Dynamic reconfigurable battery pack topology switching method for improving reliability

Abstract

The invention discloses a dynamic reconfigurable battery pack topology switching method for improving reliability, and belongs to the technical field of battery management and energy system control. According to the method, multi-dimensional parameters such as voltage, current, temperature and internal resistance of a single battery are periodically acquired through multi-source fusion sensing, and intelligent fault early warning and positioning are carried out based on a health degree score; according to the method, a topology switching candidate scheme with high reliability, optimal energy balance and good redundancy is generated by adopting an ant colony-immune hybrid algorithm of bionic group structure evolution, and optimal topology switching is dynamically realized by utilizing a multi-target hierarchical reinforcement learning decision model, so that the safety and reliability of operation of a battery pack are improved. Empirical self-learning is realized by constructing a case knowledge base, and intelligent self-adaptive operation of the battery pack under variable working conditions is guaranteed.

Description

A method for dynamically reconfigurable battery pack topology switching to improve reliability Technical Field

[0001] This invention belongs to the field of battery management and energy system control technology, and more specifically relates to a dynamic reconfigurable battery pack topology switching method for improving reliability. Background Technology

[0002] Existing Battery Management Systems (BMS) serve as the core support for energy storage and power systems, primarily undertaking key functions such as battery health monitoring, energy balance control, safety protection, and charge / discharge management. With the rapid development of applications such as new energy vehicles, renewable energy grid connection, and distributed energy storage, battery packs are becoming increasingly larger and operating environments more complex, placing higher demands on system reliability, range, and safety. However, traditional BMSs often employ fixed topologies and rule-driven management strategies, making it difficult to dynamically adapt to varying operating conditions such as individual cell aging, failures, or performance degradation. This can easily lead to problems such as reduced energy utilization efficiency, localized imbalances, fault propagation, and shortened system lifespan.

[0003] In recent years, reconfigurable topologies and intelligent switching control at the battery pack level have gradually become research hotspots. Theoretically, by dynamically recombining battery cells and flexibly switching operating modes, the fault tolerance and overall reliability of the system can be improved, enabling flexible isolation of failed cells and intelligent reconfiguration of energy paths. However, existing technologies still have significant shortcomings in intelligent sensing, decision optimization, and switching execution. On the one hand, real-time high-precision sensing of health distribution and multi-objective adaptive optimization under complex operating conditions still rely on predefined models and empirical parameters, making it difficult to balance reliability, energy efficiency, and switching risks. On the other hand, the safe execution of switching operations, anomaly detection, and closed-loop protection mechanisms are still imperfect, easily leading to safety hazards such as inrush current and system instability. Furthermore, the lack of efficient operational data archiving and knowledge accumulation mechanisms makes it difficult for the system to achieve continuous optimization and self-evolution, and to effectively cope with continuously changing or even unknown operating conditions and fault modes.

[0004] In summary, there is an urgent need for a battery pack reconfigurable topology and intelligent control method with intelligent sensing, multi-objective adaptive optimization, reliable switching control, and self-evolution capabilities, in order to break through existing technological bottlenecks, improve the long-term reliability, energy utilization efficiency, and safety level of battery systems, and meet the development needs of future intelligent energy storage and power systems. Summary of the Invention

[0005] This invention aims to address the technical problems of existing battery pack reconfigurable topology and management systems, such as insufficient accuracy in health status perception, limited multi-objective adaptive optimization capabilities, imperfect fault and anomaly handling, and lack of continuous self-evolution and knowledge accumulation mechanisms. This will improve the health management, self-healing capabilities, energy utilization efficiency, and operational safety of battery systems under complex and variable operating conditions.

[0006] To achieve the above objectives, the present invention employs the following technical solution: The method includes: fusion perception of battery cell operating status, performing fusion perception of the operating status of all cells within the battery pack, and periodically collecting multi-dimensional parameters of battery cell voltage, current, temperature, and internal resistance; fault early warning and location, based on health score time series, using soft mutation detection and spatial topology multi-scale association algorithms for fault early warning and location; adaptive topology switching candidate generation, based on the current health distribution and required output performance, using a biomimetic population structure evolution virtual ant colony-immune hybrid algorithm to generate multiple sets of reconfigurable topology switching candidate schemes with high reliability, excellent energy balance, and good redundancy; dynamic decision-making and intelligent switching control, based on the generated candidate topologies, using a multi-objective hierarchical self-reinforcement learning decision model, considering reliability improvement, energy utilization efficiency, and switching risks; and constructing a case knowledge base to update perception and decision parameters in real time, enabling the battery pack to gradually optimize the perception, early warning, decision-making, and switching processes under different operating scenarios.

[0007] In one approach, the fusion sensing of the battery cell operating status includes: periodically collecting multi-dimensional parameters such as voltage, current, temperature, and internal resistance from all cells in the battery pack; dynamically adjusting the sampling frequency based on the anomaly probability of each cell to achieve adaptive asynchronous redundant acquisition. The collected data is weighted according to historical accuracy, real-time operating conditions, and noise levels; multi-source data is jointly inferred using a hierarchical generalized Bayesian fusion method; and priors are dynamically adjusted based on historical health status and environmental factors to achieve adaptive assessment of the health probability distribution.

[0008] In one scheme, the fault early warning and location positioning includes: dynamically monitoring the evolution trend of the health of each individual cell based on the time series of the cell health score, adopting a soft mutation detection strategy, and using a nonlinear mutation probability discrimination algorithm and a sensitivity adaptive threshold; when a cell enters a potential failure risk state, further combining the spatial structure and topological connection information of the battery pack to construct a health similarity map, and conducting multi-scale correlation analysis on abnormal nodes to calculate the impact index under different topological radii, and identify the failure propagation path and impact range; at the same time, the detection window length and criterion weight parameters are dynamically adjusted according to the operating conditions to achieve adaptive improvement of early warning sensitivity and positioning accuracy.

[0009] In one approach, the adaptive topology switching candidate generation includes: based on the spatiotemporal distribution of battery cell health scores and current output performance parameters, combined with physical feasibility and objective constraints, a virtual ant colony-immune hybrid algorithm is employed. First, the pheromone distribution in the topology search graph is adjusted using health scores to guide path selection towards high-health nodes while suppressing low-health nodes from participating in critical energy flows. Subsequently, a virtual immune mechanism is used to evaluate candidate topology schemes across multiple objectives. Based on health reliability, energy balance, and redundancy indices, structural mutation, selection, and cloning operations are performed on low-activity schemes to enhance diversity and robustness. Combined with a health evolution prediction model, the adaptability and risk of candidate topologies in future time periods are assessed, achieving iterative optimization and forward-looking fusion. Finally, a set of highly reliable, well-balanced, and highly redundant topology switching candidates is output for subsequent dynamic decision-making.

[0010] In one scheme, the dynamic decision-making and intelligent switching control includes: for the multiple options output during the adaptive topology switching candidate generation stage, introducing a multi-objective hierarchical self-reinforcement learning decision model, comprehensively considering the current health distribution, historical switching records, load requirements, and candidate topology characteristics, and adaptively balancing reliability, energy balance, switching risk, and switching frequency through a hierarchical reward mechanism; utilizing a local environment self-backtracking reward mechanism to automatically retrieve similar historical scenarios, reinforce excellent decision-making experience, suppress ineffective switching, and improve global optimization capabilities; after selecting the optimal switching scheme, the control strategy layer issues a safe timing switch command, monitors key physical quantities in real time, sets an anomaly detection threshold, and automatically triggers backoff or redundancy protection in case of anomalies, ensuring electrical safety and smooth transition during the structural reconfiguration process, and realizing intelligent and adaptive topology switching control.

[0011] In one approach, the construction of a case knowledge base includes: archiving the health distribution, decision basis, switching control sequence, operational effects, and detailed information on abnormal events for each topology switch and operation process, establishing a structured case knowledge base, and conducting multi-dimensional indicator analysis and effect evaluation; based on the knowledge base, identifying operational scenarios in real time and migrating and optimizing relevant strategies to achieve rapid adaptation to new operating conditions and parameter optimization.

[0012] In one approach, the health score can be comprehensively evaluated based on multiple parameters such as voltage, current, temperature, internal resistance, and capacity decay. Multimodal fusion modeling and hierarchical heterogeneous representation are used to improve the accuracy and anti-interference capability of health perception.

[0013] According to claim 1, a dynamic reconfigurable battery pack topology switching method for improving reliability is characterized in that: the ant colony-immune hybrid algorithm updates the pheromone distribution by adjusting the health score, and introduces a mutation, cloning, and elimination mechanism for candidate topology schemes, and adopts a multi-objective index to comprehensively evaluate the quality of the scheme, including reliability, energy balance, and redundancy.

[0014] The beneficial effects of this invention are as follows: By introducing a health-adaptive reconfigurable topology switching method and an intelligent optimization control system, this invention achieves high-precision real-time perception and multi-objective adaptive decision optimization of the battery pack's health status. It can dynamically adjust the system structure based on changes in the performance and fault conditions of individual battery cells, effectively isolating failed cells and optimizing energy flow paths. This solution not only significantly improves the system's fault tolerance and safety, and extends the battery pack's lifespan, but also increases energy utilization efficiency. Furthermore, the intelligent control system possesses self-evolution and knowledge accumulation capabilities, continuously optimizing management strategies through learning and analysis of ongoing operational data, enabling the system to have stronger environmental adaptability and continuous reliable operation. Therefore, this invention overcomes the shortcomings of existing technologies in health management, fault-tolerant switching, and long-term self-optimization, meeting the urgent needs of intelligent energy storage and power battery systems for efficient, safe, and intelligent operation. Attached Figure Description

[0015] Figure 1 is a flowchart of the method of the present invention. Detailed Implementation

[0016] 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.

[0017] 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.

[0018] As shown in Figure 1, a dynamic reconfigurable battery pack topology switching method for improving reliability specifically includes: Step 1, battery cell operating status fusion perception. First, the operating status of all cells in the battery pack is fused and perceived using an adaptive asynchronous redundant acquisition and confidence-weighted multimodal fusion algorithm: Voltage, current, temperature, and internal resistance of the battery cells are periodically acquired, and the acquisition frequency of each sensor is dynamically adjusted according to historical anomaly probabilities, prioritizing the allocation of limited acquisition resources to cells that may fail. Through a multi-source data confidence dynamic allocation mechanism, each sensor data source is assigned differentiated trust weights under different operating conditions. Using a hierarchical generalized Bayesian fusion method, a health score and functional availability score for each battery cell are finally generated, providing a multi-dimensional state basis for subsequent decisions.

[0019] In the detailed implementation of the battery cell operating status fusion sensing, firstly, for all cells in the battery pack, data including voltage ( ), current ( ),temperature( ), internal resistance ( ) multidimensional parameters, among which This indicates the current cell number. To improve data collection efficiency and focus, an adaptive asynchronous redundant data collection mechanism is adopted, which analyzes the anomaly probability of each cell in different cycles based on historical operating data. If the probability of a particular individual exhibiting anomalies increases recently, then the sampling frequency of its sensor will be increased. ,in This is the sensitivity adjustment coefficient, which dynamically focuses limited acquisition resources on potentially faulty cells.

[0020] The multidimensional heterogeneous data collected by each sensor are dynamically weighted according to their historical accuracy, real-time operating conditions, and noise levels using a confidence-weighted allocation mechanism. Specifically, for the first... Sensor data (voltage, temperature), at time The confidence level is set to Then the original observation data It will be weighted as The dynamic allocation of confidence scores can be represented as a function. ,in This represents an evaluation of the stability of the current data source. Indicates the reliability of historical performance. and This is an adjustable parameter used for weight fusion of different data sources under different operating conditions.

[0021] In the multi-source data processing layer, a hierarchical generalized Bayesian fusion method is used to jointly infer multimodal data. Specifically, it is assumed that the health of a single battery cell can be determined by a set of latent variables. Characterization, the various types of observational data collected As an observation, the Bayesian inference formula is: Where, the likelihood function Different data sources are modeled using a hierarchical structure; for example, voltage and temperature are assigned to different sub-layers under their respective distributions, and then the data is merged into an individual health probability distribution. The prior distribution... This allows for dynamic adjustments based on historical health status, known failure mechanisms, and environmental factors, achieving adaptability and generalization in health status assessment.

[0022] Finally, based on the above fusion inference, a battery cell health score is defined. It can be represented by the expected or maximum a posteriori value of the observed posterior, i.e. Functional availability score This can be constructed by comparing the lower bound of the confidence interval of the health distribution with the current task requirements. in Indicates the lower bound of the confidence interval. This serves as the boundary for task availability. Together, these two scores provide a multi-dimensional and quantifiable state basis for subsequent multi-objective decision-making and control processes, achieving high robustness and accuracy at the perception layer.

[0023] Step 2, Fault Early Warning and Location: Based on the health score time series, a soft mutation detection and spatial topological multi-scale association algorithm is used for fault early warning and location. This algorithm first performs nonlinear mutation probability discrimination on the individual cell health change trend, and then constructs a health similarity map by combining battery pack structural information. Multi-scale association analysis is then conducted on suspected abnormal nodes to identify potential failure paths and their impact range. By dynamically adjusting the detection window and weight parameters, the timeliness and accuracy of early warning and fault location are improved.

[0024] First, the battery cell health score time series obtained in step 1 is used. Based on this, the health evolution trend of each individual is dynamically monitored. To accurately capture early signs of failure, this method employs a soft mutation detection strategy, that is, for... A nonlinear mutation probability discrimination algorithm is applied.

[0025] Specifically, the sliding window method is used to fit the short-term trend line of health status changes, and the first and second differences are calculated: By combining weighted exponential moving average and Z-score normalization, the sensitivity threshold for mutation detection is dynamically adjusted. When the rate or acceleration of change in the health of a particular monomer becomes abnormal, the soft mutation criterion is met. ,in If the threshold is set to adaptive, the monomer is determined to be in a state of potential failure risk, and the corresponding mutation probability is calculated. .

[0026] At the same time, taking into full account the spatial structure and topology of the battery pack, the health scores of all individual cells are calculated at the current moment. Construct a health similarity map Each node For a single battery cell, the edge This represents the topological adjacency relationship between two nodes in terms of physical or electrical connection. The health similarity between nodes can be measured using Euclidean distance or a Gaussian kernel function. .

[0027] For suspected anomalous nodes (i.e., monomers with a significantly increased mutation probability), a multi-scale association analysis method is used at different topological radii. Within, examine the synergy and linkage between its health evolution and that of surrounding nodes. Computable nodes The multiscale influence index is in Represented by node Centered on, topological distance is The set of neighboring nodes. By merging the changing trends of the influence index at different scales, the propagation path of failure and its possible scope of influence can be effectively identified.

[0028] To further improve the sensitivity and accuracy of early warning, the detection window length is dynamically adjusted based on changes in the health score sequence and the spatiotemporal characteristics of typical failure modes. and the weight parameters of the soft mutation criterion This adaptive adjustment method allows for automatic shortening of the detection window and increase of weights when the battery pack's operating state changes rapidly, thereby improving real-time performance. Conversely, during stable phases, the detection frequency is appropriately increased to suppress false alarms and false alarms, thus balancing the sensitivity of early fault warnings with the robustness of overall fault location.

[0029] By combining soft mutation detection with topological multi-scale spatial analysis, we can achieve efficient early warning and accurate location of early failures in battery pack operation, and provide key data and path basis for subsequent topology switching and active defense strategies.

[0030] Step 3: Adaptive topology switching candidate generation. Based on the current health distribution and the required output performance, a virtual ant colony-immune hybrid innovative algorithm based on biomimetic population structure evolution is used to generate multiple sets of reconfigurable topology switching candidate schemes with high reliability, good energy balance and good redundancy.

[0031] In the virtual ant colony algorithm, pheromone distribution is not only affected by topological feasibility but also coupled with nodes with high health scores reinforcing pheromones while weakly healthy nodes rapidly decay pheromones, enhancing the search capability for the global optimum. Next, a virtual immune mechanism is integrated to dynamically generate topological antibodies for candidate solutions with low reliability, subjecting their structures to mutation and selection to eliminate solutions showing deterioration trends, further improving the robustness and innovativeness of the candidates. Simultaneously, the health evolution trend under each candidate topology is predicted, enhancing the foresight and adaptability of the solutions.

[0032] During the adaptive topology switching candidate generation stage, based on the aforementioned battery cell health scores... Considering the spatiotemporal distribution of the battery pack and the current required output performance parameters (desired voltage, total current, and energy output balance requirements), and taking into account the physical feasibility and objective constraints of the reconfigurable topology, a virtual ant colony-immune hybrid innovation algorithm is adopted to generate and optimize candidate solutions. First, based on the current health distribution and historical fault correlation graph, all feasible topology configurations are abstracted as a search graph. Each node represents a sub-topological state, and edges represent reachable switching paths. The virtual ant colony algorithm deploys a large number of virtual ants in parallel on this graph. During path selection, the decision probability of each ant is influenced not only by the regular pheromone concentration... and heuristic functions The influence is also coupled with the current node health score as a moderating factor: for nodes with high health, the pheromone of their traversed paths is enhanced, and the update rule is... Among them, pheromone increment , As a pheromone volatile factor, This is used as an adjustment coefficient; however, if a node's health is low, the path pheromone decays rapidly to suppress its participation in important energy flows. Each ant, in each iteration, has a probability... Select a topology switching path. Controlling pheromones and heuristic weights.

[0033] After each iteration, a virtual immune mechanism is integrated to evaluate all currently generated candidate topology schemes based on their reliability indicators under a healthy distribution. Energy balance index and redundancy index A multi-objective evaluation function automatically identifies candidates with low reliability or deteriorating evolutionary trends. It mutates the structure through a topological antibody generation mechanism (replacing critical series and parallel paths, adding or deleting redundant nodes), and its selection probability is influenced by the historical performance of the current candidate scheme. Formally, let the candidate topology scheme be... Its antibody activity It can be represented as in , , The weighting coefficients are multi-objective and automatically adjusted to adapt to the current operating conditions. For schemes with activity below the threshold, antibody evolution operators are executed, including mutation (structural perturbation), selection (survival of the fittest), and clonal amplification processes, to promote the diversity and robustness of the solution space.

[0034] In addition, for each group of candidate topologies A health evolution prediction model is introduced, utilizing historical health change patterns and known dynamic mechanisms. Based on Bayesian time series prediction or recurrent neural network methods, the health distribution of each node in the future period after the scheme is put into use is estimated. By combining the prediction results with a reassessment of the overall fitness of the proposed solutions, dynamic integration of old and new information is achieved, enhancing the long-term effectiveness and risk resilience of candidate solutions. After several iterative optimizations, multiple sets of reconfigurable topology switching candidates with high reliability, excellent energy balance, and good redundancy are finally output, providing a rich set of feasible solutions for subsequent dynamic decision-making.

[0035] Step 4: Dynamic Decision-Making and Intelligent Switching Control. Based on the generated candidate topologies, a multi-objective hierarchical self-reinforcement learning decision-making model is employed to comprehensively consider reliability improvement, energy utilization efficiency, and switching risks. This model innovatively applies a local environment self-backtracking reward mechanism, evaluating not only the immediate effectiveness of the current switching scheme but also incorporating source analysis of historical switching results in similar scenarios. This automatically and dynamically adjusts the decision-making strategy, thereby avoiding repeated ineffective switching or local optima. Switching control commands are issued for the optimal topology scheme, driving a controllable switching matrix through a safe timing sequence to achieve smooth reconstruction of the battery pack's physical structure. Simultaneously, electrical shocks and anomalies during the switching process are monitored in real time to ensure switching safety and stability.

[0036] First, for the multiple candidate reconfigurable topology schemes generated in step 3, a hierarchical multi-objective self-reinforcement learning (HMOSRL) decision model is introduced, aiming to comprehensively improve reliability, energy utilization efficiency, and strictly control switching risks. At the high level of the model, the state space... Based on the current health distribution The action space A, consisting of historical switching records, output load requirements, and candidate topology feature vectors, represents the set of topology switching schemes to be selected; a multi-objective reward function... The design employs a hierarchical structure, where the primary reward measures health improvement and energy balance, while secondary rewards characterize switching frequency, switching energy consumption, and electrical shock. The overall structure is as follows: in Indicates reliability contribution, For energy balance, Indicates the risk and cost of the switching operation, This represents the frequency of switching within a short historical period. The weight parameters are dynamically adjusted, enabling adaptive adjustment of decision-making tendencies.

[0037] The local environment self-backtracking reward mechanism enables the agent to perform each switching action. Afterwards, in addition to immediately receiving the current reward, it also automatically searches the historical database to compare the current state with the action pair. By backtracking similar previous scenarios, we can analyze the overall performance of these switching strategies in subsequent runtime cycles. This process can be formalized as: backtracking reward adjustment. in To retrospectively assess reward weights, It is the cumulative reward of similar historical scenarios. Based on this, the agent strengthens the experience of good historical decisions by updating the policy gradient or Q value, suppresses repeated ineffective switching or getting trapped in local optima, and significantly improves decision robustness and global optimization ability.

[0038] When the agent selects the optimal topology switching scheme based on the current state The control strategy layer sends switching sequence commands to the underlying controllable switch matrix. To ensure electrical safety and a smooth transition during structural reconfiguration, safe timing logic is employed to dynamically constrain the minimum interval between switches. Sequential activation logic is used to avoid inrush current caused by momentary circuit disconnection or closing. During the switching process, real-time voltage, current, and temperature physical quantities of the switching nodes are continuously collected, and anomaly detection thresholds are set. Once a parameter jump is detected exceeding the threshold ( This immediately triggers an emergency rollback or redundancy protection strategy.

[0039] This multi-objective hierarchical reinforcement learning-self-backtracking reward model enables the agent to dynamically and adaptively optimize the topology reconfiguration strategy, significantly improving the long-term reliability and energy efficiency of the battery pack while ensuring safety. The high coupling of data flow, decision flow, and execution flow throughout the entire process provides a solid intelligent control foundation for subsequent health management and performance self-healing closed loop.

[0040] Step 5: Continuously accumulate data from each topology switch and operation to build a case knowledge base, enabling the battery pack to gradually optimize the perception, early warning, decision-making, and switching processes under different operating scenarios.

[0041] When encountering novel fault modes or extreme operating conditions never seen before, the algorithm automatically activates a dual-mode collaborative algorithm that combines rapid exploration and conservative switching. This maximizes the speed of adaptation and the efficiency of knowledge transfer while ensuring safe and reliable operation, thereby continuously improving the reliability and innovation capabilities of the battery pack.

[0042] During the continuous optimization and intelligent evolution phase, detailed information on each topology switch and its corresponding operational process—including health distribution, decision-making basis, switch control sequence, subsequent operational effects, environmental conditions, and fault types and manifestations—is comprehensively archived and gradually accumulated to form a structured case knowledge base. This knowledge base not only retains rich raw data but also organizes and summarizes multi-dimensional indicator analysis, decision-making effect evaluation, and abnormal event handling processes, providing high-value training samples and experience foundations for subsequent intelligent perception, modeling, reasoning, and decision-making.

[0043] Based on a knowledge base, transfer learning is introduced to update and optimize perception and decision-making parameters in real time. This model can intelligently identify the characteristics of the current operating scenario, transfer and optimize relevant strategies and model parameters from similar historical cases, achieving rapid adaptation and performance improvement for typical operating conditions. Simultaneously, for new scenarios with detected minor drift, the model automatically adjusts the perception threshold, feature weights, and decision preferences, ensuring that the prediction and response to health anomalies, energy imbalances, and potential faults remain sensitive and accurate.

[0044] When encountering novel fault modes or extreme operating conditions not previously covered in the past, the system automatically activates a rapid exploration-conservative switching dual-mode collaborative algorithm. Its basic principle is to employ an exploratory strategy, under the premise of safe and reliable operation, to try diverse sensing and control parameter settings, atypical topology switching operations, and appropriate model mutations, rapidly accumulating preliminary experience and feedback in new scenarios. Simultaneously, a concurrent conservative strategy is implemented, imposing stricter constraints on key processes and high-risk actions, prioritizing the use of verified past knowledge to ensure that even if unknown anomalies occur during exploration, redundancy protection and safety strategies can mitigate risks. The two modes work in real-time, dynamically adjusting the exploration intensity and conservatism based on real-time data performance, balancing innovation capabilities with operational assurance, and maximizing the speed of adaptation to new environments and the efficiency of knowledge transfer.

[0045] The case knowledge base and the hybrid learning model form a tight positive feedback loop, where every instance of perception, warning, decision-making, and switching is accumulated and fed back into the model's evolution. With the continuous accumulation of operational instances and the dynamic optimization of the model, the battery pack's health management and self-healing capabilities under varying operating conditions and even unknown scenarios are constantly improved.

[0046] Example: For a certain electric vehicle power battery pack, a practical application process of a dynamic reconfigurable battery pack topology switching method is given, and experimental data is used to demonstrate the effect of the present invention in improving the reliability of the battery pack.

[0047] 1. System Configuration and Test Platform This embodiment uses 24 lithium-ion battery cells with a rated capacity of 60Ah and a nominal voltage of 3.7V, assembled into an 8S3P (8 series 3 parallel) structure. A reconfigurable switch matrix supports mixed series and parallel operation, local isolation, and cross-group switching capabilities. The test platform is equipped with a high-precision acquisition module (ΔV error ≤1mV, ΔT error ≤0.2℃), an intelligent management unit (MCU+FPGA), a communication bus, and a temperature-controlled environmental chamber.

[0048] 2. Data Acquisition and Fusion Sensing Process: The basic sampling period is set at 1 minute. Assuming that the 7th individual cell exhibits a slight anomaly in the initial detection, the system increases its sampling frequency to once every 10 seconds based on its anomaly probability. The table below shows the key parameter sampling data (statistical average) of some individual cells within a 30-minute monitoring period, as well as the confidence-weighted health score (H-score).

[0049] Table 1 The health distribution prior of the 7th monomer is dynamically adjusted to identify the risk of degradation of this monomer.

[0050] 3. Fault warning and location analysis are combined with health time series data to perform soft mutation detection on the continuous decline of the H-score of monomer 7. At the 20-minute mark, the system triggers an warning based on a sudden increase in the abnormal probability and a drop in health, and locates the fault in monomer 7.

[0051] The historical case database retrieves the preferred local reconstruction schemes under similar environments and primary anomalies (e.g., 3 parallel to 2 parallel, temporary isolation of weak individual units), and suppresses schemes with a high probability of invalid switching.

[0052] 4. The topology switching candidate scheme generation system calls the biomimetic ant colony-immune hybrid algorithm to generate the following three sets of candidate topology switching schemes by comprehensively considering the current health distribution and energy requirements: Table 2 During the evolutionary process, health scores are used for pheromone weighting, and degraded monomers are automatically isolated first.

[0053] 5. Switching Decision and Control: A multi-objective reinforcement learning decision model was used, incorporating the following weights: reliability improvement (0.5), energy utilization efficiency (0.3), and switching risk (0.2). Option B was ultimately chosen: the 7th unit was isolated, and the 8S2P structure continued to operate. The following table compares key performance characteristics before and after the switchover: Table 3 The switching control command is issued through a safe timing sequence, dynamically monitoring current, voltage, and temperature anomalies, setting a 5% deviation threshold, and automatically triggering a scheme rollback if an abnormal situation is detected during the switching transition.

[0054] All detailed data, decision-making logic, timing instructions, and effect evaluation results of this topology switchover process have been archived in the knowledge base, forming the following records: Table 4 Through multi-dimensional index analysis and comparison, the system automatically evaluated the case as "excellent" and will prioritize migrating this optimization strategy in similar degradation scenarios in the future.

[0055] This embodiment simulates the entire process of intelligent sensing, fault warning, decision switching, and self-learning optimization of a reconfigurable battery pack using real data. The results show that the method of this invention can promptly detect degraded individual cells, automatically plan the optimal topology, effectively improve system reliability and energy utilization efficiency, and achieve operational experience accumulation and strategy transfer through a case knowledge base. It is significantly superior to traditional static topology management methods, providing a new intelligent and practical approach for the safe and reliable operation of power batteries.

[0056] 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.

[0057] 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 method for dynamically reconfigurable battery pack topology switching to improve reliability, characterized in that: The method includes: fusion perception of battery cell operating status, performing fusion perception of the operating status of all cells in the battery pack, and periodically collecting multi-dimensional parameters of battery cells such as voltage, current, temperature, and internal resistance; fault early warning and location, based on health score time series, using soft mutation detection and spatial topology multi-scale association algorithm for fault early warning and location; adaptive topology switching candidate generation, based on the current health distribution and required output performance, using a biomimetic population structure evolution virtual ant colony-immune hybrid algorithm to generate multiple sets of reconfigurable topology switching candidate schemes with high reliability, excellent energy balance, and good redundancy; dynamic decision-making and intelligent switching control, based on the generated candidate topologies, using a multi-objective hierarchical self-reinforcement learning decision model, considering reliability improvement, energy utilization efficiency, and switching risks; and building a case knowledge base to update perception and decision parameters in real time, enabling the battery pack to gradually optimize the perception, early warning, decision-making, and switching processes under different operating scenarios.

2. The method for dynamically reconfigurable battery pack topology switching to improve reliability according to claim 1, characterized in that: The aforementioned battery cell operating status fusion sensing includes: periodically collecting multi-dimensional parameters such as voltage, current, temperature, and internal resistance of all cells in the battery pack; dynamically adjusting the sampling frequency according to the anomaly probability of each cell to achieve adaptive asynchronous redundant acquisition; the collected data is weighted by confidence based on historical accuracy, real-time operating conditions, and noise levels; multi-source data is jointly inferred through a hierarchical generalized Bayesian fusion method; and priors are dynamically adjusted by combining historical health status and environmental factors to achieve adaptive assessment of the health probability distribution.

3. The method for dynamically reconfigurable battery pack topology switching to improve reliability according to claim 1, characterized in that: The aforementioned fault early warning and location tracking includes: dynamically monitoring the evolution trend of the health status of each individual cell based on the time series of the cell's health score; employing a soft mutation detection strategy based on a nonlinear mutation probability discrimination algorithm and a sensitivity-adaptive threshold; when a cell enters a potential failure risk state, further combining the battery pack's spatial structure and topological connection information to construct a health similarity map, and conducting multi-scale correlation analysis on abnormal nodes to calculate the impact index under different topological radii, identifying the failure propagation path and impact range; simultaneously, the detection window length and criterion weight parameters are dynamically adjusted according to the operating conditions to achieve adaptive improvement in early warning sensitivity and location accuracy.

4. The method for dynamically reconfigurable battery pack topology switching to improve reliability according to claim 1, characterized in that: The adaptive topology switching candidate generation includes: based on the spatiotemporal distribution of battery cell health scores and current output performance parameters, combined with physical feasibility and objective constraints, a virtual ant colony-immune hybrid algorithm is used. First, the pheromone distribution in the topology search graph is adjusted using health scores to guide path selection towards high-health nodes while suppressing low-health nodes from participating in critical energy flows. Subsequently, a virtual immune mechanism is used to evaluate candidate topology schemes for multiple objectives. Based on health reliability, energy balance, and redundancy indicators, structural mutation, selection, and cloning operations are performed on low-activity schemes. Combined with a health evolution prediction model, the adaptability and risk of candidate topologies in future periods are assessed to achieve optimization iteration and forward-looking fusion, ultimately outputting a set of highly reliable and well-balanced topology switching candidates for subsequent dynamic decision-making.

5. A method for dynamically reconfigurable battery pack topology switching to improve reliability according to claim 1, characterized in that: The aforementioned dynamic decision-making and intelligent switching control includes: for the multiple options output during the adaptive topology switching candidate generation stage, a multi-objective hierarchical self-reinforcement learning decision model is introduced, comprehensively considering the current health distribution, historical switching records, load requirements, and candidate topology characteristics. A hierarchical reward mechanism adaptively balances reliability, energy balance, switching risk, and switching frequency. A local environment self-backtracking reward mechanism is used to automatically retrieve similar historical scenarios, reinforce excellent decision-making experience, suppress ineffective switching, and improve global optimization capabilities. After selecting the optimal switching scheme, the control strategy layer issues a safe timing switch command, monitors key physical quantities in real time, sets an anomaly detection threshold, and automatically triggers backoff or redundancy protection in case of anomalies, ensuring electrical safety and smooth transition during structural reconfiguration, thus achieving intelligent and adaptive topology switching control.

6. The method for dynamically reconfigurable battery pack topology switching to improve reliability according to claim 1, characterized in that: The construction of the case knowledge base includes: archiving the health distribution, decision basis, switching control sequence, operation effect and abnormal event details of each topology switch and operation process, establishing a structured case knowledge base, and conducting multi-dimensional indicator analysis and effect evaluation; based on the knowledge base, identifying operation scenarios in real time and migrating and optimizing relevant strategies to achieve rapid adaptation to new operating conditions and parameter optimization.

7. The method for dynamically reconfigurable battery pack topology switching to improve reliability according to claim 1, characterized in that: The health score can be comprehensively evaluated based on multiple parameters such as voltage, current, temperature, internal resistance, and capacity decay. Multimodal fusion modeling and hierarchical heterogeneous expression are used to improve the accuracy and anti-interference of health perception.

8. A method for dynamically reconfigurable battery pack topology switching to improve reliability according to claim 1, characterized in that: The ant colony-immune hybrid algorithm updates pheromone distribution by adjusting health scores and introduces mutation, cloning, and elimination mechanisms for candidate topology schemes. It also uses multi-objective indicators to comprehensively evaluate the quality of the schemes, including reliability, energy balance, and redundancy.

Images