Battery energy storage system dynamic reconstruction and equalization method for fault tolerance

Abstract

The invention discloses a battery energy storage system dynamic reconstruction and equalization method oriented to fault tolerance, and belongs to the field of energy storage battery management. According to the method, multi-dimensional state parameters such as voltage, current, temperature and internal resistance of a single battery or a module are collected in real time through a distributed multi-point sensor, multi-source data are fused, a time sequence neural network and a fuzzy rule are used for jointly evaluating the health state of the battery, and fault detection and type judgment are achieved in combination with time sequence clustering and Bayesian inference. And for the detected fault, a genetic-game hybrid reconstruction algorithm of global loss constraint is adopted to dynamically adjust the system topology, and fault module isolation and optimal reconstruction of the remaining modules are realized. The system is configured with multi-stage bidirectional DC-DC energy scheduling, bidirectional collaborative balance of the state of charge and the health degree is realized, and robust disturbance compensation is introduced to improve the stability. Based on model prediction and reinforcement learning, safety early warning and strategy self-optimization are provided, and the fault-tolerant capability of the battery energy storage system is remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of energy storage battery management, and more specifically relates to a dynamic reconfiguration and balancing method for fault-tolerant battery energy storage systems. Background Technology

[0002] With the rapid development of new energy and smart grid technologies, battery energy storage systems have become core equipment supporting renewable energy access, grid peak and frequency regulation, and distributed energy management. In recent years, the scale of energy storage systems has continued to expand, their structures have become increasingly complex, and their operating conditions have become more variable, posing significant challenges to the system's safety, reliability, and efficient operation and maintenance. Traditional energy storage systems often rely on static structures or single fault tolerance measures, lacking the responsiveness for dynamic reconfiguration, energy balancing, and intelligent safety management. Once battery failure, module malfunction, or external disturbance occurs, it is often difficult to quickly isolate the fault and achieve efficient allocation of remaining resources. The overall health and lifespan of the system are also limited by the weaknesses in critical nodes.

[0003] Especially in large-scale integrated application scenarios, inconsistencies in individual battery cells, dynamic changes in their health status, and complex external operating disturbances can all lead to energy distribution imbalances, increased risk of thermal runaway, and decreased system robustness. Although academia and industry have proposed some safety protection schemes based on passive balancing, simple redundancy, or fixed strategies, these methods generally suffer from shortcomings such as slow response, weak global coordination, low resource utilization efficiency, and poor scalability, making it difficult to meet the engineering requirements of next-generation high-safety, high-reliability, and long-life energy storage systems. Therefore, there is an urgent need to develop an innovative battery energy storage system that combines intelligent sensing, dynamic reconfiguration, collaborative balancing, and self-learning safety optimization capabilities to achieve full-process safety early warning, intelligent decision-making, and efficient energy dispatch, effectively improving the overall performance and lifecycle value of energy storage systems in complex application scenarios. Summary of the Invention

[0004] This invention aims to address the lack of efficient and intelligent fault-tolerant dynamic reconfiguration and energy balancing mechanisms in existing battery energy storage systems when facing faults in individual battery cells or modules. Addressing technical challenges such as untimely fault response, low energy dispatch efficiency, inaccurate balancing control, and insufficient safety warnings and robustness, this invention proposes an energy storage system control method capable of achieving multi-dimensional state fusion perception, fault self-learning discrimination, dynamic system structure reconfiguration, bidirectional energy balancing and robust disturbance compensation, and possessing self-optimizing safety warning capabilities. This significantly improves the system's global fault tolerance, energy utilization efficiency, and operational safety.

[0005] To achieve the above objectives, the present invention employs the following technical solution: the method comprises: Real-time acquisition and fusion of multi-dimensional state parameters; deployment of distributed multi-point sensors to collect key parameters such as voltage, current, temperature and internal resistance of each battery cell / module in real time. Adaptive health status assessment based on temporal fusion utilizes a dynamic temporal neural network-fuzzy rule joint assessment algorithm to perform in-depth analysis of historical and currently collected data; Fault detection and type self-learning discrimination adopts the health score temporal clustering-Bayes confidence inference algorithm. It can detect abnormal cells in advance through temporal clustering and distinguish the battery fault type by combining Bayesian probability inference. Dynamic reconfiguration decision optimization and fault-tolerant control: A genetic-game hybrid reconfiguration algorithm with global loss constraints is proposed to make dynamic reconfiguration decisions for the system based on fault location results; Bidirectional collaborative energy balancing and robust disturbance compensation are achieved by introducing a bidirectional collaborative energy balancing algorithm: a multi-level bidirectional DC-DC energy dispatch structure is adopted to dynamically allocate the energy flow direction and rate; the controller adjusts the balancing power based on the SOC and health score of each individual cell. Safety early warning and optimization adopts a model prediction safety threshold determination + reinforcement learning self-learning optimization mechanism: predict the critical impact of key parameters, and activate safety protection if the limit is exceeded.

[0006] In one scheme, the real-time acquisition and data fusion of multi-dimensional state parameters includes: deploying a distributed multi-point sensor network, installing high-precision voltage, current, temperature and internal resistance measurement sensors at key locations of each battery cell or module, and combining a multi-channel adaptive sampling mechanism to achieve real-time monitoring of each measurement point; The collected raw multi-source data is time-aligned and synchronized, and outliers are screened through redundancy check, cross-validation and consistency analysis, and noise is filtered out, thereby obtaining high-quality fused data.

[0007] In one scheme, the adaptive health status assessment based on time-series fusion includes: extracting features from the multidimensional historical and current state parameters after data fusion processing, constructing a multidimensional feature sequence, using a long short-term memory network in a dynamic time-series neural network to train and infer the time-series feature matrix of a battery cell or module, and outputting predicted values ​​of state parameters for several future moments. By introducing fuzzy inference rules and dynamically correcting and scoring the neural network prediction results based on the impact of key parameters on health status under actual working conditions, a health status scoring function is established. Combined with fuzzy membership functions and expert knowledge, a forward-looking and adaptive dynamic adjustment of health status is achieved.

[0008] In one scheme, the fault detection and type self-learning discrimination includes: based on the time series data of health status scores, using a time series clustering analysis method to group the health score sequences of battery cells or modules to achieve multi-stage anomaly detection; For individual clusters with anomalous clustering, a Bayesian confidence inference model is used to intelligently identify the fault type of the observed feature vectors, and the fault type is determined based on the maximum a posteriori probability principle. By using incremental learning or online Bayesian update methods, the conditional probability model and prior distribution are dynamically updated to improve the accuracy and adaptability of fault detection and discrimination, thereby achieving efficient self-learning identification and type differentiation of faults.

[0009] In one scheme, the dynamic reconfiguration decision optimization and fault-tolerant control includes: determining the location and type of the faulty unit or module based on the intelligent fault detection results, and then using a genetic-game hybrid reconfiguration algorithm with global loss constraints to optimize the scheduling of the connection structure of the energy storage system. First, a genetic algorithm is used to perform a global search for feasible reconfiguration schemes. Then, a fitness function is used to introduce multi-objective constraints such as power loss, system remaining capacity and overall health to screen the optimal scheme set. By introducing a game theory mechanism to simulate inter-module collaboration, the optimal collaboration choice under different reconstruction strategies is analyzed using collaborative game theory. Based on the optimal solution output by the hybrid algorithm, the system structure is dynamically configured and the system structure is adjusted to isolate faulty modules and optimize the reconstruction of remaining modules, thereby improving global fault tolerance and energy scheduling efficiency.

[0010] In one scheme, the bidirectional collaborative energy balancing and robust disturbance compensation includes: after completing the topology adjustment of the optimal reconfiguration scheme, starting a multi-level bidirectional DC-DC energy dispatch structure, realizing bidirectional current flow between energy units through a controllable bidirectional DC-DC module, real-time detection of state of charge and health score, and dynamically adjusting the balancing power allocation according to the target balancing state to achieve bidirectional comprehensive balancing of SOC and health. The balanced scheduling controller achieves global balanced optimization by iteratively adjusting the duty cycle of each DC-DC module in real time. A robust disturbance compensation module is set up, which uses a disturbance observer or sliding mode variable structure robust control to estimate and compensate for external disturbances and model biases in real time.

[0011] In one scheme, the aforementioned safety early warning and optimization includes: performing future time-domain rolling predictions of key parameters such as voltage, current, and temperature based on a model predictive control framework, comparing the prediction results with dynamic safety thresholds, and triggering active safety links such as reconfiguration, bypass, circuit breaking, and power reduction when the limit is exceeded, thereby achieving real-time early warning and risk suppression; Meanwhile, the collected real-time data represents the environmental status, enabling various actions such as decision reconstruction, energy balancing, and early warning threshold adjustment. The strategy is continuously optimized using global operational reliability, lifetime utilization, and energy utilization rate as reward functions, thereby achieving optimization of reconstruction schemes, balanced scheduling, and multi-level safety protection.

[0012] Beneficial effects of this invention: This invention achieves precise perception of the operating status of individual battery cells and modules through real-time acquisition of multi-dimensional state parameters and high-quality data fusion, laying a solid foundation for fault diagnosis and subsequent control. Employing an adaptive health status assessment and self-learning fault discrimination mechanism based on time-series data fusion, it can not only promptly detect and accurately distinguish battery fault types but also improve the foresight and accuracy of fault detection.

[0013] Building upon this foundation, an innovative genetic-game hybrid dynamic reconstruction algorithm with global loss constraints is introduced, enabling the system to rapidly and adaptively complete structural optimization and reorganization after a fault occurs, thereby improving the system's fault tolerance and energy dispatch efficiency. Simultaneously, a bidirectional collaborative energy balancing and robust disturbance compensation strategy is employed, achieving dynamic balance between SOC and health through a multi-level bidirectional DC-DC energy dispatch structure, effectively enhancing the overall battery pack lifespan and energy utilization. Furthermore, an active safety warning and self-optimization mechanism based on model prediction and reinforcement learning enables early prediction of key operational risks and multi-level safety protection, further ensuring the system's operational safety and intelligence. In summary, this invention significantly enhances the robustness, intelligence, and full life-cycle performance of energy storage systems under complex operating conditions and fault environments, possessing broad engineering application and promotional value. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart of the dynamic reconstruction decision optimization and fault-tolerant control process of this invention. Detailed Implementation

[0015] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0016] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. To facilitate understanding, the invention will now be described more fully with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0017] like Figure 1 As shown, a dynamic reconfiguration and equalization method for fault-tolerant battery energy storage systems includes: Step 1: Real-time acquisition and fusion of multi-dimensional state parameters Distributed multi-point sensors are deployed to collect key parameters such as voltage, current, temperature, and internal resistance of each battery cell / module in real time. A multi-channel adaptive sampling and asynchronous time-series data fusion algorithm is employed to perform redundancy checks and noise suppression on the raw data, ensuring the accuracy and integrity of the input data and providing a solid foundation for subsequent intelligent evaluation and decision-making.

[0018] In the process of realizing real-time acquisition and data fusion of multi-dimensional state parameters, a distributed multi-point sensor network needs to be deployed within the battery energy storage system. By installing high-precision voltage, current, temperature, and internal resistance measurement sensors at key locations in each battery cell or module, comprehensive state information during battery operation can be captured. These sensor nodes are connected to the data acquisition unit via wired or wireless means to achieve real-time monitoring of each measurement point. To ensure the flexibility and reliability of sampling, the system adopts a multi-channel adaptive sampling mechanism, which intelligently adjusts the sampling period and channel priority according to the response speed and data change rate of different sensors, thereby improving acquisition efficiency while reducing data redundancy and system resource consumption.

[0019] After the collected raw multi-source data is transmitted to the central processing unit, it undergoes unified processing using an asynchronous time-series data fusion algorithm. This algorithm first performs time alignment and synchronization on the multiple asynchronously arriving data streams, eliminating misalignment issues caused by network latency or differences in sampling times between different sensors. Subsequently, the system performs redundancy checks on the data, screening outliers and invalid measurements through cross-validation and consistency analysis to improve data reliability. Simultaneously, using time-domain and frequency-domain noise suppression techniques, it effectively filters out random errors introduced by electromagnetic interference, line noise, and environmental fluctuations, further improving data accuracy and stability. The high-quality data after fusion and preprocessing not only provides a solid data foundation for subsequent health status assessments and fault detection but also provides reliable input for intelligent decision-making, thereby significantly enhancing the operational safety and intelligence level of the battery energy storage system.

[0020] Step 2: Adaptive Health Status Assessment Based on Temporal Fusion A joint evaluation algorithm combining dynamic temporal neural networks and fuzzy rules is used to perform in-depth analysis of historical and currently collected data. The specific process is as follows: Predict future battery performance trends using temporal neural networks (such as LSTM); integrate fuzzy inference rules to dynamically adjust health status scores, achieving this for battery cells / modules. Prospective and adaptive health assessments.

[0021] In the adaptive health status assessment process based on time-series fusion, the first step is to extract features from the multidimensional historical and current state parameters after data fusion processing. This involves constructing a multidimensional feature sequence from the time-series data of battery cells / modules, including voltage (V), current (I), temperature (T), and internal resistance (R). A dynamic temporal neural network, specifically a Long Short-Time Memory (LSTM) network, is then used to process the input time-series feature matrix. To conduct training and reasoning. Here, This represents the multidimensional state parameters of the i-th battery cell or module at the current time t and its n previous times. The LSTM model, by recursively processing sequential data, can capture the dynamic changes in battery parameters and output predicted values ​​of the state parameters for several future times. Quantitative analysis of the trend of performance degradation.

[0022] Based on this, fuzzy inference rules are introduced to dynamically correct and score the prediction results output by the neural network according to the comprehensive influence of various parameters on the battery health status under actual operating conditions. Let the health status scoring function be... The value range is [0, 1], where 1 represents optimal health and 0 represents complete failure. Based on process experience and expert knowledge, a fuzzy membership function is designed. The system categorizes key battery parameters into fuzzy levels such as health, warning, and failure, and combines the weights of actual parameters and predicted values ​​to form a fuzzy inference rule base. The comprehensive score can be expressed as: in, These are the weighting coefficients for the fuzzy scoring of each parameter. This represents the future capacity or lifetime metrics predicted by LSTM. We assign it a fuzzy membership function. Finally, by combining historical trends, current state, and future predictions, we achieve a health status score. The system features forward-looking and adaptive dynamic adjustments. This process not only overcomes the lag and uncertainty of single-threshold judgment, but also improves the sensitivity to battery degradation and early-stage faults under complex operating conditions, thus providing reliable support for subsequent fault tolerance and system reconfiguration.

[0023] Step 3: Fault Detection and Type Self-Learning Determination Based on the evaluation results in step 2, a health score temporal clustering-Bayesian confidence inference algorithm is designed to achieve the following functions: multi-stage anomaly detection: abnormal individual cells are detected in advance through temporal clustering; intelligent type discrimination: the battery fault type is identified by combining Bayesian probability inference, and the model is continuously trained using running data to improve the accuracy of detection and discrimination in a self-learning manner.

[0024] Firstly, based on the aforementioned health status score The time-series data is used to perform time-series clustering analysis to achieve multi-stage anomaly detection. Specifically, the health score sequence of each battery cell or module at consecutive time points is collected, denoted as... Clustering algorithms such as K-means, DBSCAN, or Gaussian Mixture Model (GMM) are used to group the dynamic evolution of these health score sequences in the feature space. Groups with large fluctuations in cluster centers or edges usually correspond to potential anomalous individuals. When their health scores deviate significantly from other individuals in the same group, they are identified as anomalous targets, enabling sensitive detection of early anomalies. For clustered anomalous individuals, a Bayesian confidence inference model is further used for intelligent fault type identification.

[0025] The specific method is to have a predefined set of fault types. By collecting relevant features (such as abnormal voltage drop, sudden increase in internal resistance, and excessive temperature), an observation feature vector is formed. According to Bayes' theorem, the prior probabilities of each fault type are obtained based on historical data training. Conditional probability of feature observation Calculate the posterior probability in, This indicates the presence of observed feature vectors. At that time, the confidence level belongs to the j-th type of fault. Based on the maximum a posteriori probability (MAP) principle, the system classifies the anomalous unit as having the highest confidence level. The system identifies fault types to achieve intelligent differentiation. Simultaneously, as the system operates and accumulates more real-world fault samples, incremental learning or online Bayesian update methods are used to continuously update the conditional probability model and prior distribution, improving the accuracy and adaptability of fault detection and discrimination. Through this process, the entire fault diagnosis system not only possesses high timeliness and accuracy but also strong self-learning capabilities, providing reliable support for subsequent fault-tolerant reconstruction and energy balance control.

[0026] Step 4: Dynamic Reconfiguration Decision Optimization and Fault Tolerance Control like Figure 2 As shown, a genetic-game hybrid reconstruction algorithm with global loss constraints is proposed to make dynamic reconstruction decisions for the system based on fault location results. The algorithm flow is as follows: S401 uses a genetic algorithm to globally search for multiple reconstruction schemes (bypass, switchover, and serial-parallel recombination). In the process of dynamic reconfiguration decision optimization and fault-tolerant control, the specific location and type of the faulty individual unit or module are first determined based on the results of preceding intelligent fault detection. Then, the connection structure of the entire energy storage system is optimized and scheduled using the proposed genetic-game hybrid reconfiguration algorithm with global loss constraints. Specifically, the system first uses a genetic algorithm (GA) to perform a global search for feasible reconfiguration schemes. Assuming the system contains N modules, each reconfiguration scheme can be encoded using chromosomes. It means that, among them The connection state of the i-th module (such as normal, bypass, standby startup, series or parallel, etc.) is encoded, and all feasible solutions constitute the initial population.

[0027] S402 introduces power loss, module health, and remaining capacity as constraint targets; Genetic algorithms use fitness functions To evaluate the merits of each reconfiguration scheme, power loss is introduced into the fitness function. System remaining capacity With overall health Multiple objective constraints: in, , , These are the weighting coefficients for each indicator. and These represent the rated capacity and the maximum allowable power consumption, respectively. Genetic operations include selection, crossover, and mutation, resulting in a set of highly fit candidate reconstruction schemes through multiple generations of evolution.

[0028] S403 uses a game theory mechanism to simulate the collaborative effects between modules, achieving fault tolerance and optimal reconfiguration of the entire system.

[0029] After obtaining the optimal set of solutions, a game theory mechanism is introduced to simulate the collaborative behavior among the modules. Each module is considered a player in the game, and their current health score is used to determine the outcome. Remaining capacity and power sharing capabilities participate in system-level fault-tolerant reconfiguration decisions. Define the payoff function for each participant. Taking into account its own load pressure, health risks, and cooperation goals, the system utilizes Nash equilibrium or cooperative game theory to analyze the optimal cooperation choices among modules under different reconfiguration strategies, ensuring that the game outcome maximizes the overall fault tolerance and operational efficiency of the system. Finally, based on the optimal solution output by the genetic-game hybrid algorithm, the system dynamically configures switches, switches to backups, and adjusts series and parallel connections to isolate faulty modules and optimize the reconfiguration of remaining modules, ensuring fault-tolerant operation and optimal energy scheduling under global power loss, health, and capacity constraints. This algorithm not only enhances the adaptive recovery capability after a fault but also maximizes the overall safe and reliable operation cycle of the system.

[0030] Step 5: Two-way Cooperative Energy Equilibrium and Robust Disturbance Compensation While the reconfiguration scheme is being implemented, a bidirectional collaborative energy balancing algorithm is introduced: a multi-level bidirectional DC-DC energy scheduling structure is adopted to dynamically allocate the energy flow direction and rate; the controller adjusts the balancing power based on the SOC and health score of each unit; and an embedded robust disturbance compensation module is used to eliminate the impact of external or model uncertainties on the balancing effect in real time, thereby improving the balancing speed and safety.

[0031] The balanced scheduling controller aims to achieve a balanced state (system average SOC). and weighted health balance goals Using as a reference, calculate the power scheduling command for each energy channel: in, This represents the equilibrium power flow direction and magnitude from cell i to j at time t. , These are the weighting coefficients. For regulator sensitivity parameters, To ensure the direction of energy flow, the sign function works in conjunction with the health score difference to achieve a comprehensive two-way balance between SOC and health. Throughout the scheduling process, the controller iteratively adjusts the duty cycle and commands of each DC-DC module in real time to globally minimize SOC differences and health score imbalances. in, To balance the weights of health scores, and This represents the average SOC and health score of each individual unit in the system.

[0032] To enhance the stability and security of the equalization effect against disturbances, robust disturbance compensation is introduced. A disturbance observer (DOB) with uncertainties or sliding mode variable structure robust control (SMC) is employed to estimate and compensate for external disturbances in real time. And model bias. Specifically, the controller makes synchronous corrections when updating the equalization power: in, This provides disturbance compensation for the output of the disturbance observer, thereby minimizing the impact of factors such as temperature changes, sampling errors, and load fluctuations on the energy balancing effect. Through the organic combination of the aforementioned bidirectional collaborative energy balancing algorithm and real-time robust compensation mechanism, not only is the balancing speed and accuracy of system SOC and health significantly improved, but electrical safety and system response robustness are also effectively guaranteed during the balancing process, laying a solid foundation for the safe, economical, and efficient operation of the energy storage system after a fault.

[0033] Step 6: Security Early Warning and Optimization During the system operation and reconfiguration balancing process, full-process safety monitoring is achieved. A model-predicted safety threshold determination + reinforcement learning self-learning optimization mechanism is adopted: predicting critical impacts of key parameters (voltage, current, etc.), and initiating safety protection if limits are exceeded; continuously collecting operational data, and iteratively optimizing reconfiguration, balancing, and protection strategies through a reinforcement learning agent to maximize system reliability and lifespan.

[0034] In the process of implementing safety early warning and self-learning closed-loop optimization, the system achieves end-to-end safety monitoring of the entire process operation status and establishes a multi-layer protection mechanism. First, for key operating parameters such as voltage, current, and temperature, a Model Predictive Control (MPC) framework is constructed. Using a predictive model trained with physical modeling and historical data, the system performs rolling predictions of parameter trends in the future time domain. For each parameter... (such as voltage) Current ), predicting its future h-steps as ( ), and with dynamic security thresholds Compare: Once the predicted curve is about to exceed the safety limit, the system will trigger multiple proactive safety mechanisms, including reconfiguration, bypass, emergency circuit breaking, and power reduction, to minimize the risk of critical impact. Simultaneously, the system continuously and frequently collects real-time data from each operating unit and module, using this data as the environmental state. Reinforcement learning agents (such as Deep Q-Network (DQN) and policy gradients) are then introduced to drive intelligent decision-making and policy optimization.

[0035] The reinforcement learning agent treats system reconfiguration, energy balancing, early warning threshold adjustment, and protective actions as optional actions, using global operational reliability, maximum lifetime utilization of individual units, and system energy utilization rate as reward functions. Through interaction and trial and error, the agent continuously explores optimal behavioral strategies. By establishing a closed loop of state-action-reward-current feedback, the agent continuously refines its estimation of system behavior and adaptively embeds dynamic environmental changes and fault evolution characteristics into the policy network, achieving self-learning cascade optimization of reconfiguration schemes, balanced scheduling, and multi-level safety protection. With the accumulation of operating cycles and data, the agent autonomously summarizes and evolves optimal safety control measures for different fault modes and operating conditions, ultimately driving continuous improvement in the overall safety, fault tolerance, and long lifespan of the energy storage system, demonstrating the core value of the self-learning closed-loop optimization mechanism.

[0036] Example: To verify the effectiveness of the fault-tolerant dynamic reconfiguration and equalization method for battery energy storage systems proposed in this invention, a prototype energy storage system comprising eight battery modules (each module consisting of 12 lithium-ion batteries connected in series) was designed and built, and experimental tests were conducted. The system is configured with a distributed multi-point sensor network, specifically including voltage, current, temperature, and internal resistance measurement modules installed in each module, achieving synchronous multi-dimensional data acquisition with a full sampling period of 1 second. The control system is based on an embedded multi-core controller, integrating algorithms for dynamic reconfiguration decision-making, energy equalization, and safety early warning.

[0037] During the experiment, key parameters of each module were first collected under normal operating conditions, and a voltage anomaly was artificially simulated in one module to demonstrate a module failure. The system then used the method of this invention to detect and identify the fault type, subsequently automatically decided on a dynamic reconfiguration scheme to implement isolation, and initiated bidirectional energy balancing and robust disturbance compensation for the remaining modules. Finally, the global energy distribution and safety status of the system were observed.

[0038] Table 1. Data collection and changes of key parameters for each module during the experiment. At the 10-minute mark of the experiment, through time-series fusion analysis and health scoring, the system accurately detected anomalies in module 3, including high temperature, high internal resistance, and a rapid voltage drop. The controller then executed a dynamic reconfiguration decision, automatically disconnecting and isolating the abnormal module. The remaining seven modules underwent energy scheduling through a bidirectional DC-DC balancing structure. The balancing controller adaptively allocated balancing power based on each module's real-time SOC and health score, ensuring that their state of charge and health were consistent, significantly improving energy utilization efficiency.

[0039] Furthermore, the robust disturbance compensation module monitors changes in external load and sudden environmental disturbances in real time, and compensates for them using a sliding mode variable structure method, ensuring that the entire system can still operate stably under fault conditions. During the experiment, the model predicted the safety threshold and issued early warnings of abnormal risks, promptly initiating measures such as reconfiguration and power reduction to avoid the escalation of faults and potential safety hazards.

[0040] Experimental results show that, using the method of this invention, the system can quickly complete anomaly detection, identification, isolation, and reconstruction within 2 minutes when a single module fails, achieving energy redistribution. A comparison with traditional methods is as follows: Table 2. Method Comparison (System operation data 20 minutes after simulated fault occurrence) It is evident that this invention can significantly improve the fault-tolerant response speed, energy balance efficiency, and operational safety of the system, and has important practical value for engineering applications.

[0041] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0042] It should be understood that the above detailed description of the technical solutions of the present invention with reference to preferred embodiments is illustrative and not restrictive. Those skilled in the art can modify the technical solutions described in the embodiments or make equivalent substitutions for some of the technical features based on reading this specification; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic reconfiguration and equalization method for a fault-tolerant battery energy storage system, characterized in that: The method comprises: Real-time acquisition of multi-dimensional state parameters and data fusion, deployment of distributed multi-point sensors, real-time acquisition of key parameters such as voltage, current, temperature and internal resistance of each battery monomer / module; Adaptive health state evaluation based on time sequence fusion, using a dynamic time sequence neural network-fuzzy rule joint evaluation algorithm, the historical and current collected data are deeply analyzed; Fault detection and type self-learning discrimination, using a health score time sequence clustering-Bayesian confidence inference algorithm, the abnormal monomers are found in advance through time sequence clustering, and the fault types are distinguished in combination with Bayesian probability inference; Dynamic reconstruction decision optimization and fault-tolerant control, a global loss constraint genetic-game hybrid reconstruction algorithm is proposed, and dynamic reconstruction decision of the system is made according to the fault positioning result; Bidirectional collaborative energy equalization and robust disturbance compensation, a bidirectional collaborative energy equalization algorithm is introduced: a multi-stage bidirectional DC-DC energy scheduling structure is adopted to dynamically adjust the energy flow direction and rate; the controller adjusts the equalization power according to the SOC and health score of each monomer; Safety warning and optimization, a model prediction safety threshold judgment + reinforcement learning self-learning optimization mechanism is adopted: the critical impact of key parameters is predicted, and safety protection is started if it is out of limit.

2. The fault-tolerant battery energy storage system dynamic reconfiguration and balancing method according to claim 1, characterized in that: The real-time acquisition of multi-dimensional state parameters and data fusion comprises: deploying a distributed multi-point sensor network, installing high-precision voltage, current, temperature and internal resistance measuring sensors at key positions of each battery monomer or module, combining a multi-channel adaptive sampling mechanism to realize real-time monitoring of each measuring point; The collected original multi-source data are time-aligned and synchronized, and abnormal values are screened through redundancy check, cross-validation and consistency analysis, and noise filtering, thereby obtaining high-quality fusion data.

3. The fault-tolerant battery energy storage system dynamic reconfiguration and balancing method of claim 1, wherein: The adaptive health state evaluation based on time sequence fusion comprises: feature extraction is performed on the multi-dimensional historical and current state parameters processed by data fusion, and a multi-dimensional feature sequence is constructed; a long short-term memory network in a dynamic time sequence neural network is used to train and reason the time sequence feature matrix of the battery monomer or module, and the predicted values of the state parameters at future time points are outputted; Fuzzy inference rules are introduced, the neural network prediction results are dynamically corrected and scored according to the influence of each key parameter on the health state under actual working conditions, a health state score function is established, and the health state is realized in advance and adaptively adjusted in combination with a fuzzy membership function and expert knowledge.

4. The fault-tolerant battery energy storage system dynamic reconfiguration and balancing method of claim 1, wherein: The fault detection and type self-learning discrimination comprises: based on the health state score time sequence data, a time sequence clustering analysis method is used to group the health score sequence of the battery monomer or module, and multi-stage abnormal detection is realized; For the clustered abnormal monomers, a Bayesian confidence inference model is used to intelligently discriminate the fault type of the observed feature vector, and the fault type is determined according to the maximum a posteriori probability principle; Through the incremental learning or online Bayesian updating method, the conditional probability model and the prior distribution are dynamically updated, the accuracy and adaptability of fault detection and discrimination are improved, and efficient self-learning identification and type discrimination of faults are realized.

5. The fault-tolerant battery energy storage system dynamic reconfiguration and balancing method of claim 1, wherein: The dynamic reconstruction decision optimization and fault-tolerant control comprises the following steps: determining the location and type of a fault cell or module according to the intelligent fault detection result, and then adopting a global loss constraint genetic-game hybrid reconstruction algorithm to optimize the connection structure of the energy storage system; Firstly, a genetic algorithm is used to perform global search on feasible reconstruction schemes, and a fitness function is introduced to introduce multi-objective constraints of power loss, system residual capacity and comprehensive health degree, so as to select and optimize the scheme set; A game mechanism is introduced to simulate the cooperation between modules, the optimal cooperation selection under different reconstruction strategies is analyzed by means of a cooperative game theory, and according to the optimal solution output by the hybrid algorithm, switches are dynamically configured and the system structure is adjusted to realize isolation of the fault module and optimal reconstruction of the remaining modules, thereby improving the global fault-tolerant capability and energy scheduling efficiency.

6. The fault-tolerant battery energy storage system dynamic reconfiguration and balancing method of claim 1, wherein: The bidirectional collaborative energy balancing and robust disturbance compensation comprises the following steps: after completing the topology adjustment of the optimal reconstruction scheme, starting a multi-stage bidirectional DC-DC energy scheduling structure, realizing bidirectional current flow between energy units through controllable bidirectional DC-DC modules, detecting the state of charge and health score in real time, and dynamically adjusting the balancing power distribution according to the target balancing state to realize bidirectional comprehensive balancing of SOC and health degree. The balancing scheduling controller realizes global balancing optimization by adjusting the duty cycle of each DC-DC module in real time. A robust disturbance compensation module is set, and an interference observer or a sliding mode variable structure robust control is adopted to estimate and compensate external disturbances and model deviations in real time.

7. The fault-tolerant battery energy storage system dynamic reconfiguration and balancing method of claim 1, wherein: The safety warning and optimization comprises the following steps: based on a model predictive control framework, future time domain rolling prediction is performed on key parameters such as voltage, current and temperature, the prediction result is compared with a dynamic safety threshold, and when the threshold is exceeded, a linkage trigger is performed on reconstruction, bypass, circuit breaking and power downlink active safety chain to realize real-time warning and risk suppression; At the same time, the collected real-time data are used for environment state, decision reconstruction, energy balancing, warning threshold adjustment and multiple actions, and the global operation reliability, life utilization and energy utilization rate are used as reward functions to continuously optimize the strategy, so as to realize optimization of the reconstruction scheme, balancing scheduling and multi-stage safety protection.

Images