New energy battery capacity prediction method

Abstract

The invention belongs to the technical field of new energy batteries, and discloses a new energy battery capacity prediction method, which comprises the following steps: acquiring the number of charge-discharge cycles of a new energy battery, capacity data of each charge-discharge cycle and use time; determining the standard charge-discharge cycle index of the new energy battery according to the charge-discharge cycle index and the use time; constructing a first network model and a second network model for predicting the capacity of the new energy battery; when the standard charge-discharge cycle number is # imgabs0 #, the historical capacity sequence of the previous N charge-discharge cycles is used as the input parameter of the first network model, and the first network model outputs the capacity corresponding to the subsequent multiple charge-discharge cycles; when the number of standard charge and discharge cycles is # imgabs1 #, decomposing the historical capacity sequence of the first N charge and discharge cycles into a fluctuation component and a trend component by using a modal decomposition method, inputting the fluctuation component and the trend component into a second network model, and obtaining the capacity corresponding to the subsequent charge and discharge cycles through the second network model; and # imgabs2 represents a threshold value of the number of times of charging and discharging.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy batteries, and in particular relates to a new energy battery capacity prediction method. Background Art

[0002] The existing technology for predicting new energy batteries basically uses the historical battery capacity sequence of charge and discharge cycles to predict the battery capacity after future charge and discharge cycles. However, the existing technology generally only considers the impact of the historical charge and discharge times on the battery capacity, and does not consider the impact of the usage time (total usage time) on the battery capacity. This results in a large error in the predicted capacity of new energy batteries with the same number of charge and discharge cycles but different usage time (total usage time). Summary of the invention

[0003] The purpose of the present invention is to provide a new energy battery capacity prediction method, which can improve the accuracy of the prediction result by converting the number of new energy charge and discharge cycles into the standard charge and discharge cycle number and predicting the capacity of the new energy battery according to the standard charge and discharge cycle number.

[0004] The technical solution provided by the present invention is: A new energy battery capacity prediction method, comprising: Obtain the number of charge and discharge cycles of the new energy battery, the capacity data of each charge and discharge cycle, and the usage time of the new energy battery; Determine the standard number of charge and discharge cycles of the new energy battery according to the number of charge and discharge cycles and the usage time; Constructing a first network model and a second network model for predicting the capacity of new energy batteries; When the standard charge and discharge cycle number When the historical capacity sequence of the previous N charge-discharge cycles is used as an input parameter of the first network model, the first network model outputs the capacity corresponding to the subsequent multiple charge-discharge cycles; When the standard charge and discharge cycle number When the historical capacity sequence of the previous N charge-discharge cycles is decomposed into a fluctuation component and a trend component by using a modal decomposition method, the fluctuation component and the trend component are respectively used as input parameters of the second network model, and the capacity corresponding to the subsequent multiple charge-discharge cycles is obtained by the second network model; in, Indicates the charge and discharge times threshold.

[0005] Preferably, a BP neural network is used to determine the standard charge and discharge cycle number of the new energy battery; The input parameters of the BP neural network are the number of charge and discharge cycles and the usage time of the new energy battery, and the output parameter is the standard number of charge and discharge cycles of the new energy battery.

[0006] Preferably, the value range of the charge and discharge times threshold is .

[0007] Preferably, the first network model adopts an LSTM network model.

[0008] Preferably, the second network model adopts the Holt-Winters network model.

[0009] Preferably, the CEEMDAN method is used to decompose the historical capacity sequence of the previous N charge and discharge cycles into a fluctuation component and a trend component.

[0010] Preferably, the method for constructing the training sample set of the BP neural network is: Obtain the capacity of multiple conventional new energy batteries with different charge and discharge cycle times and different usage times; and obtain the capacity of standard new energy batteries with different charge and discharge cycle times; The standard new energy battery refers to a new energy battery with a charge and discharge cycle frequency of a standard frequency; Find a standard new energy battery with the same capacity as the conventional new energy battery, and use the number of charge and discharge cycles of the standard new energy battery as the standard number of charge and discharge cycles of the corresponding conventional new energy battery; Among them, the conventional new energy battery refers to a new energy battery that does not limit the battery charging and discharging frequency; The training samples are obtained by taking the number of charge and discharge cycles and the usage time of conventional new energy batteries as input sample sets and taking the standard number of charge and discharge cycles of conventional new energy batteries as output sample sets.

[0011] Preferably, the standard frequency is a charge and discharge cycle number of 10 times / month to 12 times / month.

[0012] The beneficial effects of the present invention are: The new energy battery capacity prediction method provided by the present invention can improve the accuracy of the prediction result by converting the number of new energy charge and discharge cycles into the standard charge and discharge cycle number, and using different prediction methods to predict the capacity of the new energy battery according to the standard charge and discharge cycle number. DETAILED DESCRIPTION

[0013] The present invention is described in further detail below so that those skilled in the art can implement it according to the description.

[0014] 1. Determine the standard charge and discharge cycle times of new energy batteries The number of charge and discharge cycles of the new energy battery, the corresponding battery capacity data of each charge and discharge cycle, and the usage time of the new energy battery are obtained; the corresponding battery capacity data of the obtained charge and discharge cycles constitute a historical capacity sequence. Among them, the number of charge and discharge cycles of the new energy battery refers to the number of historical charge and discharge cycles that have been completed; the usage time of the new energy battery refers to the total usage time of the new energy battery, that is, the time difference between the current moment and the moment when the new energy battery starts to be activated and used. In this embodiment, the usage time of the new energy battery is counted in "months", and the part that is less than a whole month is calculated by rounding off. Afterwards, the standard number of charge and discharge cycles of the new energy battery is determined based on the number of charge and discharge cycles and the usage time.

[0015] Since the actual use of new energy batteries is different, the actual use time of new energy batteries with the same number of charge and discharge cycles may vary a lot. For example, with the same number of charge and discharge cycles of 100 times, a new energy battery with a high frequency of use may be used for 6 months, and a new energy battery with a low frequency of use may be used for 24 months or even longer. Although the number of charge and discharge cycles is the same, the performance of the battery is quite different due to the different actual use time, that is, the different charge and discharge frequencies (times). For example, some batteries have a charge and discharge number of 100 times, but the actual use time is 24 months. Long-term placement causes the battery performance to decline, making the battery capacity decay much greater than the capacity decay of the standard new energy battery with more than 100 charge and discharge cycles; and too many short-term charge and discharge cycles will also lead to a decline in battery performance. Therefore, the present invention corrects the charge and discharge times of the new energy battery to obtain the standard charge and discharge cycle times of the new energy battery; then, according to the standard charge and discharge cycle times of the new energy battery, different prediction methods are selected for capacity prediction, which can avoid the prediction error caused by the difference in battery use frequency and improve the accuracy of new energy battery capacity prediction.

[0016] The standard new energy battery refers to a new energy battery with a charge and discharge cycle frequency of a standard frequency. The standard frequency is determined according to the common usage of the battery. In this embodiment, the standard frequency is set to: the number of charge and discharge cycles is 10 times / month to 12 times / month.

[0017] Since the influence of battery usage time on the standard charge and discharge cycle number is nonlinear, the error of using formula fitting is not easy to control. In this embodiment, BP neural network is used to determine the standard charge and discharge cycle number of new energy batteries. Among them, the input parameters of the BP neural network are the charge and discharge cycle number and usage time of the new energy battery, and the output parameter is the standard charge and discharge cycle number of the new energy battery. The specific process is: construct a three-layer BP neural network model, including an input layer, a hidden layer and an output layer; then construct a training sample set of the BP neural network, and the construction method is: obtain the capacity of conventional new energy batteries with multiple different charge and discharge cycle numbers and different usage times; and obtain the capacity of standard new energy batteries with different charge and discharge cycle numbers; wherein the standard new energy battery refers to a new energy battery with a charge and discharge cycle frequency of a standard frequency; find out the charge and discharge number corresponding to the standard new energy battery with the same capacity as the conventional new energy battery, and use the charge and discharge cycle number of the standard new energy battery as the standard charge and discharge cycle number of the corresponding conventional new energy battery; use the charge and discharge cycle number and usage time of the conventional new energy battery as the input sample set, and use the standard charge and discharge cycle number of the conventional new energy battery as the output sample set to obtain the training sample. The training samples are divided into a training set and a test set in a ratio of 8:2. The training set is used to train the BP neural network. After the training is completed, the trained BP neural network is tested using the test set. If the test result meets the acceptable error, the optimized BP neural network is obtained. After that, the optimized BP neural network can be used to determine the standard charge and discharge cycle number of the new energy battery.

[0018] 2. Determine and predict the capacity of new energy batteries based on the number of standard charge and discharge cycles Through research, it is found that most new energy batteries will be accompanied by capacity regeneration during the decline process, while the capacity regeneration phenomenon is not obvious in the initial use of new energy batteries. Therefore, the present invention does not consider the capacity regeneration phenomenon in the capacity prediction of the battery in the initial use, but considers the capacity regeneration phenomenon in the capacity prediction of new energy batteries after the initial use, which can improve the accuracy of capacity prediction while improving the prediction efficiency. It was explained earlier that due to the inequality between the number of battery charge and discharge cycles and the usage time, errors will be caused in the judgment of battery capacity (decline), so the standard number of charge and discharge cycles is used as the basis for judging the battery usage.

[0019] In this embodiment, in order to take into account both the prediction efficiency and accuracy of the battery capacity, two network models are used to predict the battery capacity.

[0020] In the early stage of battery use, that is, when the standard charge and discharge cycle number of new energy batteries When the historical capacity sequence of the previous N charge-discharge cycles is used as an input parameter of the first network model, the first network model outputs the capacity corresponding to the subsequent multiple charge-discharge cycles.

[0021] As a preference, the first network model adopts an LSTM network model.

[0022] After the initial use of the battery, that is, when the standard charge and discharge cycle number In order to avoid the influence of capacity regeneration on the prediction results, the historical capacity sequence of the previous N charge and discharge cycles is decomposed into a fluctuation component and a trend component using the modal decomposition method. The fluctuation component and the trend component are respectively used as input parameters of the second network model, and the second network model outputs the capacity corresponding to the subsequent multiple charge and discharge cycles.

[0023] As a preference, the second network model adopts the Holt-Winters network model.

[0024] in, Indicates the charge and discharge times threshold. Studies have shown that when the battery is charged and discharged within 150 times, the capacity regeneration phenomenon is not obvious. The value range is set to .

[0025] Preferably, the CEEMDAN method is used to decompose the historical capacity sequence of the previous N charge and discharge cycles into a fluctuation component and a trend component. The specific process is as follows.

[0026] Battery capacity sequence Add an adaptive white noise sequence that follows a normal distribution , then The capacity sequence after adding white noise is: ; In the formula, is the signal-to-noise ratio, Indicates the number of times white noise is added.

[0027] Using EMD conduct The modal components are decomposed into ;right The average is calculated to obtain the first modal component of CEEMDAN decomposition. , and then use the original battery capacity sequence minus Get the first residual sequence ,Right now: ; ; For the first residual sequence Add white noise again to form a new sequence, decompose it using EMD, and then take the average value to get the second modal component and the second residual sequence .

[0028] The above-mentioned method for obtaining components and residual sequences is used to continue to obtain the modal components and residual signals of the next stage. The decomposition is completed when the obtained residual signal cannot be further decomposed or meets the EMD decomposition termination condition (the residual component has at most two extreme points). At this time, the original capacity sequence is decomposed into IMF components and a residual sequence ,Right now: ; In the formula, Indicates modal components.

[0029] The Holt-Winters model is an improved time series forecasting model based on the exponential smoothing model. Its core idea is to decompose the nonlinear series into long-term trends, seasonal or periodic fluctuation series and random fluctuation changes, and then combine it with the exponential smoothing method to build a forecasting model to achieve extrapolated forecasts of each decomposed series. It is mostly used for time series forecasting with obvious changing trends and quasi-periodic changes. Common Holt-Winters models include seasonality-free models, additive models and multiplicative models.

[0030] In this embodiment, the trend component ( ) The Holt-Winters non-seasonal model is used for forecasting, and the Holt-Winters additive model is used for forecasting each fluctuation component (IMF). The forecast results of each component are then reconstructed to obtain the final capacity forecast result.

[0031] The formula for calculating the initial value of the Holt-Winters seasonless model is: ; Where: For the sequence The trend component at the moment can be understood as the nonlinear downward trend component of the battery capacity sequence at the The size of the sample points; For the battery capacity sequence The stable component at the moment can be understood as the remaining random fluctuation component in the lithium battery capacity sequence at the The size of the sample points; for The observed value at time; The length of the season.

[0032] The prediction formula is: ; In the formula, , is the smoothing parameter, and its value is between; for The predicted value at the moment; Indicates the number of forecast periods; For the sequence The trend component of the moment, For the battery capacity sequence The stable component of the moment.

[0033] The initial value calculation formula of the Holt-Winters additive model is: ; In the formula, For the sequence The seasonal component at the time can be understood as the quasi-periodic change component caused by the capacity recovery phenomenon of the lithium battery capacity sequence at the The size of the sample points.

[0034] The prediction formula is: ; In the formula, , , is the smoothing parameter, and its value is between; and are the seasonal components of the sequence at t-L and t-L+m, respectively, which can be understood as the values ​​of the quasi-periodic variation components caused by the capacity recovery phenomenon of the battery capacity sequence at the t-L and t-L+m sample points.

[0035] After the first network model and the second network model are trained with a sample set of pre-built battery charge and discharge sequences, the first network model and the second network model can be used to predict the battery capacity. The training process uses the capacity sequence of standard new energy batteries as training samples. For the battery data set decomposed by CEEMDAN, the prediction results of each component output by the second network model are superimposed to obtain the final capacity prediction result; for the battery capacity sequence in the early stage of use that has not been decomposed, its prediction result is directly output through the first network model.

[0036] The present invention is further described below in conjunction with the embodiments and comparative examples. Due to the long charge and discharge cycle, the embodiments and comparative examples both use data of the same model with different charging frequencies uploaded by the online RTM system of the automobile as the sample set. Example

[0037] The battery capacity is predicted by dividing the battery into standard charge and discharge cycles at the initial stage of use.

[0038] Obtain the number of charge and discharge cycles of the new energy battery, the capacity data of each charge and discharge cycle, and the use time of the new energy battery; the standard frequency is set to: the number of charge and discharge cycles is 10 times / month; use the BP neural network to determine the standard number of charge and discharge cycles of the new energy battery. In the initial stage of use, 15 capacity sequences are randomly selected before 120 times, and the historical capacity sequences of the previous 10 charge and discharge cycles are used as the input parameters of the LSTM network model. The first network model outputs the capacity corresponding to the next 5 charge and discharge cycles. When the capacity sequence of 15 times is randomly selected after 120 times, the historical capacity sequence of the first 10 charge-discharge cycles is decomposed into a fluctuation component and a trend component by using the modal decomposition method. The fluctuation component and the trend component are respectively used as the input parameters of the Holt-Winters model, and the capacity corresponding to the next 5 charge-discharge cycles is obtained from the output of the Holt-Winters model.

[0039] Comparative Example The actual number of charge and discharge cycles is used to divide the battery capacity in the initial stage of use and predict the battery capacity.

[0040] Get the number of charge and discharge cycles of new energy batteries, the capacity data of each charge and discharge cycle, and the usage time of new energy batteries. In the initial stage of use, 15 capacity sequences are randomly selected before 120 times, and the historical capacity sequences of the previous 10 charge and discharge cycles are used as the input parameters of the LSTM network model. The LSTM network model outputs the capacity corresponding to the next 5 charge and discharge cycles. When the capacity sequence of 15 times is randomly selected after 120 times, the historical capacity sequence of the first 10 charge-discharge cycles is decomposed into a fluctuation component and a trend component by using the modal decomposition method. The fluctuation component and the trend component are respectively used as the input parameters of the Holt-Winters model, and the capacity corresponding to the next 5 charge-discharge cycles is obtained from the output of the Holt-Winters model.

[0041] 400 groups of tests were conducted in the embodiment and the comparative example, respectively, of which 200 groups were conducted in the initial use of the battery and 200 groups were conducted after the initial use of the battery. Mean absolute error (MAE) and root mean square error (RMSE) were used as evaluation criteria for the battery capacity prediction results. The test results are shown in Table 1.

[0042] Table 1 Test results

[0043] As can be seen from Table 1, the prediction result error of the embodiment using the method of the initial use of the battery provided by the present invention to divide the use cycle and then conduct the test is significantly smaller than the test error of the comparative example. Moreover, the error result of the initial use of the battery in the embodiment is closer to the error result after the initial use. In the comparative example, the error result of the initial use of the battery is greater than the result after the initial use. This may be due to the fact that the selected battery sequence data has a low charge and discharge frequency, resulting in the fact that the new energy battery with a small number of charge and discharge cycles but a long actual use time in the comparative example is judged to be in the initial use of the battery. The capacity regeneration phenomenon that accompanies the battery degradation process is not taken into account during the prediction, resulting in a large error in the prediction result.

[0044] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the embodiments shown and described herein.

Claims

1. A new energy battery capacity prediction method, characterized in that: include: Obtain the number of charge and discharge cycles of new energy batteries, the capacity data of each charge and discharge cycle, and the usage time of new energy batteries; Determine the standard number of charge and discharge cycles of the new energy battery according to the number of charge and discharge cycles and the usage time; Constructing a first network model and a second network model for predicting the capacity of new energy batteries; When the standard charge and discharge cycle number When the historical capacity sequence of the previous N charge-discharge cycles is used as an input parameter of the first network model, the first network model outputs the capacity corresponding to the subsequent multiple charge-discharge cycles; When the standard charge and discharge cycle number When the historical capacity sequence of the previous N charge-discharge cycles is decomposed into a fluctuation component and a trend component by using a modal decomposition method, the fluctuation component and the trend component are respectively used as input parameters of the second network model, and the capacity corresponding to the subsequent multiple charge-discharge cycles is obtained by the second network model; in, Indicates the charge and discharge times threshold.

2. The new energy battery capacity prediction method according to claim 1, characterized in that: Use BP neural network to determine the standard charge and discharge cycle times of new energy batteries; The input parameters of the BP neural network are the number of charge and discharge cycles and the usage time of the new energy battery, and the output parameter is the standard number of charge and discharge cycles of the new energy battery.

3. The new energy battery capacity prediction method according to claim 2, characterized in that: The value range of the charge and discharge times threshold is .

4. The new energy battery capacity prediction method according to claim 3, characterized in that: The first network model adopts the LSTM network model.

5. The new energy battery capacity prediction method according to claim 3 or 4, characterized in that: The second network model adopts the Holt-Winters network model.

6. The new energy battery capacity prediction method according to claim 5, characterized in that: The CEEMDAN method is used to decompose the historical capacity series of the previous N charge and discharge cycles into a fluctuation component and a trend component.

7. The new energy battery capacity prediction method according to claim 2, characterized in that: The method for constructing the training sample set of the BP neural network is: Obtain the capacity of multiple conventional new energy batteries with different charge and discharge cycle times and different usage times; and obtain the capacity of standard new energy batteries with different charge and discharge cycle times; The standard new energy battery refers to a new energy battery with a charge and discharge cycle frequency of a standard frequency; Find a standard new energy battery with the same capacity as the conventional new energy battery, and use the number of charge and discharge cycles of the standard new energy battery as the standard number of charge and discharge cycles of the corresponding conventional new energy battery; Among them, the conventional new energy battery refers to a new energy battery that does not limit the battery charging and discharging frequency; The training samples are obtained by taking the number of charge and discharge cycles and the usage time of conventional new energy batteries as input sample sets and taking the standard number of charge and discharge cycles of conventional new energy batteries as output sample sets.

8. The new energy battery capacity prediction method according to claim 7, characterized in that: The standard frequency is that the number of charge and discharge cycles is 10 times / month to 12 times / month.

Images