A lithium battery pack chip equalization control method and its control system

Abstract

The present invention provides a lithium battery pack chip equalization control method and its control system. The method includes: obtaining battery cell data; matching the health score and health level of the battery cell to obtain the final equalization control amount of the battery cell; inputting the equalization intensity of each battery cell and the health score of the battery cell into a SOM network to obtain an adaptive equalization strategy for the battery cell; the battery health prediction model is used to capture the health change trend of the battery cell on a long time scale; based on the error between the predicted health state of the battery cell and the actual health state of the battery cell, it is judged whether it is abnormal to obtain the final equalization strategy of the battery cell. The present invention effectively solves the problems of static equalization strategy, inefficient equalization control and lack of intelligent management in the prior art, not only improves the equalization efficiency of the battery pack, prolongs the service life of the battery, but also enhances the adaptability and robustness of the system in complex and dynamic environments.

Description

Technical Field

[0001] The present invention belongs to the field of lithium battery pack chip equalization control, and particularly relates to a lithium battery pack chip equalization control method and its control system. Background Art

[0002] With the rapid development of fields such as electric vehicles, wearable devices, and large-scale energy storage systems, lithium batteries, as a power source option with high energy density and long life, have been widely used in multiple fields. To ensure the stability and safety of lithium batteries, the battery management system (BMS) has become a core component for guaranteeing the performance of lithium batteries. In the battery management system, battery equalization control is a key technology, whose function is to ensure the equalization of the power of each single battery in the battery pack, prevent some single batteries from being damaged due to overcharging or over-discharging, thereby extending the service life of the battery pack and improving its overall efficiency.

[0003] Currently, there are mainly two types of battery equalization control technologies: passive equalization and active equalization. The passive equalization technology dissipates the excess electrical energy in the battery in the form of heat through a simple resistor heat dissipation method, thereby achieving the equalization of the power between batteries. Although this method is simple to implement technically and has a low cost, its efficiency is low because the excess energy in the battery is directly converted into heat and wasted, resulting in relatively serious energy loss. The active equalization technology transfers the excess energy of the single battery with a higher power in the battery pack to the single battery with a lower power through energy conversion, transfer, or storage, which can effectively reduce energy waste and improve the equalization efficiency. However, the active equalization technology has a higher system complexity, a higher cost, and requires more precise control algorithms and hardware support.

[0004] The existing battery equalization control methods, although they can solve the problem of uneven power of battery monomers to a certain extent, still have several technical bottlenecks that need to be solved urgently. First of all, most of the existing technologies still rely on static equalization strategies and lack the ability to dynamically adapt to the real-time health status and usage environment of the battery pack. This means that during the operation of the battery pack, when the health status of the battery monomer changes or the external environment fluctuates, the system often cannot adjust the equalization strategy in time, resulting in a decrease in equalization efficiency and even potential safety hazards. Secondly, the existing battery equalization systems generally lack intelligent adaptive control and cannot be optimized and adjusted according to multiple factors such as the usage status, aging condition, and temperature change of the battery monomer. Although some advanced control algorithms (such as adaptive algorithms, machine learning, etc.) have been proposed, most of these technologies face great challenges in terms of hardware implementation, real-time performance, and computational resource requirements. Moreover, in some battery pack usage scenarios, such as during long-term dormancy or without data communication, the traditional battery equalization technology cannot effectively maintain the equalization state, resulting in battery performance degradation and affecting the battery life.

[0005] In addition, the battery balancing control strategy lacks deep integration with the overall health management of the battery pack. Existing technologies are usually limited to the balancing of battery power, ignoring the comprehensive impacts of other key factors such as battery health status, charge and discharge efficiency, etc., and cannot achieve comprehensive battery management. More importantly, traditional balancing systems fail to effectively combine advanced sensor technologies, big data analysis, and artificial intelligence algorithms, which makes the system unable to perform efficient and accurate monitoring and management in the face of large-scale and distributed battery packs. Especially in complex and dynamic working environments, the robustness and adaptability of the system are poor. Summary of the Invention

[0006] The objective of the present invention is to propose a lithium battery pack chip balancing control method and its control system. By precisely monitoring the state data of each battery cell, comprehensive management of battery health is achieved. Also, by adjusting the balancing strategy in real time, the adaptability and robustness of the system in complex environments are improved. Through dynamic stratification and adaptive balancing of battery cells, the balancing strategy can be intelligently adjusted according to factors such as the health status of battery cells, temperature changes, and load fluctuations, maximizing the balancing efficiency while avoiding battery performance loss.

[0007] To achieve the above objective, in the first aspect of the present invention, a lithium battery pack chip balancing control method is provided. The method includes the following steps:

[0008] Obtain battery cell data, construct a machine learning model based on the battery cell data to evaluate the battery health status, and obtain the health score of the corresponding battery cell; the machine learning model is trained using the feature vector after fusing the battery cell data;

[0009] Match the health score of the battery cell with the health level to determine the health level of each battery cell. Based on the health score of the battery cell, through an optimized decision model, determine the balancing intensity of each battery cell, and perform a first regularization operation on the balancing intensity of the battery cell to obtain the final balancing control quantity of the battery cell ;

[0010] Input the balancing intensity of each battery cell and the health score of the battery cell into the SOM network to obtain the adaptive balancing strategy of the battery cell; perform a second regularization operation on the adaptive balancing strategy of the battery cell based on the final balancing control quantity of the battery cell to obtain the final balancing strategy ; The SOM network maps the battery cells into similar groups according to the balancing intensity of each battery cell and the health score of the battery cell, and is trained through unsupervised learning;

[0011] Input the final balancing strategy, the health score of each battery cell, and the balancing intensity of each battery cell as a data set into the battery health prediction model based on the long short-term memory network to obtain the predicted health status of the battery cell; the battery health prediction model is used to capture the health change trend of the battery cell on a long time scale;

[0012] Judge whether it is abnormal based on the error between the predicted health status of the battery cell and the actual health status of the battery cell. If the error exceeds the threshold, trigger an alarm and enter the self-recovery mechanism, and make adaptive adjustments according to the actual situation, so as to reduce the risk of battery damage; the self-recovery mechanism adjusts the balancing strategy based on the current health status and balancing intensity of the battery cell to obtain the final balancing strategy of the battery cell 。

[0013] Further, the battery cell data includes voltage, current, temperature and internal resistance;

[0014] When obtaining the battery cell data, preprocessing is also performed:

[0015] Adopt the weighted moving average method to smooth each battery cell data to obtain the smoothed data;

[0016] Perform standardization processing on the smoothed data to obtain the standardized data;

[0017] Perform fusion analysis on the standardized data and combine them into a feature vector.

[0018] Further, the machine learning model is trained based on the feature vector to obtain the health score of the battery cell; the health score of the battery cell ranges from [0,1], where 0 indicates that the battery health status is extremely poor and 1 indicates that the battery is in the best state.

[0019] Further, the health levels include health level 1, health level 2, and health level 3;

[0020] Among them, the health level 1 indicates that the health status of the battery cell is the worst and needs to be balanced with the highest priority; the health level 2 indicates that the health status of the battery cell is medium and conventional balancing is performed; the health level 3 indicates that the health status of the battery cell is the best and the balancing requirement is the least;

[0021] The optimization decision model determines the balancing intensity of each battery cell, expressed as:

[0022] ;

[0023] Among them, represents the balancing intensity of the battery cell at time ; and are weight parameters, controlling the flexibility of the balancing strength; is the health score of the battery cell at time ;

[0024] The first regularization term , used to control the excessive intervention during the balancing process, is expressed as:

[0025] ;

[0026] Among them, is the first regularization term during the balancing process, used to prevent overbalancing and ensure that the health state of the battery is not overly disturbed; is the regularization coefficient, controlling the degree of intervention, usually taking values in ;

[0027] Furthermore, the training objective of the SOM network is to minimize the error between neurons, which is to map each battery cell to the neuron closest to its health state and balancing strength. During the training process, the weight of each neuron will be updated according to the input information to form the corresponding battery cell characteristics; the training is expressed as:

[0028] ;

[0029] Among them, is the updated weight; is the weight representing neuron at time , reflecting the balancing demand and health state of the battery cell; is the learning rate, controlling the amplitude of each update; is the neighborhood function, representing the influence range of each neuron; is the input data, including the health score and balancing strength of the battery.

[0030] Furthermore, the trained SOM network will generate an adaptive balancing strategy for each battery according to the health state and balancing demand of the battery cell , expressed as:

[0031] ;

[0032] Among them, is the adaptive balancing strategy representing battery cell at time ; is the neuron weight in the SOM network, reflecting the characteristics of the battery cell; is the health score of the battery cell; is the equalization intensity of the battery cell.

[0033] Further, the second regularization operation is used to control the equalization operation of each battery from being too drastic, and is defined as:

[0034] ;

[0035] where represents the second regularization term of the battery cell at time ; is the regularization coefficient, which is used to adjust the range of the equalization strategy; represents the ideal equalization strategy, which is usually set based on the optimal battery model.

[0036] Further, the error between the predicted health state of the battery cell and the actual health state of the battery cell is measured based on an error metric function; the error metric function is expressed as:

[0037] ;

[0038] where is the error metric value, is the actual health state of the battery cell, is the predicted health state of the battery cell.

[0039] Further, the self-recovery mechanism designs a mild anomaly strategy and a severe anomaly strategy according to the anomaly degree of the battery cell;

[0040] wherein, the mild anomaly strategy relieves the problem by adjusting the equalization strategy of the battery to make the battery return to the normal working state; the severe anomaly strategy will pause or reduce the equalization operation of the battery cell, and optimize the performance of the battery pack by reallocating the load.

[0041] In the second aspect of the present invention, a lithium battery pack chip equalization control system is provided, and the system includes:

[0042] A battery cell data acquisition unit, configured to acquire battery cell data, construct a machine learning model based on the battery cell data to evaluate the health state of the battery, and obtain the health score of the corresponding battery cell; the machine learning model is trained using the feature vector after fusing the battery cell data;

[0043] The battery cell health analysis unit is used to match the health score and health level of the battery cell, determine the health level of each battery cell, determine the equalization intensity of each battery cell through an optimized decision model based on the health score of the battery cell, and perform a first regularization operation on the equalization intensity of the battery cell to obtain the final equalization control quantity of the battery cell ;

[0044] The strategy training unit is used to input the equalization intensity of each battery cell and the health score of the battery cell into the SOM network to obtain the adaptive equalization strategy of the battery cell; perform a second regularization operation on the adaptive equalization strategy of the battery cell based on the final equalization control quantity of the battery cell to obtain the final equalization strategy ; The SOM network maps the battery cells into similar groups according to the equalization intensity of each battery cell and the health score of the battery cell, and is trained through unsupervised learning

[0045] The fault prediction unit is used to input the final equalization strategy, the health score of the battery cell, and the equalization intensity of each battery cell as a data set into the battery health prediction model based on the long short-term memory network to obtain the predicted health state of the battery cell; the battery health prediction model is used to capture the health change trend of the battery cell on a long time scale

[0046] The control optimization unit is used to judge whether it is abnormal based on the error between the predicted health state of the battery cell and the actual health state of the battery cell. If the error exceeds the threshold, it triggers an alarm and enters the self-recovery mechanism, and makes an adaptive adjustment according to the actual situation, so as to reduce the risk of battery damage; the self-recovery mechanism adjusts the equalization strategy based on the current health status and equalization intensity of the battery cell to obtain the final equalization strategy of the battery cell 。

[0047] The beneficial technical effects of the present invention are at least as follows:

[0048] Adaptive hierarchical battery equalization control strategy: This method first monitors the health status (such as voltage, current, temperature, capacity, etc.) of each battery cell in real time, and divides the battery cells into different levels according to these parameters. For different health states of the battery cells, different equalization strategies are adopted. For example, batteries in good health only need minor equalization adjustments, while batteries with severe aging or abnormal temperature require more refined active equalization strategies. This hierarchical control strategy not only reduces the intervention of batteries with relatively balanced power in the battery pack, but also optimizes the management of aging batteries and improves the overall efficiency of equalization

[0049] Self-organizing neural network (SOM) self-optimizing battery equalization algorithm: The present invention combines the self-organizing neural network (SOM) algorithm. Through the real-time monitoring of the health data and working status of battery cells, it conducts dynamic learning and self-optimization. SOM analyzes the similarity of battery cells through clustering and groups the batteries according to similarity, and adjusts its equalization strategy based on the characteristics of each group. This innovation effectively overcomes the static nature of traditional equalization systems and provides an adaptive and dynamically optimized equalization control method, which can respond to changes in the battery pack in real time, improving the equalization efficiency and accuracy.

[0050] Prediction management and self-recovery mechanism: The present invention also introduces a battery health prediction algorithm based on machine learning, which predicts the future state of the battery by combining the historical data of the battery. When the system predicts a decline in the performance of certain battery cells, it will initiate corresponding equalization adjustments in advance to avoid excessive attenuation of battery performance. In addition, the self-recovery mechanism designed in the present invention can timely adjust the equalization strategy and restore the normal working state of the system when an abnormality occurs during the battery equalization process, ensuring the long-term stable operation of the battery pack.

[0051] Through the above innovations, the present invention effectively solves the problems existing in the prior art, such as static equalization strategies, inefficient equalization control, and lack of intelligent management. It not only improves the equalization efficiency of the battery pack, extends the battery life, but also enhances the adaptability and robustness of the system in complex and dynamic environments. Brief Description of the Drawings

[0052] The present invention will be further described with reference to the drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the following drawings without creative efforts.

[0053] Figure 1 Flowchart of a lithium battery pack chip equalization control method disclosed in an embodiment of the present invention

[0054] Figure 2 Framework diagram of a lithium battery pack chip equalization control system disclosed in an embodiment of the present invention. Detailed Description of the Embodiments

[0055] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0056] Embodiment 1

[0057] As Figure 1As shown in the figure, a lithium battery pack chip balancing control method provided by an embodiment of the present invention includes the following:

[0058] S101. Obtain battery cell data, construct a machine learning model based on the battery cell data to evaluate the battery health state, and obtain the health score of the corresponding battery cell; the machine learning model is trained using the feature vector after fusing the battery cell data.

[0059] Specifically, in the prior art, the health monitoring of batteries is usually discontinuous or lacks sufficient accuracy. The health status of the battery is not comprehensively evaluated based on multi-dimensional data of the battery (such as voltage, current, temperature, etc.), resulting in inaccurate balancing control. Therefore, this solution proposes a new method to achieve precise monitoring of the battery health state through real-time analysis and fusion of multi-dimensional data.

[0060] Among them, the battery management system (BMS) collects data of battery cells in real time through various sensors (such as voltage, current, temperature, internal resistance sensors, etc.). Set the multi-dimensional data of each battery cell as:

[0061] Voltage , current , temperature , internal resistance ;

[0062] These data are the basis for evaluating the battery health state. Assume that these data are collected at time , expressed as:

[0063] ;

[0064] Furthermore, since the collected data may contain noise, signal processing is required for smoothing. The weighted moving average (WMA) method is used to smooth each feature (such as voltage, current, etc.) to make the data more reliable. The specific formula is as follows:

[0065] ;

[0066] Where is the smoothed voltage data; is the voltage data of battery cell at time; is the weighting coefficient, usually using an exponential decay function: , where is the weighting coefficient, , making the weight of recent data higher. This method can eliminate the fluctuations caused by instantaneous noise and improve the data quality.

[0067] Furthermore, to avoid the influence of features with different dimensions on model training, it is necessary to standardize the data. Suppose is the voltage data of a single battery cell at time , the following standardization formula is used:

[0068] ;

[0069] where and are the minimum and maximum voltage values of the single battery cell respectively; is the standardized voltage data. The same standardization is performed on all features to ensure that their ranges are all between [0,1], which helps the training of subsequent machine learning models.

[0070] Furthermore, since the battery health status is jointly determined by multiple factors, it is necessary to perform a fusion analysis on the multi-dimensional data of the battery. First, the standardized data of each battery is combined into a feature vector:

[0071] ;

[0072] This feature vector contains the standardized voltage, current, temperature, and internal resistance data of each single battery cell. By jointly analyzing these multi-dimensional data, the health status of the battery can be comprehensively reflected.

[0073] Furthermore, a machine learning model (such as random forest regression, support vector regression, etc.) is used to construct a health assessment model to evaluate the battery health status. Based on the standardized multi-dimensional data, the model is trained to predict the battery health score . The formula is as follows:

[0074] ;

[0075] where is the health score of the single battery cell at time ; is the trained machine learning model (such as random forest regression); are the parameters of the model. This health score is generated based on the real-time data of the battery and can reflect the current health status of the battery.

[0076] Furthermore, the final output is the health score of each single battery cell, which reflects the current state of the battery. Usually, the value range is [0,1], where 0 indicates extremely poor battery health status and 1 indicates the best state. The health score can be used for subsequent equalization control and warning systems.

[0077] ;

[0078] Among them, is the set of health scores of all battery cells.

[0079] This step provides a battery health assessment method based on multi-dimensional data. It accurately assesses the health status of the battery by real-time monitoring of characteristic data such as the voltage, current, temperature, and internal resistance of the battery, and combining data denoising, smoothing, standardization, and fusion analysis, and using a machine learning model. This process effectively solves the problems of discontinuous battery health monitoring and lack of accuracy in the prior art, can provide accurate battery health scores, and provides a decision-making basis for subsequent equalization control and warning systems.

[0080] S102. Match the health scores of the battery cells with the health levels, determine the health levels of each battery cell, determine the equalization intensity of each battery cell through an optimized decision model based on the health scores of the battery cells, and perform a first regularization operation on the equalization intensity of the battery cells to obtain the final equalization control quantity of the battery cells .

[0081] In the prior art, the equalization control of the battery usually depends on a single equalization strategy and it is difficult to accurately adjust according to the actual health status of the battery. Moreover, the health conditions of different battery cells vary greatly, which may lead to low equalization efficiency and even shorten the service life of the battery. Therefore, this step proposes a hierarchical management method based on the battery health status score, which optimizes the equalization process and improves the overall equalization efficiency through hierarchical processing and differential control strategies.

[0082] Further, perform stratification of the battery health status:

[0083] According to the health scores of each battery cell obtained in step S101 of , divide all battery cells into different health levels. The specific stratification criteria are set according to the range of the battery health scores:

[0084] Health level 1 ( ): Indicates that the health condition of the battery cell is poor and it needs to be equalized preferentially.

[0085] Health level 2 ( ): Indicates that the health condition of the battery cell is medium and it can be equally balanced routinely.

[0086] Health level 3 ( ): Indicates that the health condition of the battery cell is good and the equalization requirement is less.

[0087] Through this hierarchical method, a personalized balancing strategy can be implemented for the health status of each battery cell.

[0088] Furthermore, different balancing control strategies are designed for batteries in different health levels. Specifically:

[0089] Health level 1 ( ): For batteries with poor health, stronger balancing is required. High-speed discharging and charging regulation are adopted to quickly restore the energy balance of the batteries. The current can be increased step by step to ensure the batteries are fast and stable.

[0090] Health level 2 ( ): For batteries with medium health, a medium-speed balancing strategy is adopted. Through fine current regulation, the batteries are gradually brought closer to the ideal state.

[0091] Health level 3 ( ): For batteries with good health, a low-speed balancing strategy is adopted to reduce the interference to the batteries during the balancing process. Balancing is only performed when there is an obvious difference in the battery charge levels.

[0092] Furthermore, by introducing an optimization decision model, according to the health score of each battery cell , the balancing intensity of each cell is determined . The output of this decision model is the balancing strategy and adjustment amount for each battery cell. To dynamically adjust the balancing strategy, a weighted optimization model based on the health score is designed:

[0093] ;

[0094] where, represents the balancing intensity of battery cell at time ; and are weight parameters, , controlling the flexibility of the balancing intensity; is the health score of battery cell at time . This formula indicates that the balancing intensity decreases as the battery health score increases, and vice versa. By adjusting the weights and , the balancing strategy can be flexibly adjusted according to actual needs.

[0095] Furthermore, based on the aforementioned balancing intensity , balancing is performed by adjusting the current and charge-discharge rate. To further improve the balancing efficiency, an additional first regularization term is introduced , for controlling excessive intervention during the equalization process:

[0096] ;

[0097] Among them, is the first regularization term during the equalization process, used to prevent excessive equalization and ensure that the health state of the battery is not overly disturbed; is the regularization coefficient, controlling the degree of intervention, usually taking values within .

[0098] The final equalization decision and execution are determined by weighted consideration of the equalization intensity and the regularization term to avoid unnecessary damage to the battery while ensuring the equalization effect.

[0099] ;

[0100] Among them, is the final equalization control quantity of the battery cell .

[0101] In this step, through a hierarchical management method based on the battery health state, a differentiated equalization control strategy is designed to adapt to the health conditions of different battery cells. Through a weighted optimization model and regularization method, a dynamic and intelligent equalization process can be achieved, significantly improving the equalization efficiency and effectively avoiding excessive intervention in the battery. This method can significantly optimize the energy management of the battery pack, extend the service life of the battery, and solve the problems of insufficient refinement and low efficiency in equalization control in the existing technology.

[0102] S103. Input the equalization intensity of each battery cell and the health score of the battery cell into the SOM network to obtain the adaptive equalization strategy of the battery cell; perform a second regularization operation on the adaptive equalization strategy of the battery cell based on the final equalization control quantity of the battery cell to obtain the final equalization strategy ; The SOM network maps the battery cells into similar groups according to the equalization intensity of each battery cell and the health score of the battery cell, and is trained through unsupervised learning.

[0103] Specifically, the input of this step is the output of step S102: that is, the battery hierarchical data generated by the battery health state and equalization intensity in step S102 and , and these data reflect the health conditions of the battery cells and their equalization requirements at each moment .

[0104] In this step, the SOM network maps these input data into a high-dimensional space and clusters the characteristics of battery cells through unsupervised learning methods. This step is crucial because the present invention no longer simply adjusts the balance according to fixed rules, but through the adaptive ability of SOM, enabling the network to dynamically adjust the strategy according to the health changes of the battery. The specific input data are as follows:

[0105] ;

[0106] These data provide sufficient input information for SOM to enable it to quickly respond to the health status of each battery cell.

[0107] Furthermore, the self-organizing neural network (SOM network) is trained through unsupervised learning. The core objective is to map battery cells into similar groups according to the balancing requirements and health conditions of each battery cell. This process is based on the competitive learning mechanism of SOM, and each battery cell finds its most suitable category in the SOM network.

[0108] The training objective of SOM is to minimize the error between neurons. Specifically, the objective is to map each battery cell to the neuron closest to its health status and balancing intensity. During this process, the weights of each neuron will be updated according to the input information to form the corresponding characteristics of the battery cell.

[0109] The training formula is:

[0110] ;

[0111] Where represents the weight of neuron at time , reflecting the balancing requirements and health status of the battery cell; is the learning rate, controlling the amplitude of each update; is the neighborhood function, representing the influence range of each neuron; is the input data, including the health score and balancing intensity of the battery. This training process ensures that the neural network can cluster according to the real-time health status of the battery, and as the training progresses, the adaptability of the network to different battery cells gradually increases, ensuring the self-adaptability of the balancing strategy.

[0112] Furthermore, the trained SOM network will generate the best balancing strategy for each battery according to the health status and balancing requirements of the battery cell. In practical applications, the balancing strategy of the battery usually manifests as its charging or discharging rate , which is the operation intensity required by each battery cell at a specific moment.

[0113] The innovation of this step lies in generating an adaptive strategy based on SOM that can dynamically adjust according to the actual health status and balancing requirements of battery cells. In order to enable each battery cell to obtain the most appropriate balancing intensity , the present invention defines a mapping function:

[0114] ;

[0115] wherein, represents the adaptive balancing strategy of battery cell at time ; is the neuron weight in the SOM network, reflecting the characteristics of the battery cell; is the health score of the battery cell; is the balancing intensity of the battery cell. This function combines the weight information trained by the SOM network and can accurately generate a balancing strategy suitable for the current state of the battery cell.

[0116] Furthermore, in order to avoid excessive interference with the normal operation of the battery cell, this step introduces a second regularization term , aiming to control the balancing operation of each battery from being too drastic. The regularization term can not only control the range of the balancing strategy, but also reduce the repeated adjustment of the balancing operation and avoid excessive interference with the normal discharge or charge state of the battery cell. The definition of the regularization term is:

[0117] ;

[0118] wherein, represents the regularization term of battery cell at time ; is the regularization coefficient, used to adjust the range of the balancing strategy; is the ideal balancing strategy, usually set based on the optimal battery model. The final balancing strategy is:

[0119] ;

[0120] By introducing the second regularization term , the present invention ensures the stability of the balancing operation and avoids excessive interference with the normal working state of the battery.

[0121] Through the adaptive equalization optimization based on the self-organizing neural network (SOM), the present invention breaks through the bottleneck of traditional battery equalization control: traditional equalization methods rely on rules and static algorithms, while this solution uses the self-organizing neural network to perform intelligent adjustment according to the health status and dynamic requirements of the battery. The equalization strategy for each battery cell will change dynamically according to the real-time monitored health status and equalization requirements, thus improving the equalization efficiency and battery life.

[0122] In addition, the unsupervised learning ability of the SOM network enables this method to have strong adaptability, capable of coping with the changes of the battery under different usage environments. Especially in situations such as sudden changes in health status and load changes, it can quickly adjust the equalization strategy to avoid overcharging and overdischarging of battery cells.

[0123] S104. Input the final equalization strategy, the health score of the battery cell, and the equalization intensity of each battery cell into the battery health prediction model based on the long short-term memory network to obtain the predicted health status of the battery cell; the battery health prediction model is used to capture the health change trend of the battery cell on a long time scale.

[0124] Specifically, the input of this step comes from the output of step S103, that is, the adaptive equalization strategy of each battery cell generated based on the SOM network. . These data include the equalization strategy of each battery cell and its health status information and the equalization intensity .

[0125] These data can not only be used to judge the real-time health status of the battery cell, but also provide important input information for future prediction models. In order to improve the accuracy of prediction management and anomaly detection, the present invention combines historical data and current state data for time series analysis. The input data is:

[0126] ;

[0127] Furthermore, the construction of the battery health prediction model:

[0128] The prediction of battery health is based on historical data and the current state. In order to improve the prediction accuracy, the present invention adopts a time series prediction model based on the long short-term memory network (LSTM). This model can capture the health change trend of the battery on a long time scale, especially suitable for devices such as batteries with long-term change trends. The output of the LSTM model is the predicted value of the battery health score . The prediction formula is:

[0129]

[0130] Among them, is used to represent the predicted health state of the battery cell at time ; is used to represent the input data at time , including the equalization strategy, health state, and equalization intensity. Through the training of LSTM, the present invention can accurately capture the long-term change pattern of the battery health state and predict the health trend of the battery in the future for a period of time.

[0131] S105. Determine whether it is abnormal based on the error between the predicted health state of the battery cell and the actual health state of the battery cell. If the error exceeds the threshold, trigger an alarm and enter the self-recovery mechanism, and perform adaptive adjustment according to the actual situation, so as to reduce the risk of battery damage; the self-recovery mechanism adjusts the equalization strategy based on the current health condition and equalization intensity of the battery cell to obtain the final equalization strategy of the battery cell .

[0132] Specifically, during the operation of the battery pack, the occurrence of abnormal conditions will have a significant impact on the health of the battery. Therefore, it is necessary to detect abnormalities in a timely manner and give early warnings. The anomaly detection algorithm is based on the predicted health state and the actual battery health state to determine whether an abnormality has occurred.

[0133] Define an error metric function , which is used to calculate the difference between the predicted value and the actual value:

[0134] ;

[0135] When exceeds the set threshold , the system will trigger an anomaly alarm and enter the self-recovery mechanism:

[0136] If , trigger an alarm, indicating that there is an abnormality in the battery health state;

[0137] is a preset threshold, which can be dynamically adjusted according to the actual situation.

[0138] Furthermore, once an abnormal situation is detected, the system will automatically enable the self-recovery mechanism. The core idea of self-recovery is to adjust the equalization strategy based on the current health condition and equalization demand of the battery. According to the abnormal degree of the battery cell, the present invention adopts the following strategy:

[0139] Mild abnormality: When the abnormality is relatively mild, the system adjusts the equalization strategy of the battery to alleviate the problem and return the battery to the normal working state.

[0140] Severe anomaly: When the anomaly is severe, the system will pause or reduce the equalization operation on the battery cell to avoid further damage, and optimize the performance of the battery pack by redistributing the load.

[0141] The equalization strategy adjustment formula is:

[0142] ;

[0143] Wherein, is the equalization strategy of the battery cell at time ; is the change amount of the battery cell health error, indicating the change in the health state error from the previous time to the current time; is the adjustment coefficient, which controls the amplitude of the equalization strategy adjustment. This adjustment strategy can ensure that when an anomaly occurs, the operation of the battery can be adaptively adjusted according to the actual situation, thereby reducing the risk of battery damage.

[0144] By introducing a prediction management, anomaly detection and self - recovery mechanism, the present invention adds the ability of self - regulation and self - repair to the battery management system. Under the prediction of the battery health change trend, the battery management system can detect potential problems in advance and correct them through the self - recovery mechanism. This automated and adaptive method makes the battery management more intelligent and can effectively avoid the problem of slow response of traditional battery management systems when facing battery anomalies.

[0145] Embodiment 2

[0146] As Figure 2 shown, the embodiment of the present invention also provides a lithium - battery pack chip equalization control system, and the system includes:

[0147] A battery cell data acquisition unit 101, configured to acquire battery cell data, construct a machine - learning model based on the battery cell data to evaluate the battery health state, and obtain the health score of the corresponding battery cell; the machine - learning model is trained using the feature vector after fusing the battery cell data;

[0148] A battery cell health analysis unit 102, configured to match the health score of the battery cell with the health level, determine the health level of each battery cell, determine the equalization intensity of each battery cell through an optimization decision model based on the health score of the battery cell, and perform a first regularization operation on the equalization intensity of the battery cell to obtain the final equalization control amount of the battery cell ;

[0149] A strategy training unit 103 is configured to input the equalization intensity of each battery cell and the health score of the battery cell into a SOM network to obtain an adaptive equalization strategy for the battery cell; perform a second regularization operation on the adaptive equalization strategy of the battery cell based on the final equalization control amount of the battery cell to obtain a final equalization strategy. The SOM network maps the battery cells into similar groups according to the equalization intensity of each battery cell and the health score of the battery cell, and is trained through unsupervised learning.

[0150] A fault prediction unit 104 is configured to input the final equalization strategy, the health score of the battery cell, and the equalization intensity of each battery cell as a data set into a battery health prediction model based on a long short-term memory network to obtain the predicted health state of the battery cell; the battery health prediction model is used to capture the health change trend of the battery cell on a long time scale.

[0151] A control optimization unit 105 is configured to determine whether it is abnormal based on the error between the predicted health state of the battery cell and the actual health state of the battery cell. If the error exceeds the threshold, an alarm is triggered and a self-recovery mechanism is entered, and adaptive adjustment is performed according to the actual situation, so as to reduce the risk of battery damage; the self-recovery mechanism adjusts the equalization strategy based on the current health condition and equalization intensity of the battery cell to obtain the final equalization strategy of the battery cell. 。

[0152] The specific embodiments of the present specification are described above, and other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily have to be performed in the particular order or continuous order shown to achieve the desired results. In certain implementations, multitasking and parallel processing are also possible or may be advantageous.

[0153] The systems, devices, modules, or units illustrated in the above embodiments may be specifically implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0154] For convenience of description, the above devices are described by dividing their functions into various units. Of course, when implementing this specification, the functions of each unit may be implemented in one or more software and / or hardware.

[0155] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0156] This specification is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of this specification. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in Figure 1 one or more of the flows Figure 1 or multiple flows and / or blocks

[0157] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the specified functions in Figure 1 one or more of the flows Figure 1 or multiple flows and / or blocks

[0158] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the specified functions in Figure 1 one or more of the flows Figure 1 or multiple flows and / or blocks

[0159] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0160] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash memory (flashRAM). The memory is an example of computer-readable media.

[0161] A computer-readable medium includes both permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0162] It should also be noted that the term "comprises", "comprising", or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, commodity, or device that comprises a series of elements includes not only those elements but also other elements not expressly listed, or elements that are inherent to such process, method, commodity, or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity, or device that comprises the element.

[0163] This specification can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0164] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, they are described relatively simply, and the relevant parts can be referred to the description of the method embodiments.

[0165] Finally, it should be noted that: What is disclosed by a lithium battery pack chip equalization control platform disclosed in the embodiments of the present invention is only the preferred embodiments of the present invention, which are only used to illustrate the technical solutions of the present invention, rather than to limit it; Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for balancing control of a lithium battery pack chip, characterized in that, The method includes the following steps: Obtain battery cell data, construct a machine learning model based on the battery cell data to evaluate the battery health state, and obtain the health score of the corresponding battery cell; the machine learning model is trained using the feature vector after fusing the battery cell data; Match the health score and health level of the battery cell to determine the health level of each battery cell. Based on the health score of the battery cell, optimize the decision model to determine the equalization strength of each battery cell, and perform a first regularization operation on the equalization strength of the battery cell to obtain the final equalization control amount of the battery cell ; Input the equalization intensity of each battery cell and the health score of the battery cell into the SOM network to obtain the adaptive equalization strategy of the battery cell; perform a second regularization operation on the adaptive equalization strategy of the battery cell based on the final equalization control amount of the battery cell to obtain the final equalization strategy ; The SOM network maps the battery cells into similar groups according to the equalization intensity of each battery cell and the health score of the battery cell, and is trained through unsupervised learning Use the final balancing strategy, the health score of the battery cell, and the balancing strength of each battery cell as a data set to input into a battery health prediction model based on a long short-term memory network to obtain the predicted health state of the battery cell; the battery health prediction model is used to capture the health change trend of the battery cell on a long time scale; Judge whether it is abnormal based on the error between the predicted health state of the battery cell and the actual health state of the battery cell. If the error exceeds the threshold, trigger an alarm and enter the self-recovery mechanism, and make adaptive adjustments according to the actual situation, so as to reduce the risk of battery damage; the self-recovery mechanism adjusts the equalization strategy based on the current health status and equalization strength of the battery cell to obtain the final equalization strategy of the battery cell 。 2. A lithium battery pack chip equalization control method according to claim 1, characterized in that, The battery cell data includes voltage, current, temperature, and internal resistance; When obtaining the battery cell data, preprocessing is also performed: Adopt a weighted moving average method to smooth each battery cell data to obtain the smoothed data; Perform normalization processing on the smoothed data to obtain the normalized data; Perform fusion analysis on the normalized data and combine them into a feature vector.

3. A method for balancing control of a lithium battery pack chip according to claim 2, characterized in that, The machine learning model is trained based on the feature vector to obtain the health score of the battery cell; the value range of the health score of the battery cell is [0,1], where 0 indicates that the battery health state is extremely poor, and 1 indicates that the battery is in the best state.

4. A method for balancing control of a lithium battery pack chip according to claim 1, characterized in that, The health levels include health level 1, health level 2, and health level 3; Among them, the health level 1 indicates that the health condition of the battery cell is the worst and needs to be balanced with the highest priority; the health level 2 indicates that the health condition of the battery cell is medium and routine balancing is performed; the health level 3 indicates that the health condition of the battery cell is the best and the balancing requirement is the least; The optimization decision model determines the balancing strength of each battery cell, which is expressed as: ； Among them, represents the equalization strength of the battery cell at time ; and are weight parameters, which control the flexibility of the equalization strength; is the health score of the battery cell at time ; The first regularization term , which is used to control excessive intervention during the balancing process, is expressed as: ； Among them, is the first regularization term in the balancing process, which is used to prevent over-balancing and ensure that the health state of the battery is not overly disturbed; is the regularization coefficient that controls the degree of intervention and takes values within between.

5. A method for balanced control of a lithium battery pack chip according to claim 1, characterized in that, The training objective of the SOM network is to minimize the error between neurons, so that each battery cell is mapped to the neuron closest to its health state and balancing strength. During the training process, the weight of each neuron will be updated according to the input information to form the corresponding battery cell characteristics; The training is expressed as: ； Among them, is the updated weight; represents the neuron at time whose weight reflects the equalization demand and health status of the battery cell; is the learning rate that controls the magnitude of each update; is the neighborhood function that represents the influence range of each neuron; is the input data that includes the health score and equalization intensity of the battery.

6. A method for balancing control of a lithium battery pack chip according to claim 5, characterized in that, The trained SOM network generates an adaptive equalization strategy for each battery according to the health state and equalization requirements of the battery cell , expressed as: ； Among them, represents the adaptive equalization strategy of the battery cell at time ; is the neuron weight in the SOM network, reflecting the characteristics of the battery cell; is the health score of the battery cell; is the equalization intensity of the battery cell.

7. A lithium battery pack chip balancing control method according to claim 5, characterized in that, The second regularization operation is used to control the balancing operation of each battery from being too drastic, which is defined as: ； Among them, represents a battery cell at the moment the second regularization term; is the regularization coefficient, used to adjust the range of the balancing strategy; represents the ideal balancing strategy, set based on the optimal battery model.

8. A method for balancing control of a lithium battery pack chip according to claim 1, characterized in that, The error between the predicted health state of the battery cell and the actual health state of the battery cell is measured based on an error metric function; the error metric function is expressed as: ； wherein, is the error metric value, is the actual state of health of the battery cell, is the predicted state of health of the battery cell.

9. A method for balancing control of a lithium battery pack chip according to claim 1, characterized in that The self-recovery mechanism designs a mild anomaly strategy and a severe anomaly strategy according to the anomaly degree of the battery cell; Among them, the mild anomaly strategy alleviates the problem by adjusting the battery's balancing strategy to return the battery to its normal operating state; the severe anomaly strategy suspends or reduces the balancing operation of the battery cell and optimizes the performance of the battery pack by reallocating the load.

10. A lithium battery pack chip balancing control system, characterized in that, The system includes: A battery cell data acquisition unit, which is used to obtain battery cell data, construct a machine learning model based on the battery cell data to evaluate the battery health state, and obtain the health score of the corresponding battery cell; the machine learning model is trained using the feature vector after fusing the battery cell data; The battery cell health analysis unit is used to match the health score and health level of the battery cell, determine the health level of each battery cell, determine the equalization intensity of each battery cell through an optimization decision model based on the health score of the battery cell, and perform a first regularization operation on the equalization intensity of the battery cell to obtain the final equalization control quantity of the battery cell ; A strategy training unit, configured to input the equalization intensity of each battery cell and the health score of the battery cell into a SOM network to obtain an adaptive equalization strategy for the battery cell; perform a second regularization operation on the adaptive equalization strategy of the battery cell based on the final equalization control amount of the battery cell to obtain a final equalization strategy ; the SOM network maps the battery cells into similar groups according to the equalization intensity of each battery cell and the health score of the battery cell, and is trained through unsupervised learning; A fault prediction unit, which is used to use the final balancing strategy, the health score of the battery cell, and the balancing strength of each battery cell as a data set to input into a battery health prediction model based on a long short-term memory network to obtain the predicted health state of the battery cell; the battery health prediction model is used to capture the health change trend of the battery cell on a long time scale; A control optimization unit is used to determine whether there is an abnormality based on the error between the predicted health state of the battery cell and the actual health state of the battery cell. If the error exceeds the threshold, an alarm is triggered and the self-recovery mechanism is entered, and adaptive adjustment is performed according to the actual situation, so as to reduce the risk of battery damage; the self-recovery mechanism adjusts the equalization strategy based on the current health status and equalization intensity of the battery cell to obtain the final equalization strategy of the battery cell .

Images