditions are met, proceed to step (d); when the optimization conditions are not met, proceed to step (c2).

7. The method for monitoring and predicting the life of an intelligent battery according to claim 6, characterized in that: The PSO algorithm optimization is specifically as follows: (c1) Define the particle population and initialize the particle speed and position to random values; each particle represents a BI-LSTM network model parameter combination; (c2) Train a BI-LSTM model using the current parameters of each particle and evaluate its performance according to the fitness function; (c3) Record the best parameter pbest of each particle in history, and select the particle with the highest fitness among all particles as the best parameter gbest of the group; (c4) Calculate the average optimal position mbest, and then update the speed and position of the particle; then calculate the fitness value of each particle according to the selected fitness function, compare the current individual optimal value and the current global optimal value with their previous values, and obtain a smaller fitness value, thereby adjusting the individual optimal position and the global optimal position of the particle; (c5) Determine whether the optimization result of the PSO algorithm reaches the set target or satisfies the iterative loop termination condition based on the fitness value; when the optimization condition is met, proceed to step (d); when the optimization condition is not met, proceed to step (c2).

Images