A method for predicting state of charge of lithium-ion batteries based on multi-feature screening

Abstract

The present invention relates to the prediction of the state of charge of a lithium-ion battery, and in particular to a method for predicting the state of charge of a lithium-ion battery based on multi-feature quantity screening. A method for predicting the state of charge of a lithium-ion battery based on multi-feature quantity screening includes collecting external characteristic parameters of the lithium-ion battery for predicting the battery state of charge (SOC) during the discharge process of the lithium-ion battery. The beneficial effects of the present invention are: the state of charge of the lithium-ion battery is predicted by combining various characteristic quantities of battery charging and discharging at the same time; in data extraction, in order to reduce sensor accuracy and environmental impact, the calculation error is reduced by using finite differences; in the data screening part, the capacity increment curve, temperature change curve, equal time voltage drop curve and equal time current drop curve are grouped and screened to select the optimal data combination; in data fitting, Gaussian smoothing is used to smooth the characteristic curve; in SOC prediction, the state of charge of the lithium-ion battery is predicted using an optimized neural network.

Description

Technical Field

[0001] The present invention relates to the prediction of the state of charge of a lithium-ion battery, and in particular to a method for predicting the state of charge of a lithium-ion battery based on multi-feature quantity screening. Background Art

[0002] With the continuous depletion of fossil fuels and the worsening environmental problems caused by them, lithium-ion batteries are widely used in various fields as a symbol of new energy. However, after repeated charge and discharge cycles, the capacity of lithium-ion batteries will gradually decrease. Battery aging will affect the safety and reliability of equipment, and may even cause serious accidents, such as spontaneous combustion of cars and explosion of mobile phones. All countries are vigorously advocating the development of electric vehicle technology and promoting the commercialization of electric vehicles. Power batteries are the only power source for pure electric vehicles, and the performance of electric vehicles depends largely on the characteristics of power batteries. Lithium-ion batteries are favored by the industry for their high density, long life, and low self-discharge rate. They currently occupy the main market of power batteries. In order to ensure the normal operation of electric vehicles, real-time monitoring and management of lithium-ion battery status have become a top priority. Therefore, battery management system (BMS) has gradually become a hot topic of research.

[0003] The battery management system is the "brain" of the power battery system, providing the necessary conditions to ensure the normal operation of the battery module. The indicators predicted by the battery management system mainly include SOC (battery state of charge), capacity, energy state and power state. Among them, the battery SOC and capacity are the two most critical indicators to be monitored by the battery management system. Because accurate SOC estimation is of great significance in preventing overcharging or over-discharging, improving battery energy utilization, and ensuring the safety and stability of electric vehicle battery systems. However, SOC, as an internal characteristic of lithium-ion batteries, cannot be directly measured during vehicle operation, and can only be predicted through external parameters that can be directly measured, such as voltage, current, temperature, and internal resistance. Therefore, this paper proposes a lithium-ion battery state of charge prediction method based on surface temperature and capacity increment.

[0004] The existing battery SOC prediction methods include the ampere-hour integration method, the open circuit voltage method, the model method and the data-driven method. The ampere-hour integration method is the simplest SOC estimation method. The ampere-hour integration method integrates the current in the time dimension during the charging and discharging process to obtain the battery SOC, but the ampere-hour integration method has strong limitations. On the one hand, this method requires an accurate initial SOC value, otherwise, it is easy to introduce initial errors; on the other hand, this method is an open-loop method and lacks feedback correction capabilities. As time goes by, the cumulative error will become larger and larger.

[0005] The open circuit voltage method establishes the relationship between open circuit voltage and SOC based on the offline open circuit voltage test data of the battery. Commonly used test methods mainly include low current test method and pulse discharge test method. The low current test method usually needs to wait for the battery to reach balance before measuring the open circuit voltage when the battery is balanced, which usually takes a long time, so it is also difficult to apply to electric vehicles.

[0006] Model methods can be divided into SOC estimation methods based on electrochemical models and SOC estimation methods based on equivalent circuit models according to the modeling form. The electrochemical modeling model is too complex and is not suitable for real-time measurement battery prediction. The equivalent circuit requires accurate modeling of the battery and parameter identification of the parameters of the equivalent circuit. The accuracy of the fixed parameter equivalent circuit model obtained by the offline parameter identification algorithm is usually not high, and real-time calculation also puts forward new requirements for calculation.

[0007] The data-driven method is the mainstream method now. With the continuous improvement of computer hardware performance, on the one hand, the storage capacity has been significantly improved, and a large amount of data can be stored; on the other hand, the computing power has been significantly enhanced, which has promoted the development and application of machine learning algorithms. The improvement of storage capacity of large-scale data and the progress of machine learning algorithms have greatly promoted the application of data-driven methods. Among them, the data-driven methods used in the SOC estimation process can be divided into three categories: fuzzy logic, support vector machine, and neural network. Neural networks have strong nonlinear modeling capabilities and are suitable for large-scale data processing. However, the existing SOC estimation methods mainly focus on voltage and current characteristics, rarely consider temperature changes during battery degradation, and lack unified consideration of voltage, current, and temperature. There are also fluctuations caused by changes in measurement sensor accuracy and environment, which also put forward requirements for the accuracy of battery SOC prediction. Summary of the invention

[0008] The technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a method for predicting the state of charge of a lithium-ion battery based on multi-feature quantity screening.

[0009] The present invention is achieved through the following technical solutions:

[0010] A method for predicting the state of charge of a lithium-ion battery based on multi-feature quantity screening comprises the following steps:

[0011] S1: During the discharge process of the lithium-ion battery, external characteristic parameters of the lithium-ion battery are collected for predicting the battery state of charge SOC;

[0012] S2: Determine the sampling time interval of the finite difference data according to the error of the collected data, so as to determine the capacity increment and surface temperature change;

[0013] S3: After the sampling interval is established, the capacity increment curve IC and the surface temperature change curve DT are obtained through Gaussian smoothing curve, and the voltage change curve and current change curve are recorded during discharge;

[0014] S4: extract the following characteristic quantities from the IC and DT curves respectively: select the peak and valley coordinates of the two curves, the peak-to-peak distance, the peak-to-valley distance, and the peak-to-peak area, calculate the equal-time voltage drop value for the voltage curve, and record the equal-time current drop value for the current curve;

[0015] S5: screening and grouping the feature quantities for prediction, comparing the prediction results with the algorithm calculation speed and accuracy, and selecting the optimal feature quantity group;

[0016] S6: Import the optimal characteristic quantity of the battery under different battery states of charge (SOC) into the neural network prediction model to obtain the prediction result, and analyze the prediction result and prediction error.

[0017] According to the above technical solution, preferably, in the step S1, during the operation of the electric vehicle, the external characteristic parameters of the battery are collected in real time, including the voltage U, current I, temperature T, charge and discharge time t and the real value of the battery state of charge SOC.

[0018] According to the above technical solution, preferably, in step S3, the calculation formula of the capacity increment curve and the surface temperature change curve at any time k is:

[0019]

[0020] .

[0021] According to the above technical solution, preferably, in the step S5, the feature quantities are screened and grouped:

[0022] The first group: capacity increment curve, temperature change curve, equal time voltage drop curve, equal time current drop curve;

[0023] The second group: capacity increment curve, temperature change curve;

[0024] The third group: capacity increment curve, temperature change curve, and equal time pressure drop curve;

[0025] The fourth group: capacity increment curve, temperature change curve, and equal time current drop curve;

[0026] The battery state of charge (SOC) prediction is performed for each of the four sets of data.

[0027] According to the above technical solution, preferably, in step S5, the selected optimal feature quantity group includes: the height value of the two peaks of the IC curve, the voltage value corresponding to the two peaks of the IC curve, the integral value of the 3.5V-3.9V curve of the IC curve, the height value of the first valley of the DT curve, the height value of the first peak of the DT curve, the voltage value corresponding to the first peak of the DT curve, the height value of the second valley of the DT curve, the voltage value corresponding to the second valley of the DT curve, and the distance between the first peak and the second valley of the DT curve.

[0028] According to the above technical solution, preferably, the optimal feature quantity group selected in step S5 is divided into a training set and a test set.

[0029] According to the above technical solution, preferably, in step S6, the neural network prediction model includes: an input layer, a hidden layer, and an output layer, the input layer is a selected optimal feature quantity combination, and the input layer selects 9 nodes; the hidden layer is an information processing end responsible for information transformation, and the hidden layer selects 6 nodes; the output layer is the battery state of charge SOC, and 1 node is selected.

[0030] According to the above technical solution, preferably, the process of constructing the neural network prediction model is: importing the training set into the neural network, allowing the neural network to complete the optimization selection through the optimization algorithm, and the optimized neural network predicts the test set.

[0031] According to the above technical solution, preferably, the analysis prediction results and prediction errors include mean square error MSE, root mean square error RMSE, mean absolute error MAE, and mean absolute percentage error MAPE.

[0032] According to the above technical solution, preferably, the specific calculation formula of the mean square error MSE is:

[0033] ;

[0034] The specific calculation formula of the root mean square error RMSE is:

[0035] ;

[0036] The specific calculation formula of the mean absolute error MAE is:

[0037] ;

[0038] The mean absolute percentage error MAPE is:

[0039] .

[0040] The beneficial effects of the present invention are as follows: the state of charge of a lithium-ion battery is predicted by combining various characteristic quantities of battery charging and discharging; in data extraction, in order to reduce sensor accuracy and environmental impact, finite differences are used to reduce calculation errors; in the data screening part, the capacity increment curve, the temperature change curve, the equal time voltage drop curve and the equal time current drop curve are grouped and screened to select the optimal data combination; Gaussian smoothing is used to smooth the characteristic curve in data fitting; and the state of charge of the lithium-ion battery is predicted by using an optimized neural network in SOC prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The battery IC curves under different degradation cycles of the embodiment of the present invention are shown.

[0042] Figure 2 The DT curves of the batteries under different decay cycles of the embodiments of the present invention are shown.

[0043] Figure 3 A flow chart of a prediction algorithm according to an embodiment of the present invention is shown.

[0044] Figure 4 The flowchart of the prediction method structure of the embodiment of the present invention is shown.

[0045] Figure 5 A schematic flow chart of a method for predicting the state of charge of a lithium-ion battery according to an embodiment of the present invention is shown.

[0046] Figure 6 The battery degradation test set results of the embodiment of the present invention are shown.

[0047] Figure 7 A fitness curve for an embodiment of the present invention is shown.

[0048] Figure 8 The prediction error distribution of the battery degradation data set according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and the best embodiments.

[0050] As shown in the figure, the method for predicting the state of charge of a vehicle-mounted lithium-ion battery based on battery multi-feature quantity screening of the present invention includes the following steps:

[0051] S1: During the discharge process of the lithium-ion battery, external characteristic parameters of the lithium-ion battery are collected for predicting the battery SOC;

[0052] S2: Determine the sampling time interval of the finite difference data according to the error of the collected data , thereby determining the capacity increment and temperature change;

[0053] S3: After the sampling interval is established, the data is processed, and the capacity increment curve IC and the surface temperature change curve DT are obtained through the Gaussian smoothing curve. The voltage change curve and the current change curve are recorded during discharge.

[0054] S4: Extract characteristic quantities for IC and DT curves respectively, select the peak and valley coordinates, peak-to-peak distance, peak-to-valley distance, and peak-to-peak area of ​​the two curves, calculate the equal-time voltage drop value for the voltage curve, and record the equal-time current drop value for the current curve;

[0055] S5: Screen the extracted features, group them for prediction, compare the prediction results with the algorithm calculation speed and accuracy, consider the actual engineering conditions, and select the optimal feature group under the premise of ensuring prediction accuracy;

[0056] S6: Import the optimal characteristic quantity of the battery under different SOC conditions into the neural network to obtain the prediction result, and analyze the prediction result and prediction error.

[0057] Wherein, in the step S1, during the battery charging and discharging process, the external characteristic parameters of the battery are collected in real time, including voltage (U), current (I), temperature (T), charging and discharging time (t) and the real value of the battery SOC.

[0058] Among them, in the step S2, considering the limited accuracy of the temperature sensor, if it is directly calculated, it is easily affected by external noise and produces a large deviation. Therefore, in order to avoid such problems, the finite difference method is used to approximate the temperature change rate. It can be selected according to the actual situation. Here, L=30 is selected.

[0059] Wherein, in step S3, the calculation formula of the capacity increment curve and the surface temperature change curve at any time k is:

[0060]

[0061]

[0062] Wherein, in the step S4, the required characteristic quantity is obtained by drawing the capacity increment curve and the temperature change curve under different SOC. Figure 1 The capacity increment curves are given for charge and discharge cycles of 500, 1000, 2000, 3000, 5000, 6000, 7000 and 8100 times. Extract two peak coordinates, one valley coordinate, one peak-to-peak distance value, and the peak integral area value. Figure 2The temperature change curves are given for charge and discharge cycles of 500, 1000, 3000, 5000, 7000 and 8100 times. The first valley coordinate, the first peak coordinate, the second valley coordinate, the maximum coordinate, and the peak-to-peak and peak-to-valley distance values ​​of the curve are extracted. The equal time voltage drop value is determined as the voltage change data corresponding to 60s, and the equal time current drop value is determined as the voltage change data corresponding to 60s.

[0063] Among them, in step S5, the characteristic value data of the above-mentioned capacity increment curve, temperature change curve, equal time voltage drop curve and equal time current drop curve are screened and grouped:

[0064] 1: Capacity increment curve, temperature change curve, equal time voltage drop curve, equal time current drop curve;

[0065] 2. Capacity increment curve and temperature change curve;

[0066] 3. Capacity increment curve, temperature change curve, and equal time pressure drop curve;

[0067] 4: Capacity increment curve, temperature change curve, and equal time current decrease curve.

[0068] SOC prediction was performed on four sets of data respectively. It was found that considering the capacity increment curve and temperature change curve can reduce the calculation and optimization algorithm iteration time as much as possible while ensuring the prediction accuracy, which is beneficial to real-time SOC monitoring of the battery and has better engineering significance.

[0069] Among them, in step S5, the determined capacity increment curve can see that the first peak gradually decreases with the cycle, and it is basically unrecognizable after 3000 times. The second peak also decreases, but it is still clearly visible, so the second peak height value of the IC curve and the integral area of ​​the curve between the two peaks of 3.5V-3.9V are selected to characterize the characteristic quantity. It can be seen that the first valley height value of the temperature change curve gradually decreases with the cycle, and the peak point gradually shifts to the right, the first valley value of the DT curve gradually decreases with the cycle, and shifts to the right, the second valley value of the DT curve increases with the cycle, and shifts to the left, the distance between the first peak point and the second valley point of the DT curve gradually decreases, and the highest point of the right curve and the highest point voltage value are very chaotic. After the Pearson correlation coefficient is calculated, they are all lower than 0.6, lacking a good linear correlation.

[0070] Through the above description, it is determined that the 9 parameters selected as feature quantities are selected, including the height value of the second peak of the IC curve, the voltage value corresponding to the second peak of the IC curve, the integral value of the 3.5V-3.9V curve of the IC curve, the height value of the first valley of the DT curve, the height value of the first peak of the DT curve, the voltage value corresponding to the first peak of the DT curve, the height value of the second valley of the DT curve, the voltage value corresponding to the second valley of the DT curve, and the distance between the first peak and the second valley of the DT curve. The Pearson correlation parameters between each parameter and SOC are calculated with reference to the battery degradation data set:

[0071] Characteristic parameter value IC curve second peak height The voltage value corresponding to the second peak of IC curve IC curve 3.5V-3.9V curve integral value DT curve valley height value DT curve peak height The voltage value corresponding to the peak of the DT curve DT curve second valley height value The voltage value corresponding to the second valley of the DT curve The distance between the peak and valley of the DT curve Pearson correlation coefficient -0.9485 0.9501 0.9959 0.6255 0.9465 -0.8587 -0.6495 0.8858 0.9675

[0072] It shows that these 9 parameters are strongly correlated with the battery SOC. The SOC determined by comprehensive judgment of 9 sets of data will be more accurate, which is conducive to the prediction of SOC of vehicle batteries.

[0073] Among them, in step S5, the training set and the test set are divided by dividing the calculated characteristic parameter data set, and the training set and the test set are divided according to 85% and 15%. The decay data set used in this paper is three batteries of the same model, with a total of 200 sets of data, including 170 training sets and 30 test sets. The neural network belongs to the artificial intelligence algorithm, which has a three-layer structure, including: input layer, hidden layer, output layer, such as Figure 4 As shown. Among them:

[0074] The input layer is a combination of different input feature parameter data sets in step S4. Since there are 9 input feature quantities, the input layer has 9 nodes;

[0075] The hidden layer is responsible for information transformation at the information processing end. According to the requirements of information transformation capability, the middle layer can be designed as a single hidden layer or a multi-hidden layer structure. The output of the hidden layer is calculated by the following formula:

[0076]

[0077] Where: is the activation function; is the threshold corresponding to the output layer; , They refer to the connection weight value between the hidden layer and the output layer and the output value of the hidden layer respectively. It is a nonlinear transformation function - Sigmoid function (also known as S function). The number of hidden layers is 6.

[0078] The output layer is the predicted SOC value of the lithium-ion battery, and the output value of each node in the output layer is:

[0079]

[0080] Where: is the threshold of the hidden layer node of the neural network; is the connection weight value between the input layer and the hidden layer of the network. Since the output is the battery SOC, the output layer is selected as 1. , and connection threshold , In order to solve the problem with a greater impact, a neural network with optimized weights was used, and an optimization algorithm was introduced to complete the optimization selection for the initialization parameters. The network prediction algorithm flow of the optimized parameters is as follows: Figure 3 The specific steps of the lithium-ion battery prediction model based on battery temperature and capacity changes are as follows:

[0081] (1) Put the multiple training sets in step S4 into the optimized neural network respectively.

[0082] (2) Use the training set to train the neural network until the requirements for training error and number of training times are met.

[0083] (2) Run the trained neural network on the test set to obtain the predicted value.

[0084] (3) Perform error analysis on the predicted SOC values ​​of different lithium-ion batteries and the actual values ​​of the test set.

[0085] (4) If the error is good, the trained model can be applied to the SOC prediction under this type of battery working condition, and by increasing the training set, the prediction result will further conform to the true value, greatly improving the accuracy of predicting the battery SOC. Finally, the model is established with the battery degradation data set, and the prediction results are shown in Figure 6 As shown, the prediction error of 30 groups of results is Figure 8 As shown, the maximum error is 9% and the average absolute error is around 3.6%.

[0086] Wherein, in the step S5, the accuracy analysis of the predicted value includes: mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Wherein, MSE can evaluate the degree of change of the data. The smaller the MSE value, the better the accuracy of the prediction model. RMSE is the arithmetic square root of MSE. MAE can better reflect the actual situation of the prediction value error. The smaller the MAE value, the better the accuracy of the prediction model. The smaller the MAPE value, the better the accuracy of the prediction model. Wherein:

[0087] The specific calculation formula of the mean square error MSE is:

[0088] .

[0089] The specific calculation formula of the root mean square error RMSE is:

[0090] .

[0091] The specific calculation formula of the mean absolute error MAE is:

[0092] .

[0093] The mean absolute percentage error MAPE is:

[0094] .

[0095] The present invention takes into account the practical problem that it is difficult to directly predict the SOC of lithium-ion batteries in electric vehicles under actual conditions, and proposes a method for predicting the state of charge of lithium-ion batteries based on the screening of multiple battery feature quantities. The influence of different input parameter combinations on the SOC value is analyzed, and data minimization is achieved while ensuring prediction accuracy, thereby reducing calculation time. The battery state of charge is estimated using an optimized neural network model. This model has a fast calculation speed and a small error for the prediction results of the lithium-ion state of charge, and meets the requirements for the prediction results of the lithium-ion battery state of charge under different operating conditions. It has high usability, and the capacity prediction model results are as follows: Figure 6 As shown, the errors after prediction are as follows:

[0096] Error category Mean Square Error (MSE) Root mean square error RMSE Mean absolute error (MAE) Mean absolute percentage error MAPE Error value 0.0019564 0.044231 0.036792 9.3051%

[0097] Through the final prediction results and error analysis, it can be concluded that the lithium-ion battery state of charge prediction method based on multi-feature quantity screening has good prediction accuracy and introduces temperature changes, which is beneficial to the grasp of the battery working state and prevents the occurrence of battery thermal runaway; the present invention successfully establishes a relatively accurate prediction model through changes in external characteristic quantities such as voltage, current, temperature, and charging and discharging time. It has good prediction accuracy for the prediction of lithium-ion battery SOC and has certain application value.

[0098] The beneficial effects of the present invention are as follows: the state of charge of a lithium-ion battery is predicted by combining various characteristic quantities of battery charging and discharging; in data extraction, in order to reduce sensor accuracy and environmental impact, finite differences are used to reduce calculation errors; in the data screening part, the capacity increment curve, the temperature change curve, the equal time voltage drop curve and the equal time current drop curve are grouped and screened to select the optimal data combination; Gaussian smoothing is used to smooth the characteristic curve in data fitting; and the state of charge of the lithium-ion battery is predicted by using an optimized neural network in SOC prediction.

[0099] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting the state of charge of a lithium-ion battery based on multi-feature screening, characterized in that: The following steps are involved: S1: During the discharge process of the lithium-ion battery, external characteristic parameters of the lithium-ion battery are collected for predicting the battery state of charge SOC; S2: Determine the sampling time interval of the finite difference data according to the error of the collected data, so as to determine the capacity increment and surface temperature change; S3: After the sampling interval is established, the capacity increment curve IC and the surface temperature change curve DT are obtained through Gaussian smoothing curve, and the voltage change curve and current change curve are recorded during discharge; S4: extract the following characteristic quantities from the IC and DT curves respectively: select the peak and valley coordinates of the two curves, the peak-to-peak distance, the peak-to-valley distance, and the peak-to-peak area, calculate the equal-time voltage drop value for the voltage curve, and record the equal-time current drop value for the current curve; S5: screening and grouping the feature quantities for prediction, comparing the prediction results with the algorithm calculation speed and accuracy, and selecting the optimal feature quantity group; S6: Import the optimal characteristic quantity of the battery under different battery states of charge (SOC) into the neural network prediction model to obtain the prediction result, and analyze the prediction result and prediction error.

2. The method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 1, characterized in that: In the step S1, during the operation of the electric vehicle, external characteristic parameters of the battery are collected in real time, including voltage U, current I, temperature T, charge and discharge time t, and the real value of the battery state of charge SOC.

3. The method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 2, characterized in that: In step S3, the calculation formula of the capacity increment curve and the surface temperature change curve at any time k is: 。 4. The method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 3, characterized in that: In the step S5, the feature quantities are filtered and grouped: The first group: capacity increment curve, temperature change curve, equal time voltage drop curve, equal time current drop curve; The second group: capacity increment curve, temperature change curve; The third group: capacity increment curve, temperature change curve, and equal time pressure drop curve; The fourth group: capacity increment curve, temperature change curve, and equal time current drop curve; The battery state of charge (SOC) prediction is performed for each of the four sets of data.

5. The method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 4, characterized in that: In step S5, the selected optimal feature quantity group includes: the height value of the second peak of the IC curve, the voltage value corresponding to the second peak of the IC curve, the integral value of the 3.5V-3.9V curve of the IC curve, the height value of the first valley of the DT curve, the height value of the first peak of the DT curve, the voltage value corresponding to the first peak of the DT curve, the height value of the second valley of the DT curve, the voltage value corresponding to the second valley of the DT curve, and the distance between the first peak and the second valley of the DT curve.

6. The method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 5, characterized in that: The optimal feature quantity group selected in step S5 is divided into a training set and a test set.

7. A method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 6, characterized in that: In step S6, the neural network prediction model includes: an input layer, a hidden layer, and an output layer. The input layer is the selected optimal feature quantity combination, and the input layer selects 9 nodes; the hidden layer is the information processing end responsible for information transformation, and the hidden layer selects 6 nodes; the output layer is the battery state of charge SOC, and selects 1 node.

8. The method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 7, characterized in that: The process of constructing the neural network prediction model is as follows: the training set is introduced into the neural network, the neural network is optimized through the optimization algorithm, and the optimized neural network predicts the test set.

9. The method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 8, characterized in that: The analysis of prediction results and prediction errors includes mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE).

10. The method for predicting the state of charge of a lithium-ion battery based on multi-feature screening according to claim 9, characterized in that: The specific calculation formula of the mean square error MSE is: ； The specific calculation formula of the root mean square error RMSE is: ； The specific calculation formula of the mean absolute error MAE is: ； The mean absolute percentage error MAPE is: 。

Images