A power battery retirement prediction method based on capacity decay

Abstract

The present invention provides a power battery retirement prediction method based on capacity decay, constructs a capacity decay algorithm prediction model based on LSTM network and a retirement statistical prediction model based on statistics of retired batteries, which are respectively used as two basic models of the method, and are used to decide which basic model to use according to the actual situation of the data sample. When the number of samples is small and the data quality is poor, a relatively rough prediction result is obtained through the retirement statistical prediction model. When the number of samples is sufficient and the quality is high, the capacity decay algorithm prediction model can achieve a more accurate retirement time prediction for batteries of different materials and scenarios. This method can achieve a wider coverage of existing new energy vehicles, which is conducive to the effective big data analysis of retired batteries and batteries to be retired in the future, and thus has many beneficial effects that the existing technology does not have.

Description

Technical Field

[0001] The present invention belongs to the technical field of residual value assessment of power battery capacity for new energy vehicles, and specifically relates to a power battery retirement prediction method based on capacity attenuation. Background Art

[0002] With the increase in the number of new energy vehicles at home and abroad, the peak of retired power batteries is about to come. How to achieve effective recycling of retired batteries is of great significance to the protection of the environment, resources and sustainable development. At this stage, the problem of life prediction and subsequent sorting and utilization of retired power batteries in this field can only meet the test and rough statistics of a small number of batteries. It is not possible to obtain more accurate prediction results by performing multi-dimensional parameter index analysis on various factors that may cause battery degradation in large quantities. It also hinders the acquisition of the specific conditions of each vehicle's power battery, which is not conducive to planning the later use, retirement and recycling of each battery in advance. Summary of the invention

[0003] In view of this, in view of the technical problems existing in the above-mentioned field, the present invention provides a method for predicting power battery retirement based on capacity decay, which specifically includes the following steps:

[0004] Step 1: Select sample vehicles and extract their charging and discharging history data throughout their life cycle;

[0005] Step 2: Divide the power batteries into 6 types according to the material types of the sample vehicle power batteries: ternary material batteries, lithium iron phosphate batteries, other types of batteries, and vehicle usage types: operating vehicles and non-operating vehicles;

[0006] Step 3: Calculate the charging capacity and discharging capacity samples corresponding to the daily charging and discharging segments during the use of the battery using the historical data of different power battery types;

[0007] Step 4: According to the daily proportion of the charging segment and the discharging segment, the weights of the charging capacity and discharging capacity sample values ​​calculated in step 2 are determined respectively, and the weighted average is calculated to obtain the maximum available capacity of the battery corresponding to different cumulative usage times;

[0008] Step 5: Use the maximum available capacity and accumulated usage time of different power battery types to construct a training sample set, train the long short-term memory network through the back propagation training method, and obtain the capacity decay algorithm prediction model corresponding to the six power battery types;

[0009] Step 6: Multiply the initial maximum available capacity in the historical data of the sample vehicle by the capacity decay rate η when the same type of battery is retired to obtain the retirement condition of the capacity decay algorithm prediction model; Statistically calculate the different cumulative usage time and capacity decay rate η change trends of each retired battery, and combine the relationship between the average cumulative usage time and the average capacity decay rate of the same type of battery to establish a retirement statistical prediction model for power batteries;

[0010] Step 7. Integrate the established capacity decay algorithm prediction model and retirement statistical prediction model to predict vehicle capacity decay trend and retirement time; use the number of sample vehicles and the quality of historical data as the basis for deciding which of the two prediction models to use for retirement prediction.

[0011] Furthermore, the historical data extracted in step 1 specifically includes the following parameter indicators:

[0012] Charging current, charging time, discharging current, discharging time and battery state of charge SOC.

[0013] Furthermore, in step 2, power batteries are specifically divided into the following 6 types:

[0014] Ternary operating vehicles (Class 1), ternary non-operating vehicles (Class 2), lithium iron phosphate operating vehicles (Class 3), lithium iron phosphate non-operating vehicles (Class 4), other types of battery operating vehicles (Class 5), other types of battery non-operating vehicles (Class 6).

[0015] Furthermore, in step three, the charging capacity and discharging capacity samples are respectively obtained in the following manner:

[0016] For each charging segment, the charging capacity sample value C is calculated using the following formula: C :

[0017]

[0018] In the above formula, I is the charging current, t 0 is the initial time of the charging segment, t 1 Charging segment end time, SOC 0 With SOC 1 are the SOC values ​​at the beginning and end of the charging segment, respectively;

[0019] For each discharge segment, the discharge capacity sample value C is calculated using the following formula: d :

[0020]

[0021] In the above formula, I is the discharge current, t 0 is the initial time of the discharge segment, t 1The end time of the discharge segment, SOC 0 With SOC 1 They are the SOC values ​​at the beginning and end of the discharge segment, respectively.

[0022] Furthermore, in step 4, the specific c charging segments and n d The time proportions of the discharge segments are α% and β%, and the maximum available capacity C of the day is calculated by weighted average, which is specifically expressed as:

[0023]

[0024] Among them, C c1 ……C cn are the sample values ​​of each charging capacity on the day, C d1 ……C dn They are the sample values ​​of discharge capacity for each time on that day.

[0025] Furthermore, the core of the long short-term memory network described in step 5 is composed of an input gate, a forget gate, and an output gate; the specific calculation formula of the network is:

[0026] i t =σ(W i ·[y t-1 ,x t ]+b i )

[0027] f t =σ(W f ·[y t-1 ,x t ]+b f )

[0028] o t =σ(W o ·[y t-1 ,x t ]+b o )

[0029]

[0030] y t =o t tanh(C t )

[0031] In the formula, x represents the input vector, i.e., the capacity and cumulative time in the historical samples, y represents the output vector, i.e., the predicted results of the capacity and cumulative time, i, f, o, C represent the input gate, forget gate, output gate, and cell state, respectively, and C t-1 For long-term memory, is the current memory, the subscript t represents the t-th state, the matrices W and b are the weight parameters and bias terms to be trained respectively (where the initial values ​​of the W matrix and b can be set to any value between 0 and 1), and σ(·) is the sigmoid nonlinear function;

[0032] Define the error term S and weight gradient loss of each cell in the LSTM network to update the network parameters:

[0033]

[0034] In the formula, N is the number of samples, y t is the predicted value, y t * is the true value.

[0035] Furthermore, the specific process of training the long short-term memory network includes:

[0036] a. Input the calculated capacity sample data of sample vehicles;

[0037] b. Initialize the matrices W and b, with specific values ​​​​selected arbitrarily between 0 and 1;

[0038] c. Perform calculations through the LSTM network, calculate the loss value at the same time, and automatically optimize the values ​​of W and b according to the loss value;

[0039] d. Output the calculated value C t and t ;

[0040] e. C t and t As the parameter of the next training, it is brought into the t+1th training;

[0041] f. Repeat step be until all training data are trained;

[0042] g. After the training is completed, a trained capacity decay algorithm prediction model is obtained.

[0043] By executing the above training process, a capacity attenuation algorithm prediction model applicable to the above 6 different types of power batteries can be obtained.

[0044] Furthermore, the specific process of establishing the retirement statistical prediction model described in step 6 includes:

[0045] The different cumulative usage time and capacity decay rate η changing trends of retired batteries are counted, and the relationship between the average cumulative usage time and the average capacity decay rate of batteries of the same type is combined to form the retirement statistical prediction model;

[0046] The retirement condition of the capacity decay algorithm prediction model is obtained by multiplying the initial maximum available capacity in the historical data of the sample vehicle by the capacity decay rate η of the same type of battery when it is retired.

[0047] Furthermore, in step seven, it is specifically determined whether the number of sample vehicles reaches a predetermined value, whether the number of charge and discharge times reaches more than 200 times, whether the interval of each frame of historical data is within 0 to 30 seconds, and whether each frame of historical data contains all: charging current, charging time, discharge current, discharge time, and SOC parameters; if the judgment is yes, the capacity decay algorithm prediction model is used to predict the retirement time in combination with the retirement conditions, and the model is continuously updated; if the judgment is no, the retirement statistical prediction model is used to predict the retirement time.

[0048] The power battery retirement prediction method based on capacity decay provided by the present invention above constructs a capacity decay algorithm prediction model based on LSTM network and a retirement statistical prediction model based on statistics of retired batteries, which are respectively used as two basic models of the method to decide which basic model to use according to the actual situation of the data sample. When the number of samples is small and the data quality is poor, a relatively rough prediction result is obtained through the retirement statistical prediction model. When the number of samples is sufficient and the quality is high, the capacity decay algorithm prediction model can achieve a more accurate retirement time prediction for batteries of different materials and scenarios. This method can achieve a wider coverage of existing new energy vehicles, which is conducive to the effective big data analysis of retired batteries and batteries to be retired in the future, and thus has many beneficial effects that the existing technology does not have. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of the training process of the LSTM network in the method provided by the present invention;

[0050] Figure 2 It is a schematic diagram of retirement prediction results in an example based on the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] The present invention provides a method for predicting power battery retirement based on capacity attenuation, which specifically includes the following steps:

[0053] Step 1: Select a sample vehicle and extract its charging and discharging history data throughout its life cycle. The historical data includes the following parameter indicators: charging current, charging time, discharging current, discharging time, and battery state of charge SOC;

[0054] Step 2: According to the material type of the sample vehicle power battery: ternary material battery, lithium iron phosphate battery, other types of batteries, and the vehicle use type: operating vehicles, non-operating vehicles, the power batteries are divided into 6 types, such as: ternary operating vehicles (category 1), ternary non-operating vehicles (category 2), lithium iron phosphate operating vehicles (category 3), lithium iron phosphate non-operating vehicles (category 4), other types of battery operating vehicles (category 5), other types of battery non-operating vehicles (category 6). Those skilled in the art should know that other classification rules can also be implemented according to actual needs for the subsequent establishment of capacity attenuation algorithm prediction models for each battery type;

[0055] Step 3: Calculate the charging capacity and discharging capacity samples corresponding to the daily charging and discharging segments during the use of the battery using the historical data of different power battery types;

[0056] Step 4: According to the daily proportion of the charging segment and the discharging segment, the weights of the charging capacity and discharging capacity sample values ​​calculated in step 2 are determined respectively, and the weighted average is calculated to obtain the maximum available capacity of the battery corresponding to different cumulative usage times;

[0057] Step 5: Use the maximum available capacity and accumulated usage time of different power battery types to construct a training sample set, train the long short-term memory network through the back propagation training method, and obtain the capacity decay algorithm prediction model corresponding to the six power battery types;

[0058] Step 6: Multiply the initial maximum available capacity in the historical data of the sample vehicle by the capacity decay rate η when the same type of battery is retired to obtain the retirement condition of the capacity decay algorithm prediction model; Statistically calculate the different cumulative usage time and capacity decay rate η change trends of each retired battery, and combine the relationship between the average cumulative usage time and the average capacity decay rate of the same type of battery to establish a retirement statistical prediction model for power batteries;

[0059] Step 7. Integrate the established capacity decay algorithm prediction model and retirement statistical prediction model to predict the vehicle capacity decay trend and retirement time; use the number of sample vehicles and the quality of historical data as the basis for deciding which of the two prediction models to use for retirement prediction.

[0060] In a preferred embodiment of the present invention, the charging capacity and discharging capacity samples are respectively charged and discharged in the following manner in step 3:

[0061] For each charging segment, the charging capacity sample value C is calculated using the following formula: C :

[0062]

[0063] In the above formula, I is the charging current, t 0 is the initial time of the charging segment, t 1 Charging segment end time, SOC 0 With SOC 1 are the SOC values ​​at the beginning and end of the charging segment, respectively;

[0064] For each discharge segment, the discharge capacity sample value C is calculated using the following formula: d :

[0065]

[0066] In the above formula, I is the discharge current, t 0 is the initial time of the discharge segment, t 1 The end time of the discharge segment, SOC 0 With SOC 1 They are the SOC values ​​at the beginning and end of the discharge segment, respectively.

[0067] In a preferred embodiment of the present invention, in step 4, the n c charging segments and n d The time of the discharge segments accounts for 40% and 60%, and the maximum available capacity C of the day is calculated by weighted average, which is specifically expressed as:

[0068]

[0069] Among them, C c1 ……C cn are the sample values ​​of each charging capacity on the day, C d1 ……C dn They are the sample values ​​of discharge capacity for each time on that day.

[0070] In a preferred embodiment of the present invention, the core of the long short-term memory network described in step 5 is composed of an input gate, a forget gate and an output gate; the specific calculation formula of the network is:

[0071] i t =σ(W i ·[y t-1 ,x t ]+b i )

[0072] f t =σ(Wf ·[y t-1 ,x t ]+b f )

[0073] o t =σ(W o ·[y t-1 ,x t ]+b o )

[0074]

[0075] y t =o t tanh(C t )

[0076] In the formula, x represents the input vector, y represents the output vector, i, f, o, and C represent the input gate, forget gate, output gate, and cell state respectively, and C t-1 For long-term memory, is the current memory, the subscript t represents the t-th state, the matrices W and b are the weight parameters and bias terms to be trained respectively (where the initial values ​​of the W matrix and b can be set to any value between 0 and 1), and σ(·) is the sigmoid nonlinear function;

[0077] Define the error term S and weight gradient loss of each cell in the LSTM network to update the network parameters:

[0078]

[0079]

[0080] In the formula, N is the number of samples, y t is the predicted value, y t * is the true value.

[0081] like Figure 1 As shown in Figure 1, the specific process of training the long short-term memory network includes:

[0082] a. Input the calculated capacity sample data of sample vehicles;

[0083] b. Initialize the matrices W and b, with specific values ​​​​selected arbitrarily between 0 and 1;

[0084] c. Perform calculations through the LSTM network, calculate the loss value at the same time, and automatically optimize the values ​​of W and b according to the loss value;

[0085] d. Output the calculated value C t and t ;

[0086] e. C t and t As the parameter of the next training, it is brought into the t+1th training;

[0087] f. Repeat step be until all training data are trained;

[0088] g. After the training is completed, a trained capacity decay algorithm prediction model is obtained.

[0089] By executing the above training process, a capacity attenuation algorithm prediction model applicable to the above 6 different types of power batteries can be obtained.

[0090] In a preferred embodiment of the present invention, the specific process of establishing the retirement statistical prediction model in step 6 includes:

[0091] The different cumulative usage time and capacity decay rate η changing trends of retired batteries are counted, and the relationship between the average cumulative usage time and the average capacity decay rate of batteries of the same type is combined to form the retirement statistical prediction model;

[0092] The retirement condition of the capacity decay algorithm prediction model is obtained by multiplying the initial maximum available capacity in the historical data of the sample vehicle by the capacity decay rate η of the same type of battery when it is retired. The retirement time of such batteries can be obtained by the following steps:

[0093] a. First, the capacity sample mean calculated from the vehicle's previous 10 historical data is used as the initial capacity C 0 ;

[0094] b. Calculate the maximum available capacity C at the time of retirement by combining the statistical capacity decay rate η Re ;

[0095] c. Based on the results and the capacity decay trend predicted by the capacity decay algorithm prediction model, the cumulative usage time before retirement is predicted.

[0096] In a preferred embodiment of the present invention, in step seven, it is specifically determined whether the number of sample vehicles reaches a predetermined value, whether the number of charge and discharge times reaches more than 200 times, whether the interval of each frame of historical data is within 0 to 30s, and whether each frame of historical data contains all: charging current, charging time, discharge current, discharge time, and SOC parameters; if the judgment is yes, the capacity decay algorithm prediction model is used to predict the retirement time in combination with the retirement conditions, and the model is continuously updated; if the judgment is no, it means that the historical data samples provided by the sample vehicles or the specific parameters contained therein are not sufficient to construct an accurate LSTM network model. Therefore, the retirement time is predicted using the retirement statistical prediction model, and a rough battery retirement prediction result can be obtained in the initial stage. If necessary, the fine division of battery types may not be performed temporarily. When the number of samples and the data quality meet the requirements, classification statistics are performed according to actual needs, and the corresponding LSTM network models are trained respectively, so that the algorithm can be continuously improved during use.

[0097] After selecting any applicable basic model through the above decision, it is also necessary to consider the cumulative usage time before the sale, so the final predicted retirement time fin_t is:

[0098] fin_t=t sell +t 退役

[0099] Among them, t 退役 is the predicted retirement time result obtained by using the capacity decay algorithm prediction model or retirement statistical prediction model, t sell It is the cumulative usage time of the target vehicle when it is first sold.

[0100] In an example according to the present invention, the corresponding retirement prediction is achieved by executing the following process:

[0101] 1. Select the historical charging and discharging data of a passenger car and calculate the charging capacity and discharging capacity of each time. Among them, the charging capacity is 832 times and the discharging capacity is 1045 times.

[0102] 2. Take the natural day as the time period and take the weighted average value of the charge and discharge capacity of each day to form a capacity data sample.

[0103] 3. The retirement prediction model determines that the vehicle belongs to a ternary operating vehicle (class 1) based on the basic information of the vehicle, and its historical charging and discharging data meets the conditions of the capacity attenuation prediction model. Therefore, the data model of class 1, LSTM class 1, is used for prediction.

[0104] 4. Set the initial values ​​of the LSTM class 1 model parameters and import sample data for model training.

[0105] 5. Take the values ​​of W and b corresponding to the lowest point of LSTM class 1 as the final parameter optimization values ​​of the model.

[0106] 6. Calculate the initial value of the sample vehicle’s battery capacity, that is, the average value of the first 10 capacity samples is 108.348Ah.

[0107] 7. Calculate the capacity value of the battery when it reaches retirement conditions. The capacity decay rate of the battery of this type of vehicle when it reaches retirement conditions is 25%. That is, the final retirement capacity is 81.261Ah.

[0108] 8. Use the LSTM network model to predict battery capacity attenuation. The attenuation trend is as follows: Figure 2 shown.

[0109] 9. Through model calculation, it can be predicted that the vehicle battery is expected to reach its capacity retirement value on March 21, 2022, and corresponding retirement planning will be carried out.

[0110] It should be understood that the size of the serial numbers of the steps in the embodiment of the present invention does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0111] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A power battery retirement prediction method based on capacity decay, Features: The specific steps include: Step 1: Select sample vehicles and extract their charging and discharging history data throughout their life cycle; Step 2: Divide the power batteries into 6 types according to the material types of the sample vehicle power batteries: ternary material batteries, lithium iron phosphate batteries, other types of batteries, and vehicle usage types: operating vehicles and non-operating vehicles; Step 3: Calculate the charging capacity and discharging capacity samples corresponding to the daily charging and discharging segments during the use of the battery using the historical data of different power battery types; Step 4: According to the daily proportion of the charging segment and the discharging segment, the weights of the charging capacity and discharging capacity sample values ​​calculated in step 3 are determined respectively, and the weighted average is calculated to obtain the maximum available capacity of the battery corresponding to different cumulative usage times; Step 5: Use the maximum available capacity and accumulated usage time of different power battery types to construct a training sample set, train the long short-term memory network through the back propagation training method, and obtain the capacity decay algorithm prediction model corresponding to the six power battery types; Step 6: Multiply the initial maximum available capacity in the historical data of the sample vehicle by the capacity decay rate η when the same type of battery is retired to obtain the retirement condition of the capacity decay algorithm prediction model; Statistically calculate the different cumulative usage time and capacity decay rate η change trends of each retired battery, and combine the relationship between the average cumulative usage time and the average capacity decay rate of the same type of battery to establish a retirement statistical prediction model for power batteries; Step 7: Integrate the established capacity decay algorithm prediction model and retirement statistical prediction model to predict the vehicle capacity decay trend and retirement time; The number of sample vehicles and the quality of historical data are used as the basis for deciding which of the two prediction models to use for retirement prediction.

2. The method according to claim 1, Features: The historical data extracted in step 1 specifically includes the following parameter indicators: Charging current, charging time, discharging current, discharging time and battery state of charge SOC.

3. The method according to claim 1, Features: In step 2, power batteries are specifically divided into the following 6 types: Ternary operating vehicles, ternary non-operating vehicles, lithium iron phosphate operating vehicles, lithium iron phosphate non-operating vehicles, other types of battery operating vehicles, other types of battery non-operating vehicles.

4. The method according to claim 1, Features: In step 3, the following methods are used to respectively charge and discharge capacity samples: For each charging segment, the charging capacity sample value C is calculated using the following formula: C : In the above formula, I is the charging current, t 0 is the initial time of the charging segment, t 1 Charging segment end time, SOC 0 With SOC 1 are the SOC values ​​at the beginning and end of the charging segment, respectively; For each discharge segment, the discharge capacity sample value C is calculated using the following formula: d : In the above formula, I is the discharge current, t 0 is the initial time of the discharge segment, t 1 The end time of the discharge segment, SOC 0 With SOC 1 They are the SOC values ​​at the beginning and end of the discharge segment, respectively.

5. The method according to claim 1, Features: In step 4, the specific c charging segments and n d The time proportions of the discharge segments are α% and β%, and the maximum available capacity C of the day is calculated by weighted average, which is specifically expressed as: Among them, C c1 ……C cn are the sample values ​​of each charging capacity on the day, C d1 ……C dn They are the sample values ​​of discharge capacity for each time on that day.

6. The method according to claim 1, Features: The core of the long short-term memory network described in step 5 is composed of an input gate, a forget gate, and an output gate; the specific calculation formula of the network is: i t =σ(W i ·[y t-1 ,x t ]+b i ) f t =σ(W f ·[y t-1 ,x t ]+b f ) the t =σ(W o ·[y t-1 ,x t ]+b o ) y t =o t fishy(C) t ) In the formula, x represents the input vector, y represents the output vector, i, f, o, and C represent the input gate, forget gate, output gate, and cell state respectively, and C t-1 For long-term memory, is the current memory, the subscript t represents the t-th state, the matrices W and b are the weight parameters and bias terms to be trained, respectively, and σ(·) is the sigmoid nonlinear function; Define the error term S and weight gradient loss of each cell in the long short-term memory network to update the network parameters: In the formula, N is the number of samples, y t is the predicted value, y t * is the true value.

7. The method according to claim 6, Features: The specific process of training the long short-term memory network includes: a. Input the calculated capacity sample data of sample vehicles; b. Initialize the matrices W and b, with specific values ​​​​selected arbitrarily between 0 and 1; c. Calculate the loss value by using the long short-term memory network, and automatically optimize the values ​​of W and b according to the loss value; d. Output the calculated value C t and t ; e. C t and t As the parameter of the next training, it is brought into the t+1th training; f. Repeat step be until all training data are trained; g. After the training is completed, a trained capacity decay algorithm prediction model is obtained.

8. The method according to claim 1, Features: In step 7, it is specifically determined whether the number of sample vehicles reaches a predetermined value, whether the number of charge and discharge times reaches more than 200 times, whether the interval of each frame of historical data is within 0 to 30 seconds, and whether each frame of historical data contains all: charging current, charging time, discharge current, discharge time, and SOC parameters; If the judgment is yes, the capacity decay algorithm prediction model is used to predict the retirement time in combination with the retirement conditions, and the model is continuously updated; if the judgment is no, the retirement statistical prediction model is used to predict the retirement time.

Images