Lithium ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization

Abstract

The invention discloses a lithium ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization, and relates to the field of lithium ion batteries, and the method comprises the steps: obtaining the historical data of the operation of a plurality of battery packs of the same specification, and processing the historical data; constructing a regression equation according to historical data; according to the regression equation, obtaining a regression coefficient in combination with historical data; acquiring a loss function according to the regression equation and the regression coefficient; updating the regression coefficient according to the loss function; acquiring current data of the plurality of to-be-predicted battery packs; according to the multiple groups of current data, the regression coefficient and the updated regression equation, the predicted capacity of the battery pack to be predicted is obtained; according to the method, the current data of the to-be-predicted battery pack can be obtained in real time, the predicted capacity is rapidly calculated in combination with the updated regression equation, the method is suitable for lithium ion battery pack SOH prediction in various actual scenes, and errors of prediction results are reduced.

Description

Technical Field

[0001] The present invention relates to the field of lithium-ion batteries, and in particular to a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization. Background Art

[0002] Lithium batteries are a type of battery that uses lithium metal or lithium alloy as positive / negative electrode materials and non-aqueous electrolyte solutions. Due to the highly active chemical properties of lithium metal, the processing, storage, and use of lithium metal have very high environmental requirements. With the development of science and technology, lithium batteries have become mainstream;

[0003] The SOH (State of Health) of a lithium-ion battery pack refers to the health status of the battery pack, which indicates the ratio of the performance parameters of the battery pack after a period of use to the nominal parameters.

[0004] Traditional battery SOH prediction methods often rely on a single parameter or a simple model, such as evaluating the health status of a battery only by the number of cycles, internal resistance change or capacity decay. However, these methods ignore the complexity and diversity of the battery in actual use, such as the impact of factors such as charge and discharge rate, temperature fluctuation, and use environment on battery performance. Therefore, traditional prediction methods often fail to accurately reflect the true health status of the battery, resulting in large errors in the prediction results. Therefore, the present invention proposes a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization. Summary of the invention

[0005] In order to solve the above technical problems, a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization is provided to solve the problem of large errors in the above prediction results.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0007] A lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization, comprising:

[0008] S100, obtaining historical operation data of multiple battery packs with the same specifications, and processing the historical data;

[0009] S200, constructing a regression equation based on historical data;

[0010] S300, obtaining a regression coefficient based on the regression equation and combining historical data;

[0011] S400, obtaining a loss function according to the regression equation and the regression coefficient;

[0012] S500, updating the regression coefficient according to the loss function;

[0013] S600, obtaining current data of multiple groups of battery groups to be predicted;

[0014] S700, obtaining a predicted capacity of the battery group to be predicted based on multiple groups of current data, regression coefficients and an updated regression equation;

[0015] S800: Obtain the current temperature of the battery pack to be predicted, and correct the predicted capacity according to the temperature to obtain the actual capacity.

[0016] Preferably, the step of obtaining the historical operation data of multiple battery packs of the same specification and processing the historical data comprises the following steps:

[0017] S101, obtaining historical operation data of multiple battery packs with the same specifications;

[0018] S102, obtaining abnormal data in historical records and marking them;

[0019] S103, obtaining the position of abnormal data during the operation period of the battery pack;

[0020] S104, performing mean processing on abnormal data based on data of battery packs with the same specifications and in the same period;

[0021] S105, filling abnormal data with mean value;

[0022] The mean calculation formula is:

[0023]

[0024] In the formula, is the mean, S q The data are for battery packs of the same specification in the same period, and L is the number of battery packs of the same specification.

[0025] According to the claim, a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization is characterized in that: the construction of a regression equation based on historical data includes the following steps:

[0026] S201, calculating the remaining capacity of the corresponding battery pack based on historical data;

[0027] S202, extracting parameters of the charge and discharge rate, current voltage and internal resistance of the lithium battery pack after each charging in the historical data;

[0028] S203, constructing a multiple regression equation based on the parameters of remaining capacity, charge and discharge rate, current voltage and internal resistance;

[0029] Among them, the specific formula of the multiple regression equation is:

[0030] Y j =β0+β1C j +β2U j +β3R j +∈ k ;

[0031] In the formula, β0, β1, β2 and β3 are regression coefficients, C j is the charge and discharge rate, U j is the current voltage, R j is the internal resistance, ∈ k is the error term, Y j is the remaining capacity.

[0032] Preferably, the step of calculating the remaining capacity of the corresponding battery pack based on historical data comprises the following steps:

[0033] S2011, extracting the total capacity of the battery pack from the historical data;

[0034] S2012, obtaining the start time and end time of battery pack charging;

[0035] S2013, obtaining a battery pack discharge equation;

[0036] S2014, obtaining the corresponding remaining capacity according to the discharge equation, the start time and the end time;

[0037] The remaining capacity calculation formula is:

[0038]

[0039] In the formula, Q r is the remaining capacity, Q al is the rated capacity at the factory, f(a) is the discharge equation of the battery pack, t0 is the start time, and t1 is the end time.

[0040] Preferably, the step of obtaining the regression coefficient based on the regression equation and combining historical data comprises the following steps:

[0041] S301, extracting parameters of charge and discharge rates, current voltages, internal resistances, and corresponding remaining capacities of multiple battery packs in historical data;

[0042] S302, obtaining an equation of an error term, and standardizing the error term;

[0043] S303, constructing a matrix according to the multiple regression equation;

[0044] S304, obtaining regression coefficients according to the matrix;

[0045] The matrix is ​​in the form of:

[0046]

[0047] Among them, the specific calculation formula of the regression coefficient is:

[0048]

[0049] Where Y j×1 is the matrix of remaining capacity, X j(j+1) is the matrix of dependent variables, β (j+1)×1 is the matrix of regression coefficients, e j×1 is the matrix of standard error terms, Q(β (i+1)×1 ) is the calculated regression coefficient.

[0050] Preferably, the step of obtaining the loss function based on the regression equation and the regression coefficient comprises the following steps:

[0051] S401, extracting the number of battery packs;

[0052] S402, extracting the remaining capacity of the corresponding battery pack;

[0053] S403, obtaining a loss function according to the regression equation and the regression coefficient;

[0054] The loss function is calculated as follows:

[0055]

[0056] Where J(β) is the corresponding loss function and m is the number of battery packs.

[0057] Preferably, updating the regression coefficient according to the loss function comprises the following steps:

[0058] S501, extracting regression coefficients from known information;

[0059] S502, obtaining the partial derivative of the loss function;

[0060] S503, assigning values ​​to partial derivatives;

[0061] S504, updating the regression coefficient according to the assignment, the regression coefficient and the partial derivative;

[0062] Among them, the calculation formula of the updated regression coefficient is:

[0063]

[0064] Where β' e is the updated regression coefficient, β e is the regression coefficient to be updated, α is the assignment, g i β eThe corresponding independent variable.

[0065] Preferably, obtaining the predicted capacity of the battery group to be predicted based on multiple groups of current data, regression coefficients and updated regression equations includes the following steps:

[0066] S701, extracting the loss function and the updated regression equation;

[0067] S702, determining whether the loss function is a convergence function;

[0068] S703, if the loss function is a convergence function, there is no need to loop S401 to S504;

[0069] S704, if the loss function is not a convergent function, loop S401 to S504 until the loss function is a convergent function;

[0070] S705 , bringing the current data of the multiple groups of battery groups to be predicted into the updated regression equation to obtain the predicted capacity of the battery groups to be predicted.

[0071] Preferably, the step of obtaining the current temperature of the battery pack to be predicted, correcting the predicted capacity according to the temperature, and obtaining the actual capacity comprises the following steps:

[0072] S801, obtaining a standard error according to an error function;

[0073] S802, obtaining a prediction interval based on the standard error;

[0074] S803, comparing the predicted capacity with the predicted interval;

[0075] S804. If the predicted capacity belongs to the prediction interval, the prediction is accurate;

[0076] S805. If the predicted capacity does not belong to the prediction interval, recalculate the error function;

[0077] S806, obtaining the relationship between temperature and battery capacity based on historical data;

[0078] S807, based on the relationship and in combination with the current temperature, the predicted capacity is corrected to obtain the actual capacity;

[0079] The calculation formula for the standard error is:

[0080]

[0081] The formula for the prediction interval is:

[0082]

[0083] In the formula, is the standard error; x0 is the sample, is the sample mean, X is the sample matrix, t is the distribution critical value, is the predicted value.

[0084] Compared with the prior art, the advantages of the present invention are: the present invention constructs a multivariate linear regression equation to describe the relationship between the characteristics and the remaining capacity, extracts multiple groups of characteristic values ​​and the corresponding remaining capacity from historical data, constructs a matrix equation, obtains the regression coefficient by calculation, and calculates the difference between the predicted value and the actual value based on the regression equation and the regression coefficient, that is, the loss function. By calculating the loss function and iteratively updating the regression coefficient, the method continuously optimizes the regression equation using reinforcement learning optimization technology, so that the prediction result is closer to the actual value, thereby improving the accuracy and efficiency of the prediction. The method can obtain the current data of the battery pack to be predicted in real time, and quickly calculate the predicted capacity in combination with the updated regression equation. It is suitable for SOH prediction of lithium-ion battery packs in various practical scenarios and reduces the error of the prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 A schematic flow chart of steps S100-S800 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention;

[0086] Figure 2 A schematic flow chart of steps S101-S105 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention;

[0087] Figure 3 A schematic flow chart of steps S201-S203 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention;

[0088] Figure 4 A flowchart of steps S2011-S2014 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention;

[0089] Figure 5 A flowchart of steps S301-S304 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention;

[0090] Figure 6 A schematic flow chart of steps S401-S403 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention;

[0091] Figure 7A flowchart of steps S501-S504 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention;

[0092] Figure 8 A flowchart of steps S701-S701 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention;

[0093] Fig. 9 This is a flow chart of steps S801-S807 in a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization proposed by the present invention. DETAILED DESCRIPTION

[0094] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0095] Reference Figure 1-Figure 9 As shown, a lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization includes:

[0096] S100, obtaining historical operation data of multiple battery packs with the same specifications, and processing the historical data;

[0097] S200, constructing a regression equation based on historical data;

[0098] S300, obtaining a regression coefficient based on the regression equation and combining historical data;

[0099] S400, obtaining a loss function according to the regression equation and the regression coefficient;

[0100] S500, updating the regression coefficient according to the loss function;

[0101] S600, obtaining current data of multiple groups of battery groups to be predicted;

[0102] S700, obtaining a predicted capacity of the battery group to be predicted based on multiple groups of current data, regression coefficients and an updated regression equation;

[0103] S800, obtaining the current temperature of the battery pack to be predicted, and correcting the predicted capacity according to the temperature to obtain the actual capacity;

[0104] Those skilled in the art can understand that, the charge and discharge rate, current voltage and internal resistance are used as independent variables. Since the charge and discharge rate reflects the speed of battery charge and discharge, it has a direct impact on the life and performance of the battery; the current voltage directly reflects the current energy storage state of the battery; the internal resistance is the obstacle to the flow of current inside the battery, and its change often indicates a change in the health state of the battery. Historical operation data is obtained from multiple battery packs with the same specifications, abnormal data is identified and processed, and the data mean of the same period of battery packs with the same specifications is used for filling to reduce the impact of data deviation on the prediction results. The charge and discharge rate, current voltage and internal resistance are selected as the features of the regression equation, and a multivariate regression equation is established to describe the relationship between the features and the remaining capacity. Multiple groups of eigenvalues ​​and the corresponding remaining capacity are extracted from the historical data, and a matrix equation is constructed. The regression coefficient is obtained by calculation. Based on the regression equation and regression coefficient, calculate the difference between the predicted value and the actual value, that is, the loss function, calculate the partial derivative of the loss function, iteratively update the regression coefficient to minimize the loss function, obtain the current data of the battery pack to be predicted, substitute the current data into the updated regression equation, obtain the predicted capacity, calculate the standard error and the prediction interval, judge the accuracy of the predicted capacity, and correct the predicted capacity according to the relationship between temperature and battery capacity in historical data to obtain a more accurate actual capacity. By calculating the loss function and iteratively updating the regression coefficient, this method uses reinforcement learning optimization technology to continuously optimize the regression equation, so that the prediction result is closer to the actual value, which improves the prediction accuracy and efficiency. This method can obtain the current data of the battery pack to be predicted in real time, and quickly calculate the predicted capacity in combination with the updated regression equation. It is suitable for SOH prediction of lithium-ion battery packs in various practical scenarios.

[0105] Obtaining the historical data of multiple battery packs with the same specifications and processing the historical data includes the following steps:

[0106] S101, obtaining historical operation data of multiple battery packs with the same specifications;

[0107] S102, obtaining abnormal data in historical records and marking them;

[0108] S103, obtaining the position of abnormal data during the operation period of the battery pack;

[0109] S104, performing mean processing on abnormal data based on data of battery packs with the same specifications and in the same period;

[0110] S105, filling abnormal data with mean value;

[0111] The mean calculation formula is:

[0112]

[0113] In the formula, is the mean, S q is the data of the same specification battery pack in the same period, L is the number of battery packs of the same specification;

[0114] It will be understood by those skilled in the art that by acquiring historical operating data of multiple battery packs with the same specifications, a rich data sample is provided for subsequent data analysis, model building and prediction. In the data set, abnormal data may have a negative impact on model training, resulting in inaccurate prediction results. For data marked as abnormal, mean processing is performed using data from the same period of battery packs with the same specifications, and the calculated mean is used to fill in abnormal data points, which can make the data set more complete and consistent. The mean interpolation method does not require complex calculations or advanced statistical knowledge, and is simple and fast to operate. The interpolated values ​​will be within a reasonable range of the data, reducing the risk of inserting abnormal values. Since the mean interpolation method does not require complex calculations or iterative processes, the processing speed is faster.

[0115] Constructing a regression equation based on historical data includes the following steps:

[0116] S201, extracting parameters of the charge and discharge rate, current voltage and internal resistance of the lithium battery pack after each charging in the historical data;

[0117] S202, constructing a multiple regression equation based on the parameters of charge and discharge rate, current voltage and internal resistance;

[0118] Among them, the specific formula of the multiple regression equation is:

[0119] Y j =β0+β1C j +β2U j +β3R j +∈ k ;

[0120] In the formula, β0, β1, β2 and β3 are regression coefficients, C j is the charge and discharge rate, U j is the current voltage, R j is the internal resistance, ∈ k is the error term, Y j is the remaining capacity;

[0121] It can be understood by those skilled in the art that the charge and discharge rate, current voltage and internal resistance are important indicators of lithium battery performance, which directly reflect the working state and health of the battery. By extracting these parameters, we can obtain detailed performance data of the lithium battery after each charge, and provide necessary input variables for the subsequent construction of the regression equation. The change pattern of these parameters contains information about battery aging and performance degradation, and is an important basis for predicting battery SOH. The multivariate regression equation can describe the mathematical relationship between these parameters and the remaining capacity (SOH) of the battery. By fitting historical data, we can obtain a set of regression coefficients, which reflect the degree and direction of the influence of each parameter on the battery SOH, which helps us to have a deeper understanding of the mechanism of battery performance degradation. At the same time, the equation can be used as part of the battery management system to monitor and predict the health of the battery in real time.

[0122] According to the regression equation, the regression coefficient is obtained by combining historical data, including the following steps:

[0123] S301, extracting parameters of charge and discharge rates, current voltages, internal resistances, and corresponding remaining capacities of multiple battery packs in historical data;

[0124] S302, obtaining an equation of an error term, and standardizing the error term;

[0125] S303, constructing a matrix according to the multiple regression equation;

[0126] S304, obtaining regression coefficients according to the matrix;

[0127] The matrix is ​​in the form of:

[0128]

[0129] Among them, the specific calculation formula of the regression coefficient is:

[0130]

[0131] Where Y j×1 is the matrix of remaining capacity, X j(j+1) is the matrix of dependent variables, β (j+1)×1 is the matrix of regression coefficients, e j×1 is the matrix of standard error terms, Q(β (i+1)×1 ) is the calculated regression coefficient;

[0132] It is understandable to those skilled in the art that by extracting historical data of multiple groups of battery packs, including parameters such as charge and discharge rate, current voltage, internal resistance, and the remaining battery capacity (SOH) corresponding to these parameters, we provide a rich data set for subsequent regression analysis. In regression analysis, the error term is an important indicator to measure the difference between the model prediction value and the actual value. By obtaining the equation of the error term, we can quantitatively evaluate the prediction accuracy of the model. At the same time, standardizing the error term can eliminate the influence between different dimensions, making the error term more comparable. By constructing a matrix based on the multivariate regression equation, we can organize the independent variables and dependent variables in the data set in the form of a matrix, which is convenient for subsequent calculations and analysis. By calculating based on the matrix, we can solve the specific values ​​of the regression coefficients. These regression coefficients are not only used to construct the final regression model, but also can be used to analyze the contribution of each variable to the remaining battery capacity.

[0133] Obtaining the loss function based on the regression equation and regression coefficient includes the following steps:

[0134] S401, extracting the number of battery packs;

[0135] S402, extracting the remaining capacity of the corresponding battery pack;

[0136] S403, obtaining a loss function according to the regression equation and the regression coefficient;

[0137] The loss function is calculated as follows:

[0138]

[0139] Where J(β) is the corresponding loss function, and m is the number of battery packs;

[0140] It will be understood by those skilled in the art that, by extracting the number of battery packs, we can clearly define the range of battery packs involved in subsequent analysis or prediction. The remaining capacity (SOH) is an important indicator for measuring the health of a battery. By extracting the remaining capacity of the corresponding battery pack, we can obtain the performance data of the battery pack during actual use. In the regression analysis, we can calculate the predicted value of the model based on the constructed regression equation and the solved regression coefficient. Then, we compare these predicted values ​​with the actually observed remaining capacity of the battery pack and construct a loss function based on the degree of difference. The size of the loss function directly reflects the prediction accuracy of the model. By minimizing the loss function, we can optimize the regression model to improve the accuracy and reliability of the prediction.

[0141] Updating the regression coefficients according to the loss function includes the following steps:

[0142] S501, extracting regression coefficients from known information;

[0143] S502, obtaining the partial derivative of the loss function;

[0144] S503, assigning values ​​to partial derivatives;

[0145] S504, updating the regression coefficient according to the assignment, the regression coefficient and the partial derivative;

[0146] Among them, the calculation formula of the updated regression coefficient is:

[0147]

[0148] Where β' e is the updated regression coefficient, β e is the regression coefficient to be updated, α is the assignment, g i β e The corresponding independent variable;

[0149] Those skilled in the art can understand that the loss function is a mathematical expression that measures the difference between the model's predicted value and the actual value, and the partial derivative is the rate of change of the loss function with respect to the regression coefficient. After obtaining the partial derivatives of the loss function, we need to assign values ​​to these partial derivatives for use in subsequent update steps, and the size of the assigned value can control the size of the updated regression coefficient. Based on the assigned value (i.e., the value of the partial derivative), the current regression coefficient and the partial derivative itself, we can use some optimization algorithm (such as the gradient descent method) to update the regression coefficient. The purpose of the update is to reduce the value of the loss function, thereby improving the prediction accuracy of the model.

[0150] According to multiple sets of current data, regression coefficients and updated regression equations, obtaining the predicted capacity of the battery pack to be predicted includes the following steps:

[0151] S701, extracting the loss function and the updated regression equation;

[0152] S702, determining whether the loss function is a convergence function;

[0153] S703, if the loss function is a convergence function, there is no need to loop S401 to S504;

[0154] S704, if the loss function is not a convergent function, loop S401 to S504 until the loss function is a convergent function;

[0155] S705, bringing the current data of the multiple groups of battery groups to be predicted into the updated regression equation to obtain the predicted capacity of the battery groups to be predicted;

[0156] It will be understood by those skilled in the art that, by extracting the current loss function and the updated regression equation, we obtain the current performance status and prediction ability of the model. By judging whether the loss function is a convergence function, we can determine whether the model has reached the optimal state or whether further optimization is needed. If the loss function has converged, it means that the prediction performance of the model has stabilized and the optimization process can be stopped. When the loss function has converged, it means that the prediction performance of the model is good enough and no further optimization is required. When the loss function has not yet converged, it means that the prediction performance of the model still has room for improvement. Steps S401 to S504 are continued to be executed in a loop, by continuously updating the regression coefficient and optimizing the model until the loss function converges. By bringing the current data of multiple groups of battery packs to be predicted into the updated regression equation, we can obtain the predicted capacity of these battery packs.

[0157] Obtaining the current temperature of the battery pack to be predicted, and correcting the predicted capacity according to the temperature, and obtaining the actual capacity includes the following steps:

[0158] S801, obtaining a standard error according to an error function;

[0159] S802, obtaining a prediction interval based on the standard error;

[0160] S803, comparing the predicted capacity with the predicted interval;

[0161] S804. If the predicted capacity belongs to the prediction interval, the prediction is accurate;

[0162] S805: If the predicted capacity does not belong to the prediction interval, recalculate the error function;

[0163] S806, obtaining the relationship between temperature and battery capacity based on historical data;

[0164] S807, based on the relationship and in combination with the current temperature, the predicted capacity is corrected to obtain the actual capacity;

[0165] The calculation formula for the standard error is:

[0166]

[0167] The formula for the prediction interval is:

[0168]

[0169] In the formula, is the standard error; x0 is the sample, is the sample mean, X is the sample matrix, t is the distribution critical value, is the predicted value;

[0170] Those skilled in the art can understand that by calculating the standard error based on the error function, we can quantify the degree of difference between the model prediction value and the actual value. By calculating the prediction interval based on the standard error, we can provide a confidence range for the model's prediction results. By comparing the predicted capacity with the prediction interval, we can determine whether the predicted value falls within the confidence range. When the predicted capacity falls within the prediction interval, we can consider that the model's prediction is accurate. When the predicted capacity does not fall within the prediction interval, it means that there is a large difference between the model's prediction result and the actual situation. By analyzing the relationship between temperature and battery capacity based on historical data, we can understand the influence of temperature on battery performance. Based on the relationship between temperature and battery capacity and combined with the current actual temperature, we can correct the predicted capacity to obtain an actual capacity that is closer to the actual situation.

[0171] In summary, the advantages of the present invention are: by constructing a multivariate linear regression equation to describe the relationship between the characteristics and the remaining capacity, multiple groups of characteristic values ​​and the corresponding remaining capacity are extracted from historical data, a matrix equation is constructed, and the regression coefficient is obtained by calculation. Based on the regression equation and the regression coefficient, the difference between the predicted value and the actual value, that is, the loss function, is calculated. By calculating the loss function and iteratively updating the regression coefficient, the method uses reinforcement learning optimization technology to continuously optimize the regression equation, so that the prediction result is closer to the actual value, thereby improving the accuracy and efficiency of the prediction. The method can obtain the current data of the battery pack to be predicted in real time, and quickly calculate the predicted capacity in combination with the updated regression equation. It is suitable for lithium-ion battery pack SOH prediction in various practical scenarios and reduces the error of the prediction result.

[0172] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A lithium-ion battery pack SOH prediction method based on multi-source data fusion and reinforcement learning optimization, characterized in that: include: S100, obtaining historical operation data of multiple battery packs with the same specifications, and processing the historical data; S200, constructing a regression equation based on historical data; S300, obtaining a regression coefficient based on the regression equation and combining historical data; S400, obtaining a loss function according to the regression equation and the regression coefficient; S500, updating the regression coefficient according to the loss function; S600, obtaining current data of multiple groups of battery groups to be predicted; S700, obtaining a predicted capacity of the battery group to be predicted based on multiple groups of current data, regression coefficients and an updated regression equation; S800: Obtain the current temperature of the battery pack to be predicted, and correct the predicted capacity according to the temperature to obtain the actual capacity.

2. The method for predicting SOH of a lithium-ion battery pack based on multi-source data fusion and reinforcement learning optimization according to claim 1, characterized in that: The method of obtaining the historical operation data of multiple battery packs with the same specifications and processing the historical data includes the following steps: S101, obtaining historical operation data of multiple battery packs with the same specifications; S102, obtaining abnormal data in historical records and marking them; S103, obtaining the position of abnormal data during the operation period of the battery pack; S104, performing mean processing on abnormal data based on data of battery packs with the same specifications and in the same period; S105, filling abnormal data with mean value; The mean calculation formula is: ; In the formula, is the mean, This is the data of the same specification battery pack in the same period. It is the number of battery packs with the same specifications.

3. The method for predicting SOH of a lithium-ion battery pack based on multi-source data fusion and reinforcement learning optimization according to claim 1, characterized in that: The construction of the regression equation based on historical data comprises the following steps: S201, calculating the remaining capacity of the corresponding battery pack based on historical data; S202, extracting parameters of the charge and discharge rate, current voltage and internal resistance of the lithium battery pack after each charging in the historical data; S203, constructing a multiple regression equation based on the parameters of remaining capacity, charge and discharge rate, current voltage and internal resistance; Among them, the specific formula of the multiple regression equation is: ; In the formula, , , and are regression coefficients, is the charge and discharge rate, is the current voltage, is the internal resistance, is the error term, is the remaining capacity.

4. The method for predicting SOH of a lithium-ion battery pack based on multi-source data fusion and reinforcement learning optimization according to claim 3 is characterized in that: Calculating the remaining capacity of the corresponding battery pack according to the historical data includes the following steps: S2011, extracting the total capacity of the battery pack from the historical data; S2012, obtaining the start time and end time of battery pack charging; S2013, obtaining a battery pack discharge equation; S2014, obtaining the corresponding remaining capacity according to the discharge equation, the start time and the end time; The remaining capacity calculation formula is: ; In the formula, is the remaining capacity, is the rated capacity at the factory. is the discharge equation of the battery pack, is the starting time, is the end time.

5. The method for predicting SOH of a lithium-ion battery pack based on multi-source data fusion and reinforcement learning optimization according to claim 1, characterized in that: The step of obtaining the regression coefficient based on the regression equation and combining historical data includes the following steps: S301, extracting parameters of charge and discharge rates, current voltages, internal resistances, and corresponding remaining capacities of multiple battery packs in historical data; S302, obtaining an equation of an error term, and standardizing the error term; S303, constructing a matrix according to the multiple regression equation; S304, obtaining regression coefficients according to the matrix; The matrix is ​​in the form of: ; Among them, the specific calculation formula of the regression coefficient is: ; In the formula, is the matrix of remaining capacity, is the matrix of dependent variables, is the matrix of regression coefficients, is the matrix of standardized error terms, is the calculated regression coefficient.

6. The method for predicting SOH of a lithium-ion battery pack based on multi-source data fusion and reinforcement learning optimization according to claim 1, characterized in that: The step of obtaining the loss function based on the regression equation and the regression coefficient comprises the following steps: S401, extracting the number of battery packs; S402, extracting the remaining capacity of the corresponding battery pack; S403, obtaining a loss function according to the regression equation and the regression coefficient; The loss function is calculated as follows: ； In the formula, is the corresponding loss function, is the number of battery packs.

7. The method for predicting SOH of a lithium-ion battery pack based on multi-source data fusion and reinforcement learning optimization according to claim 1, characterized in that: The updating of the regression coefficient according to the loss function comprises the following steps: S501, extracting regression coefficients from known information; S502, obtaining the partial derivative of the loss function; S503, assigning values ​​to partial derivatives; S504, updating the regression coefficient according to the assignment, the regression coefficient and the partial derivative; Among them, the calculation formula of the updated regression coefficient is: ; In the formula, is the updated regression coefficient, is the regression coefficient to be updated, For assignment, for The corresponding independent variable.

8. The method for predicting SOH of a lithium-ion battery pack based on multi-source data fusion and reinforcement learning optimization according to claim 1, characterized in that: S701, extracting the loss function and the updated regression equation; S702, determining whether the loss function is a convergence function; S703, if the loss function is a convergence function, there is no need to loop S401 to S504; S704, if the loss function is not a convergent function, loop S401 to S504 until the loss function is a convergent function; S701, bringing the current data of multiple groups of battery groups to be predicted into the updated regression equation to obtain the predicted capacity of the battery groups to be predicted.

9. The method for predicting SOH of a lithium-ion battery pack based on multi-source data fusion and reinforcement learning optimization according to claim 1, characterized in that: The method of obtaining the predicted capacity of the battery pack to be predicted based on multiple sets of current data, regression coefficients and updated regression equations includes the following steps: S801, obtaining a standard error according to an error function; S802, obtaining a prediction interval based on the standard error; S803, comparing the predicted capacity with the predicted interval; S804. If the predicted capacity belongs to the prediction interval, the prediction is accurate; S805. If the predicted capacity does not belong to the prediction interval, recalculate the error function; S806, obtaining the relationship between temperature and battery capacity based on historical data; S807, based on the relationship and in combination with the current temperature, the predicted capacity is corrected to obtain the actual capacity; The calculation formula for the standard error is: ; The formula for the prediction interval is: ; In the formula, is the standard error; For the sample, is the sample mean, is the sample matrix, is the critical value of the distribution, is the predicted value.

Images