Machine learning based hybrid energy storage device coordinated control method and system

Abstract

The present application relates to the technical field of energy storage control, in particular to a hybrid energy storage device cooperative control method and system based on machine learning. First, the expected workload spectrum is obtained according to historical operation data and service period, then the dominant control mode is determined and the mode health degree is generated according to the expected workload spectrum and the current health state of the system, the real-time state characteristics are extracted to input the reinforcement learning controller branch to obtain power distribution instructions, the life consumption increment and the comprehensive quality coefficient are calculated, and finally the effectiveness of the instructions is evaluated according to the quality coefficient threshold to realize control mode switching or controller parameter updating. By implementing the present application, the optimal balance of asset health and comprehensive benefit of the mobile storage and charging device hybrid energy storage system in the whole life cycle can be realized.

Description

Technical Field

[0001] This invention relates to the field of energy storage control technology, and in particular to a collaborative control method and system for hybrid energy storage devices based on machine learning. Background Technology

[0002] In the field of hybrid energy storage collaborative control, existing technologies are mostly geared towards fixed scenarios. For example, patent application number CN112260296A discloses an optimization method for hybrid energy storage systems to smooth intermittent loads in fixed distribution networks. Its core lies in power allocation through signal decomposition and capacity configuration based on static constraints, but it does not consider the impact of periodic changes in device operating modes on the long-term lifespan of energy storage components. Another patent application number CN120601424A involves the revenue optimization of commercial energy storage power stations, using deep reinforcement learning to balance economic benefits with equipment lifespan. However, its target is fixed power stations participating in electricity market dispatch, whose operation is continuous and driven by electricity prices.

[0003] However, mobile energy storage and charging devices, such as mobile charging robots, exhibit distinct tidal and intermittent operation characteristics: they operate at overload during peak service periods such as holidays, while remaining idle or underloaded on weekdays. Existing control strategies for fixed energy storage systems, whether based on static filtering-based power allocation or market-price-based scheduling optimization, are unable to adapt to this unique operating condition caused by drastic, periodic fluctuations in service demand over time. This leads to hybrid energy storage systems in mobile charging and charging devices experiencing accelerated lifespan degradation when pursuing peak-period service performance, while lacking proactive health maintenance mechanisms during idle periods, making it difficult to achieve an optimal balance between asset health and overall benefits throughout their entire lifecycle. Therefore, a dedicated control strategy capable of adaptive lifespan balancing for intermittent operation characteristics is urgently needed. Summary of the Invention

[0004] This invention addresses the problem in existing mobile energy storage devices where hybrid energy storage systems tend to experience accelerated lifespan degradation when pursuing peak service performance, while lacking proactive health maintenance mechanisms during idle periods, making it difficult to achieve an optimal balance between asset health and overall benefits throughout the entire lifecycle. The invention provides a machine learning-based collaborative control method and system for hybrid energy storage devices to solve this problem.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a machine learning-based collaborative control method for hybrid energy storage devices, comprising: predicting expected operating data for a future preset period based on historical operating datasets of a mobile energy storage and charging device; and analyzing and obtaining the expected workload spectrum of the mobile energy storage and charging device within the future preset period based on the expected operating data and the preset service period of the mobile energy storage and charging device; wherein the historical operating dataset includes health status indicators of the hybrid energy storage system; and the hybrid energy storage system includes a main energy storage battery and power-type energy storage elements. Based on the expected workload spectrum and the current health status of the hybrid energy storage system, the dominant control mode of the mobile energy storage and charging device in the next adaptive control cycle is determined through predefined mode decision rules, and the corresponding mode health is generated. The dominant control mode includes at least the performance priority mode, the life balance mode, the maintenance mode, and the economic benefit mode. Extract the real-time state features of the hybrid energy storage system and obtain N real-time optimization controller branches based on reinforcement learning training. Input the real-time state features and the dominant control mode parameters into the corresponding real-time optimization controller branches to obtain the real-time power allocation instruction set and the corresponding strategy output confidence of each component in the hybrid energy storage system. The system is controlled to operate according to the real-time power distribution instruction set, and the theoretical lifetime loss increment generated in the current control cycle is calculated based on the preset lifetime loss model. Based on preset rules, the initial overall quality coefficient of the device is calculated according to the dominant control mode, real-time power allocation instruction set and theoretical lifetime loss increment. The initial overall quality coefficient is then corrected using the mode health and the strategy output confidence to obtain the final overall quality coefficient. Based on the preset quality coefficient threshold and the final comprehensive quality coefficient, the effectiveness of the current real-time power allocation command is evaluated, and a decision is made to trigger the control mode or to update the controller parameters in real time.

[0006] Optionally, based on the historical operating dataset of the mobile energy storage and charging device, the expected operating data for a future preset period is predicted. Based on the expected operating data and the preset service period of the mobile energy storage and charging device, the expected workload spectrum of the mobile energy storage and charging device in the future preset period is analyzed and obtained, including: Obtain the historical operation dataset of the mobile energy storage and charging device. The historical operation dataset includes at least calendar time, load power sequence, and corresponding health status indicators of the hybrid energy storage system. Feature extraction is performed on the load power sequence arranged by timestamp in the historical operating dataset to obtain the historical working load spectrum; Based on time series analysis, and combining calendar features with preset service periods, the historical workload spectrum is correlated and predicted with the time tags of future preset periods. The average load power and load fluctuation intensity within the future preset period are output as the expected working load spectrum.

[0007] Optionally, based on the expected workload spectrum and the current health status of the hybrid energy storage system, the dominant control mode of the mobile energy storage and charging device in the next adaptive control cycle is determined through predefined mode decision rules, including: The peak intensity of the expected workload spectrum, the health status deviation between the current health status and the rated health status of the hybrid energy storage system, and whether the future preset period includes preset high-demand calendar events are used as input conditions. The input conditions are matched with a predefined pattern decision rule base, which stores the mapping relationship between different combinations of input conditions and recommended control modes. Based on the matching results, the dominant control mode is output.

[0008] The dominant control mode is output based on the matching results, including: The peak intensity of the expected workload spectrum is compared with a preset high intensity threshold, the current health status deviation is compared with a preset degradation threshold, and it is determined whether the future preset period includes holidays or preset high electricity price periods. Based on the above comparison and judgment results, mapping is performed according to the following rules: If the peak intensity exceeds the high intensity threshold, it is mapped to the performance priority mode; If the peak intensity does not exceed the high intensity threshold but the health status deviation exceeds the degradation threshold, then it is mapped to the maintenance mode; If the peak intensity does not exceed the high intensity threshold, the health status deviation does not exceed the degradation threshold, and the future preset period includes holidays or preset high electricity price periods, then it is mapped to the economic benefit model. If none of the above conditions are met, the default mapping will be to the lifetime balancing mode.

[0009] Optionally, the dominant control mode includes at least a performance-priority mode, a lifespan balancing mode, a maintenance mode, and an economic efficiency mode, including: When in performance priority mode, the upper limit of the output power of the power-type energy storage element is increased to its rated power, while the maximum output power of the main energy storage battery is limited. When in life balance mode, the optimization objective is to minimize the theoretical life loss increment of the main energy storage battery and the power energy storage element under the current operating conditions, and the power output ratio of the main energy storage battery and the power energy storage element is dynamically adjusted. When in maintenance mode, the state of charge of the main energy storage battery is controlled within a preset maintenance window range, and the power-type energy storage element is used to bear the external load fluctuations within the maintenance window range. When in economic benefit mode, the charging power and discharging power of the main energy storage battery are dynamically set according to the real-time electricity price signal, wherein the charging power is negatively correlated with the electricity price and the discharging power is positively correlated with the electricity price; at the same time, the power-type energy storage element is used to supplement the discharging power difference.

[0010] The generation of the corresponding pattern health score includes: Obtain the rule satisfaction score of the pattern decision rule base during the matching process. The rule satisfaction score is calculated based on the degree of matching between the input conditions and the rules corresponding to the selected dominant control mode. Obtain the prediction confidence level of the expected workload spectrum; Based on the rule satisfaction score and the prediction confidence, a quantified value between 0 and 1 is generated through weighted calculation as the health of the pattern.

[0011] Optionally, based on a preset lifetime loss model, the theoretical lifetime loss increment generated within the current control cycle is calculated, including: A first lifetime loss model for the main energy storage battery and a second lifetime loss model for the power-type energy storage element are established respectively. The first lifetime loss model uses current stress, temperature stress, and state-of-charge stress as input parameters, while the second lifetime loss model uses power stress and temperature stress as input parameters. At the end of each control cycle, the operating stress time-series data of the main energy storage battery and the power energy storage element during that control cycle are collected; The operating stress time series data are input into the corresponding first lifetime loss model and second lifetime loss model respectively, and the instantaneous value of the lifetime decay rate of the main energy storage battery and the power type energy storage element within the control cycle is calculated by the corresponding lifetime loss model. The instantaneous value of the lifetime decay rate is integrated over the time range of the control cycle to obtain the theoretical lifetime loss increment of the main energy storage battery and the power energy storage element during the control cycle.

[0012] Optionally, based on preset rules, the initial overall quality coefficient of the device is calculated according to the dominant control mode, the real-time power allocation instruction set, and the theoretical lifetime loss increment. The initial overall quality coefficient is then corrected using the mode health and the strategy output confidence level to obtain the final overall quality coefficient, including: The corresponding weight coefficient combination is determined based on the current dominant control mode. The weight coefficient combination includes load tracking accuracy weight, system efficiency weight, and lifetime loss cost weight. Calculate the load tracking accuracy score and system efficiency score based on the actual execution results of the real-time power allocation instruction set; The life loss cost score is calculated based on the theoretical life loss increment and the preset economic conversion factor. The initial comprehensive quality coefficient is obtained by weighting and summing the load tracking accuracy score, system efficiency score, and lifetime loss cost score using the weighted coefficient combination. The initial comprehensive quality coefficient is corrected based on the pattern health and the policy output confidence to obtain the final comprehensive quality coefficient.

[0013] Optionally, based on a preset quality coefficient threshold and the final overall quality coefficient, the effectiveness of the current real-time power allocation command is evaluated, and a decision is made to trigger a control mode or to update the controller parameters in real time, including: Two quality coefficient thresholds are preset, including a first threshold T1 and a second threshold T2, where T1 <T2; When the average value of the final comprehensive quality coefficient for X consecutive control cycles is lower than T1, the process of redetermining the dominant control mode is triggered; where X is a positive integer greater than or equal to 1. When the final comprehensive quality coefficient is higher than T2 for X consecutive control cycles, the online update of the parameters of the current real-time optimization controller branch is triggered. When the final overall quality coefficient is between T1 and T2, the current control strategy remains unchanged, and only the operating data is recorded for subsequent analysis.

[0014] Secondly, the present invention provides a machine learning-based collaborative control system for hybrid energy storage devices, comprising: The workload spectrum acquisition module is used to predict the expected operating data for a future preset period based on the historical operating dataset of the mobile energy storage and charging device. Based on the expected operating data and the preset service period of the mobile energy storage and charging device, the module analyzes and obtains the expected workload spectrum of the mobile energy storage and charging device in the future preset period. The historical operating dataset includes health status indicators of the hybrid energy storage system. The hybrid energy storage system includes a main energy storage battery and a power-type energy storage element. The dominant control mode determination module is used to determine the dominant control mode of the mobile energy storage device in the next adaptive control cycle based on the expected workload spectrum and the current health status of the hybrid energy storage system, and to generate the corresponding mode health status through predefined mode decision rules. The dominant control mode includes at least a performance priority mode, a life balance mode, a maintenance mode, and an economic benefit mode. The power allocation instruction acquisition module is used to extract the real-time state features of the hybrid energy storage system and obtain N real-time optimization controller branches based on reinforcement learning training. The real-time state features and the dominant control mode parameters are input into the corresponding real-time optimization controller branches to obtain the real-time power allocation instruction set and the corresponding strategy output confidence of each component in the hybrid energy storage system. The lifetime loss increment calculation module is used to control the operation of the hybrid energy storage system according to the real-time power distribution instruction set, and to calculate the theoretical lifetime loss increment generated in the current control cycle based on the preset lifetime loss model. The overall quality coefficient calculation module is used to calculate the initial overall quality coefficient of the device based on preset rules, according to the dominant control mode, real-time power distribution instruction set and theoretical lifetime loss increment, and to correct the initial overall quality coefficient using the mode health and the strategy output confidence to obtain the final overall quality coefficient. The system update and optimization module is used to evaluate the effectiveness of the current real-time power allocation command based on the preset quality coefficient threshold and the final comprehensive quality coefficient, and to decide whether to trigger the control mode or update the controller parameters in real time.

[0015] By implementing this invention, it is possible to predict expected operating data for a future preset period based on historical operating datasets of mobile energy storage and charging devices. Based on the expected operating data and the preset service period of the mobile energy storage and charging devices, the expected workload spectrum of the mobile energy storage and charging devices in the future preset period can be analyzed and obtained. The historical operating dataset includes health status indicators of the hybrid energy storage system, which includes a main energy storage battery and power-type energy storage elements. This breaks through the limitation of traditional control strategies that lack advance load perception, providing objective and quantitative load basis for accurate decision-making of subsequent control modes, and allowing the control strategy to be formulated in accordance with the actual future operating needs of the device, rather than just based on the current static operating conditions.

[0016] By implementing this invention, based on the expected workload spectrum and the current health status of the hybrid energy storage system, and through predefined mode decision rules, the dominant control mode of the mobile energy storage and charging device in the next adaptive control cycle can be determined, and a corresponding mode health score can be generated. The dominant control mode includes at least a performance-priority mode, a lifespan balancing mode, a maintenance mode, and an economic benefit mode. This enables adaptive selection of the control mode, allowing flexible switching between different modes such as performance-priority and lifespan balancing according to load characteristics and equipment health status, adapting to the tidal and intermittent operating characteristics of the mobile energy storage and charging device. The generation of the mode health score provides a quantitative reference for subsequent control effect evaluation at the mode selection level, improving the scientific nature of mode decision-making.

[0017] By implementing this invention, it is possible to extract the real-time state features of the hybrid energy storage system and obtain N real-time optimization controller branches based on reinforcement learning training. The real-time state features and the parameters of the dominant control mode are input into the corresponding real-time optimization controller branches to obtain the real-time power allocation instruction set and corresponding strategy output confidence of each component in the hybrid energy storage system. Leveraging the optimization capabilities of the reinforcement learning model, the power allocation instructions are made to align with the core objectives of the real-time operating conditions and the dominant control mode, achieving refined and real-time power allocation of energy storage components. The multi-controller branch design can adapt to different modes and operating conditions, and the strategy output confidence provides a quantitative indicator at the model output level for subsequent strategy effectiveness evaluation.

[0018] By implementing this invention, it is possible to control the operation of a hybrid energy storage system according to the real-time power allocation instruction set, and calculate the theoretical lifetime loss increment generated in the current control cycle based on a preset lifetime loss model; to achieve precise quantitative monitoring of the lifetime loss of energy storage components, so that the execution effect of the control strategy can be intuitively fed back from the perspective of equipment lifetime, breaking the problem that traditional strategies cannot perceive lifetime loss in real time, and providing core data on lifetime loss dimension for subsequent comprehensive quality assessment and strategy optimization.

[0019] By implementing this invention, it is possible to calculate the initial overall quality coefficient of the device based on preset rules, according to the dominant control mode, real-time power allocation instruction set, and theoretical lifetime loss increment. The initial overall quality coefficient is then corrected using the mode health and the strategy output confidence level to obtain the final overall quality coefficient. This allows for a comprehensive evaluation of the control strategy's execution effect from multiple dimensions, including load tracking, system efficiency, and lifetime loss cost, rather than a single dimension. Furthermore, by correcting the mode health and strategy output confidence level, the impact of uncertainties in mode selection and model output on the evaluation results is eliminated, enabling the overall quality coefficient to more realistically and comprehensively reflect the actual performance of the current control strategy.

[0020] By implementing this invention, it is possible to evaluate the effectiveness of the current real-time power allocation command based on a preset quality coefficient threshold and the final comprehensive quality coefficient, and decide whether to trigger the control mode or update the controller parameters in real time. This establishes a closed-loop optimization mechanism for the control strategy, enabling dynamic evaluation and adaptive optimization of the control strategy's effectiveness. When the control effect is poor, the control mode is switched in a timely manner; when the control effect is excellent, the controller parameters are optimized, allowing the control strategy to continuously adapt to changes in the device's operating conditions, avoiding the drawbacks of traditional strategies that remain fixed once formulated.

[0021] In summary, by implementing this invention, it is possible to effectively suppress the lifespan loss of energy storage components while meeting the core requirements of performance and economy under different operating scenarios, and to achieve the optimal balance between asset health and comprehensive benefits throughout the entire life cycle of the mobile energy storage and charging device hybrid energy storage system. Attached Figure Description

[0022] Figure 1 A flowchart illustrating the collaborative control method for hybrid energy storage devices based on machine learning provided by this invention. Figure 2 A schematic diagram of the structure of the machine learning-based hybrid energy storage device collaborative control system provided by the present invention.

[0023] In the attached diagram, the components represented by each number are as follows: The system includes a workload spectrum acquisition module 11, a dominant control mode determination module 12, a power distribution command acquisition module 13, a life loss increment calculation module 14, a comprehensive quality coefficient calculation module 15, and a system update and optimization module 16. Detailed Implementation

[0024] Example 1, as Figure 1 As shown, embodiments of the present invention provide a collaborative control method and system for hybrid energy storage devices based on machine learning, including: S100: Based on the historical operation dataset of the mobile energy storage and charging device, predict the expected operation data for a future preset period. Based on the expected operation data and the preset service period of the mobile energy storage and charging device, analyze and obtain the expected workload spectrum of the mobile energy storage and charging device in the future preset period. The historical operation dataset includes health status indicators of the hybrid energy storage system. The hybrid energy storage system includes a main energy storage battery and a power-type energy storage element. S200: Based on the expected workload spectrum and the current health status of the hybrid energy storage system, the dominant control mode of the mobile energy storage and charging device in the next adaptive control cycle is determined through predefined mode decision rules, and the corresponding mode health is generated. The dominant control mode includes at least performance priority mode, life balance mode, maintenance mode and economic benefit mode. S300: Extract the real-time state features of the hybrid energy storage system and obtain N real-time optimization controller branches based on reinforcement learning training. Input the real-time state features and the dominant control mode parameters into the corresponding real-time optimization controller branches to obtain the real-time power allocation instruction set and the corresponding strategy output confidence of each component in the hybrid energy storage system. S400: Controls the operation of the hybrid energy storage system according to the real-time power distribution instruction set, and calculates the theoretical lifetime loss increment generated in the current control cycle based on the preset lifetime loss model. S500: Based on preset rules, according to the dominant control mode, real-time power allocation instruction set and theoretical lifetime loss increment, calculate the initial comprehensive quality coefficient of the device, and use the mode health and the strategy output confidence to correct the initial comprehensive quality coefficient to obtain the final comprehensive quality coefficient. S600: Based on the preset quality coefficient threshold and the final comprehensive quality coefficient, evaluate the effectiveness of the current real-time power allocation command and decide whether to trigger the control mode or update the controller parameters in real time.

[0025] In step S100 of this application embodiment, the expected operating data for a future preset period is predicted based on the historical operating dataset of the mobile energy storage and charging device. Based on the expected operating data and the preset service period of the mobile energy storage and charging device, the expected workload spectrum of the mobile energy storage and charging device in the future preset period is analyzed and obtained, including: Obtain the historical operation dataset of the mobile energy storage and charging device. The historical operation dataset includes at least calendar time, load power sequence, and corresponding health status indicators of the hybrid energy storage system. Feature extraction is performed on the load power sequence arranged by timestamp in the historical operating dataset to obtain the historical working load spectrum; Based on time series analysis, and combining calendar features with preset service periods, the historical workload spectrum is correlated and predicted with the time tags of future preset periods. The average load power and load fluctuation intensity within the future preset period are output as the expected working load spectrum.

[0026] In this embodiment of the application, the purpose of step S100 is to provide accurate load basis for determining the dominant control mode of the hybrid energy storage system of the mobile energy storage and charging device, predict the operating load characteristics in the future preset period in advance, adapt to the tidal and intermittent operation characteristics of the mobile energy storage and charging device, avoid the adaptability problem of the fixed energy storage control strategy, and make the subsequent control mode selection and power allocation more in line with the actual operation needs, so as to lay a data foundation for achieving the optimal balance between asset health and comprehensive benefits throughout the entire life cycle.

[0027] To achieve the above objectives, it is first necessary to obtain the historical operation dataset of the mobile energy storage and charging device. The historical operation dataset includes at least calendar time, load power sequence, and corresponding health status indicators of the hybrid energy storage system. This involves directly collecting relevant historical operating data accumulated during the operation of mobile energy storage and charging devices. The historical operating dataset must include three core elements: calendar time, load power sequence, and corresponding health status indicators of the hybrid energy storage system.

[0028] For example, calendar time can be collected as specific timestamps such as 9:00 on February 1, 2026, 10:00 on February 2, 2026; load power sequence can be collected as the load power value per second within a certain period, such as 10kW, 12kW, 9kW, etc.; health status indicators of hybrid energy storage system can be collected as the state of charge of the main energy storage battery at 90% and the temperature of the power storage element at 25℃, etc.

[0029] Then, feature extraction is required from the load power sequence arranged by timestamp in the historical operating dataset to obtain the historical workload spectrum; Specifically, for load power sequences arranged in order of timestamps, feature extraction methods are used to mine the load patterns, and the regularized load features are integrated into a historical working load spectrum. This spectrum is a quantitative summary of the past load operation characteristics of the device.

[0030] For example, extracting features such as the average load power of 15kW during the morning peak hours of 8-10 am, the average load power of 18kW during the evening peak hours of 18-20 pm, and the average load power of 25kW throughout the weekend on weekdays, and integrating them to form the historical workload spectrum for that week.

[0031] Next, based on time series analysis methods, combined with calendar features and preset service periods, it is necessary to correlate and predict the historical workload spectrum with the time tags of the future preset period. This involves using time series analysis to analyze historical load patterns, and combining calendar characteristics and preset service periods as two key conditions to establish a correlation between historical workload spectrum and future preset cycle time labels, thereby enabling the prediction of future loads.

[0032] For example, calendar features can refer to time attributes such as holidays, weekdays, and weekends; the preset service period can be set as the operating period of the mobile storage and charging device for the next 3 months; the time label of the future preset period can be each time period of the first week of March 2026. Through time series analysis, the historical pattern of high load on weekends and flat load on weekdays can be associated with the corresponding time label of the first week of March 2026 to predict the load.

[0033] Finally, the average load power and load fluctuation intensity within the future preset period are output as the expected working load spectrum. The system will output two core load indicators for the future preset period, namely average load power and load fluctuation intensity. These two indicators together constitute the expected workload spectrum and become the core input basis for subsequent mode decision-making.

[0034] For example, if the output for the future preset period from March 1st to 7th, 2026 is an average daily load power of 20kW and a load fluctuation intensity of ±5kW, this data is the expected working load spectrum for that period.

[0035] In step S200 of this embodiment, based on the expected workload spectrum and the current health status of the hybrid energy storage system, the dominant control mode of the mobile energy storage and charging device in the next adaptive control cycle is determined through predefined mode decision rules, including: The peak intensity of the expected workload spectrum, the health status deviation between the current health status and the rated health status of the hybrid energy storage system, and whether the future preset period includes preset high-demand calendar events are used as input conditions. The input conditions are matched with a predefined pattern decision rule base, which stores the mapping relationship between different combinations of input conditions and recommended control modes. Based on the matching results, the dominant control mode is output.

[0036] In this embodiment of the application, the purpose of step S200 is to determine the appropriate dominant control mode based on the expected workload spectrum and the real-time health status of the hybrid energy storage system obtained in the early stage through standardized rule matching. This allows the hybrid energy storage system of the mobile energy storage and charging device to select targeted control strategies such as performance priority and life balance according to future load characteristics and its own health status, adapting to its tidal and intermittent operation characteristics, avoiding the problem of excessively rapid life loss or low efficiency caused by a single control mode, and providing a clear control direction for subsequent power distribution and system optimization.

[0037] To achieve the above objectives, the peak intensity of the expected workload spectrum, the health status deviation between the current health status and the rated health status of the hybrid energy storage system, and whether the future preset period includes preset high-demand calendar events are all taken as input conditions. Three sets of key data were extracted and integrated as input conditions for the model decision: the peak intensity of the expected workload spectrum, the health status deviation of the hybrid energy storage system, and the judgment results of high-demand calendar events in the future preset period. The health status deviation is the difference between the current health status and the rated health status of the hybrid energy storage system. The preset high-demand calendar events include holidays, preset high electricity price periods, etc.

[0038] For example, if the peak intensity of the expected workload spectrum is 30kW, the current health status of the main energy storage battery is 80%, and the rated health status is 100%, then the health status deviation is 20%. If the future preset cycle includes the Spring Festival holiday, it is determined that there is a preset high demand calendar event.

[0039] Next, the input conditions are matched with a predefined pattern decision rule base, which stores the mapping relationship between different combinations of input conditions and recommended control modes; This involves retrieving a predefined pattern decision rule base, which stores a one-to-one mapping between different combinations of input conditions and recommended control patterns, serving as the standardized basis for pattern decision-making. The three types of input condition combinations actually extracted are then compared one by one with the preset condition combinations in the pattern decision rule base to complete the rule matching operation.

[0040] Then, based on the matching results, the dominant control mode is output.

[0041] In step S200 of this application embodiment, the dominant control mode is output according to the matching result, including: The peak intensity of the expected workload spectrum is compared with a preset high intensity threshold, the current health status deviation is compared with a preset degradation threshold, and it is determined whether the future preset period includes holidays or preset high electricity price periods. Based on the above comparison and judgment results, mapping is performed according to the following rules: If the peak intensity exceeds the high intensity threshold, it is mapped to the performance priority mode; If the peak intensity does not exceed the high intensity threshold but the health status deviation exceeds the degradation threshold, then it is mapped to the maintenance mode; If the peak intensity does not exceed the high intensity threshold, the health status deviation does not exceed the degradation threshold, and the future preset period includes holidays or preset high electricity price periods, then it is mapped to the economic benefit model. If none of the above conditions are met, the default mapping will be to the lifetime balancing mode.

[0042] In this embodiment, the purpose of step S200 is to transform the initial input conditions into a specific dominant control mode through quantitative threshold comparison and clear calendar event determination. This allows the mode mapping process to have standardized and implementable judgment criteria, ensuring that the output dominant control mode can accurately match the future load characteristics of the mobile energy storage and charging device and the health status of the hybrid energy storage system. This avoids the subjectivity of mode selection and provides a clear and unique execution direction for the subsequent power allocation and operation control of the hybrid energy storage system. It adapts to the tidal and intermittent operation characteristics of the device and takes into account system operation performance, equipment life, maintenance needs, and economic benefits.

[0043] To achieve the above objectives, it is first necessary to compare the peak intensity of the expected workload spectrum with the preset high intensity threshold, compare the current health status deviation with the preset degradation threshold, and determine whether the future preset period includes holidays or preset high electricity price periods. That is, three independent condition determination tasks are carried out. The first group compares the peak intensity of the expected workload spectrum with the preset high intensity threshold. The second group compares the current health status deviation of the hybrid energy storage system with the preset degradation threshold. The third group makes a qualitative judgment on the time attributes of the future preset period to determine whether it includes holidays or preset high electricity price periods. The results of the three determinations serve as the core basis for subsequent mode mapping.

[0044] For example, the preset high-intensity threshold is 28kW, the expected peak intensity of the working load spectrum is 35kW, and the comparison shows that the peak intensity exceeds the high-intensity threshold; the preset degradation threshold is 15%, the current health status deviation of the hybrid energy storage system is 20%, and the comparison shows that the health status deviation exceeds the degradation threshold; the preset future cycle includes the National Day holiday, and the judgment shows that it includes preset high-demand calendar events.

[0045] Next, based on the above comparison and judgment results, mapping needs to be performed according to preset rules. That is, based on the judgment results of the three conditions, the preset mapping rules with progressive priority are strictly followed to complete the corresponding matching of input conditions to the dominant control mode. During the rule execution process, the conditions are judged in sequence according to their satisfaction. If the preceding conditions are satisfied, the corresponding mode is directly mapped. If not satisfied, the subsequent conditions are judged. If no conditions are satisfied, the default mapping is executed.

[0046] Specifically, if the peak intensity exceeds the high intensity threshold, it is mapped to the performance priority mode; For example, if the peak intensity is 32kW, which exceeds the preset high intensity threshold of 28kW, no other conditions need to be determined, and it is directly mapped to the performance priority mode; If the peak intensity does not exceed the high intensity threshold but the health status deviation exceeds the degradation threshold, then it is mapped to the maintenance mode; For example, if the peak intensity is 25kW, which does not exceed the high intensity threshold of 28kW, but the health status deviation is 18%, which exceeds the preset degradation threshold of 15%, it is mapped to the maintenance mode. If the peak intensity does not exceed the high intensity threshold, the health status deviation does not exceed the degradation threshold, and the future preset period includes holidays or preset high electricity price periods, then it is mapped to the economic benefit model. For example, if the peak intensity is 24kW, which does not exceed the high intensity threshold, the health status deviation is 10%, which does not exceed the degradation threshold, and the future preset period includes the high electricity price period in industrial and commercial parks, it is mapped to the economic benefit model. If none of the above conditions are met, the default mapping will be to the lifetime balancing mode; For example, if the peak intensity is 23kW, which does not exceed the high intensity threshold, the health status deviation is 12%, which does not exceed the degradation threshold, and the future preset cycle is only ordinary working days without high electricity prices, and no prerequisite conditions are met, it will be mapped to the lifetime balance mode by default.

[0047] Finally, based on the final mapping result, the dominant control mode of the mobile energy storage and charging device in the next adaptive control cycle is determined, and the mode output is completed. The output mode is the only one among the performance priority mode, maintenance mode, economic benefit mode, and life balance mode. This result will be directly transmitted to the subsequent power distribution stage as the core guiding basis for the operation and control of the hybrid energy storage system.

[0048] For example, if the mapping determines that the mode is performance-priority, then this mode will be output directly. Subsequently, the hybrid energy storage system will adjust the power output rules of the main energy storage battery and power-type energy storage elements according to the requirements of the performance-priority mode.

[0049] In step S200 of this application embodiment, the dominant control mode includes at least a performance priority mode, a lifespan balancing mode, a maintenance mode, and an economic benefit mode, including: When in performance priority mode, the upper limit of the output power of the power-type energy storage element is increased to its rated power, while the maximum output power of the main energy storage battery is limited. When in life balance mode, the optimization objective is to minimize the theoretical life loss increment of the main energy storage battery and the power energy storage element under the current operating conditions, and the power output ratio of the main energy storage battery and the power energy storage element is dynamically adjusted. When in maintenance mode, the state of charge of the main energy storage battery is controlled within a preset maintenance window range, and the power-type energy storage element is used to bear the external load fluctuations within the maintenance window range. When in economic benefit mode, the charging power and discharging power of the main energy storage battery are dynamically set according to the real-time electricity price signal, wherein the charging power is negatively correlated with the electricity price and the discharging power is positively correlated with the electricity price; at the same time, the power-type energy storage element is used to supplement the discharging power difference.

[0050] In step S200 of this application embodiment, the purpose of the above steps is to clarify the operation control rules of the four dominant control modes, so that the hybrid energy storage system with different load characteristics and system health states has corresponding standardized operation strategies, and to specifically adapt to the tidal and intermittent operation characteristics of mobile energy storage and charging devices, so as to achieve the goals of peak period performance guarantee, normal operating condition life protection, fault precursor proactive maintenance, and electricity price-driven revenue improvement, and ultimately enable the hybrid energy storage system to achieve the optimal balance between asset health and comprehensive benefits throughout its entire life cycle, while providing a clear control execution basis for the generation of subsequent power allocation commands.

[0051] To achieve the above objectives, the following is specifically required: When in performance priority mode, the upper limit of the output power of the power-type energy storage element is increased to its rated power, while the maximum output power of the main energy storage battery is limited. The core principle is to formulate rules to improve the output performance of hybrid energy storage systems. The upper limit of the output power of power-type energy storage components is directly set to their rated power to give full play to their power response advantages. At the same time, the maximum output power of the main energy storage battery is limited to reduce its loss under high load and ensure the load supply capacity during peak periods.

[0052] For example, the rated power of a power-type energy storage element is 50kW, and its output power is set to a maximum of 50kW. The original maximum output power of the main energy storage battery is 40kW, but it is limited to 20kW. The high load demand is mainly met by the power-type energy storage element.

[0053] When in life balance mode, the optimization objective is to minimize the theoretical life loss increment of the main energy storage battery and the power energy storage element under the current operating conditions, and the power output ratio of the main energy storage battery and the power energy storage element is dynamically adjusted. The core optimization objective is to minimize the theoretical lifespan loss increment of the main energy storage battery and the power-type energy storage element under the current operating conditions. The algorithm calculates the lifespan loss of the two elements in real time and dynamically adjusts the power output ratio of the two elements to make the lifespan loss of the two elements tend to be balanced and avoid excessive loss of a single element.

[0054] For example, under the current operating conditions, the theoretical lifespan loss increment of the main energy storage battery outputting 1kW of power is 0.002%, and the theoretical lifespan loss increment of the power-type energy storage element outputting 1kW of power is 0.001%. In order to minimize the overall loss, the power output ratio of the two is adjusted to 60% for the main energy storage battery and 40% for the power-type energy storage element.

[0055] When in maintenance mode, the state of charge of the main energy storage battery is controlled within a preset maintenance window range, and the power-type energy storage element is used to bear the external load fluctuations within the maintenance window range. This involves conducting proactive health maintenance on the main energy storage battery. First, a fixed state of charge (SCC) maintenance window is set, and the SCC is strictly controlled within this window range to isolate the influence of external load fluctuations. At the same time, the power-type energy storage element fully bears the external load fluctuations during the maintenance period, ensuring that the main energy storage battery is in a stable working state.

[0056] For example, the state of charge maintenance window of the main energy storage battery can be set to 50%-60%, and its state of charge can be maintained within this range in real time. When the external load suddenly increases from 10kW to 18kW, the power-type energy storage element can supplement the fluctuating load of 8kW.

[0057] When in economic benefit mode, the charging power and discharging power of the main energy storage battery are dynamically set according to the real-time electricity price signal, wherein the charging power is negatively correlated with the electricity price and the discharging power is positively correlated with the electricity price; at the same time, the power-type energy storage element is used to supplement the discharging power difference. This involves formulating power regulation rules based on real-time electricity price signals, linking the charging and discharging power of the main energy storage battery with the electricity price. Charging power is negatively correlated with the electricity price, while discharging power is positively correlated with the electricity price. At the same time, power-type energy storage elements supplement the discharge power difference of the main energy storage battery, balancing revenue and power supply stability.

[0058] For example, when the real-time electricity price is 0.5 yuan / kWh, the charging power of the main energy storage battery is set to 30kW and the discharging power is set to 10kW; when the real-time electricity price rises to 1.2 yuan / kWh, the charging power is reduced to 5kW and the discharging power is increased to 40kW. When the load demand is 45kW, the power-type energy storage element supplements the 5kW discharging power difference.

[0059] In step S200 of this application embodiment, generating the corresponding mode health score includes: Obtain the rule satisfaction score of the pattern decision rule base during the matching process. The rule satisfaction score is calculated based on the degree of matching between the input conditions and the rules corresponding to the selected dominant control mode. Obtain the prediction confidence level of the expected workload spectrum; Based on the rule satisfaction score and the prediction confidence, a quantified value between 0 and 1 is generated through weighted calculation as the health of the pattern.

[0060] In this embodiment of the application, the purpose of step S200 is to quantitatively evaluate the selected dominant control mode and generate a mode health quantification value between 0 and 1, thereby reflecting the degree of fit between the selected dominant control mode and the actual input conditions and expected load forecast results. This provides a key quantitative basis for the subsequent correction of the comprehensive quality coefficient, making the control strategy evaluation of the hybrid energy storage system more accurate and more in line with the actual operating conditions, and avoiding the failure of the control strategy due to mode matching deviation or load forecast error.

[0061] To achieve the above objectives, it is first necessary to obtain the rule satisfaction score of the pattern decision rule base during the matching process. The rule satisfaction score is calculated based on the degree of matching between the input conditions and the rules corresponding to the selected dominant control mode. This refers to the rule satisfaction score generated by the extraction mode decision rule base during the matching process between input conditions and the dominant control mode. The rule satisfaction score can be calculated based on the degree of matching between the actual input conditions and the corresponding rules of the selected dominant control mode. The rule satisfaction score result can reflect the fit between the input conditions and the mode rules, and the score is usually a value between 0 and 1.

[0062] For example, if the actual peak intensity far exceeds the preset high intensity threshold, and there are no other conditions interfering, the performance priority mode is directly matched. The input conditions and the corresponding rules are completely consistent, and the rule satisfaction score can be counted as 0.95. If the peak intensity just exceeds the high intensity threshold and the health status deviation is close to the degradation threshold, there are some condition conflicts when matching the performance priority mode, and the rule satisfaction score can be counted as 0.6.

[0063] Next, it is necessary to obtain the prediction confidence level of the expected workload spectrum; This refers to the confidence level of the expected workload spectrum generated when predicting future loads based on historical operating datasets. This confidence level reflects the reliability of the load prediction results and is output by the prediction methods themselves, such as time series analysis. The score is also a value between 0 and 1, with a higher value indicating a more reliable prediction result.

[0064] For example, if a large amount of historical operational data is used and there are no special unforeseen calendar events in the future, the predicted expected workload spectrum is highly reliable, and the prediction confidence level can be counted as 0.9; if there are uncertain temporary high demand events in the future, the uncertainty of the load forecast increases, and the prediction confidence level can be counted as 0.7.

[0065] Then, based on the rule satisfaction score and the prediction confidence, a quantitative value between 0 and 1 needs to be generated through weighted calculation as the health of the pattern. This involves assigning reasonable weighting coefficients to the rule satisfaction score and the expected workload spectrum prediction confidence level, and then calculating the two values ​​by weighted summation to generate a quantitative value between 0 and 1, which is the corresponding model health.

[0066] For example, if the rule satisfaction score is weighted at 0.6 and the expected workload spectrum prediction confidence score is weighted at 0.4, then if the rule satisfaction score is 0.9 and the prediction confidence score is 0.8, the weighted model health score is calculated as 0.9 × 0.6 + 0.8 × 0.4 = 0.86; if the rule satisfaction score is 0.7 and the prediction confidence score is 0.6, the weighted model health score is calculated as 0.7 × 0.6 + 0.6 × 0.4 = 0.66.

[0067] In step S300 of this application embodiment, it is necessary to extract the real-time state features of the hybrid energy storage system and obtain N real-time optimization controller branches based on reinforcement learning training. The real-time state features and the dominant control mode parameters are input into the corresponding real-time optimization controller branches to obtain the real-time power allocation instruction set and the corresponding strategy output confidence of each component in the hybrid energy storage system. The system combines the real-time operating status of the hybrid energy storage system with the selected dominant control mode. Relying on a dedicated controller branch trained by reinforcement learning, it generates precise power allocation commands adapted to the current operating conditions. At the same time, it provides a strategy output confidence level to reflect the reliability of the commands. This provides specific and executable power control basis for the real-time operation of each component of the hybrid energy storage system, ensuring that the power allocation not only meets the core objectives of the dominant control mode but also fits the real-time status of the system and adapts to the intermittent operation characteristics of mobile energy storage and charging devices.

[0068] To achieve the above objectives, it is first necessary to extract the real-time status characteristics of the hybrid energy storage system. This involves collecting the operating status data of each core component in the hybrid energy storage system in real time and integrating them to form real-time status characteristics. These real-time status characteristics are the core data reflecting the current operating status of the system, and at least include the state of charge and temperature of the aqueous sodium-ion battery, the state of charge of the power storage components, and the current load power demand.

[0069] For example, the extracted real-time status features are: the aqueous sodium-ion battery has a state of charge of 75% and a temperature of 28°C; the power storage element has a state of charge of 80%; and the current load requires 30kW of power.

[0070] Next, it is necessary to obtain N real-time optimization controller branches based on reinforcement learning training; This involves retrieving N real-time optimized controller branches that have been pre-trained using reinforcement learning algorithms. These real-time optimized controller branches are specifically trained for different operating scenarios and dominant control modes of the hybrid energy storage system. They can accurately output appropriate power allocation strategies based on the input parameters. The value of N can be set according to actual control requirements.

[0071] The method for obtaining the N real-time optimization controller branches based on reinforcement learning training described here is an existing mature technology, well known to those skilled in the art, and will not be elaborated here.

[0072] For example, based on four dominant control modes—performance priority, lifespan balance, maintenance, and economic benefits—four corresponding real-time optimization controller branches are trained, i.e., N=4. Each branch completes algorithm training for the control objective of its corresponding mode.

[0073] Finally, parameters need to be input to the corresponding controller branch to obtain the real-time power allocation instruction set and policy output confidence level; Based on the selected dominant control mode, the corresponding real-time optimization controller branch is matched. The extracted real-time state characteristics of the hybrid energy storage system, together with the relevant parameters of the dominant control mode, are input into this controller branch. After calculation by the controller branch, the real-time power allocation instruction set of the main energy storage battery and power-type energy storage element in the hybrid energy storage system is output. At the same time, the corresponding strategy output confidence level is output. The confidence level is a quantized value between 0 and 1, reflecting the reliability of the output power allocation strategy.

[0074] For example, if the selected dominant control mode is the performance-priority mode, the corresponding performance-priority controller branch is matched. The real-time state characteristics of the aqueous sodium-ion battery (75% state of charge, 28°C), the power storage element (80% state of charge), and the current load power demand (30kW) are input into this branch along with the performance-priority mode parameters (power storage element rated power 50kW, main energy storage battery maximum output power limit 20kW). After calculation, the real-time power allocation instruction set is obtained: power storage element output 30kW, main energy storage battery output 0kW, and the output strategy confidence level is 0.92.

[0075] In step S400 of this application embodiment, it is necessary to control the operation of the hybrid energy storage system according to the real-time power distribution instruction set, and calculate the theoretical lifetime loss increment generated in the current control cycle based on the preset lifetime loss model. In step S400 of this application embodiment, based on a preset lifetime loss model, the theoretical lifetime loss increment generated within the current control cycle is calculated, including: A first lifetime loss model for the main energy storage battery and a second lifetime loss model for the power-type energy storage element are established respectively. The first lifetime loss model uses current stress, temperature stress, and state-of-charge stress as input parameters, while the second lifetime loss model uses power stress and temperature stress as input parameters. At the end of each control cycle, the operating stress time-series data of the main energy storage battery and the power energy storage element during that control cycle are collected; The operating stress time series data are input into the corresponding first lifetime loss model and second lifetime loss model respectively, and the instantaneous value of the lifetime decay rate of the main energy storage battery and the power type energy storage element within the control cycle is calculated by the corresponding lifetime loss model. The instantaneous value of the lifetime decay rate is integrated over the time range of the control cycle to obtain the theoretical lifetime loss increment of the main energy storage battery and the power energy storage element during the control cycle.

[0076] In this embodiment, the purpose of step S400 is, on the one hand, to control the coordinated operation of each component of the hybrid energy storage system according to precise power allocation instructions, ensuring that the system conforms to the dominant control mode to achieve the operating target; on the other hand, through a preset dedicated lifetime loss model, to quantitatively calculate the theoretical lifetime loss increment of the main energy storage battery and the power-type energy storage element in the current control cycle, providing core lifetime loss data for subsequent comprehensive quality coefficient calculation and control strategy effectiveness evaluation, realizing accurate monitoring of the lifetime loss of energy storage elements, and avoiding blind optimization of control strategies due to the lack of quantified loss data.

[0077] To achieve the above objectives, it is first necessary to control the operation of the hybrid energy storage system according to the real-time power distribution instruction set. This means that the real-time power distribution command set output by the S300 stage is used as the basis for execution to precisely regulate the power output of the main energy storage battery and power-type energy storage elements in the hybrid energy storage system, so that the two can work together according to the power values ​​of the command set, and ensure that the system operation status matches the core requirement of the dominant control mode.

[0078] For example, if the real-time power distribution instruction set is 25kW output from the power storage element and 10kW output from the main energy storage battery, then both will be directly controlled to operate at that power value to meet the current load demand of 35kW, while also conforming to the corresponding control mode requirements.

[0079] Then, it is necessary to establish a first lifetime loss model for the main energy storage battery and a second lifetime loss model for the power energy storage element. The first lifetime loss model uses current stress, temperature stress, and state-of-charge stress as input parameters, while the second lifetime loss model uses power stress and temperature stress as input parameters. This involves building different lifetime loss models for two types of energy storage components. The main energy storage battery corresponds to the first lifetime loss model, with input parameters including current stress, temperature stress, and state of charge stress, accurately matching the loss influencing factors of the main energy storage battery. The power-type energy storage component corresponds to the second lifetime loss model, with input parameters including power stress and temperature stress, which fits the loss characteristics of its power-type operation.

[0080] Specifically, for the first life loss model: multiple accelerated aging experiments with different current stress, temperature stress, and state of charge stress can be designed for the main energy storage battery. The battery's life characteristics data such as capacity decay and internal resistance change under various stress combinations can be continuously collected. Through methods such as multiple regression and machine learning, the functional relationship between the three types of stress and the battery life decay rate can be fitted. After determining the model parameters, the final first life loss model can be obtained.

[0081] For the second lifetime loss model: Accelerated aging experiments with different power stresses and temperature stresses can be designed for power-type energy storage elements to collect data such as element cycle life and power characteristic decay. By fitting the data, a quantization function of the two types of stresses and the element lifetime decay rate can be established, and after calibration, a suitable second lifetime loss model can be obtained.

[0082] After modeling, the model will be continuously revised based on actual operating data to improve its adaptability to actual working conditions.

[0083] For example, the first lifetime loss model built for aqueous sodium-ion batteries uses current stresses such as charging current of 10A and discharging current of 8A, temperature stresses such as 25℃ and 30℃, and state-of-charge stresses such as 60% and 70% as inputs; the second lifetime loss model built for supercapacitor-type power energy storage devices uses power stresses such as 20kW and 30kW, and temperature stresses such as 26℃ and 29℃ as inputs.

[0084] Next, at the end of each control cycle, the operating stress timing data of the main energy storage battery and the power energy storage element during that control cycle are collected; This involves setting a fixed adaptive control cycle, and after each cycle is completed, accurately collecting time-series data on the current stress, temperature stress, and state-of-charge stress of the main energy storage battery, as well as the power stress and temperature stress of the power storage element. The time-series data must include stress values ​​at different time points within the cycle to reflect the stress change process.

[0085] For example, if the control cycle is set to 1 hour, after the cycle ends, collect time-series data such as the current value of the main energy storage battery (10A-9A-10A), temperature value (28℃-29℃-28℃), and state of charge (70%-69%-68%) per minute during that hour; and collect time-series data such as the power value of the power storage element (25kW-26kW-25kW) and temperature value (27℃-28℃-27℃) per minute.

[0086] Furthermore, the operating stress time series data are input into the corresponding first lifetime loss model and second lifetime loss model respectively, and the instantaneous values ​​of the lifetime decay rate of the main energy storage battery and the power type energy storage element within the control cycle are calculated through the corresponding lifetime loss model. The time-series data of the main energy storage battery's operating stress is input into the first lifetime loss model, and the time-series data of the power energy storage element's operating stress is input into the second lifetime loss model. Through the algorithm calculation of the model, the instantaneous values ​​of the lifetime decay rate of the two types of energy storage elements at different time points within the control cycle are calculated respectively. This value reflects how fast the lifetime decays of the element at a certain instant.

[0087] For example, after inputting the stress time series data of the main energy storage battery into the first model, the instantaneous value of its lifetime decay rate in the 10th minute of the cycle is calculated to be 0.0005% / min and in the 20th minute to be 0.0004% / min; after inputting the stress time series data of the power energy storage element into the second model, the instantaneous value of its lifetime decay rate in the 10th minute is calculated to be 0.0002% / min and in the 20th minute to be 0.0003% / min.

[0088] Finally, the instantaneous value of the lifetime decay rate is integrated over the time range of the control cycle to obtain the theoretical lifetime loss increment of the main energy storage battery and the power energy storage element during the control cycle. That is, with time as the horizontal axis and the instantaneous value of the life decay rate as the vertical axis, the instantaneous values ​​of the life decay rate of the main energy storage battery and the power energy storage element are integrated within the time range of the control cycle. The integration result is the theoretical life loss increment of each type of energy storage element in the current control cycle. This value is the total life decay of the element within the cycle.

[0089] For example, by integrating the instantaneous value of the life decay rate of the main energy storage battery within 1 hour, we obtain its theoretical life loss increment of 0.025%; by integrating the instantaneous value of the power energy storage element within 1 hour, we obtain its theoretical life loss increment of 0.012%.

[0090] In step S500 of this embodiment, based on preset rules, the initial overall quality coefficient of the device is calculated according to the dominant control mode, the real-time power allocation instruction set, and the theoretical lifetime loss increment. The initial overall quality coefficient is then corrected using the mode health and the strategy output confidence level to obtain the final overall quality coefficient, including: The corresponding weight coefficient combination is determined based on the current dominant control mode. The weight coefficient combination includes load tracking accuracy weight, system efficiency weight, and lifetime loss cost weight. Calculate the load tracking accuracy score and system efficiency score based on the actual execution results of the real-time power allocation instruction set; The life loss cost score is calculated based on the theoretical life loss increment and the preset economic conversion factor. The initial comprehensive quality coefficient is obtained by weighting and summing the load tracking accuracy score, system efficiency score, and lifetime loss cost score using the weighted coefficient combination. The initial comprehensive quality coefficient is corrected based on the pattern health and the policy output confidence to obtain the final comprehensive quality coefficient.

[0091] In this embodiment of the application, the purpose of step S500 is to quantitatively evaluate the operating effect of the current control strategy from three dimensions: load tracking, system efficiency, and lifetime loss cost. First, the initial comprehensive quality coefficient is calculated, and then it is corrected by combining the mode health and strategy output confidence to obtain a final comprehensive quality coefficient that is more in line with the actual working conditions. This provides a core quantitative basis for subsequent evaluation of the effectiveness of power allocation commands, decision on control mode adjustment or controller parameter update, and makes the strategy evaluation more comprehensive and accurate.

[0092] To achieve the above objectives, it is first necessary to determine the corresponding weight coefficient combination based on the current dominant control mode. The weight coefficient combination includes load tracking accuracy weight, system efficiency weight, and lifetime loss cost weight. Different dominant control modes correspond to different evaluation focuses, and are matched with preset weight coefficient combinations accordingly. The combinations include load tracking accuracy weight, system efficiency weight, and life loss cost weight. The sum of the three weight values ​​is 1, and the proportion is allocated as needed.

[0093] For example, the performance-first mode focuses on load tracking, with a weight of 0.5 for load tracking accuracy, 0.3 for system efficiency, and 0.2 for lifetime loss cost; the lifetime balancing mode focuses on lifetime loss, with the three weights set to 0.2, 0.3, and 0.5 respectively.

[0094] Next, the load tracking accuracy score and system efficiency score are calculated based on the actual execution results of the real-time power allocation instruction set; Then, the life loss cost score is calculated based on the theoretical life loss increment and the preset economic conversion coefficient. Based on the actual execution results of the real-time power allocation command set, a load tracking accuracy score and a system efficiency score are calculated, both quantified values ​​ranging from 0 to 100. A lifetime loss cost score is calculated by combining the theoretical lifetime loss increment and a preset economic conversion factor, also quantified value from 0 to 100, with lower losses resulting in higher scores. The load tracking accuracy score can be calculated by comparing the total output power of the real-time power allocation command with the actual load demand power, according to preset scoring rules; higher matching results in higher scores. The system efficiency score can be calculated by determining the ratio of the actual output electrical energy of the hybrid energy storage system to the total input electrical energy, i.e., the operating efficiency, and mapping the efficiency value to the corresponding score.

[0095] For example, the total output power of the command is 30kW, the actual tracked load power is 29.5kW, and the load tracking accuracy score is 98; the actual operating efficiency of the system is 92%, and the system efficiency score is 92; the main energy storage battery loss increment is 0.02%, the power component loss increment is 0.01%, the economic conversion factor is 10000, and the calculated life loss cost score is 95.

[0096] Furthermore, the load tracking accuracy score, system efficiency score, and lifetime loss cost score are weighted and summed using the weighted coefficient combination to obtain the initial comprehensive quality coefficient; For example, the scores of the three dimensions can be multiplied by their corresponding weight coefficients, and then the products can be added together to obtain the initial comprehensive quality coefficient, which reflects the basic operational quality of the current control strategy.

[0097] For example, calculated using the performance priority mode with weights of 0.5, 0.3, and 0.2, the initial comprehensive quality coefficient = 98 × 0.5 + 92 × 0.3 + 95 × 0.2 = 49 + 27.6 + 19 = 95.6.

[0098] Finally, the initial comprehensive quality coefficient is corrected based on the pattern health and the policy output confidence to obtain the final comprehensive quality coefficient.

[0099] For example, the initial comprehensive quality coefficient can be used as the base, and multiplied by the pattern health and the policy output confidence. Both are quantitative values ​​between 0 and 1. After correction, the final comprehensive quality coefficient is obtained, so that the evaluation result takes into account both pattern matching and policy reliability.

[0100] For example, with an initial overall quality coefficient of 95.6, a model health score of 0.86, and a strategy output confidence score of 0.92, the final overall quality coefficient is 95.6 × 0.86 × 0.92 = 75.37.

[0101] In step S600 of this embodiment, based on a preset quality coefficient threshold and the final comprehensive quality coefficient, the effectiveness of the current real-time power allocation command is evaluated, and a decision is made to trigger a control mode or update the real-time optimized controller parameters, including: Two quality coefficient thresholds are preset, including a first threshold T1 and a second threshold T2, where T1 <T2; When the average value of the final comprehensive quality coefficient for X consecutive control cycles is lower than T1, the process of redetermining the dominant control mode is triggered; where X is a positive integer greater than or equal to 1. When the final comprehensive quality coefficient is higher than T2 for X consecutive control cycles, the online update of the parameters of the current real-time optimization controller branch is triggered. When the final overall quality coefficient is between T1 and T2, the current control strategy remains unchanged, and only the operating data is recorded for subsequent analysis.

[0102] In the embodiments of the present application, the purpose of the above step S600 is to scientifically evaluate the actual effectiveness of the current power distribution instruction based on the final comprehensive quality coefficient, and in combination with the preset double quality coefficient thresholds, and accordingly dynamically decide the adjustment method of the control strategy, so as to achieve the re-matching of the control mode or the online optimization of the controller parameters, enabling the control strategy of the hybrid energy storage system to continuously optimize adaptively to the working condition changes, and ensuring that the system is always in the optimal operating state.

[0103] To achieve the above purpose, first, two quality coefficient thresholds need to be preset, including the first threshold T1 and the second threshold T2, where T1 < T2; That is, two quality coefficient thresholds with different values are set in advance. The first threshold T1 is the lower limit value of the control strategy effectiveness, and the second threshold T2 is the upper limit value, and the numerical relationship of T1 < T2 is strictly followed as the core standard for coefficient determination.

[0104] For example, preset T1 = 60 and T2 = 90, and it is clear that the coefficient lower than 60 is invalid, higher than 90 is excellent, and between the two is qualified.

[0105] Then, set the number of control cycles X for continuous determination. X is a positive integer greater than or equal to 1. Calculate the average value of the final comprehensive quality coefficients of X consecutive control cycles, and compare it with T1 and T2 to determine the interval to which it belongs.

[0106] When the average value of the final comprehensive quality coefficients of X consecutive control cycles is lower than T1, trigger the process of re-determining the dominant control mode; where X is a positive integer greater than or equal to 1; For example, the final comprehensive quality coefficients of 3 consecutive control cycles are 52, 58, and 55 respectively, and the average value is 55, which is lower than T1 = 60. It is determined that the current dominant control mode does not match the working condition, trigger the process of re-determining the dominant control mode, and re-match the appropriate control mode in combination with the expected workload spectrum and the system health status.

[0107] When the average value of the final comprehensive quality coefficients of X consecutive control cycles is continuously higher than T2, trigger the online update of the parameters of the current real-time optimization controller branch; For example, the final comprehensive quality coefficients of 3 consecutive control cycles are 93, 95, and 92 respectively, and the average value is 93, which is higher than T2 = 90. It is determined that the current control strategy has excellent operation effect, trigger the online update of the parameters of the current real-time optimization controller branch, and make the controller parameters more suitable for the actual working condition through iterative optimization to further improve the control effect.

[0108] When the final comprehensive quality coefficient is between T1 and T2, keep the current control strategy unchanged, and only record the operation data for subsequent analysis.

[0109] For example, the final comprehensive quality coefficients for three consecutive control cycles are 78, 82, and 75, respectively, with an average of 78. This value falls between T1=60 and T2=90. The current control strategy is determined to be suitable for the operating conditions. The existing dominant control mode and power allocation strategy are maintained unchanged, and only the system operation data and power allocation commands within this cycle are synchronously recorded to provide data support for subsequent controller training and rule optimization.

[0110] Example 2, as Figure 2 As shown, based on the same inventive concept as the machine learning-based hybrid energy storage device collaborative control method provided in Embodiment 1, this embodiment of the invention also provides a machine learning-based hybrid energy storage device collaborative control system, including: The workload spectrum acquisition module 11 is used to predict the expected operating data for a future preset period based on the historical operating dataset of the mobile energy storage and charging device, and to analyze and obtain the expected workload spectrum of the mobile energy storage and charging device in the future preset period based on the expected operating data and the preset service period of the mobile energy storage and charging device; wherein, the historical operating dataset includes health status indicators of the hybrid energy storage system; the hybrid energy storage system includes a main energy storage battery and a power-type energy storage element. The dominant control mode determination module 12 is used to determine the dominant control mode of the mobile energy storage device in the next adaptive control cycle based on the expected workload spectrum and the current health status of the hybrid energy storage system, and to generate the corresponding mode health status through predefined mode decision rules. The dominant control mode includes at least a performance priority mode, a life balance mode, a maintenance mode, and an economic benefit mode. The power allocation instruction acquisition module 13 is used to extract the real-time state features of the hybrid energy storage system and obtain N real-time optimization controller branches based on reinforcement learning training. The real-time state features and the dominant control mode parameters are input into the corresponding real-time optimization controller branches to obtain the real-time power allocation instruction set and the corresponding strategy output confidence of each component in the hybrid energy storage system. The lifetime loss increment calculation module 14 is used to control the operation of the hybrid energy storage system according to the real-time power distribution instruction set, and calculate the theoretical lifetime loss increment generated in the current control cycle based on the preset lifetime loss model. The comprehensive quality coefficient calculation module 15 is used to calculate the initial comprehensive quality coefficient of the device based on preset rules, according to the dominant control mode, real-time power distribution instruction set and theoretical lifetime loss increment, and to correct the initial comprehensive quality coefficient using the mode health and the strategy output confidence to obtain the final comprehensive quality coefficient. The system update and optimization module 16 is used to evaluate the effectiveness of the current real-time power allocation command based on the preset quality coefficient threshold and the final comprehensive quality coefficient, and to decide whether to trigger the control mode or update the real-time optimized controller parameters.

[0111] Furthermore, the workload spectrum acquisition module 11 includes the following execution steps: Obtain the historical operation dataset of the mobile energy storage and charging device. The historical operation dataset includes at least calendar time, load power sequence, and corresponding health status indicators of the hybrid energy storage system. Feature extraction is performed on the load power sequence arranged by timestamp in the historical operating dataset to obtain the historical working load spectrum; Based on time series analysis, and combining calendar features with preset service periods, the historical workload spectrum is correlated and predicted with the time tags of future preset periods. The average load power and load fluctuation intensity within the future preset period are output as the expected working load spectrum.

[0112] Furthermore, the dominant control mode determination module 12 includes the following execution steps: The peak intensity of the expected workload spectrum, the health status deviation between the current health status and the rated health status of the hybrid energy storage system, and whether the future preset period includes preset high-demand calendar events are used as input conditions. The input conditions are matched with a predefined pattern decision rule base, which stores the mapping relationship between different combinations of input conditions and recommended control modes. Based on the matching results, the dominant control mode is output.

[0113] The dominant control mode is output based on the matching results, including: The peak intensity of the expected workload spectrum is compared with a preset high intensity threshold, the current health status deviation is compared with a preset degradation threshold, and it is determined whether the future preset period includes holidays or preset high electricity price periods. Based on the above comparison and judgment results, mapping is performed according to the following rules: If the peak intensity exceeds the high intensity threshold, it is mapped to the performance priority mode; If the peak intensity does not exceed the high intensity threshold but the health status deviation exceeds the degradation threshold, then it is mapped to the maintenance mode; If the peak intensity does not exceed the high intensity threshold, the health status deviation does not exceed the degradation threshold, and the future preset period includes holidays or preset high electricity price periods, then it is mapped to the economic benefit model. If none of the above conditions are met, the default mapping will be to the lifetime balancing mode.

[0114] The dominant control mode includes at least a performance-priority mode, a lifespan balancing mode, a maintenance mode, and an economic benefit mode, including: When in performance priority mode, the upper limit of the output power of the power-type energy storage element is increased to its rated power, while the maximum output power of the main energy storage battery is limited. When in life balance mode, the optimization objective is to minimize the theoretical life loss increment of the main energy storage battery and the power energy storage element under the current operating conditions, and the power output ratio of the main energy storage battery and the power energy storage element is dynamically adjusted. When in maintenance mode, the state of charge of the main energy storage battery is controlled within a preset maintenance window range, and the power-type energy storage element is used to bear the external load fluctuations within the maintenance window range. When in economic benefit mode, the charging power and discharging power of the main energy storage battery are dynamically set according to the real-time electricity price signal, wherein the charging power is negatively correlated with the electricity price and the discharging power is positively correlated with the electricity price; at the same time, the power-type energy storage element is used to supplement the discharging power difference.

[0115] The generation of the corresponding pattern health score includes: Obtain the rule satisfaction score of the pattern decision rule base during the matching process. The rule satisfaction score is calculated based on the degree of matching between the input conditions and the rules corresponding to the selected dominant control mode. Obtain the prediction confidence level of the expected workload spectrum; Based on the rule satisfaction score and the prediction confidence, a quantified value between 0 and 1 is generated through weighted calculation as the health of the pattern.

[0116] Furthermore, the lifetime loss increment calculation module 14 includes the following execution steps: A first lifetime loss model for the main energy storage battery and a second lifetime loss model for the power-type energy storage element are established respectively. The first lifetime loss model uses current stress, temperature stress, and state-of-charge stress as input parameters, while the second lifetime loss model uses power stress and temperature stress as input parameters. At the end of each control cycle, the operating stress time-series data of the main energy storage battery and the power energy storage element during that control cycle are collected; The operating stress time series data are input into the corresponding first lifetime loss model and second lifetime loss model respectively, and the instantaneous value of the lifetime decay rate of the main energy storage battery and the power type energy storage element within the control cycle is calculated by the corresponding lifetime loss model. The instantaneous value of the lifetime decay rate is integrated over the time range of the control cycle to obtain the theoretical lifetime loss increment of the main energy storage battery and the power energy storage element during the control cycle.

[0117] Furthermore, the comprehensive quality coefficient calculation module 15 includes the following execution steps: The corresponding weight coefficient combination is determined based on the current dominant control mode. The weight coefficient combination includes load tracking accuracy weight, system efficiency weight, and lifetime loss cost weight. Calculate the load tracking accuracy score and system efficiency score based on the actual execution results of the real-time power allocation instruction set; The life loss cost score is calculated based on the theoretical life loss increment and the preset economic conversion factor. The initial comprehensive quality coefficient is obtained by weighting and summing the load tracking accuracy score, system efficiency score, and lifetime loss cost score using the weighted coefficient combination. The initial comprehensive quality coefficient is corrected based on the pattern health and the policy output confidence to obtain the final comprehensive quality coefficient.

[0118] Furthermore, the system update and optimization module 16 includes the following execution steps: Two quality coefficient thresholds are preset, including a first threshold T1 and a second threshold T2, where T1 <t2；When the average value of the final comprehensive quality coefficient for X consecutive control cycles is lower than T1, the process of redetermining the dominant control mode is triggered; where X is a positive integer greater than or equal to 1. When the final comprehensive quality coefficient is higher than T2 for X consecutive control cycles, the online update of the parameters of the current real-time optimization controller branch is triggered. When the final overall quality coefficient is between T1 and T2, the current control strategy remains unchanged, and only the operating data is recorded for subsequent analysis.

Claims

1. A collaborative control method for hybrid energy storage devices based on machine learning, characterized in that, include: Obtain the expected workload spectrum of the mobile energy storage and charging device within a future preset period; Determine the dominant control mode of the mobile storage and charging device in the next adaptive control cycle, and generate the corresponding mode health status; The mode health score is used to measure the degree of fit between the selected dominant control mode and the current operating conditions. The dominant control modes include at least: performance priority mode, which increases the upper limit of the output power of the hybrid energy storage system; life balancing mode, which controls charging and discharging with the goal of reducing the life loss of the hybrid energy storage system; maintenance mode, which controls charging and discharging based on a preset maintenance window range; and economic benefit mode, which controls charging and discharging based on real-time electricity prices. Obtain the real-time power allocation instruction set and corresponding strategy output confidence level of each component in the hybrid energy storage system; The system is controlled to operate according to the real-time power distribution instruction set, and the theoretical lifetime loss increment generated in the current control cycle is calculated based on the preset lifetime loss model. The final overall quality coefficient is calculated based on the dominant control mode, real-time power distribution instruction set, and theoretical lifetime loss increment. Based on the preset quality coefficient threshold and the final comprehensive quality coefficient, the effectiveness of the current real-time power allocation command is evaluated, and a decision is made to trigger the control mode or to update the controller parameters in real time. Specifically, based on the expected workload spectrum and the current health status of the hybrid energy storage system, the dominant control mode of the mobile energy storage and charging device in the next adaptive control cycle is determined through predefined mode decision rules, including: The peak intensity of the expected workload spectrum, the health status deviation between the current health status and the rated health status of the hybrid energy storage system, and whether the future preset period includes preset high-demand calendar events are used as input conditions. The input conditions are matched with a predefined pattern decision rule base, which stores the mapping relationship between different combinations of input conditions and recommended control modes. Based on the matching results, the dominant control mode is output; The generation of the corresponding pattern health score includes: Obtain the rule satisfaction score of the pattern decision rule base during the matching process. The rule satisfaction score is calculated based on the degree of matching between the input conditions and the rules corresponding to the selected dominant control mode. Obtain the prediction confidence level of the expected workload spectrum; Based on the rule satisfaction score and the prediction confidence, a quantitative value between 0 and 1 is generated by weighted calculation as the health of the pattern. Specifically, based on preset rules, the initial overall quality coefficient of the device is calculated according to the dominant control mode, real-time power allocation instruction set, and theoretical lifetime loss increment. The initial overall quality coefficient is then corrected using the mode health and strategy output confidence levels to obtain the final overall quality coefficient, including: The corresponding weight coefficient combination is determined based on the current dominant control mode. The weight coefficient combination includes load tracking accuracy weight, system efficiency weight, and lifetime loss cost weight. Calculate the load tracking accuracy score and system efficiency score based on the actual execution results of the real-time power allocation instruction set; The life loss cost score is calculated based on the theoretical life loss increment and the preset economic conversion factor. The initial comprehensive quality coefficient is obtained by weighting and summing the load tracking accuracy score, system efficiency score, and lifetime loss cost score using the weighted coefficient combination. The initial comprehensive quality coefficient is corrected based on the pattern health and the policy output confidence to obtain the final comprehensive quality coefficient.

2. The machine learning-based collaborative control method for hybrid energy storage devices according to claim 1, characterized in that, Based on historical operational datasets of mobile energy storage and charging devices, the expected operational data for a future preset period is predicted. Based on this expected operational data and the preset service period of the mobile energy storage and charging devices, the expected workload spectrum of the mobile energy storage and charging devices within the future preset period is analyzed and obtained, including: Obtain the historical operation dataset of the mobile energy storage and charging device. The historical operation dataset includes at least calendar time, load power sequence, and corresponding health status indicators of the hybrid energy storage system. Feature extraction is performed on the load power sequence arranged by timestamp in the historical operating dataset to obtain the historical working load spectrum; Based on time series analysis, and combining calendar features with preset service periods, the historical workload spectrum is correlated and predicted with the time tags of future preset periods. The average load power and load fluctuation intensity within the future preset period are output as the expected working load spectrum.

3. The machine learning-based collaborative control method for hybrid energy storage devices according to claim 1, characterized in that, Based on the matching results, the dominant control mode is output, including: The peak intensity of the expected workload spectrum is compared with a preset high intensity threshold, the current health status deviation is compared with a preset degradation threshold, and it is determined whether the future preset period includes holidays or preset high electricity price periods. Based on the above comparison and judgment results, mapping is performed according to the following rules: If the peak intensity exceeds the high intensity threshold, it is mapped to the performance priority mode; If the peak intensity does not exceed the high intensity threshold but the health status deviation exceeds the degradation threshold, then it is mapped to the maintenance mode; If the peak intensity does not exceed the high intensity threshold, the health status deviation does not exceed the degradation threshold, and the future preset period includes holidays or preset high electricity price periods, then it is mapped to the economic benefit model. If none of the above conditions are met, the default mapping will be to the lifetime balancing mode.

4. The machine learning-based collaborative control method for hybrid energy storage devices according to claim 1, characterized in that, The dominant control modes include at least performance-priority mode, lifespan balancing mode, maintenance mode, and economic efficiency mode, including: When in performance priority mode, the upper limit of the output power of the power-type energy storage element is increased to its rated power, while the maximum output power of the main energy storage battery is limited. When in life balance mode, the optimization objective is to minimize the theoretical life loss increment of the main energy storage battery and the power energy storage element under the current operating conditions, and the power output ratio of the main energy storage battery and the power energy storage element is dynamically adjusted. When in maintenance mode, the state of charge of the main energy storage battery is controlled within a preset maintenance window range, and the power-type energy storage element is used to bear the external load fluctuations within the maintenance window range. When in economic benefit mode, the charging power and discharging power of the main energy storage battery are dynamically set according to the real-time electricity price signal, wherein the charging power is negatively correlated with the electricity price and the discharging power is positively correlated with the electricity price; at the same time, the power-type energy storage element is used to supplement the discharging power difference.

5. The machine learning-based collaborative control method for hybrid energy storage devices according to claim 1, characterized in that, Based on a pre-defined lifetime loss model, the theoretical lifetime loss increment generated within the current control cycle is calculated, including: A first lifetime loss model for the main energy storage battery and a second lifetime loss model for the power-type energy storage element are established respectively. The first lifetime loss model uses current stress, temperature stress, and state-of-charge stress as input parameters, while the second lifetime loss model uses power stress and temperature stress as input parameters. At the end of each control cycle, the operating stress time-series data of the main energy storage battery and the power energy storage element during that control cycle are collected; The operating stress time series data are input into the corresponding first lifetime loss model and second lifetime loss model respectively, and the instantaneous value of the lifetime decay rate of the main energy storage battery and the power type energy storage element within the control cycle is calculated by the corresponding lifetime loss model. The instantaneous value of the lifetime decay rate is integrated over the time range of the control cycle to obtain the theoretical lifetime loss increment of the main energy storage battery and the power energy storage element during the control cycle.

6. The machine learning-based collaborative control method for hybrid energy storage devices according to claim 1, characterized in that, Based on a preset quality coefficient threshold and the final overall quality coefficient, the effectiveness of the current real-time power allocation command is evaluated, and a decision is made to trigger a control mode or update the controller parameters in real time, including: Two quality coefficient thresholds are preset, including a first threshold T1 and a second threshold T2, where T1 <T2; When the average value of the final comprehensive quality coefficient for X consecutive control cycles is lower than T1, the process of redetermining the dominant control mode is triggered; where X is a positive integer greater than or equal to 1. When the final comprehensive quality coefficient is higher than T2 for X consecutive control cycles, the online update of the parameters of the current real-time optimization controller branch is triggered. When the final overall quality coefficient is between T1 and T2, the current control strategy remains unchanged, and only the operating data is recorded for subsequent analysis.

7. A machine learning-based collaborative control system for hybrid energy storage devices, characterized in that, The machine learning-based hybrid energy storage device collaborative control system is used to implement the machine learning-based hybrid energy storage device collaborative control method as described in any one of claims 1-6, including: The workload spectrum acquisition module is used to predict the expected operating data for a future preset period based on the historical operating dataset of the mobile energy storage and charging device. Based on the expected operating data and the preset service period of the mobile energy storage and charging device, the module analyzes and obtains the expected workload spectrum of the mobile energy storage and charging device in the future preset period. The historical operating dataset includes health status indicators of the hybrid energy storage system. The hybrid energy storage system includes a main energy storage battery and a power-type energy storage element. The dominant control mode determination module is used to determine the dominant control mode of the mobile energy storage device in the next adaptive control cycle based on the expected workload spectrum and the current health status of the hybrid energy storage system, and to generate the corresponding mode health status through predefined mode decision rules. The dominant control mode includes at least a performance priority mode, a life balance mode, a maintenance mode, and an economic benefit mode. The power allocation instruction acquisition module is used to extract the real-time state features of the hybrid energy storage system and obtain N real-time optimization controller branches based on reinforcement learning training. The real-time state features and the dominant control mode parameters are input into the corresponding real-time optimization controller branches to obtain the real-time power allocation instruction set and the corresponding strategy output confidence of each component in the hybrid energy storage system. The lifetime loss increment calculation module is used to control the operation of the hybrid energy storage system according to the real-time power distribution instruction set, and to calculate the theoretical lifetime loss increment generated in the current control cycle based on the preset lifetime loss model. The overall quality coefficient calculation module is used to calculate the initial overall quality coefficient of the device based on preset rules, according to the dominant control mode, real-time power distribution instruction set and theoretical lifetime loss increment, and to correct the initial overall quality coefficient using the mode health and the strategy output confidence to obtain the final overall quality coefficient. The system update and optimization module is used to evaluate the effectiveness of the current real-time power allocation command based on the preset quality coefficient threshold and the final comprehensive quality coefficient, and to decide whether to trigger the control mode or update the controller parameters in real time.

Images