A battery SOC prediction method based on BP neural network

Abstract

The present invention is applicable to the field of battery testing technology. The present invention provides a battery SOC prediction method based on a BP neural network, which is used for the rapid prediction of the charge state of a lead-acid battery. The prediction method adopts a conventional single hidden layer BP neural network model; the experimental objects are lead-acid batteries of the same type and model with different newness and oldness. A battery testing system is used to discharge these batteries in a cycle using a composite periodic pulse until the cut-off voltage, and the discharge voltage change data and discharge quantity change data of these batteries are obtained. Then, the voltage discharge data of the battery is plotted into a curve and divided into several cycles, and the curve of each cycle is fitted with a mathematical interpolation function, and the parameters of the fitted function are the model characteristics. Compared with the battery standard capacity method, the prediction method provided by the present invention has the advantages of being simple and fast, and takes into account the aging of the battery, and has better applicability.

Description

Technical Field

[0001] The invention belongs to the technical field of battery testing, and in particular relates to a battery SOC prediction method based on BP neural network. Background Art

[0002] In recent years, the booming communications industry has led to an increasing demand for the reliability of base stations. For base stations, power supply maintenance is crucial, so each base station is equipped with emergency backup batteries in case of emergency. Batteries are an important part of ensuring the safe and stable operation of base stations. Most of the existing base stations are lead-acid batteries or lithium batteries. With the increasing professional concentration of the development of communication technology and Internet of Things technology, base stations have higher and higher requirements for power supply reliability. Batteries are not often used, so it is necessary to regularly maintain the emergency backup power supply of base stations. Ensuring timely investment and adequate protection during emergency use is the primary task of the battery management system. It is difficult to find actual problems with batteries through routine inspections once or several times a year. The need for real-time online and precise management is becoming more and more prominent.

[0003] In order to ensure the safe, reliable and efficient operation of the battery system, the battery status needs to be monitored, controlled and managed regularly. Among them, accurate battery capacity (SOH) estimation and battery state of charge (SOC) estimation are the core tasks that support the overall function of the battery management system. However, the battery capacity and battery charge state cannot be accurately obtained through existing sensors. The conventional capacity verification and state estimation of communication base station batteries are based on the standard capacity verification method of the battery. After the battery is fully charged and left to stand for two hours, it is discharged with a current of 0.1C. Although the result obtained in this way is very accurate, due to the large number of batteries in each communication base station, the time required for capacity verification is relatively long and the efficiency is also very low. Summary of the invention

[0004] The purpose of the present invention is to provide a battery SOC prediction method based on BP neural network, aiming to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions.

[0006] A battery SOC prediction method based on BP neural network, the prediction method comprising the following steps:

[0007] Step S101: leaving the test battery to stand for one hour, then discharging the test battery for several cycles using a composite pulse discharge method, obtaining discharge voltage data of the test battery during the discharge process using a battery tester, and processing the discharge voltage data;

[0008] Step S102: using a mathematical interpolation function fitting method to obtain characteristic parameters of several cycles;

[0009] Step S103: Substitute the characteristic parameters of several cycles into the prediction model and output the prediction results.

[0010] In a preferred embodiment provided by the present invention, the method for establishing the prediction model comprises the following steps:

[0011] Step S201: Fully charge all the experimental batteries, and then let them stand for several hours. Perform a periodic discharge experiment on the batteries using a composite pulse discharge method, and use a battery tester to obtain the voltage, current, power change and cycle number of the battery discharge process.

[0012] Step S202: Divide the collected voltage data into periods, and fit the voltage variation curve of each period using a mathematical interpolation function;

[0013] Step S203: The obtained parameters of the best fitting function are used as an input set;

[0014] Step S204: The corresponding power percentage at the beginning of each cycle is used as an output set;

[0015] Step S205: Establish a BP neural network model, and use the input set and the output set to obtain a trained prediction model.

[0016] In a preferred embodiment provided by the present invention, the composite pulse discharge method comprises the following steps:

[0017] Step S301: fully charge the experimental battery and then let it stand for a period of time;

[0018] Step S302: the experimental battery is discharged with a current of magnitude I1 for t1 minute, and then left to stand for t2 minutes;

[0019] Step S303: discharging the experimental battery with a current of I2 for t1 minute, and then leaving it at rest for t2 minutes;

[0020] Step S304: discharging the experimental battery with a current of I3 for t1 minute, and then leaving it at rest for t2 minutes;

[0021] Step S305: Repeat steps S302, S303 and S304 until the battery voltage is discharged to the cut-off voltage of the battery and the discharge is stopped; then the battery discharge data including voltage, current and discharge capacity are exported from the battery tester.

[0022] In a preferred embodiment provided by the present invention, the sizes of I1, I2 and I3 are different.

[0023] In a preferred embodiment provided by the present invention, the selection range of t1 and t2 is 30 seconds to 3 minutes.

[0024] In a preferred embodiment provided by the present invention, the selected values ​​of t1 and t2 are 1 minute.

[0025] In a preferred embodiment provided by the present invention, the method for extracting characteristic parameters comprises the following steps:

[0026] After collecting the battery discharge data, the battery data needs to be processed to extract key features from the battery voltage change curve.

[0027] In a preferred embodiment of the present invention, the step of processing the battery data and extracting key features from the battery voltage change curve includes:

[0028] The voltage discharge curve of the battery is divided into several complete cycles. When the last cycle is complete, the voltage change curve of a single cycle is fitted, and the parameters of the fitted interpolation function are used as the characteristics of the single cycle.

[0029] Obtain the instantaneous voltage drop of the battery when discharging and stopping discharging in each cycle, and obtain the internal resistance of the battery at that moment through the instantaneous voltage drop;

[0030] Obtain the curve of the voltage drop or rise trend of the battery when discharging and stopping discharging, and then use the fourth-order interpolation function to fit the curve part of the voltage drop and recovery in each cycle; the five parameters corresponding to the fitted fourth-order interpolation function are recorded as features; each cycle has three discharges and three rests; one discharge or one rest will produce one internal resistance, and five function parameters for a total of six features; each cycle will generate 36 features; these features will be used as the input set of the neural network model.

[0031] In a preferred embodiment provided by the present invention, the voltage discharge curve of the battery is divided into several complete cycles, and when the last cycle is incomplete, no processing is performed.

[0032] In a preferred embodiment provided by the present invention, the prediction method is implemented based on a test system, and the test system includes the following modules:

[0033] An electronic load module, wherein the electronic load module leads to two interfaces, namely a positive electrode and a negative electrode, the positive electrode is connected to the positive electrode of the test battery pack, and the negative electrode is connected to the negative electrode of the battery pack to be tested; the electronic load module performs a discharge experiment on the test battery based on a preset composite pulse signal;

[0034] Data acquisition module: each battery in the test experiment will be connected to the data acquisition module through its own data interface, and the changes in voltage and power data in the test experiment will be transmitted to the data acquisition module. The data acquisition module will transmit the received data to the system control module through another interface;

[0035] The system control module has two connection interfaces, one of which is connected to the data acquisition module; receives the battery data transmitted from the data acquisition module and processes the battery data; the other interface is connected to the electronic load module; the system control module is also used to preset the composite pulse signal of the electronic load module.

[0036] Unlike the prior art, the prediction method provided by the present invention has the advantages of being simple and fast compared to the battery standard capacity verification method, and takes into account the battery aging and has better applicability; and the prediction method of the present invention is based on the aging characteristics of the battery, and the voltage change curves obtained by batteries with different aging degrees for the same discharge pulse contain different characteristics, and the characteristic factors affecting the battery aging are in the differences between these curves, which makes up for the poor applicability of the existing prediction method caused by only considering the discharge characteristics of new batteries; and because the voltage curves obtained by discharging the battery with currents of different sizes are different, the composite pulse discharge method of the present invention is helpful to identify more characteristics inside the battery and can improve the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0038] Figure 1 A schematic diagram of a flow chart of a battery SOC prediction method based on a BP neural network according to the present invention;

[0039] Figure 2 A sub-flow chart of the battery SOC prediction method provided by the present invention;

[0040] Figure 3 Another sub-flow chart of the battery SOC prediction method provided by the present invention;

[0041] Figure 4 A structural block diagram of a test system provided by an embodiment of the present invention;

[0042] Figure 5 It is the voltage discharge curve during the battery discharge process;

[0043] Figure 6 It is the battery voltage single cycle variation curve;

[0044] Figure 7 It is a schematic diagram of interpolation function fitting of battery voltage curve;

[0045] Figure 8 Schematic diagram of single-cycle fitting of battery voltage curve. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] At present, in order to ensure the safe, reliable and efficient operation of the battery system, it is necessary to regularly monitor, control and manage the battery status. Among them, accurate battery capacity (SOH) estimation and battery state of charge (SOC) estimation are the core tasks that support the overall function of the battery management system. However, the battery capacity and battery charge state cannot be accurately obtained through existing sensors. The conventional capacity verification and state estimation of communication base station batteries are based on the standard capacity verification method of the battery. After the battery is fully charged and left to stand for two hours, it is discharged with a current of 0.1C. Although the result obtained in this way is very accurate, due to the large number of batteries in each communication base station, the time required for capacity verification is relatively long and the efficiency is also very low.

[0048] To solve the above problems, the present invention provides a method for quickly predicting the state of charge (SOC) of a lead-acid battery. The algorithm adopts a conventional single hidden layer BP neural network model; the experimental objects are lead-acid batteries of the same type and model with different newness and oldness, and the battery testing system is used to discharge these batteries with composite periodic pulses until the cut-off voltage, and the discharge voltage change data and discharge power change data of these batteries are obtained, and then the voltage discharge data of the battery is plotted into a curve and divided into several cycles, and the curve of each cycle is fitted with a mathematical interpolation function, and the parameters of the fitted function are the model features, which are used as the input set of the BP neural network model, and then the initial power corresponding to each cycle is used as the output set for model training.

[0049] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0050] The present invention proposes a battery SOC prediction method and a test system based on a BP neural network, which are used for predicting batteries of the same type and model with different aging degrees and can provide reliable results.

[0051] like Figure 1 As shown, in one embodiment provided by the present invention, a battery SOC prediction method based on a BP neural network comprises the following steps:

[0052] Step S101: leaving the test battery to stand for one hour, then discharging the test battery for several cycles using a composite pulse discharge method, obtaining discharge voltage data of the test battery during the discharge process using a battery tester, and processing the discharge voltage data;

[0053] Step S102: using a mathematical interpolation function fitting method to obtain characteristic parameters of several cycles;

[0054] Step S103: Substitute the characteristic parameters of several cycles into the prediction model and output the prediction results.

[0055] Among them, Figure 2 As shown, in the specific implementation of step S103, the method for establishing the prediction model used in step S103 includes the following steps:

[0056] Step S201: Fully charge all the experimental batteries, and then let them stand for several hours. Perform a periodic discharge experiment on the batteries using a composite pulse discharge method, and use a battery tester to obtain the voltage, current, power change and cycle number of the battery discharge process.

[0057] Step S202: Divide the collected voltage data into periods, and fit the voltage variation curve of each period using a mathematical interpolation function;

[0058] Step S203: The obtained parameters of the best fitting function are used as an input set;

[0059] Step S204: The corresponding power percentage at the beginning of each cycle is used as an output set;

[0060] Step S205: Establish a BP neural network model, and use the input set and the output set to obtain a trained prediction model.

[0061] Further, such as Figure 3 As shown, in the above step S101, the composite pulse discharge method adopted in step S101 includes the following steps:

[0062] Step S301: fully charge the experimental battery and then let it stand for a period of time;

[0063] Step S302: the experimental battery is discharged with a current of magnitude I1 for t1 minute, and then left to stand for t2 minutes;

[0064] Step S303: discharging the experimental battery with a current of I2 for t1 minute, and then leaving it at rest for t2 minutes;

[0065] Step S304: discharging the experimental battery with a current of I3 for t1 minute, and then leaving it at rest for t2 minutes;

[0066] Step S305: Repeat steps S302, S303 and S304 until the battery voltage is discharged to the cut-off voltage of the battery and the discharge is stopped; then the battery discharge data including voltage, current and discharge capacity are exported from the battery tester.

[0067] Furthermore, in the embodiment of the present invention, the sizes of I1, I2, and I3 are different.

[0068] Furthermore, in an embodiment of the present invention, the selection range of t1 and t2 is 30 seconds to 3 minutes; preferably, the selected value of t1 and t2 is 1 minute.

[0069] Furthermore, in a preferred embodiment of the present invention, the method for extracting characteristic parameters comprises the following steps:

[0070] After collecting the battery discharge data, the battery data needs to be processed to extract key features from the battery voltage change curve.

[0071] Furthermore, in a preferred embodiment of the present invention, the step of processing the battery data and extracting key features from the battery voltage change curve includes:

[0072] The voltage discharge curve of the battery is divided into several complete cycles. When the last cycle is complete, the voltage change curve of a single cycle is fitted, and the parameters of the fitted interpolation function are used as the characteristics of the single cycle.

[0073] Obtain the instantaneous voltage drop of the battery when discharging and stopping discharging in each cycle, and obtain the internal resistance of the battery at that moment through the instantaneous voltage drop;

[0074] Obtain the curve of the voltage drop or rise trend of the battery when discharging and stopping discharging, and then use the fourth-order interpolation function to fit the curve part of the voltage drop and recovery in each cycle; the five parameters corresponding to the fitted fourth-order interpolation function are recorded as features; each cycle has three discharges and three rests; one discharge or one rest will produce one internal resistance, and five function parameters for a total of six features; each cycle will generate 36 features; these features will be used as the input set of the neural network model.

[0075] Furthermore, in a preferred embodiment of the present invention, the voltage discharge curve of the battery is divided into several complete cycles, and when the last cycle is incomplete, no processing is performed.

[0076] After collecting the battery discharge data, the battery data needs to be processed to extract key features from the battery voltage change curve.

[0077] First, the voltage discharge curve of the battery is divided into several complete cycles. If the last cycle is incomplete, it will not be processed. The voltage change curve of a single cycle is fitted, and the parameters of the fitted interpolation function are used as the characteristics of a single cycle.

[0078] like Figure 7 As shown in the figure, at the moment when the battery discharges and stops discharging in each cycle, due to the internal resistance inside the battery, the voltage will have a momentary voltage drop. Through this voltage drop, the internal resistance of the battery at that moment can be obtained, which can be used as one of the battery characteristics. After the moment when the battery discharges and stops discharging, the voltage drop or rise trend is a nonlinear curve, which is due to the performance of the polarization characteristics inside the battery. Then use the fourth-order interpolation function to fit the curve part of the voltage drop and rise in each cycle. The five parameters corresponding to the fitted fourth-order interpolation function are recorded as features. Each cycle has three discharges and three rests. One discharge or one rest will generate one internal resistance, and five function parameters for a total of six features. Each cycle will generate 36 features. These features will be used as the input set of the neural network model.

[0079] Figure 5 is the voltage change curve during battery discharge, from which the voltage curve change of one cycle is intercepted, which is Figure 6 It can be seen that three currents of different sizes are discharged for the same length of time and then rest for the same length of time.

[0080] The voltage data is collected at a certain sampling rate, so Figure 6 The medium voltage curve is actually composed of several sampling points. During the initial unit sampling time of discharge, the voltage will drop significantly due to the internal resistance of the battery. The voltage drop at the beginning of each discharge can be used to calculate the approximate battery internal resistance as one of the key features. Similarly, during the initial unit sampling time at the end of discharge, the voltage will rise significantly. This voltage can be used to calculate the approximate battery internal resistance again as one of the key features.

[0081] Figure 8The black circle in the figure is the sampling point for recording the voltage change of the battery in a single cycle. During the three discharges with different currents in this cycle, it can be clearly seen that during the discharge process, the voltage experienced a large voltage drop at the initial moment, and then slowly declined. The voltage change trend was a smooth downward curve. Similarly, at the initial moment of static state after the discharge, the voltage rose sharply, and then rose slowly. The voltage change trend was a smooth upward curve. Based on this feature, the smooth curve part of the voltage during the discharge and static process can be fitted. Because the curve is smooth, the most commonly used method is to use the polynomial function interpolation method. The cftool function in matlab can fit the calibrated data. After repeated attempts, the fitting effect using the 4th-order polynomial interpolation function meets the requirements. Even if used The 4th order polynomial function is fitted. In Matlab, the nlinfit function and polyfit function can be used to fit the calibrated data to obtain the parameters a1, a2, a3, a4, and a5 in the fitting function corresponding to each smooth curve. For example, Figure 8 In the example, the battery discharge voltage is sampled at a sampling rate of 1000ms. t1, t2, and t3 are set to 60s. The six curve segments of this cycle are fitted, and the interpolation function is as follows:

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088] The process of establishing the prediction model specifically includes the following steps:

[0089] 1) Collect data. Fully charge all the experimental batteries and then let them stand for several hours. Design a composite pulse discharge method to perform a cyclic discharge experiment on the batteries. Use a battery tester to obtain the voltage, current, power changes, and cycle times of the battery discharge process.

[0090] 2) The collected voltage data is divided into periods, and the voltage change curve of each period is fitted with a mathematical interpolation function. The parameters of the best fitting function are used as the input set, and the corresponding percentage of electricity at the beginning of each period is used as the output set.

[0091] 3) Establish a BP neural network model and use the previous input set and output set to obtain the trained model.

[0092] 4) Let the test battery stand for one hour, then discharge the battery for several cycles using the composite pulse discharge method in step 1), and process the obtained discharge voltage data. Use the mathematical interpolation function fitting method in step 2) to obtain the characteristic parameters of these cycles.

[0093] 5) Substitute the feature parameters in step 4) into the trained model to obtain the prediction results.

[0094] Furthermore, for the specific implementation of building the BP neural network model in the above process:

[0095] From the above data fitting process, it can be seen that a single battery has 6 curve sections during each discharge cycle, 3 voltage drops and 3 voltage increases. Each curve section can be fitted with a 4th-order interpolation function, and 5 function parameters can be obtained for each curve section each time it is fitted. That is, each discharge cycle of each battery will obtain 30 4th-order interpolation function parameters and 6 battery internal resistance parameters, a total of 36 parameters. Before each discharge cycle, the battery's state of charge (SOC) in this state will be recorded. These 36 parameters are the model input set, and SOC is the model output set.

[0096] The data of several batteries collected in the experiment are processed by the above processing steps, and all the collected voltage change data are processed into an input set of 36 parameters, thereby obtaining many groups of 36-parameter input sets and many groups of corresponding output sets.

[0097] The BP neural network model built is a conventional single hidden layer model. The number of hidden layer neural networks is set to 10, the learning rate is set to 0.078, the number of model training times is set to 200, and the training accuracy requirement is set to 0.00005. Then configure the input and output sets, and use the train function in matlab to train the model. Get the trained model.

[0098] Further, such as Figure 4 As shown, in a preferred embodiment of the present invention, the prediction method is implemented based on a test system, and the test system includes the following modules:

[0099] An electronic load module 401, wherein the electronic load module leads to two interfaces, namely a positive electrode and a negative electrode, wherein the positive electrode is connected to the positive electrode of the test battery pack, and the negative electrode is connected to the negative electrode of the battery pack to be tested; the electronic load module performs a discharge experiment on the test battery based on a preset composite pulse signal;

[0100] Data acquisition module 402, each battery in the test experiment will be connected to the data acquisition module through its own data interface, and the changes in voltage and power data in the test experiment will be transmitted to the data acquisition module, and the data acquisition module will transmit the received data to the system control module through another interface;

[0101] System control module 403, the system control module has two connection interfaces, one of which is connected to the data acquisition module; receives the battery data transmitted from the data acquisition module and processes the battery data; the other interface is connected to the electronic load module; the system control module is also used to preset the composite pulse signal of the electronic load module.

[0102] In summary, the prediction method provided by the embodiment of the present invention has the advantages of being simple and fast compared to the standard battery capacity verification method, and it also takes into account the aging of the battery and has better applicability; and according to the aging characteristics of the battery, the voltage change curves obtained by batteries with different aging degrees for the same discharge pulse contain different characteristics. The characteristic factors that affect battery aging are in the differences between these curves. This makes up for the poor applicability of existing prediction methods that only consider the discharge characteristics of new batteries; because the voltage curves obtained by discharging the battery with currents of different sizes are different, the use of a composite pulse discharge method helps to identify more characteristics inside the battery and can improve the accuracy of the prediction.

[0103] The above solutions are only an illustration of a preferred example, but are not limited thereto. When implementing the present invention, appropriate replacement and / or modification can be performed according to user needs.

[0104] The number of devices and processing scales described here are used to simplify the description of the present invention. Applications, modifications and variations of the present invention will be obvious to those skilled in the art.

[0105] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and the embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily realized. Therefore, without departing from the general concept defined by the claims and equivalent scope, the present invention is not limited to the specific details and the illustrations shown and described here.

Claims

1. A battery SOC prediction method based on BP neural network, characterized in that: The prediction method comprises the following steps: Step S101: leaving the test battery to stand for one hour, then discharging the test battery for several cycles using a composite pulse discharge method, obtaining discharge voltage data of the test battery during the discharge process using a battery tester, and processing the discharge voltage data; Step S102: using a mathematical interpolation function fitting method to obtain characteristic parameters of several cycles; Step S103: Substitute the characteristic parameters of several cycles into the prediction model and output the prediction results; The method for establishing the prediction model comprises the following steps: Step S201: fully charge the experimental batteries, and then let them stand for several hours, and perform a cycle discharge experiment on the batteries using a composite pulse discharge method, and use a battery tester to obtain the voltage, current, power change and cycle number of the battery discharge process; Step S202: Divide the collected voltage data into periods, and fit the voltage variation curve of each period using a mathematical interpolation function; Step S203: The obtained parameters of the best fitting function are used as an input set; Step S204: The corresponding power percentage at the beginning of each cycle is used as an output set; Step S205: Establish a BP neural network model, and use the input set and the output set to obtain a trained prediction model.

2. The battery SOC prediction method based on BP neural network according to claim 1 is characterized in that: The composite pulse discharge method comprises the following steps: Step S301: fully charge the experimental battery and then let it stand for a period of time; Step S302: the experimental battery is discharged with a current of magnitude I1 for t1 minute, and then left to stand for t2 minutes; Step S303: discharging the experimental battery with a current of I2 for t1 minute, and then leaving it at rest for t2 minutes; Step S304: discharging the experimental battery with a current of I3 for t1 minute, and then leaving it at rest for t2 minutes; Step S305: Repeat steps S302, S303 and S304 until the battery voltage is discharged to the cut-off voltage of the battery and the discharge is stopped; then the battery discharge data including voltage, current and discharge capacity are exported from the battery tester.

3. The battery SOC prediction method based on BP neural network according to claim 2 is characterized in that: The sizes of I1, I2 and I3 are different.

4. The battery SOC prediction method based on BP neural network according to claim 3 is characterized in that: The selection range of t1 and t2 is 30 seconds to 3 minutes.

5. The battery SOC prediction method based on BP neural network according to any one of claims 2 to 4, characterized in that: The selected values ​​of t1 and t2 are 1 minute.

6. The battery SOC prediction method based on BP neural network according to claim 2 is characterized in that: The method for extracting characteristic parameters comprises the following steps: After collecting the battery discharge data, the battery data needs to be processed to extract key features from the battery voltage change curve.

7. The battery SOC prediction method based on BP neural network according to claim 6 is characterized in that: The steps of processing the battery data and extracting key features from the battery voltage change curve include: The voltage discharge curve of the battery is divided into several complete cycles. When the last cycle is complete, the voltage change curve of a single cycle is fitted, and the parameters of the fitted interpolation function are used as the characteristics of the single cycle. Obtain the instantaneous voltage drop of the battery when discharging and stopping discharging in each cycle, and obtain the internal resistance of the battery at that moment through the instantaneous voltage drop; Obtain the curve of the voltage drop or rise trend of the battery when discharging and stopping discharging, and then use the fourth-order interpolation function to fit the curve part of the voltage drop and recovery in each cycle; the five parameters corresponding to the fitted fourth-order interpolation function are recorded as features; each cycle has three discharges and three rests; one discharge or one rest will produce one internal resistance, and five function parameters for a total of six features; each cycle will generate 36 features; these features will be used as the input set of the neural network model.

8. The battery SOC prediction method based on BP neural network according to claim 7 is characterized in that: The voltage discharge curve of the battery is divided into several complete cycles, and when the last cycle is incomplete, no processing is performed.

9. The battery SOC prediction method based on BP neural network according to claim 8, characterized in that: The prediction method is implemented based on a test system, which includes the following modules: An electronic load module, wherein the electronic load module leads to two interfaces, namely a positive electrode and a negative electrode, the positive electrode is connected to the positive electrode of the test battery pack, and the negative electrode is connected to the negative electrode of the battery pack to be tested; the electronic load module performs a discharge experiment on the test battery based on a preset composite pulse signal; Data acquisition module: each battery in the test experiment will be connected to the data acquisition module through its own data interface, and the changes in voltage and power data in the test experiment will be transmitted to the data acquisition module. The data acquisition module will transmit the received data to the system control module through another interface; The system control module has two connection interfaces, one of which is connected to the data acquisition module; receives the battery data transmitted from the data acquisition module and processes the battery data; the other interface is connected to the electronic load module; the system control module is also used to preset the composite pulse signal of the electronic load module.

Images