A Battery Health Assessment Method and System Based on a Dual-Channel Attention Mechanism

Abstract

This invention proposes a battery health assessment method and system based on a dual-channel attention mechanism. This method integrates deep temporal modeling and degradation trajectory learning. By constructing a dual-channel data processing structure for real-time monitoring data and historical degradation analysis, it extracts a first health score representing the short-term health status of the target battery pack and a second health score representing the long-term degradation status of the target battery pack. At the same time, combined with a dynamic fusion weight function driven by operating conditions, the weights of the first and second health scores are flexibly adjusted according to the battery's operating stage (such as charge / discharge rate fluctuations and aging process). This enables high-precision assessment of the battery pack's health status and has adaptive adjustment capabilities to meet the health judgment needs under complex environments.

Description

Technical Field

[0001] This invention relates to the field of battery health assessment technology, specifically to a battery health assessment method and system based on a dual-channel attention mechanism. Background Technology

[0002] Currently, there are two main types of battery health evaluation schemes. One type is based on expert experience and domain knowledge to construct a battery health evaluation benchmark, which determines battery health by comparing real-time battery status data with the benchmark. For example, patent CN107561452A proposes a battery health assessment method based on grayscale correlation, which calculates the correlation degree as a health indicator by comparing battery operating parameters with a preset standard sequence. However, this type of scheme relies heavily on static rule settings and lacks adaptability to fluctuations in actual operating conditions, especially under dynamic conditions such as current surges and high temperatures, where the evaluation accuracy decreases significantly.

[0003] Another approach is to model the battery's health status based on historical data, using a pre-trained model to predict battery health based on real-time battery status data. However, this method mostly employs a static weighting strategy, which cannot flexibly adjust the weighting of different operating characteristics on the final health evaluation result according to the battery's operating stage (such as fluctuations in charge and discharge rates and the aging process), resulting in insufficient evaluation accuracy. Summary of the Invention

[0004] Purpose of the invention: The present invention aims to propose a battery health assessment method and system based on a dual-channel attention mechanism, so as to at least partially overcome the deficiencies of the prior art.

[0005] Summary of the Invention: To achieve the above objectives, the present invention proposes the following technical solution:

[0006] Firstly, a battery health assessment method based on a dual-channel attention mechanism is provided, including the following steps:

[0007] Acquire short-term state characteristic data and long-term historical degradation index data of the target battery pack;

[0008] The short-term state feature data is input into a pre-trained first health evaluation model to obtain a predicted first health score related to the available capacity of the target battery pack.

[0009] The long-term historical degradation index data is input into a pre-trained second health evaluation model to obtain a predicted second health score related to the available capacity of the target battery pack.

[0010] Obtain the current load fluctuation standard deviation of the target battery pack. and monthly capacity decay rate And based on the load fluctuation standard deviation and the monthly capacity decay rate Calculate the first fusion weight and the second fusion weight:

[0011] ;

[0012] ;

[0013] in, The first fusion weight, This is the second fusion weight. and For dynamically fitted parameters, and The update is performed using the Cultural Whale Optimization Algorithm, with the goal of minimizing the mean square error between the predicted and reference values ​​of the target battery pack health score within a preset historical time window.

[0014] The first health score and the second health score are weighted and summed based on the first fusion weight and the second fusion weight to obtain the predicted current health score S of the target battery pack:

[0015] ;

[0016] in, This represents the first health score. This represents the second health score.

[0017] As an optional implementation of the method described in the first aspect, the first health assessment model is an LSTM model; the method further includes:

[0018] Obtain the first training sample, which is a short-term state feature data sample of the sample battery pack.

[0019] Obtain the first measurement value of the available capacity of the sample battery pack within the sampling time window of the short-term state feature data sample, and construct the label of the first training sample based on the first measurement value of the available capacity;

[0020] The LSTM model is trained using the first training samples until a first health assessment model that meets the requirements is obtained.

[0021] As an optional implementation of the method described in the first aspect, the second health assessment model is the LightGBM model; the method further includes:

[0022] Obtain a second training sample, which is a long-term historical degradation index data sample of the sample battery pack.

[0023] Obtain the second measurement of available capacity of the sample battery pack at the last moment of the sampling time window of the long-term historical degradation index data sample, and construct the label of the second training sample based on the second measurement of available capacity;

[0024] The LightGBM model is trained using the second training sample until a second health assessment model that meets the requirements is obtained.

[0025] As an optional implementation of the method described in the first aspect, updating the dynamic fitting parameters using the cultural whale optimization algorithm specifically includes:

[0026] initialization and The parameter range;

[0027] For each sampling time point within the historical time window, the available capacity measurement value of the target battery pack at that sampling time point is obtained, and a reference value for the health score of the target battery pack is constructed based on the available capacity measurement value.

[0028] For each sampling time point within the historical time window, acquire the short-term state characteristic data and long-term historical degradation index data of the target battery pack corresponding to that sampling time point;

[0029] Input the short-term state feature data corresponding to each sampling time point into the first health evaluation model to obtain the first health score prediction value corresponding to the sampling time point;

[0030] Input the long-term historical decline index data corresponding to each sampling time point into the second health assessment model to obtain the second health score prediction value corresponding to the sampling time point;

[0031] Based on the reference value corresponding to the sampling time point, the predicted value of the first health score, and the predicted value of the second health score, the mean square error between the predicted value and the reference value of the target battery pack health score is constructed.

[0032] Searching within the parameter range with the objective of minimizing the mean square error. and The optimal value, and and The value is updated to the corresponding optimal value.

[0033] As an optional implementation of the method described in the first aspect, the method further includes:

[0034] If the load fluctuation standard deviation If the value exceeds the preset high volatility threshold, the value of the fusion weight is set to... , ;

[0035] If the monthly capacity decay rate If the value exceeds the preset lifetime threshold, the value of the fusion weight is set to 0.6.

[0036] Secondly, a battery health assessment system based on a dual-channel attention mechanism is provided, including:

[0037] The data acquisition module is used to collect short-term state characteristic data and long-term historical degradation index data of the target battery pack.

[0038] The real-time analysis module is used to input the short-term state feature data into a pre-trained first health evaluation model to obtain a predicted first health score related to the available capacity of the target battery pack.

[0039] The historical analysis module is used to input the long-term historical degradation index data into a pre-trained second health evaluation model to obtain a predicted second health score related to the available capacity of the target battery pack.

[0040] The dynamic weight generation module is used to obtain the current load fluctuation standard deviation of the target battery pack. and monthly capacity decay rate And based on the load fluctuation standard deviation and the monthly capacity decay rate Calculate the first fusion weight and the second fusion weight:

[0041] ;

[0042] ;

[0043] in, The first fusion weight, This is the second fusion weight. and For dynamically fitted parameters, and The update is performed using the Cultural Whale Optimization Algorithm, with the goal of minimizing the mean square error between the predicted and reference values ​​of the target battery pack health score within a preset historical time window.

[0044] The health fusion module is used to perform a weighted summation of the first health score and the second health score based on the first fusion weight and the second fusion weight to obtain the predicted current health score S of the target battery pack.

[0045] ;

[0046] in, This represents the first health score. This represents the second health score.

[0047] As an optional implementation of the system described in the second aspect, the first health assessment model is an LSTM model; the system further includes a first training module, which is used for:

[0048] Obtain the first training sample, which is a short-term state feature data sample of the sample battery pack.

[0049] Obtain the first measurement value of the available capacity of the sample battery pack within the sampling time window of the short-term state feature data sample, and construct the label of the first training sample based on the first measurement value of the available capacity;

[0050] The LSTM model is trained using the first training samples until a first health assessment model that meets the requirements is obtained.

[0051] As an optional implementation of the system described in the second aspect, the second health assessment model is a LightGBM model; the system further includes a second training module, the second training module being used for:

[0052] Obtain a second training sample, which is a long-term historical degradation index data sample of the sample battery pack.

[0053] Obtain the second measurement of available capacity of the sample battery pack at the last moment of the sampling time window of the long-term historical degradation index data sample, and construct the label of the second training sample based on the second measurement of available capacity;

[0054] The LightGBM model is trained using the second training sample until a second health assessment model that meets the requirements is obtained.

[0055] As an optional implementation of the system described in the second aspect, the system further includes a parameter self-optimization module, which is used to update the dynamically fitted parameters using the Cultural Whale Optimization Algorithm. The specific update steps include:

[0056] initialization and The parameter range;

[0057] For each sampling time point within the historical time window, the available capacity measurement value of the target battery pack at that sampling time point is obtained, and a reference value for the health score of the target battery pack is constructed based on the available capacity measurement value.

[0058] For each sampling time point within the historical time window, acquire the short-term state characteristic data and long-term historical degradation index data of the target battery pack corresponding to that sampling time point;

[0059] Input the short-term state feature data corresponding to each sampling time point into the first health evaluation model to obtain the first health score prediction value corresponding to the sampling time point;

[0060] Input the long-term historical decline index data corresponding to each sampling time point into the second health assessment model to obtain the second health score prediction value corresponding to the sampling time point;

[0061] Based on the reference value corresponding to the sampling time point, the predicted value of the first health score, and the predicted value of the second health score, the mean square error between the predicted value and the reference value of the target battery pack health score is constructed.

[0062] Searching within the parameter range with the objective of minimizing the mean square error. and The optimal value, and and The value is updated to the corresponding optimal value.

[0063] As an optional implementation of the system described in the second aspect, the dynamic weight generation module is further used for:

[0064] The standard deviation of load fluctuation When the value exceeds the preset high volatility threshold, the value of the fusion weight is set to... , ;

[0065] The monthly capacity decay rate When the value exceeds the preset lifetime threshold, the value of the fusion weight is set to 0.6.

[0066] Beneficial effects: Compared with existing technologies, the battery health assessment method based on a dual-channel attention mechanism proposed in this invention has the following advantages:

[0067] This method integrates deep temporal modeling and degradation trajectory learning. By constructing a dual-channel data processing structure for real-time monitoring data and historical degradation analysis, it extracts a first health score representing the short-term health status of the target battery pack and a second health score representing the long-term degradation status of the target battery pack. At the same time, combined with a dynamic fusion weight function driven by operating conditions, the weights of the first and second health scores are flexibly adjusted according to the battery's operating stage (such as charge / discharge rate fluctuations and aging process). This enables high-precision assessment of the battery pack's health status and has adaptive adjustment capabilities to meet the health judgment needs under complex environments.

[0068] The battery health assessment system based on a dual-channel attention mechanism proposed in this invention also possesses the aforementioned beneficial effects. Attached Figure Description

[0069] Figure 1 This is a schematic flowchart of a battery health assessment method based on a dual-channel attention mechanism, as described in an embodiment.

[0070] Figure 2 This is a specific implementation scenario involved in the embodiment. First fusion weight Standard deviation of load fluctuation and monthly capacity decay rate A schematic diagram of a changing three-dimensional surface.

[0071] Figure 3 This is a specific implementation scenario involved in the embodiment. Second fusion weight Standard deviation of load fluctuation and monthly capacity decay rate A schematic diagram of a changing three-dimensional surface.

[0072] Figure 4 This is a specific implementation scenario involved in the embodiment. First fusion weight Standard deviation of load fluctuation and monthly capacity decay rate A schematic diagram of changing contour lines.

[0073] Figure 5 This is a specific implementation scenario involved in the embodiment. Second fusion weight Standard deviation of load fluctuation and monthly capacity decay rate A schematic diagram of changing contour lines.

[0074] Figure 6This is a schematic diagram of a battery health assessment system based on a dual-channel attention mechanism, as described in an embodiment. Detailed Implementation

[0075] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. However, it should be understood that the present invention can be implemented in various forms. The exemplary and non-limiting embodiments shown in the drawings and described below are not intended to limit the invention to the specific embodiments illustrated.

[0076] It should be understood that, where technically feasible, the technical features listed above for different embodiments can be combined with each other to form other embodiments within the scope of this invention. Furthermore, the specific examples and embodiments described in this invention are non-limiting, and corresponding modifications can be made to the structures, steps, and order described above without departing from the protection scope of this invention.

[0077] Please refer to Figure 1 , Figure 1 An exemplary flowchart of a battery health assessment method based on a dual-channel attention mechanism is provided. Figure 1 As shown, the method includes steps S100 to S104.

[0078] S100: Obtain short-term state characteristic data and long-term historical degradation index data of the target battery pack.

[0079] The aforementioned short-term state characteristic data can be selected from state parameters that characterize the current operating state of the target battery pack, such as voltage V, current T, and temperature T. You can select a single parameter or multiple parameters.

[0080] When collecting the aforementioned short-term state characteristic data, a time window can be set. , Where represents the current time, This represents the span of the time window and is a very small positive number. (Data acquisition time window) The state parameters of the target battery pack within the range are used to obtain a set of time-series data, which is the aforementioned short-term state characteristic data.

[0081] Taking the voltage V, current T, and temperature T of the target battery pack as state parameters as an example, a time window length of 30 seconds is selected, and sampling is performed once per second. Each sampling performs three parameters [V, I, T]. Within the 30-second time window, a total of 30 samplings are performed to obtain a set of 90-dimensional time series data. This set of time series data is the short-term state characteristic data.

[0082] The aforementioned long-term historical degradation indicator data can be selected from indicators such as capacity decay rate and internal resistance change rate, which can characterize the long-term degradation trend of the target battery pack. For the specific selection of long-term historical degradation indicator data, one or more can be selected, and this embodiment does not limit this.

[0083] Long-term historical degradation index data is a set of time-series data with a long time span. For example, the battery capacity degradation rate and battery internal resistance change rate over the past 6 months can be selected as long-term historical degradation index data. If data is collected once a month, the long-term historical degradation index data will include the battery capacity degradation rate over these 6 months. And the rate of change of battery internal resistance over these 6 months In addition, the average temperature of the target battery pack over the past 6 months may also be included. with standard deviation and frequency of use F (The cumulative number of charge-discharge cycles of the target battery pack), then the long-term historical degradation index data can be expressed as: .

[0084] S102: Input the short-term state feature data into the pre-trained first health evaluation model to obtain the predicted first health score related to the available capacity of the target battery pack.

[0085] In this embodiment, the battery health level is related to the available capacity of the target battery pack; a higher available capacity indicates a higher battery health level. Therefore, the first health score in this step is related to the available capacity of the target battery pack.

[0086] In some specific implementations, the aforementioned first health score can be designed as a normalized result of the available capacity of the target battery pack.

[0087] In order for the first health assessment model to predict the first health score of the target battery pack based on the input short-term state feature data, the first health assessment model needs to be trained in advance.

[0088] Specifically, the LSTM model can be chosen as the primary health assessment model, and its training process is as follows:

[0089] S1020: Obtain the first training sample, which is a short-term state feature data sample of the sample battery pack.

[0090] S1022: Obtain the first measurement of the available capacity of the sample battery pack within the sampling time window of the short-term state feature data sample, and construct the label of the first training sample based on the first measurement of the available capacity.

[0091] S1024: Train the LSTM model using the first training sample until a first health assessment model that meets the requirements is obtained.

[0092] During the training of the first health assessment model, after acquiring short-term state characteristic data samples of the sample battery packs, it is necessary to assign labels to these samples. These labels are related to the real-time available capacity of the sample battery packs. Since the sampling time window for the short-term state characteristic data samples is very short, any point in time within the sampling time window can be selected, and then the available capacity of the sample battery pack at that point in time can be obtained.

[0093] To obtain the usable capacity of a sample battery pack, capacity testing can be used for measurement, or approximate estimation methods such as the equivalent internal resistance method can be used for estimation.

[0094] After obtaining the available capacity of the sample battery pack, the ratio of the available capacity to the rated capacity of the sample battery pack (current available capacity / rated capacity) can be calculated to normalize the available capacity, and the calculation result can be used as a label for the short-term state characteristic data sample.

[0095] After preparing the first training sample and its labels, the first training sample is input into the LSTM model to obtain the first health score predicted by the LSTM model. Based on the first health score Labels of the samples and the first training samples Construct a first loss function and update the parameters of the LSTM model using the first loss function until a first health assessment model that meets the requirements is obtained.

[0096] After obtaining the first health assessment model, the short-term state characteristic data collected in real time is input into the first health assessment model to obtain the predicted first health score related to the available capacity of the target battery pack.

[0097] S104: Input long-term historical degradation index data into a pre-trained second health assessment model to obtain a predicted second health score related to the available capacity of the target battery pack.

[0098] The aforementioned second health score can be designed as a normalized result of the target battery pack's available capacity. For example, the available capacity can be normalized by calculating the ratio of the target battery pack's available capacity to its rated capacity (previous available capacity / rated capacity).

[0099] In order for the second health assessment model to predict the second health score of the target battery pack based on the input long-term historical degradation index data, the second health assessment model needs to be trained in advance.

[0100] Specifically, the LightGBM model can be chosen as the second health assessment model, and its training process is as follows:

[0101] S1040: Obtain the second training sample, which is a sample of long-term historical degradation index data of the sample battery pack.

[0102] S1042: Obtain the second measurement of available capacity of the sample battery pack at the last moment of the sampling time window of the long-term historical degradation index data sample, and construct the label of the second training sample based on the second measurement of available capacity.

[0103] S1044: Train the LightGBM model using the second training samples until a second health assessment model that meets the requirements is obtained.

[0104] During the training of the second health assessment model, the ratio of available capacity to rated capacity at the last moment of the sampling time window of the long-term historical decline index data sample can be directly used as the label of the long-term historical decline index data sample.

[0105] After preparing the second training samples and their labels, the second training samples are input into the LightGBM model to obtain the second health score predicted by the LightGBM model. According to the second health score Labels of the second training sample A second loss function is constructed, and the parameters of the LightGBM model are updated using the second loss function until a second health assessment model that meets the requirements is obtained.

[0106] After obtaining the second health assessment model, the collected long-term historical degradation index data is input into the second health assessment model to obtain the predicted second health score related to the available capacity of the target battery pack.

[0107] S106: Obtain the current load fluctuation standard deviation of the target battery pack and monthly capacity decay rate And based on the standard deviation of load fluctuation and monthly capacity decay rate Calculate the first fusion weight and the second fusion weight:

[0108] ;

[0109] ;

[0110] in, The first fusion weight, This is the second fusion weight. and For dynamically fitted parameters, and The update is performed using the Cultural Whale optimization algorithm, with the goal of minimizing the mean square error between the predicted and reference values ​​of the target battery pack health score within a preset historical time window.

[0111] For the current load fluctuation standard deviation The current standard deviation can be approximated by the standard deviation of the current over the past hour. Monthly capacity decay rate This is calculated by taking the initial capacity from one month ago as a baseline and then calculating the decrease in capacity over the most recent month.

[0112] Specifically, the dynamic fitting parameters are updated using the Cultural Whale optimization algorithm. and The specific steps are as follows:

[0113] S1060: Initialization and The parameter range.

[0114] S1062: For each sampling time point within the historical time window, obtain the available capacity measurement value, short-term state characteristic data, and long-term historical degradation index data of the target battery pack at the sampling time point, and construct a reference value for the health score of the target battery pack based on the available capacity measurement value.

[0115] Specifically, starting from the current moment, the past 24 hours can be selected as the historical time window, and the sampling step size can be set to 1 hour to obtain the measurement value of the available capacity of the target battery pack at each sampling time point.

[0116] S1064: Input the short-term state characteristic data corresponding to each sampling time point into the first health assessment model to obtain the first health score prediction value corresponding to the sampling time point; input the long-term historical decline index data corresponding to each sampling time point into the second health assessment model to obtain the second health score prediction value corresponding to the sampling time point.

[0117] S1066: Based on the reference value, the predicted value of the first health score, and the predicted value of the second health score corresponding to each sampling time point, construct the mean square error between the predicted value and the reference value of the target battery pack health score; with the goal of minimizing the mean square error, search within the parameter range... and The optimal value, and and The value is updated to the corresponding optimal value.

[0118] The expression for the mean squared error mentioned above is:

[0119] ;

[0120] ;

[0121] in, N This represents the total number of sampling time points. Indicates the first i The predicted first health score for each sampling time point Indicates the first i The predicted second health score for each sampling time point. It is by and The first fusion weight is calculated from the search value. and Within the parameter range, a search is performed with the objective of minimizing the MSE, yielding... optimal value and optimal value .

[0122] This step S106 uses a dynamic weighting mechanism to construct the fusion weights. This mechanism allows the system to adaptively adjust the importance weights of short-term / long-term predictions under different operating conditions.

[0123] Please refer to Figures 2 to 5 , Figures 2 to 5 This illustrates a specific implementation scenario. First fusion weight Second fusion weight Standard deviation of load fluctuation and monthly capacity decay rate The changing pattern. Among them, Figure 2 The first fusion weight is shown. Standard deviation of load fluctuation and monthly capacity decay rate A schematic diagram of a changing three-dimensional surface. Figure 3 The second fusion weight is shown. Standard deviation of load fluctuation and monthly capacity decay rate A schematic diagram of a changing three-dimensional surface. Figure 4 The first fusion weight is shown. Standard deviation of load fluctuation and monthly capacity decay rate A schematic diagram of changing contour lines. Figure 5 The second fusion weight is shown. Standard deviation of load fluctuation and monthly capacity decay rate A schematic diagram of changing contour lines.

[0124] Depend on Figure 2 and Figure 3 As can be seen in the 3D surface plot, the first fusion weight Standard deviation of load fluctuation Rapid increase upon expansion, monthly capacity decay rate The weight decreases slowly as it increases, while the second fusion weight... They exhibit completely complementary characteristics. (By...) Figure 4 and Figure 5 As can be seen, the decision boundary is clearly displayed in the contour map: when or The time is dominated by short-term state characteristics (dark area), when or The data is dominated by long-term historical recession indicators (light-colored area). Figures 2 to 5 This demonstrates that the optimized parameters achieve an adaptive fusion mechanism that enhances real-time data during operating condition fluctuations and improves historical data during equipment degradation.

[0125] S108: Based on the first fusion weight and the second fusion weight, the first health score and the second health score are weighted and summed to obtain the predicted current health score S of the target battery pack.

[0126] ;

[0127] in, This indicates the first health score. This indicates the second health score.

[0128] Optionally, in the battery health assessment method based on the dual-channel attention mechanism described above, a weight correction mechanism for abnormal operating conditions can also be set. Specifically, to improve the stability and safety of the system under extreme operating conditions, rule-based intervention logic is further introduced:

[0129] If the standard deviation of load fluctuation If the value exceeds the preset high volatility threshold, the value of the fusion weight is set to... , For example, if (High fluctuation state, such as frequent accelerated discharge), forced setting This intervention logic ensures the dominant role of the primary health assessment model and improves responsiveness.

[0130] If monthly capacity decay rate If the lifetime threshold is exceeded, the fusion weight is set to 0.6. For example, if... This indicates that the battery is nearing its critical lifespan, at which point a forced setting should be implemented. This intervention logic can weaken the impact of short-term outliers and maintain overall forecast stability.

[0131] In contrast to the battery health assessment method based on the dual-channel attention mechanism described above, this embodiment also provides a battery health assessment system based on the dual-channel attention mechanism. The aforementioned battery health assessment method based on the dual-channel attention mechanism can be implemented using this system, but is not limited to it. Please refer to... Figure 6 The system includes:

[0132] The data acquisition module is used to collect short-term state characteristic data and long-term historical degradation index data of the target battery pack.

[0133] The real-time analysis module is used to input short-term state feature data into a pre-trained first health assessment model to obtain a predicted first health score related to the available capacity of the target battery pack.

[0134] The historical analysis module is used to input long-term historical degradation index data into a pre-trained second health assessment model to obtain a predicted second health score related to the available capacity of the target battery pack.

[0135] The dynamic weight generation module is used to obtain the current load fluctuation standard deviation of the target battery pack. and monthly capacity decay rate And based on the standard deviation of load fluctuation and monthly capacity decay rate Calculate the first fusion weight and the second fusion weight:

[0136] ;

[0137] ;

[0138] in, As the first fusion weight, As the second fusion weight, and For dynamically fitted parameters, and The update is performed using the Cultural Whale optimization algorithm, with the goal of minimizing the mean square error between the predicted and reference values ​​of the target battery pack health score within a preset historical time window.

[0139] The health fusion module is used to perform a weighted summation of the first health score and the second health score based on the first fusion weight and the second fusion weight to obtain the predicted current health score S of the target battery pack.

[0140] ;

[0141] in, This indicates the first health score. This indicates the second health score.

[0142] Optionally, the first health assessment model can be an LSTM model; the system may also include a first training module, which is used for:

[0143] Obtain the first training sample, which is a short-term state feature data sample of the sample battery pack.

[0144] Obtain the first measurement of the available capacity of the sample battery pack within the sampling time window of the short-term state characteristic data sample, and construct the label of the first training sample based on the first measurement of the available capacity;

[0145] The LSTM model is trained using the first training samples until a first health assessment model that meets the requirements is obtained.

[0146] Optionally, the second health assessment model described above can be the LightGBM model; the system may also include a second training module, which is used for:

[0147] Obtain a second training sample, which is a sample of long-term historical degradation index data of the sample battery pack.

[0148] Obtain the second measurement of available capacity of the sample battery pack at the last moment of the sampling time window of the long-term historical degradation index data sample, and construct the label of the second training sample based on the second measurement of available capacity;

[0149] The LightGBM model is trained using the second training samples until a second health assessment model that meets the requirements is obtained.

[0150] Optionally, the system may also include a parameter self-optimization module, which is used to update the dynamically fitted parameters using the Cultural Whale Optimization Algorithm. The specific update steps include:

[0151] initialization and The parameter range;

[0152] For each sampling time point within the historical time window, the available capacity measurement value of the target battery pack at the sampling time point is obtained, and a reference value for the health score of the target battery pack is constructed based on the available capacity measurement value.

[0153] For each sampling time point within the historical time window, acquire the short-term state characteristic data and long-term historical degradation index data of the target battery pack corresponding to that sampling time point;

[0154] Input the short-term state feature data corresponding to each sampling time point into the first health assessment model to obtain the predicted value of the first health score corresponding to the sampling time point;

[0155] Input the long-term historical decline index data corresponding to each sampling time point into the second health assessment model to obtain the predicted value of the second health score corresponding to the sampling time point.

[0156] Based on the reference value corresponding to the sampling time point, the predicted value of the first health score, and the predicted value of the second health score, the mean square error between the predicted value and the reference value of the target battery pack health score is constructed.

[0157] The objective is to minimize the mean squared error within the parameter range. and The optimal value, and and The value is updated to the corresponding optimal value.

[0158] Optionally, the dynamic weight generation module is also specifically used for:

[0159] Standard deviation of load fluctuation When the value exceeds the preset high volatility threshold, the fusion weight value is set to... , ;

[0160] Monthly capacity decay rate When the value exceeds the preset lifetime threshold, the fusion weight will be set to 0.6.

[0161] The specific functions of each module involved in this embodiment can be referred to the corresponding steps in the battery health evaluation method based on the dual-channel attention mechanism described above, and will not be repeated here.

[0162] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0163] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A battery health assessment method based on a dual-channel attention mechanism, characterized in that, Including the following steps: Acquire short-term state characteristic data and long-term historical degradation index data of the target battery pack; The short-term state feature data is input into a pre-trained first health evaluation model to obtain a predicted first health score related to the available capacity of the target battery pack. The long-term historical degradation index data is input into a pre-trained second health evaluation model to obtain a predicted second health score related to the available capacity of the target battery pack. Obtain the current load fluctuation standard deviation of the target battery pack. and monthly capacity decay rate And based on the load fluctuation standard deviation and the monthly capacity decay rate Calculate the first fusion weight and the second fusion weight: ； in, The first fusion weight, This is the second fusion weight. and For dynamically fitted parameters, and The update is performed using the Cultural Whale Optimization Algorithm, with the goal of minimizing the mean square error between the predicted and reference values ​​of the target battery pack health score within a preset historical time window. The first health score and the second health score are weighted and summed based on the first fusion weight and the second fusion weight to obtain the predicted current health score S of the target battery pack: ； in, This represents the first health score. This represents the second health score.

2. The method according to claim 1, characterized in that, The first health assessment model is an LSTM model; the method further includes: Obtain the first training sample, which is a short-term state feature data sample of the sample battery pack. Obtain the first measurement value of the available capacity of the sample battery pack within the sampling time window of the short-term state feature data sample, and construct the label of the first training sample based on the first measurement value of the available capacity; The LSTM model is trained using the first training samples until a first health assessment model that meets the requirements is obtained.

3. The method according to claim 1, characterized in that, The second health assessment model is the LightGBM model; the method further includes: Obtain a second training sample, which is a long-term historical degradation index data sample of the sample battery pack. Obtain the second measurement of available capacity of the sample battery pack at the last moment of the sampling time window of the long-term historical degradation index data sample, and construct the label of the second training sample based on the second measurement of available capacity; The LightGBM model is trained using the second training sample until a second health assessment model that meets the requirements is obtained.

4. The method according to claim 1, characterized in that, The dynamic fitting parameters are updated using the Cultural Whale Optimization Algorithm, specifically including: initialization and The parameter range; For each sampling time point within the historical time window, the available capacity measurement value of the target battery pack at that sampling time point is obtained, and a reference value for the health score of the target battery pack is constructed based on the available capacity measurement value. For each sampling time point within the historical time window, acquire the short-term state characteristic data and long-term historical degradation index data of the target battery pack corresponding to that sampling time point; Input the short-term state feature data corresponding to each sampling time point into the first health evaluation model to obtain the first health score prediction value corresponding to the sampling time point; Input the long-term historical decline index data corresponding to each sampling time point into the second health assessment model to obtain the second health score prediction value corresponding to the sampling time point; Based on the reference value corresponding to the sampling time point, the predicted value of the first health score, and the predicted value of the second health score, the mean square error between the predicted value and the reference value of the target battery pack health score is constructed. Searching within the parameter range with the objective of minimizing the mean square error. and The optimal value, and and The value is updated to the corresponding optimal value.

5. The method according to claim 1, characterized in that, The method further includes: If the load fluctuation standard deviation If the value exceeds the preset high volatility threshold, the value of the first fusion weight is set to... , ; If the monthly capacity decay rate If the value exceeds the preset lifetime threshold, the value of the first fusion weight is set to 0.

6.

6. A battery health assessment system based on a dual-channel attention mechanism, characterized in that, include: The data acquisition module is used to collect short-term state characteristic data and long-term historical degradation index data of the target battery pack. The real-time analysis module is used to input the short-term state feature data into a pre-trained first health evaluation model to obtain a predicted first health score related to the available capacity of the target battery pack. The historical analysis module is used to input the long-term historical degradation index data into a pre-trained second health evaluation model to obtain a predicted second health score related to the available capacity of the target battery pack. The dynamic weight generation module is used to obtain the current load fluctuation standard deviation of the target battery pack. and monthly capacity decay rate And based on the load fluctuation standard deviation and the monthly capacity decay rate Calculate the first fusion weight and the second fusion weight: ； in, The first fusion weight, This is the second fusion weight. and For dynamically fitted parameters, and The update is performed using the Cultural Whale Optimization Algorithm, with the goal of minimizing the mean square error between the predicted and reference values ​​of the target battery pack health score within a preset historical time window. The health fusion module is used to perform a weighted summation of the first health score and the second health score based on the first fusion weight and the second fusion weight to obtain the predicted current health score S of the target battery pack. ； in, This represents the first health score. This represents the second health score.

7. The system according to claim 6, characterized in that, The first health assessment model is an LSTM model; the system further includes a first training module, which is used for: Obtain the first training sample, which is a short-term state feature data sample of the sample battery pack. Obtain the first measurement value of the available capacity of the sample battery pack within the sampling time window of the short-term state feature data sample, and construct the label of the first training sample based on the first measurement value of the available capacity; The LSTM model is trained using the first training samples until a first health assessment model that meets the requirements is obtained.

8. The system according to claim 6, characterized in that, The second health assessment model is the LightGBM model; the system also includes a second training module, which is used for: Obtain a second training sample, which is a long-term historical degradation index data sample of the sample battery pack. Obtain the second measurement of available capacity of the sample battery pack at the last moment of the sampling time window of the long-term historical degradation index data sample, and construct the label of the second training sample based on the second measurement of available capacity; The LightGBM model is trained using the second training sample until a second health assessment model that meets the requirements is obtained.

9. The system according to claim 6, characterized in that, The system also includes a parameter self-optimization module, which is used to update the dynamically fitted parameters using the Cultural Whale Optimization Algorithm. The specific update steps include: initialization and The parameter range; For each sampling time point within the historical time window, the available capacity measurement value of the target battery pack at that sampling time point is obtained, and a reference value for the health score of the target battery pack is constructed based on the available capacity measurement value. For each sampling time point within the historical time window, acquire the short-term state characteristic data and long-term historical degradation index data of the target battery pack corresponding to that sampling time point; Input the short-term state feature data corresponding to each sampling time point into the first health evaluation model to obtain the first health score prediction value corresponding to the sampling time point; Input the long-term historical decline index data corresponding to each sampling time point into the second health assessment model to obtain the second health score prediction value corresponding to the sampling time point; Based on the reference value corresponding to the sampling time point, the predicted value of the first health score, and the predicted value of the second health score, the mean square error between the predicted value and the reference value of the target battery pack health score is constructed. Searching within the parameter range with the objective of minimizing the mean square error. and The optimal value, and and The value is updated to the corresponding optimal value.

10. The system according to claim 6, characterized in that, The dynamic weight generation module is further used for: The standard deviation of load fluctuation When the value exceeds the preset high volatility threshold, the value of the first fusion weight is set to... ; The monthly capacity decay rate When the value exceeds the preset lifetime threshold, the value of the first fusion weight is set to 0.6.

Images