Lithium ion battery health state estimation method of variable voltage window in random discharge scene

Abstract

The invention discloses a lithium ion battery health state estimation method of a variable voltage window in a random discharge scene, and the method comprises the steps: obtaining a complete discharge voltage window through the dynamic discharge condition of an electric vehicle lithium ion battery, and dividing the complete discharge voltage window into a plurality of segmented segment voltage windows; extracting battery parameters under a voltage window of a part of segmentation fragments, taking the battery parameters as health characteristic parameters, and performing normalization processing on the health characteristic parameters; the electric vehicle lithium ion battery state discharge working condition is divided into a complete discharge voltage window and a segmented segment voltage window, and four health characteristic parameters of accumulated discharge capacity, discharge energy, charge energy and fused discharge energy are extracted from the voltage windows with different lengths. And the discharge voltage at least comprises one shortest voltage window, so that a relatively accurate SOH estimation result can be obtained. The requirement of a partial discharge voltage range can be met, and the requirement of the partial discharge voltage range generated in the random discharge process of the electric vehicle is met.

Description

Technical Field

[0001] The present invention belongs to the field of battery health state estimation and relates to a lithium-ion battery health state estimation method, specifically a lithium-ion battery health state estimation method with a variable voltage window under a random discharge scenario. Background Art

[0002] As the direct energy source for new energy vehicles, lithium-ion batteries play a decisive role in the vehicle's power, safety, and durability. Therefore, the state of health (SOH) of lithium-ion batteries has become a key focus in battery management systems. In the actual use of electric vehicles, users' charging and discharging behaviors are random. SOH estimation is typically based on historical data of lithium-ion batteries' full charge and discharge cycles. Due to the limitations of random user charging and discharging, acquiring historical data for SOH estimation is difficult, time-consuming, and has low accuracy.

[0003] Therefore, there is an urgent need for a health status estimation method for lithium-ion batteries with a variable voltage window under random discharge scenarios. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a lithium-ion battery health state estimation method with a variable voltage window under random discharge scenarios, so as to solve the technical problems in the existing technology that electric vehicles are restricted by random charging and discharging by users, and SOH estimation is time-consuming and has low accuracy.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A method for constructing a health status estimation model for lithium-ion batteries under different voltage windows includes the following steps:

[0007] Step 1: obtaining a complete discharge voltage window through a dynamic discharge condition of an electric vehicle lithium-ion battery, and dividing the complete discharge voltage window into a plurality of segmented voltage windows;

[0008] Step 2: extract the battery parameters under the voltage window of some segmented segments, use them as health characteristic parameters, and perform normalization processing on them. 80% of the processed health characteristic parameters are used as a training set; 20% are used as a test set. The health characteristic parameters C include cumulative discharge capacity, cumulative discharge energy, cumulative charge energy, and fusion discharge energy.

[0009] The following formula is used to calculate the correlation P between each health characteristic parameter and the battery health status: CH ;

[0010]

[0011] in:

[0012] C i is the value of the health characteristic parameter in the i-th cycle of the battery parameter;

[0013] is the average value of all health characteristic parameters in the battery parameters;

[0014] H i is the value of the battery health status in the battery parameters;

[0015] is the average value of the battery health status in the battery parameters;

[0016] n is the number of charge and discharge cycles;

[0017] Step 3, construct the TIB estimation model;

[0018] The TIB estimation model includes a sequentially connected feature attention mechanism, a two-layer temporal convolutional network, an informer encoder, a two-layer BiLSTM, a temporal attention mechanism, and a fully connected layer; the BiLSTM includes a forward LSTM and a backward LSTM; the temporal convolutional network includes causal convolution and dilated convolution;

[0019] The feature attention mechanism is used to dynamically adjust the weights of the input health feature parameters to highlight the relevance P CH The largest health feature parameter and the influence of irrelevant or noise features are suppressed to obtain the health feature parameter after dynamic weighting;

[0020] The correlation P CH Used to provide initial weights for the feature attention mechanism;

[0021] The temporal convolutional network is used to preliminarily extract the temporal features of the dynamically weighted health feature parameters and enrich the dimensions of the temporal features to obtain local temporal features;

[0022] The Informer encoder is used to convert local time series features into global time series features S;

[0023] The BiLSTM is used to combine the forward LSTM and backward LSTM outputs to capture the bidirectional dependency in the global time series feature S and output the global time series feature S';

[0024] The temporal attention mechanism is used to assign attention weights to the historical moment information carried in the global temporal feature S' of BiLSTM and output the global temporal feature S".

[0025] The fully connected layer is used to map the global time series feature S' into a single SOH estimation value through linear transformation and nonlinear activation, and output the SOH estimation value of the lithium-ion battery, that is, the battery health status;

[0026] Step 4: Combine the training set obtained in step 2 with the correlation P CH The health status is used as the input of the TIB estimation model and the output of the TIB estimation model, and the TIB estimation model is trained to obtain a trained TIB estimation model.

[0027] Step 1 includes the following steps:

[0028] Step 1.1, charging the electric vehicle lithium-ion battery using a constant current and constant voltage charging method to obtain a charging cut-off voltage V;

[0029] Step 1.2, charging the electric vehicle lithium-ion battery in a constant voltage manner to obtain a lower cutoff current I, stopping charging and allowing the battery to stand;

[0030] Step 1.3, using an equivalent short-time test condition to discharge the electric vehicle lithium-ion battery to the lower cut-off voltage, and then let it stand;

[0031] Step 1.4: Repeat steps 1.1 to 1.3 until the battery health state decays to 80%, and then obtain the battery parameters; calculate the actual value of the battery health state y based on the battery parameters. i ;

[0032] The battery parameters include the current maximum charging capacity of the battery;

[0033]

[0034] in:

[0035] C cur is the current maximum charging capacity;

[0036] C nom It is the factory nominal capacity;

[0037] In step 1.5, the complete discharge voltage window is divided into multiple segmented voltage windows using a 0.1 V voltage sliding window.

[0038] A method for estimating the health status of a lithium-ion battery based on different voltage windows includes:

[0039] The test set obtained in step 2 of the method for constructing the estimation model of the health status of lithium-ion batteries under different voltage windows is input into the trained TIB estimation model obtained in step 4, and the battery health status is output.

[0040] Compared with the prior art, the present invention has the following beneficial technical effects:

[0041] (I) The present invention divides the discharge condition of electric vehicle lithium-ion batteries into a complete discharge voltage window and a segmented voltage window. It extracts the cumulative discharge capacity, discharge energy, and charge energy from voltage windows of different lengths, and integrates the four health characteristic parameters of discharge energy. Since the lithium-ion battery discharge stage is segmented according to the voltage window length, the discharge voltage contains at least one minimum voltage window, which can obtain a relatively accurate SOH estimation result. This can meet the requirements of the partial discharge voltage range and solve the problem of the partial discharge voltage range generated by the random discharge process of electric vehicles.

[0042] (II) In terms of SOH estimation accuracy, the present invention constructs a TIB estimation model. This model uses a two-layer BiLSTM neural network to simultaneously learn bidirectional data information, further exploring the potential relationship between the current lithium-ion battery capacity decay data and the past and future data, thereby improving the utilization rate of lithium-ion battery capacity decay data and the accuracy of lithium-ion battery health status prediction.

[0043] (III) In terms of the time consumption of SOH estimation, the present invention constructs a TIB estimation model. The use of temporal convolutional network product and informer encoder in this model can effectively reduce the computational complexity of the model and improve the computational efficiency of the TIB estimation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of a method for constructing an estimation model of the health status of a lithium-ion battery under different voltage windows according to the present invention;

[0045] Figure 2 It is a structural diagram of the TIB estimation model in the present invention.

[0046] The specific contents of the present invention are further explained in detail below with reference to the embodiments. DETAILED DESCRIPTION

[0047] It should be noted that, unless otherwise specified, all components in the present invention are components known in the art.

[0048] Specific embodiments of the present invention are given below. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent modifications made on the basis of the technical solution of this application fall within the protection scope of the present invention.

[0049] The present invention provides a method for constructing an estimation model of the health status of lithium-ion batteries under different voltage windows, such as Figure 1 As shown, the following steps are included:

[0050] Step 1: obtaining a complete discharge voltage window through a dynamic discharge condition of an electric vehicle lithium-ion battery, and dividing the complete discharge voltage window into a plurality of segmented voltage windows;

[0051] Step 2: Extract the battery parameters under the voltage window of some segmented segments and use them as health characteristic parameters. After normalization, 80% of the processed health characteristic parameters are used as the training set; 20% are used as the test set; the average value of the battery health status in the parameters is calculated; the health characteristic parameters C include cumulative discharge capacity, cumulative discharge energy, cumulative charge energy and fusion discharge energy;

[0052] The following formula is used to calculate the correlation P between each health characteristic parameter and the battery health status: CH ;

[0053]

[0054] in:

[0055] C i is the value of the health characteristic parameter in the i-th cycle of the battery parameter;

[0056] is the average value of all health characteristic parameters in the battery parameters;

[0057] H i is the value of the battery health status in the battery parameters;

[0058] is the average value of the battery health status in the battery parameters;

[0059] n is the number of charge and discharge cycles;

[0060] Step 3, construct the TIB estimation model;

[0061] The average value of the battery health status in the parameters; the TIB estimation model includes a sequentially connected feature attention mechanism, a two-layer temporal convolutional network, an informer encoder, a two-layer BiLSTM, a temporal attention mechanism, and a fully connected layer; the average value of the battery health status in the parameters; the BiLSTM includes forward LSTM and backward LSTM; the average value of the battery health status in the parameters; the temporal convolutional network includes causal convolution and dilated convolution;

[0062] The average value of the battery health status in the parameter; the feature attention mechanism is used to dynamically adjust the weight of the input health feature parameters to highlight the correlation P CH The largest health feature parameter and the influence of irrelevant or noise features are suppressed to obtain the health feature parameter after dynamic weighting;

[0063] Average value of battery health status in parameters; correlation PCH Used to provide initial weights for the feature attention mechanism;

[0064] The average value of the battery health status in the parameters; the temporal convolutional network is used to preliminarily extract the time characteristics of the dynamically weighted health characteristic parameters and enrich the dimensions of the time characteristics to obtain local time series features;

[0065] The average value of the battery health status in the parameter; the Informer encoder is used to convert local time series features into global time series features S;

[0066] The average value of the battery health status in the parameter; BiLSTM is used to combine the forward LSTM and backward LSTM outputs to capture the bidirectional dependency in the global time series feature S and output the global time series feature S';

[0067] The average value of the battery health status in the parameter; the temporal attention mechanism is used to assign attention weights to the historical moment information carried in the global temporal feature S' of BiLSTM and output the global temporal feature S".

[0068] The average value of the battery health status in the parameters; the fully connected layer is used to map the global time series feature S" into a single SOH estimate through linear transformation and nonlinear activation, and output the SOH estimate of the lithium-ion battery, that is, the battery health status;

[0069] Step 4: Combine the training set obtained in step 2 with the correlation P CH The health status is used as the input of the TIB estimation model and the output of the TIB estimation model, and the TIB estimation model is trained to obtain a trained TIB estimation model.

[0070] In step 2, the reasons for selecting cumulative discharge capacity, cumulative discharge energy, cumulative charge energy, and fusion discharge energy as health characteristic parameters are as follows:

[0071] Capacity, as the most direct characteristic parameter to ensure battery performance degradation, is often used in the SOH estimation process. Therefore, the cumulative discharge capacity is selected as the health characteristic parameter.

[0072] As the number of aging cycle discharges increases, the cumulative discharge energy based on short-term operating conditions continues to decrease. It directly reflects the amount of energy released by the battery during the discharge stage and also characterizes the aging of the battery. Therefore, it is used as a health characteristic parameter.

[0073] The selected short-term operating conditions include the coasting-braking segments, so there is a braking energy recovery process in these two parts, which generates the charging energy. This characteristic parameter can characterize the battery health status under the aging cycle state of the lithium battery, so the accumulated charging energy is used as the health characteristic parameter.

[0074] Establish the fusion relationship between voltage V and capacity Cc under single short-time working condition, and obtain the fusion discharge energy C wp , single short-term working condition C wp As the battery ages, it shows a decreasing trend, so it can be used as a health characteristic parameter, C wp =V*C c ;

[0075] After selecting the characteristic parameters, all health characteristic parameters in the data set are normalized to the interval [0, 1] by using the characteristic parameter scaling method. The health characteristic scaling method used in the present invention is the max-min scaling method:

[0076]

[0077] x i Represents a health characteristic parameter of the i-th charge and discharge cycle;

[0078] In order to verify the correlation between the extracted health characteristic parameters and the battery health status, the present invention selects the statistical Pearson correlation coefficient (PCC) to calculate the correlation between the four health characteristic parameters and SOH under different voltage sliding windows;

[0079] The accuracy and robustness of the estimation method are verified using aging test data of batteries with different capacities.

[0080] The global time series features simultaneously learn the time series relationship between historical and future moments through forward LSTM and backward LSTM, further exploring the potential relationship between the current lithium-ion battery capacity decay data and the data of the past and future moments, thereby improving the utilization rate of lithium-ion battery capacity decay data and the accuracy of lithium-ion battery health status prediction.

[0081] The Informer encoder uses multi-head probabilistic sparse self-attention and distillation mechanisms to capture the global trend of battery health degradation while optimizing computational efficiency.

[0082] On the one hand, the temporal attention mechanism distinguishes the importance of different health feature parameters to the prediction of SOH at the current moment. On the other hand, it can autonomously extract the temporal information between the input data at historical moments and enhance the expression of health feature parameters at critical moments.

[0083] Through dense connections, the fully connected layer deeply integrates the local timing features of the temporal convolutional network (TCN), the global dependencies of the informer, the contextual information of the BiLSTM, and the key moment weights of the temporal attention mechanism to form a comprehensive representation of the battery health status.

[0084] In terms of SOH estimation accuracy, a TIB estimation model was constructed. This model uses a two-layer BiLSTM neural network to simultaneously learn two-way data information, further exploring the potential relationship between the current lithium-ion battery capacity decay data and the past and future data, thereby improving the utilization rate of lithium-ion battery capacity decay data and the accuracy of lithium-ion battery health status prediction.

[0085] In terms of the time consumption of SOH estimation, a TIB estimation model was constructed. The use of temporal convolutional network product and Informer encoder in this model can effectively reduce the computational complexity of the model and improve the computational efficiency of the TIB estimation model.

[0086] Specifically, step 1 includes the following steps:

[0087] Step 1.1, charging the electric vehicle lithium-ion battery using a constant current and constant voltage charging method to obtain a charging cut-off voltage V;

[0088] Step 1.2, charging the electric vehicle lithium-ion battery in a constant voltage manner to obtain a lower cutoff current I, stopping charging and allowing the battery to stand;

[0089] Step 1.3, using an equivalent short-time test condition to discharge the electric vehicle lithium-ion battery to the lower cut-off voltage, and then let it stand;

[0090] Step 1.4: Repeat steps 1.1 to 1.3 until the battery health state decays to 80%, and then obtain the battery parameters; calculate the actual value of the battery health state y based on the battery parameters. i ;

[0091] The average value of the battery health status in the parameters; the battery parameters include the current maximum charge capacity of the battery;

[0092]

[0093] in:

[0094] C cur is the current maximum charging capacity;

[0095] C nom It is the factory nominal capacity;

[0096] In step 1.5, the complete discharge voltage window is divided into multiple segmented voltage windows using a 0.1 V voltage sliding window.

[0097] In the above technical solution, for each charge and discharge cycle of a lithium battery, it is difficult to obtain the estimated SOH value of the lithium-ion battery over the entire discharge life cycle due to the variable operating modes of the battery and the random discharge voltage range. Therefore, compared with the current lithium battery SOH estimation, which is only applicable to a fixed voltage segment;

[0098] The discharge conditions of electric vehicle lithium-ion batteries are divided into a complete discharge voltage window and a segmented voltage window. The cumulative discharge capacity, discharge energy, and charge energy are extracted from voltage windows of different lengths, and the four health characteristic parameters of discharge energy are integrated. Since the lithium-ion battery discharge stage is segmented according to the voltage window length, the discharge voltage contains at least one minimum voltage window, which can obtain a more accurate SOH estimation result. This can meet the requirements of the partial discharge voltage range and solve the problem of the partial discharge voltage range generated by the random discharge process of electric vehicles.

[0099] In step 1.3, the equivalent short-time test conditions can be found in Jiao Zhipeng's paper "Development of a Rapid Inspection Driving Cycle for Battery Electric Vehicles Based on Operational Safety".

[0100] Preferably, constant current 0.5C rate.

[0101] In the above technical solution, causal convolution is used to utilize the local characteristics of the time series in the lithium-ion battery capacity decay data, and compared with other convolutions, it can effectively avoid the problem of historical information leakage; dilated convolution is used to reduce the computational complexity of the model and improve the computational efficiency of the TIB estimation model.

[0102] The present invention provides a method for estimating the health status of a lithium-ion battery based on different voltage windows, comprising:

[0103] The average value of the battery health status in the parameters; the test set obtained in step 2 of the method for constructing the estimation model of the lithium-ion battery health status under different voltage windows is input into the trained TIB estimation model obtained in step 4, and the battery health status is output.

[0104] In the above technical solution, three indicators, namely root mean square error (RMSE), mean absolute error (MAE) and relative percentage error (MAPE), are used to evaluate the constructed TIB estimation model.

[0105]

[0106] RMSE represents the average size of the measurement error. The smaller the value, the smaller the error. MAE represents the indicator of prediction deviation. The smaller the value, the better. MAPE represents the relative error percentage between the true value and the predicted value. The smaller the value, the better the prediction performance.

[0107] Actual measurement example:

[0108] This experimental example presents a method for estimating the state of health (SOH) of lithium-ion batteries (Li-ion batteries) under random discharge scenarios with a variable voltage window. Using steps 1 through 4 above, the SOH estimation errors for different voltage windows are shown in Table 1. The results demonstrate that the SOH estimated by the TIB estimation model is very close to the true value. In the long voltage window [4.1V, 2.5V], the SOH estimation error is kept within 0.5%. Based on the shortest voltage window [4.1V, 4.0V], the maximum error between the estimated SOH and the true value is 0.72%, and the TIB model also maintains high SOH estimation accuracy. As the voltage window length increases, the TIB model captures richer health feature information across different voltage window intervals. The TIB estimation model receives more health feature parameters as input, and its estimation accuracy increases with the voltage window length. During the actual discharge process of an electric vehicle, there's no need to analyze the battery's full lifecycle discharge behavior. Based on the SOH estimation results of the TIB model under different voltage windows, it's clear that the SOH estimation of a lithium battery can be completed using only the health characteristic parameters extracted from the shortest voltage window [4.1V, 4.0V], with an SOH estimation error within 0.72%. This significantly saves time and financial resources and enables accurate SOH estimation. The SOH estimation error is controlled within 0.5% using the full discharge voltage window [4.1V, 2.5V] for the lithium battery, and within 0.72% using the shortest voltage window [4.1V, 4.0V] for the shortest voltage window, demonstrating the model's high estimation accuracy.

[0109] Table 1 Comparison of SOH estimation errors under different voltage windows of the verification battery

[0110]

[0111]

[0112] Comparative Example:

[0113] The TIB estimation model obtained by the present invention is compared with other lithium-ion battery SOH estimation technologies published in recent years. As shown in Table 2, using the long short-term memory neural network as the SOH estimation model, the final SOH estimation error RMSE is 0.3874%, which is lower than the error level of the present invention. However, the method proposed by the present invention performs SOH estimation under the shortest voltage window, and there is no need to extract multiple health feature parameters based on the complete charge and discharge process of the battery, which saves computing efficiency and SOH estimation time. Compared with machine learning methods such as the interpretation enhancement machine (EBM) for SOH estimation, the final RMSE estimation errors are 1.49% and 0.776, which are both higher than the RMSE level of the present invention. Based on the feedforward neural network, a mapping relationship between health indicators and battery SOH is established. Finally, the SOH estimation RMSE error of all data sets is controlled within 0.99%, which is higher than the error size of the present invention.

[0114] Table 2 SOH estimation results of existing lithium-ion batteries

[0115]

Claims

1. A method for constructing an estimation model of the health status of a lithium-ion battery under different voltage windows, characterized in that: The following steps are involved: Step 1: obtaining a complete discharge voltage window through a dynamic discharge condition of an electric vehicle lithium-ion battery, and dividing the complete discharge voltage window into a plurality of segmented voltage windows; Step 2: extract the battery parameters under the voltage window of some segmented segments, use them as health feature parameters, perform normalization on them, and use 80% of the processed health feature parameters as the training set; 20% as a test set; the health characteristic parameters C include cumulative discharge capacity, cumulative discharge energy, cumulative charging energy and fusion discharge energy; The following formula is used to calculate the correlation between each health characteristic parameter and the battery health status: P CH ; in: C i is the value of the health characteristic parameter in the i-th cycle of the battery parameter; is the average value of all health characteristic parameters in the battery parameters; H i is the value of the battery health status in the battery parameters; is the average value of the battery health status in the battery parameters; n is the number of charge and discharge cycles; Step 3, construct the TIB estimation model; The TIB estimation model includes a sequentially connected feature attention mechanism, a two-layer temporal convolutional network, an informer encoder, a two-layer BiLSTM, a temporal attention mechanism, and a fully connected layer; the BiLSTM includes a forward LSTM and a backward LSTM; the temporal convolutional network includes causal convolution and dilated convolution; The feature attention mechanism is used to dynamically adjust the weights of the input health feature parameters to highlight the relevance P CH The largest health feature parameter and the influence of irrelevant or noise features are suppressed to obtain the health feature parameter after dynamic weighting; The correlation P CH Used to provide initial weights for the feature attention mechanism; The temporal convolutional network is used to preliminarily extract the temporal features of the dynamically weighted health feature parameters and enrich the dimensions of the temporal features to obtain local temporal features; The Informer encoder is used to convert local time series features into global time series features S; The BiLSTM is used to combine the forward LSTM and backward LSTM outputs to capture the bidirectional dependency in the global time series feature S and output the global time series feature S'; The temporal attention mechanism is used to assign attention weights to the historical moment information carried in the global temporal feature S' of BiLSTM and output the global temporal feature S". The fully connected layer is used to map the global time series feature S' into a single SOH estimation value through linear transformation and nonlinear activation, and output the SOH estimation value of the lithium-ion battery, that is, the battery health status; Step 4: Combine the training set obtained in step 2 with the correlation P CH The health status is used as the input of the TIB estimation model and the output of the TIB estimation model, and the TIB estimation model is trained to obtain a trained TIB estimation model.

2. The method for constructing a health status estimation model of a lithium-ion battery based on different voltage windows according to claim 1, characterized in that: Step 1 includes the following steps: Step 1.1, charging the electric vehicle lithium-ion battery using a constant current and constant voltage charging method to obtain a charging cut-off voltage V; Step 1.2, charging the electric vehicle lithium-ion battery in a constant voltage manner to obtain a lower cutoff current I, stopping charging and allowing the battery to stand; Step 1.3, using an equivalent short-time test condition to discharge the electric vehicle lithium-ion battery to the lower cut-off voltage, and then let it stand; Step 1.4: Repeat steps 1.1 to 1.3 until the battery health state decays to 80%, and then obtain the battery parameters; calculate the actual value of the battery health state y based on the battery parameters. i ; The battery parameters include the current maximum charging capacity of the battery; in: C cur is the current maximum charging capacity; C nom It is the factory nominal capacity; In step 1.5, the complete discharge voltage window is divided into multiple segmented voltage windows using a 0.1 V voltage sliding window.

3. A method for estimating the health status of a lithium-ion battery based on different voltage windows, characterized in that: include: The test set obtained in step 2 of the method for constructing an estimation model of the health status of lithium-ion batteries under different voltage windows according to claim 2 is input into the trained TIB estimation model obtained in step 4, and the battery health status is output.

Images