A method for predicting the current state health of a battery based on a BP neural network algorithm

Abstract

A method for predicting the current state health of a battery based on a BP neural network algorithm belongs to the field of battery prediction. It includes the following steps: detection of healthy batteries, data processing and definition, random testing and data processing of actual batteries under test, establishment and prediction of a BP neural network model, and calculation of the actual battery's health. This invention, based on complete battery test data, constructs complete battery charge / discharge data at different rates. By combining this with the BP neural network algorithm, it can construct dedicated neural network datasets for batteries with different state attributes and apply them to the prediction of various attributes of actual batteries, achieving high accuracy and strong practicality.

Description

Technical Field

[0001] This invention relates to the application of a BP neural network algorithm in predicting battery state of health (SOH), state of charge (SOC), and remaining useful life (RUL), belonging to the field of battery prediction. Background Technology

[0002] Batteries, as the lifeblood of the new energy field, not only help our country achieve industrial transformation, but also make outstanding contributions to reducing the use of fossil fuels.

[0003] Electric bicycles have become a convenient mode of transportation for short-distance travel, facilitating people's daily commutes. However, in recent years, due to the uneven distribution of battery models and varying safety standards, coupled with people's lack of experience in preventing battery-related accidents, the number of safety incidents involving electric bicycles has been gradually increasing and cannot be ignored, especially as electric bicycles become more widespread.

[0004] At this time, if there were a detection device that could promptly detect and alert people after they use electric bicycles, it would be possible to minimize the occurrence of spontaneous combustion and explosion of electric bicycle batteries.

[0005] Traditional battery testing requires using the voltage-time curve from the highest voltage to the cutoff voltage as the input signal, which is then fed into a backpropagation neural network for calculation to obtain various battery data. This undoubtedly increases the time cost. However, to detect the discharge curve within a short period of time, such as 30 minutes, and accurately predict the discharge cycle number of the discharge curve, the dataset needs to be processed.

[0006] This invention improves upon the existing BP neural network algorithm, significantly reducing the detection time and accuracy of electric bicycle batteries. It primarily uses three indicators to indicate SOH, SOC, and RUL, achieving a prediction accuracy of over 95%. Summary of the Invention

[0007] Building upon current research on BP neural network algorithms and electric bicycle battery testing, this paper proposes a method that utilizes a BP neural network algorithm combined with specialized data processing techniques. Compared to traditional battery testing methods, this method can successfully predict the battery's SOH, SOC, and RUL without requiring testing the complete discharge curve from the battery's highest voltage to its cutoff voltage. This solution is implemented as follows.

[0008] A computational model based on the BP neural network algorithm, and a detection method for predicting battery SOH, SOC, and RUL, characterized by comprising the following steps:

[0009] Step 1: Testing for Healthy Batteries

[0010] Healthy batteries (any type of ternary lithium battery) were subjected to constant current charge-discharge cycle tests at different rates. Since the cycle voltage range of ternary lithium batteries is usually 4.15V to 3.0V, and the sampling time is 30s, the cycle charge-discharge program is rest-constant current charging-rest-constant current discharging, and so on. The charge and discharge data of the battery SOH from 100% to 80% were collected, and the discharge time-voltage curves corresponding to different cycles under each rate condition were used to form a discharge curve set for different cycles.

[0011] Batteries are tested starting with brand new, uniform, and of the same type and brand. Before the experiment, the battery's State of Harm (SOH) should be measured at 100%. The batteries are then placed on a charge / discharge testing device for constant current charge / discharge cycle testing. The SOH will gradually decrease after a certain number of cycles. According to industry standards for electric bicycles, batteries are unusable once their SOH drops below 80%. Therefore, charge / discharge data for batteries with SOH between 100% and 80% are compiled, and the discharge data is organized into a single dataset (since some batteries may have an SOH higher than 100% after several cycles, this data also needs to be included in the dataset). Each data point in this dataset is a time-voltage curve, with an upper voltage limit of 4.15V and a lower voltage limit of 3.0V. Finally, the charging or discharging curves of the battery test data with SOH between 100% and 80% are compiled as follows: Figure 1 The data chart shown, Figure 1 This dataset contains all cycle discharge data for a 100% SOC battery, subjected to constant current charge-discharge cycles at 0.3C rate until 80% SOH is reached. Each time-voltage curve corresponds to one cycle, and the entire graph also indicates which cycle it belongs to.

[0012] Step two: Data processing and definition

[0013] After compiling all discharge datasets from constant current charge-discharge tests performed on batteries at a certain rate (e.g., 0.3C) from 100% to 80% SOH, and plotting these datasets into an image, as shown below... Figure 1 As shown, the multiplier of this dataset is 0.3C, then D rate=0.3C. It can be seen that all data in this curve set are voltage-time curves. In step one, the initial voltage value of each or each cycle curve in the discharge curve set is defined as 4.15V, corresponding to an initial time of 0s, and the termination voltage value is defined as 3.0V, corresponding to a termination time point T. END The termination time T of each or every cycle curve END They are all different; each or each cycle curve corresponds to N cycles. C The maximum number of cycles corresponding to the last or last segment of the cycle curve is denoted as N. max ;

[0014] The data corresponding to the entire curve graph set mentioned above is named Data_origin.

[0015] Step 3: Random testing and data processing of actual batteries

[0016] (1) At a certain multiple D rate The discharge curve of an actual battery at a given instant is called the actual discharge curve. This actual battery is named Battery A. Figure 2 As shown in (a), the upper limit of the voltage of this segment of the actual discharge curve of battery A is V1, where V1≤4.15V; the initial time of discharge is recorded as 0s, and the termination voltage of discharge is V2. The discharge termination time corresponding to V2≥3.0V is T. e The actual discharge curve duration is T. e .

[0017] To obtain a dataset covering a specific voltage range and use it for neural network training, all discharge curves in the Data_origin dataset corresponding to the discharge rate Xc with voltages between V1 and V2 were selected. The times corresponding to V1 in all selected samples were then changed to the initial time and initialized to 0 seconds, creating a new set of voltage-time curves for different cycles. Figure 2 As presented in (b), this new graph set is named Data_mid; it can be generated by... Figure 2 As can be seen from the curves in (b), the duration of each cycle curve in this new curve set is different.

[0018] (2) In order to reduce the error caused by different data lengths during the training of the BP neural network, it is necessary to unify the data length of all data in the curve set Data_mid, that is, to unify the discharge time of all selected discharge data as T. neat T neat The value is the longest duration of the shortest discharge time in the curve graph set Data_mid. For example... Figure 2In the dataset in (b), the shortest discharge time of the discharge curve is 390s, then T is defined as... neat =390s.

[0019] (3) For each discharge curve in the dataset Data_mid, the discharge time is greater than T. neat Delete all data and create a file like this. Figure 2 The dataset shown in (d) is named Data_final; all data in the Data_final dataset have an initial time of 0s and an ending time of T. neat In this dataset, the starting voltage for each cyclic curve is V1, but the ending voltages are different.

[0020] (4) To facilitate the training of the neural network, the actual discharge curve from step (1) needs to be displayed as follows: Figure 2 The time duration of battery A measured in (a) is also defined as T. neat Therefore, Figure 2 (a) The measured discharge curve in the image is processed: the processing time is T. neat The discharge curve starts at 0 seconds and ends at T seconds. neat The starting voltage of the actual discharge curve for this treatment is V1. Since the termination time of the curve changes, the termination voltage is no longer V2. Therefore, V3 is defined as the new termination voltage for the actual discharge curve of this treatment. Figure 2 As shown in (c).

[0021] Step 4: Establishment and Prediction of the BP Neural Network Model

[0022] All data in the dataset Data_final were used as the training set, with the number of cycles in each data set (i.e., each cycle curve) serving as the target value for training a backpropagation (BP) neural network. This network was named net_01 and ultimately had the following three characteristics: 1. It was specific to battery A; 2. The voltage range of the training set was between V1 and V2; 3. The discharge data duration was T. neat like Figure 2 As shown in (a), the actual curve processed in step (4), i.e., the corresponding termination voltage, is V3. Figure 2 (c) The discharge data is input into network net_01;

[0023] Finally, net_01 can predict the number of charge-discharge cycles that battery A is in during the instantaneous test in step (1), and this number of cycles is named N. predict Then N predict The value is the number of discharge cycles in which the discharge curve segment is predicted by the BP neural network.

[0024] Step 5: Calculation of actual battery health

[0025] Assume that the predicted cycle number of battery A is N. predict The dataset Data_origin contains N circles. predict The discharge curves were selected, with an initial voltage of 4.15V, a cutoff voltage of 3.0V, an initial time of 0s, and a termination time of T. END The curve is then named Curve_. predic Then the Nth element in the dataset Data_origin predict The state of harmonics (SOH) of a cell with multiple revolutions is:

[0026] T END ×D rate / 3600×100%=SOH (1)

[0027] Where T END For the Nth node in Data_origin predict The discharge time corresponding to the number of battery cycles, D rate Given the discharge current ratio and 3600 as the number of seconds per hour, the SOH is considered to be the SOH of battery A.

[0028] Due to the curve Curve predic Selected, at the same time Figure 2 In (a), the actual discharge curve cutoff voltage of A is V2, and the curve is... predic The point with an on-voltage value of V2 or the closest V2 is selected and named SOH_real(T). end The voltage at point V2 is V2, and the time value is T. end The SOC of battery A is calculated as follows:

[0029] (T END -T end ) / T END ×100%=SOC (2)

[0030] Where T end For point SOH_real(T) end The time data on V2) is 3600, which is the number of seconds per hour.

[0031] The remaining lifespan of battery A is calculated as follows:

[0032] N max -N predict =RUL (3)

[0033] N predictN represents the number of cycles that battery A is predicted to be in by the neural network algorithm. max RUL represents the cycle number of the last data set in the dataset Data_origin, where RUL is the remaining lifetime.

[0034] Advantages of this invention:

[0035] 1. Based on the establishment of complete battery test data, this invention constructs complete battery charge / discharge data at different rates. After combining with the BP neural network algorithm, it can build a dedicated neural network dataset for batteries with different state attributes and apply it to the prediction of various attributes of actual batteries. It has high accuracy and strong practicality.

[0036] 2. The complete set of algorithms involved in this invention can predict the SOH, SOC and RUL of batteries in any state with an accuracy of over 95%, and the testing time is short, making it easy to put into practical applications.

[0037] 3. The full description of this invention is based on a 0.3C rate, but this method is applicable not only to 0.3C but also to other rates. If a more complete and detailed database is available, the SOH, SOC, and RUL of a battery at any rate and in any state can be detected in a short time. Attached Figure Description

[0038] Figure 1 Data set images of a healthy battery corresponding to a 0.3C discharge rate, as shown in Embodiment 1 of this invention;

[0039] Figure 2 The data processing method of Embodiment 1 of the present invention is as follows: (a) is the discharge curve of the battery under test A over a period of time; (b) is the dataset Data_mid constructed after extracting the total data set of the discharge curve voltage range corresponding to battery A; (c) is the measured discharge data of the dataset with the same length as (d); (d) is the dataset Data_final obtained after extracting the shortest time length from the dataset in (b).

[0040] Figure 3 The present invention provides the following examples: (a) a training template demonstration of the BP neural network algorithm; (b) a graph of the variance data of the BP neural network algorithm; and (c) a display of the accuracy of the self-validating network dataset.

[0041] Figure 4The data processing method of Embodiment 2 of the present invention is as follows: (a) is the discharge curve of the battery B under test over a period of time; (b) is the dataset Data_mid constructed after extracting the total data set of the discharge curve voltage range corresponding to battery B; (c) is the measured discharge data of the dataset with the same length as (d); (d) is the dataset Data_final obtained after extracting the shortest time length from the dataset in (b).

[0042] Figure 5 The following are examples of Embodiment 2 of the present invention: (a) a training template demonstration of the BP neural network algorithm; (b) a graph of the variance data of the BP neural network algorithm; and (c) a display of the accuracy of the self-validating network dataset. Detailed Implementation

[0043] The essential features and significant advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments, but are not limited thereto.

[0044] Example 1

[0045] Step 1

[0046] All time-voltage curves from the battery cycle data were extracted, with an upper voltage limit of 4.15V and a lower voltage limit of 3.0V. Finally, the charging or discharging curves from the battery's SOH (State of Hypothesis) test data (100%–80%) were compiled into [the appropriate database / system]. Figure 1 In the dataset shown, the discharge rate D of this dataset rate =0.3C.

[0047] Step Two

[0048] It can be seen that all data in this dataset are voltage-time curves. The initial voltage value for each discharge curve segment in this dataset is defined as 4.15V; the initial time as 0s; the termination voltage value as 3.0V; and the termination time as T. END However, the termination time of each curve segment is different; the number of cycles corresponding to each discharge curve segment is N. C The number of cycles corresponding to the last segment of the discharge curve is N. max The dataset is named Data_origin.

[0049] Step 3

[0050] (1) To determine the SOH, SOC, and RUL of a battery, it is necessary to obtain the discharge curve of any segment of the battery and name the battery as: Battery A, such as Figure 2 As shown in (a), the upper voltage limit of this segment of the discharge curve for battery A is V1 = 3.96V; the initial discharge time is 0s; the discharge termination voltage is V2 = 3.91V; and the discharge termination time is T. e=1020s, then the duration of this discharge curve segment is T. e =1020s.

[0051] To obtain a new dataset for training the neural network, all discharge curves with voltages between 3.96V and 3.91V in the Data_origin dataset were selected, and the initial time of all data was set to 0s, preparing them as follows: Figure 2 The dataset presented in (b) is named Data_mid.

[0052] (2) Standardize the length of all data in the dataset Data_mid. Figure 2 In the dataset in (b), the discharge time of the shortest discharge curve is 390s, T neat =390s.

[0053] (3) Next, the data for the first 390 seconds of all curves in the dataset Data_mid will be filtered out and compiled into the following format: Figure 2 The dataset shown in (d) is named Data_final. All data in this dataset has an initial time of 0s and an ending time of T. neat The starting voltage of this dataset is V1 = 3.96V; however, the ending voltages are different for each dataset.

[0054] (4) Figure 2 The measured T-value of battery A in (a) neat Data with a duration of 390 seconds was selected; the curve's start time is 0 seconds and its end time is T. neat The starting voltage of this curve is V1 = 3.96V; the ending voltage of this curve is defined as V3 = 3.935V as the new ending voltage, as follows: Figure 2 As shown in (c).

[0055] Step Four

[0056] Using all data in the dataset Data_final as the training set, and the number of loops in each data set as the target value, a backpropagation (BP) neural network is trained. This network is named net_01 and has the following characteristics: 1. It is specific to battery A; 2. The voltage range of the dataset is between V1 and V2 (3.96V to 3.91V); 3. The dataset duration is 390 seconds. Figure 2 As shown in (a); will Figure 2 (c) The discharge data of battery A in a certain state, which was actually tested, is input into the network net_01.

[0057] Ultimately, net_01 can Figure 2(a) The measured data of battery A is processed to predict the number of charge-discharge cycles that battery A was in during the test. This number of cycles is named N. predict In this embodiment, the predicted value of the BP neural network is 241.93, which is rounded down to N. predict =242. Then N predict The value of 242 represents the number of discharge cycles in which the discharge curve segment is predicted by the BP neural network.

[0058] Step 5

[0059] The number of cycles for battery A is predicted to be N. predict =242, which means N circles in the dataset Data_origin predict The discharge curves were selected, with an initial voltage of 4.15V, a cutoff voltage of 3.0V, an initial time of 0s, and a termination time of T. END =12870s; then the curve is named Curve_ predic Then the Nth element in the dataset Data_origin predict The state of harmonics (SOH) of a cell with multiple revolutions is:

[0060] SOH=T END ×D rate / 3600×100%=12870s×0.3C / 3600×100%=107.25%

[0061] Therefore, the SOH is considered to be 107.25% for battery A.

[0062] Due to the curve Curve predic Selected, at the same time Figure 2 (a) The cutoff voltage of the discharge curve of battery A is V2 = 3.91V. At this time, the curve... predic The point with a voltage value of V2 = 3.91V or the closest value of 3.91V was selected and named SOH_real(204s, 3.91V). The voltage value at this point is V2 = 3.91V, and the time value is T. end =2040s, then the SOC of battery A is calculated as follows:

[0063] SOC=(T END -T end ) / T END ×100%=(12870s-2040s) / 12870s×100%

[0064] =84.149%

[0065] Therefore, SOC means that the current state of charge of battery A is 84.149%.

[0066] Since the dataset Data_origin contains 2375 sets of books, therefore N max = 2375, and the calculation process N was explained above. predict =242, then the remaining lifespan of battery A is:

[0067] RUL=N max -N predict =2375-242=2133

[0068] The RUL of this battery is 2133 cycles, meaning that the battery can be charged and discharged 2133 times at a rate of 0.3C.

[0069] The calculation process and error analysis in the BP neural network calculation process are as follows: Figure 3 As shown.

[0070] The model ultimately predicted 242 laps, and the model network's accuracy was [insert accuracy here]. Figure 3 (c) The battery with 97.249% charge has a SOC of 84.149%, a SOH of 107.25%, and a remaining lifespan of 2133 cycles. This means that the battery currently has 84.149% of its full charge capacity, and the battery's full charge capacity is 107.25% of its initial state. The battery can be charged and discharged at a rate of 0.3C for another 2133 cycles before the SOH of the battery drops to 80%.

[0071] Example 2

[0072] Step 1

[0073] All time-voltage curves from the battery cycle data were extracted, with an upper voltage limit of 4.15V and a lower voltage limit of 3.0V. Finally, the charging or discharging curves from the battery's SOH (State of Hypothesis) test data (100%–80%) were compiled into [the appropriate database / system]. Figure 1 In the dataset shown, the discharge rate D of this dataset rate =0.3C.

[0074] Step Two

[0075] It can be seen that all data in this dataset are voltage-time curves. The initial voltage value for each discharge curve segment in this dataset is defined as 4.15V (approximately 4.15V); the initial time is 0s; the termination voltage value is 3.0V; and the termination time is T. END However, the termination time of each curve segment is different; the number of cycles corresponding to each discharge curve segment is N. C The number of cycles corresponding to the last segment of the discharge curve is N.max The dataset is named Data_origin.

[0076] Step 3

[0077] (1) To determine the SOH, SOC, and RUL of a battery, it is necessary to obtain any segment of the battery's discharge curve and name the battery as: Battery B, such as Figure 4 As shown in (a), the upper voltage limit of this segment of the discharge curve for battery B is V1 = 3.889V; the initial discharge time is 0s; the discharge termination voltage is V2 = 3.805V; and the discharge termination time is T. e =1020s, then the duration of this discharge curve segment is T. e =1020s.

[0078] To obtain a new dataset for training the neural network, all discharge curves with voltages between 3.889V and 3.805V in the Data_origin dataset were selected, and the initial time of all data was initialized to 0s, resulting in a dataset as shown below. Figure 4 The dataset presented in (b) is named Data_mid.

[0079] (2) Standardize the length of all data in the dataset Data_mid. Figure 4 In the dataset in (b), the discharge time of the shortest discharge curve is 810 s, T neat =810s.

[0080] (3) Next, the data for the first 810 seconds of all curves in the dataset Data_mid will be filtered out and compiled into the following format: Figure 4 The dataset shown in (d) is named Data_final. All data in this dataset has an initial time of 0s and an ending time of T. neat The starting voltage of this dataset is V1 = 3.889V; however, the ending voltages are different for each dataset.

[0081] (4) Figure 2 The measured T of battery B in (a) neat Data with a duration of 810 seconds was selected; the curve's start time is 0 seconds and its end time is T. neat The starting voltage of this curve is V1 = 3.889V; the ending voltage of this curve is defined as V3 = 3.832V as the new ending voltage, as follows: Figure 4 As shown in (c).

[0082] Step Four

[0083] Using all data in the dataset Data_final as the training set, and the number of cycles in each data set as the target value, a backpropagation (BP) neural network is trained. This network is named net_01 and has the following characteristics: 1. It is specific to battery B; 2. The voltage range of the training set data is between V1 and V2 (3.889V to 3.805V); 3. The discharge time of the dataset is 810s. Figure 4 As shown in (a). Will Figure 4 (c) The discharge data of battery B in a certain state, which was actually tested, is input into the network net_01.

[0084] Ultimately, net_01 can Figure 4 (a) The measured data of battery B is processed to predict the number of charge-discharge cycles that battery B was in during the test. This number of cycles is named N. predict In this embodiment, the predicted value of the BP neural network is 1792.3, which is rounded down to N. predict =1793. Then N predict The value of 1793 represents the number of discharge cycles in which the discharge curve for that segment is predicted by the BP neural network.

[0085] Step 5

[0086] The number of cycles for battery A is predicted to be N. predict =1793, which means N circles in the dataset Data_origin predict The discharge curves were selected, with an initial voltage of 4.15V, a cutoff voltage of 3.0V, an initial time of 0s, and a termination time of T. END =11070s; then the curve is named Curve_ predic Then the Nth element in the dataset Data_origin predict The state of harmonics (SOH) of a cell with multiple revolutions is:

[0087] SOH=T END ×D rate / 3600×100%=11070s×0.3C / 3600×100%=92.25%

[0088] Therefore, the SOH is considered to be 92.25% for battery B.

[0089] Due to the curve Curve predic Selected, at the same time Figure 4 (a) The discharge curve cutoff voltage of battery B is V2 = 3.805V. At this time, the curve... predicThe point with a voltage value of V2 = 3.805V or the closest value of 3.805V was selected and named SOH_real(1890s, 3.805V). The voltage value at this point is V2 = 3.91V, and the time value is T. end =2040s, then the SOC of battery A is calculated as follows:

[0090] SOC=(T END -T end ) / T END ×100%=(11070s-1890s) / 11070s×100%

[0091] =82.92%

[0092] The SOC is the current state of charge of battery B, which is 82.92%.

[0093] Since the dataset Data_origin contains 2375 sets of books, therefore N max = 2375, and the calculation process N was explained above. predict =1793, then the remaining lifespan of battery B is:

[0094] RUL=N max -N predict =2375-1793=582

[0095] The RUL of this battery is 582 cycles, meaning that the battery can be charged and discharged 582 times at a rate of 0.3C.

[0096] The calculation process and error analysis in the BP neural network calculation process are as follows: Figure 4 As shown.

[0097] To verify the accuracy of this method, Figure 4 (b) Ten data points from the new dataset were used as a test set to verify the accuracy of the BP neural network algorithm.

[0098] The final predicted value of the model was 1793 cycles, the accuracy of the model network was 99.455%, the SOC of the battery was 92.25%, the SOH of the battery was 82.92%, and the remaining battery life was 582 cycles. This means that the current capacity of the battery is 82.92% of its full capacity, the full capacity of the battery is 92.25% of its initial state, and the battery can be charged and discharged at a rate of 0.3C for another 582 cycles before the SOH of the battery will decrease to 80%.

[0099] The patent specifies a charge / discharge rate of 0.3C throughout, but in practical applications, it is also applicable to other charge / discharge rates such as 0.1C, 0.5C, and 1C.

Claims

1. A method for predicting the current state health of a battery based on a BP neural network algorithm, characterized in that, Includes the following steps: Step 1: Testing for a healthy battery; Healthy batteries were subjected to constant current charge-discharge cycle tests at different rates. Since the cycle voltage range of ternary batteries is 4.15 V to 3.0 V and the sampling time is 30 s, the cycle charge-discharge program is rest-constant current charging-rest-constant current discharging, and so on. The charge and discharge data of the battery SOH at 100% to 80% were collected, and the discharge time-voltage curves corresponding to different cycles under each rate condition were used to form a discharge curve set for different cycles. Step two: Data processing and definition; After compiling all discharge data sets from constant current charge-discharge tests conducted on batteries at a certain rate from 100% to 80% SOH, the datasets were plotted as images. All data in the curve dataset are voltage-time curves. In step one, the initial voltage value of each curve or segment in the discharge curve dataset is defined as 4.15 V, the corresponding initial time as 0 s, and the termination voltage value as 3.0 V, corresponding to the termination time point T. END The termination time T of each or every cycle curve END They are all different; each or each cycle curve corresponds to N cycles. C The maximum number of cycles corresponding to the last or last segment of the cycle curve is denoted as N. max ; The data corresponding to the entire curve graph set is named Data_origin; Step 3: Random testing and data processing of the actual battery under test; (1) At a certain multiple D rate The discharge curve of a segment of an actual battery at a given instant is called the actual discharge curve. This actual battery is named Battery A. The upper limit of the voltage of the actual discharge curve of Battery A is V. 1， V1≤4.15 V; the initial discharge time of the discharge curve is recorded as 0 s, the discharge termination voltage is V2, and the discharge termination time corresponding to V2≥3.0 V is T. e The time length of the actual discharge curve mentioned above is T. e ; The data of all discharge curves with voltages between V1 and V2 corresponding to discharge rate Xc in the dataset Data_origin are filtered out, and the time corresponding to V1 of all the filtered ones is changed to the initial time and initialized to 0 s. This creates a new set of voltage-time curves for different cycles, which is named Data_mid. The duration of each cycle curve in the new set of voltage-time curves for different cycles is different. (2) In order to reduce the error caused by different data lengths during the training of the BP neural network, the data length of all data in the curve set Data_mid is unified, that is, the discharge time of all selected discharge data is denoted as T. neat T neat The value is the longest time of the shortest discharge time in the curve graph set Data_mid; (3) For each discharge curve in the dataset Data_mid, the discharge time is greater than T. neat All data is deleted, and the resulting dataset is named Data_final. All data in the Data_final dataset starts at time 0 seconds and ends at time T. neat In the Data_final dataset, each loop curve has a starting voltage of V1, but the ending voltages are different. (4) In order to facilitate the training of the neural network, the time length of battery A measured by the actual discharge curve in step (1) is also defined as T. neat The measured discharge curve was processed as follows: the processing time was T. neat The discharge curve starts at 0 s and ends at T. neat The starting voltage of the actual discharge curve is V1. Since the termination time of the processed curve has changed, the termination voltage is no longer V2. The termination voltage of the actual processed curve is defined as V3 as the new termination voltage. Step 4: Establishment and prediction of the BP neural network model; All data in the Data_final dataset were used as the training set. The number of cycles in each data set (i.e., each cycle curve) was used as the target value to train a backpropagation (BP) neural network. This target value-trained BP neural network was named net_01 and ultimately had the following three characteristics:

1. It was specific to battery A; 2. The training set voltage range was between V1 and V2; 3. The discharge data duration was T. neat The actual curve processed in step (4), i.e. the discharge data corresponding to the termination voltage V3, is input into the network net_01; Finally, net_01 predicts the number of charge / discharge cycles that battery A is in during the instantaneous test in step (1), and the value of the number of cycles is named N. predict Then N predict The value is the number of discharge cycles in which the discharge curve segment is predicted by the BP neural network. Step 5 Assume that the predicted cycle number of battery A is N. predict Set the number of circles in the Data_origin dataset to N. predict The discharge curves were selected, and the selected curves had an initial voltage of 4.15 V, a cutoff voltage of 3.0 V, an initial time of 0 s, and an termination time of T. END The curve is then named Curve_ predic Then the Nth element in the dataset Data_origin predict The state of harmonics (SOH) of a cell with multiple revolutions is: T END ×D rate / 3600×100 %=SOH (1) Where T END For the Nth node in Data_origin predict The discharge time corresponding to the number of battery cycles, D rate Where 3600 is the discharge current multiplier, 3600 is the number of seconds per hour, and N is the number of seconds per hour. predict The state of harmonics (SOH) of a cell with a certain number of revolutions is considered to be the state of harmonics of cell A. Due to the curve Curve predic It was selected, and the actual discharge curve cutoff voltage of A was V2. The curve Curve_ predic The point with an upper voltage value of V2 or the closest value of V2 is selected, and the selected point is named SOH_real(T). end The voltage value at the selected point is V2, and the time value is T. end The SOC of battery A is calculated as follows: (T END -T end ) / T END ×100 %=SOC (2) Where T end For point SOH_real(T) end The time data on V2), where 3600 represents the number of seconds per hour; The remaining lifespan of battery A is calculated as follows: N max -N predict =RUL (3) N predict N represents the number of cycles that battery A is predicted to be in by the neural network algorithm. max RUL represents the cycle number of the last data set in the dataset Data_origin, where RUL is the remaining lifetime.

Images