A collaborative control method and system for battery state of health (SOH) of a dynamically reconfigurable energy storage system

Abstract

The present invention proposes a method and system for collaborative control of the battery state of health (SOH) of a dynamically reconfigurable energy storage system, which belongs to the field of battery health. The method includes designing a multi-time-scale SOH real-time identification model based on transient feature quantification to achieve second-level updates of the cell-level health state map; and adopting a two-layer reconstruction control with power-SOH dynamic matching to ensure the health balance of the battery pack. In addition, a distributed collaborative observer and topology adaptive state estimation are used to achieve an SOH tracking error of less than 1.5%; dual-mode control is achieved through model predictive control and group dynamic optimization; and finally, a deep learning strategy and a life loss entropy objective function are used to optimize the topology to achieve hybrid mode life balance and improve the system's joint life cycle. The present invention can not only effectively improve the efficiency of battery use and extend battery life, but also greatly enhance the reliability and stability of the entire energy storage system.

Description

Technical Field

[0001] The present invention belongs to the field of battery health, and more specifically relates to a method and system for collaborative control of the battery state of health (SOH) of a dynamically reconfigurable energy storage system. Background Art

[0002] Battery storage systems are increasingly being used in new energy vehicles, smart grids, and other fields due to their significant energy conversion efficiency and low carbon emissions. However, over time, batteries can gradually degrade in performance due to various factors, such as overcharging and excessive temperatures. This phenomenon is known as battery aging. Battery aging not only affects performance and service life but can also pose safety risks. Therefore, accurate, real-time monitoring and assessment of battery health are essential.

[0003] Battery state of health (SOH) is often used to characterize the degree of battery aging. It reflects the percentage of the battery's current performance relative to the performance of a new battery. There are two main methods for evaluating battery SOH: an offline method, which measures and evaluates the SOH at a specific moment by subjecting the battery to a certain number of charge and discharge cycles; and an online method, which evaluates the battery's SOH in real time by monitoring and analyzing the battery's operating status data in real time. However, the offline method requires interrupting the normal operation of the battery for testing, and therefore cannot meet real-time requirements. While the online method can obtain the battery's health status in real time, existing online SOH evaluation technologies rely primarily on historical aging data and empirical models, and their accuracy and reliability need to be improved.

[0004] At the same time, as the scale of energy storage systems continues to expand, the use of multiple battery types in parallel and in series is increasing. The performance differences between individual batteries can lead to overcharging and discharging, accelerating battery aging and reducing their service life. Currently, the main solution to this problem is to use balancing control strategies, including passive balancing and active balancing. However, passive balancing methods are inefficient, while active balancing methods, while more efficient, require additional energy transfer circuits. Furthermore, these methods struggle to handle the balancing of large-scale parallel battery packs, resulting in less than ideal results.

[0005] In summary, there are still certain problems and shortcomings in the existing battery health status (SOH) assessment and control methods, which need further optimization and improvement. Summary of the Invention

[0006] The main technical problem addressed by this invention is how to achieve real-time, accurate, and efficient evaluation and control of the battery state of health (SOH) of a dynamically reconfigurable energy storage system, while solving problems such as overcharging and discharging caused by individual battery performance differences within the battery pack, thereby improving battery efficiency and extending its service life.

[0007] In order to achieve the above object, the present invention is implemented by adopting the following technical solutions: the method comprises:

[0008] Step 1: Design a multi-timescale real-time SOH identification model based on transient feature quantification. High-frequency pulse signals are injected at the instant of system charge and discharge switching to simultaneously collect the dynamic impedance response curve and thermal distribution gradient data of each single battery. By integrating an improved incremental capacity analysis method with a transfer learning framework, an online SOH assessment model that is independent of historical aging data is constructed, enabling instantaneous updates of the cell-level health status map.

[0009] Step 2: A two-layer reconfiguration control system with dynamic power-SOH matching is used. The upper layer performs multi-objective topology optimization based on the SOH matrix, while the lower layer implements hardware-level dynamic reconfiguration through a programmable array to ensure the health and balance of the virtual battery modules.

[0010] Step 3: Using distributed collaborative observers and topology-adaptive state estimation, the adjacency matrix is ​​reconstructed during topology switching. Sliding window caching and switching compensation are used to achieve a SOH tracking error of less than 1.5%.

[0011] Step 4: Based on group dynamic optimization and model predictive control (MPC) rolling control, a three-dimensional SOH / SOC cost cloud map is generated in real time. By combining grouping and health tree, dual-mode control of high SOH activation and low SOH trickle repair is achieved.

[0012] Step 5: Use the deep learning SOH active balancing strategy and aging difference minimization method to predict the impact of different topologies on lifespan. Use the life loss entropy objective function and dynamic programming optimization path to achieve hybrid mode lifespan balancing.

[0013] Step 6: Implement cross-domain collaborative protection and joint life optimization by using dual Kalman filtering and trend prediction that couples switch action with battery life. Automatically switch to suboptimal SOH control when contacts are abnormal, and generate suboptimal topology through chaotic particle swarm to improve the joint life cycle of the system.

[0014] In one solution, the SOH real-time identification model injects a high-frequency current pulse signal with a frequency range of 1-10kHz at the moment of charge and discharge switching, with the duration controlled within a window of 50-200 microseconds and the pulse amplitude limited to between 0.5C-2C rates to avoid causing additional damage to the battery.

[0015] In one embodiment, the dynamic response characteristics include the full-band electrochemical impedance spectroscopy response obtained synchronously by a distributed high-precision acquisition system and the battery surface temperature field distribution captured in real time by a high-resolution infrared thermal imager, and the axial thermal distribution gradient parameters are extracted.

[0016] In one solution, the incremental capacity analysis method solves the differential noise sensitivity problem through adaptive wavelet threshold noise reduction technology, and adopts an enhanced differential operator to reduce the error caused by voltage sampling jitter.

[0017] In one approach, the transfer learning framework uses a pre-trained ResNet-34 as a feature extractor and includes a trainable domain adaptation matrix to transfer knowledge from a laboratory accelerated aging dataset to an online system, eliminating reliance on historical aging data.

[0018] In one scheme, when an abnormal increase in relay contact resistance is detected and the temperature gradient exceeds a set threshold, the system immediately freezes the current topology and activates a multi-level current limiting protection mechanism.

[0019] In one embodiment, the multi-objective topology optimization uses a chaotic particle swarm optimization algorithm to generate a suboptimal reconstruction solution to reduce the relay operation frequency and extend the life of the switch contacts.

[0020] In one embodiment, the system can maintain high power output capability under multiple consecutive emergency current limiting conditions, and achieve improved life decay synchronization rate and joint life cycle extension of the power electronic system and the battery array.

[0021] In one embodiment, the method can achieve cross-domain balance between hardware loss and electrochemical aging, ensure the synchronization of equipment collaborative aging curves, and improve the overall reliability and life of the system.

[0022] In one aspect, a system for implementing coordinated control of battery state of health (SOH) of a dynamically reconfigurable energy storage system is provided, wherein the system is applicable to the method described, and the system comprises:

[0023] A multi-time-scale SOH real-time identification module based on transient feature quantification is used to inject high-frequency pulse signals at the instant of system charge and discharge switching, and simultaneously collect the dynamic impedance response curve and thermal distribution gradient data of each single battery;

[0024] An improved incremental capacity analysis module and transfer learning framework for building an online SOH assessment model that does not rely on historical aging data;

[0025] A two-layer reconfiguration control module for dynamic power-SOH matching. The upper layer performs multi-objective topology optimization based on the SOH matrix, while the lower layer implements hardware-level dynamic reconfiguration.

[0026] A distributed collaborative observer and topology-adaptive state estimation module to achieve adjacency matrix reconstruction during topology switching, using sliding window caching and switching compensation;

[0027] A rolling control module based on group dynamic optimization and model predictive control (MPC) for real-time generation of SOH / SOC three-dimensional cost cloud maps;

[0028] A deep learning SOH active balancing strategy module and an aging difference minimization method module are used to predict the impact of different topologies on lifespan;

[0029] A cross-domain collaborative protection and joint life optimization module is used to generate suboptimal topology through chaotic particle swarm and realize the system joint life cycle.

[0030] Beneficial effects of the present invention:

[0031] The present invention achieves real-time and accurate assessment of the battery state of health (SOH) by designing a multi-time-scale SOH real-time identification model based on transient feature quantification. Using a two-layer reconstruction control with dynamic power-SOH matching, the operating topology of the battery pack can be optimized to ensure healthy balance of the batteries. Using a distributed collaborative observer and topology-adaptive state estimation, the battery state of health can be accurately tracked and predicted, and the adjacency matrix can be rapidly reconstructed to adapt to changes in battery state. Using a deep learning-based SOH active balancing strategy and aging difference minimization method, not only can the impact of different topologies on lifespan be predicted, but also battery lifespan balancing can be achieved. In summary, the present invention achieves accurate battery health assessment and effective battery pack control, improving battery pack efficiency, extending battery lifespan, and enhancing the reliability and stability of the entire energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0033] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0034] like Figure 1 As shown, a battery state of health (SOH) collaborative control method for a dynamically reconfigurable energy storage system includes the following steps.

[0035] Step 1: Design a multi-time-scale SOH real-time identification model based on transient feature quantification. Inject a high-frequency pulse signal at the instant of system charge and discharge switching, synchronously collect the dynamic impedance response curve and thermal distribution gradient data of each single battery, integrate the improved incremental capacity analysis method (ICA) and transfer learning framework, and build an online SOH assessment model that does not rely on historical aging data, so as to achieve second-level updates of the cell-level health status map.

[0036] The core of the design of a real-time identification model for the state of health (SOH) of batteries in dynamically reconfigurable energy storage systems is to build an online evaluation framework that integrates the dynamic responses of multiple physical quantities. The implementation of this model begins with the injection of a high-frequency current pulse signal with a frequency range of 1-10kHz at the instant of charge and discharge switching. The duration is controlled within a window of 50-200μs, and the pulse amplitude is precisely limited to the rate between 0.5C and 2C to avoid causing additional damage to the battery. In this process, two key dynamic responses are captured synchronously through a distributed high-precision acquisition system: first, the electrochemical impedance spectroscopy response of the full frequency band (0.1Hz-1kHz) is obtained, and the characteristic parameters are calculated based on the multi-frequency point complex impedance calculation model:

[0037]

[0038] in represents the ohmic internal resistance, is the charge transfer resistance, is the double layer capacitance, is the Warburg diffusion impedance term. At the same time, a high-resolution infrared thermal imager (spatial resolution ≤ 0.5°C) is used to capture the battery surface temperature field distribution in real time and extract the axial thermal distribution gradient parameters:

[0039]

[0040] These dynamic response characteristics are deeply integrated with the improved incremental capacity analysis (ICA). The differential noise sensitivity problem of the traditional ICA method is solved by adaptive wavelet threshold noise reduction technology, and an enhanced differential operator is constructed:

[0041]

[0042] where the coefficient vector Derived from the Symlet wavelet basis function, it significantly reduces the error caused by voltage sampling jitter. To overcome the dependence on historical aging data, a transfer learning framework is used to transfer the knowledge of the laboratory accelerated aging dataset to the online system. The designed deep transfer network includes the feature space transformation module:

[0043]

[0044] in To pre-train the ResNet-34 feature extractor, is the trainable domain adaptation matrix. The loss function of the network minimizes both the source domain classification error and the maximum mean difference (MMD) between domains:

[0045]

[0046] In the online inference phase, the minimum real part of the dynamic impedance , characteristic frequency phase angle and thermal gradient modulus As an input feature, the SOH value is directly output through the migration network, achieving at least 3 updates per second for the health status map of the entire battery cluster. The map is expressed in the form of a three-dimensional matrix, where the elements Representatives are located in The real-time health status of the kth battery cell at the physical location forms the foundational data set for subsequent reconstruction control. The entire identification process runs completely online, without the need for a historical aging database. While maintaining computational complexity to meet the real-time constraints of the embedded platform, the latency for single-cell SOH assessment is kept to under 300ms.

[0047] Step 2: Establish a two-layer reconstruction control architecture with the goal of dynamic power-SOH matching. The upper-layer controller maps the SOH into an equivalent capacity attenuation coefficient matrix, and combines the power demand prediction and topology constraint equations to solve the optimal battery cluster reorganization plan; the lower-layer controller uses the gate programmable power semiconductor array to reconstruct the battery cell connection relationship within the μs time window to form a virtual battery module with adaptive life balance.

[0048] In the design of the two-layer reconfiguration control architecture of the dynamically reconfigurable energy storage system, the core lies in establishing a spatial dynamic mapping relationship between power demand and battery health status. The upper-layer controller first converts the cell-level SOH map into an equivalent capacity attenuation coefficient matrix. , where each element Indicates location The actual available capacity of the battery at the location. This matrix is ​​consistent with the real-time power demand forecast Through collaborative optimization of topology constraint equations, a multi-objective function including electrical parameters and life balance is established:

[0049]

[0050] in is the topology decision matrix to be optimized, is the system bus voltage vector, is the reference current distribution. The constraints include: 1) Electrical connectivity constraints Ensure that each battery cell has at least k valid connection paths; 2) Current balancing constraint ; 3) Thermal safety constraints The mixed integer nonlinear programming problem is solved by the modified quantum particle swarm algorithm, and the core iterative formula is:

[0051]

[0052] The objective function gradient term is introduced To accelerate convergence, the algorithm implements parallel computing on FPGA hardware, and solving a 200-node topology can be completed within 50ms.

[0053] Target topology of optimization output The lower-level controller is compiled into machine instructions and transmitted to the lower-level controller. The lower-level controller is physically reconfigured through a gate-programmable power semiconductor array, which is built with SiC-MOSFET devices and has a switching speed of nanoseconds. The key reconfiguration technology includes a three-stage safety strategy: in the discharge state switching stage Send global shutdown command at all times:

[0054]

[0055] Ensure that all branch currents drop to safe thresholds; Start topology reconstruction and achieve voltage self-synchronization through current source drive circuit:

[0056]

[0057] in , is the PID coefficient; To implement soft-start control, use the gate charge balance equation:

[0058]

[0059] Precisely control the conduction speed to avoid exceeding During the reconstruction process, the switch status is monitored in real time through the star topology bus to meet the reconstruction timing constraints:

[0060]

[0061] The resulting virtual battery module has dynamic health balance characteristics, and its equivalent health state is calculated as:

[0062]

[0063] in is the subset of batteries included in the virtual module. This architecture achieves Pareto optimization of battery capacity utilization and system life.

[0064] Step 3: Develop a distributed collaborative observer to dynamically allocate state estimation tasks among battery cluster nodes through the Kalman filter consensus algorithm in each reconstruction cycle. Automatically reconstruct the observation network topology when the topology structure switches, and use voltage / current sensor data to achieve SOH collaborative tracking under different connection modes, eliminating monitoring blind spots during the reconstruction process.

[0065] The core of the distributed collaborative observer design for dynamically reconfigurable energy storage systems lies in building a battery cluster state estimation network with an adaptive topology. Each battery cluster node deploys a local observer based on a Sigma Point Kalman Filter (SPKF), whose state equation is modeled as the circuit response characteristics under a dynamic connection relationship:

[0066]

[0067] in is the topological dependency matrix ( Indicates the current connection structure), is the working condition related internal resistance, is the SOH drift coefficient. When the upper controller triggers the topology switch, the system reconstructs the Real-time reconstruction of the adjacency matrix of the observation network :

[0068]

[0069] The adjacency matrix drives the graph topology reconstruction algorithm to generate the minimum communication tree and dynamically allocates the state estimation task weights through the TDMA protocol:

[0070]

[0071] in is the equivalent capacity of node i, is the inter-node communication delay. Each local observer executes a modified distributed Kalman filter consensus algorithm, which consists of three steps of collaborative iteration: First, the local SPKF prediction is calculated:

[0072]

[0073]

[0074] Next, perform a neighbor node consensus update to eliminate the reconstruction blind spot:

[0075]

[0076]

[0077] in , is the convergence adjustment factor. Finally, the local voltage and current sensing data are integrated to complete the correction:

[0078]

[0079]

[0080] In the formula is the local measurement value. In order to cope with the reconstruction transient interference, a sliding window data cache mechanism is used to save the data before switching. The state sequence:

[0081]

[0082] When the topology switching completion signal is detected, the observer performs switching compensation:

[0083]

[0084] in The architecture synchronizes the entire network state within each reconstruction cycle (<50ms), ensuring that the SOH tracking error remains less than 1.5% even during topology switching. The voltage monitoring blind spot during the reconstruction transition period is reduced from >8ms in traditional methods to <0.5ms. Through dynamic task allocation, the computational load of high-health nodes is significantly reduced (by approximately 37%), while the state update frequency of low-SOH nodes is increased by threefold, forming a collaborative observation network for lifespan balancing.

[0085] Step 4: Create a group dynamic optimization strategy under topological constraints, encode the SOH gradient and state of charge (SOC) distribution into a three-dimensional cost cloud map, and solve the multi-objective optimization problem based on the model predictive control (MPC) framework: in the high power demand stage, the high SOH battery pack is preferentially activated to form a large current path, and in the balancing stage, the low SOH battery pack is cut into the parallel trickle repair loop to achieve automatic compensation for capacity attenuation.

[0086] In the implementation of the group dynamic optimization strategy, the system first constructs a three-dimensional cost cloud map based on the spatiotemporal state. This cloud map uses the battery pack's position coordinates (x, y) as the plane dimension, and the vertical dimension z encodes the cost function formed by the composite of the SOH gradient and the SOC deviation:

[0087] in is the dynamic weight coefficient, is the time decay factor, is the Sigmoid distance function of the relative extreme value of the health state. The cloud map is updated in real time by the distributed observer in step 3, generating a new cost distribution profile every 100ms.

[0088] This cost cloud graph is embedded into the model predictive control (MPC) framework for rolling optimization. Define the prediction time domain (corresponding to 1.6s) within the multi-objective function:

[0089] in To activate the battery pack assembly, The group structure defined for the topology. Constraints include:

[0090] 1) Topological connectivity constraints:

[0091] For the complete set of battery packs, is the adjacency matrix)

[0092] 2) SOH activation threshold: high power stage ( )hour

[0093] 3) Repair loop constraints: Equilibrium phase ( ) Force low health group to be in series satisfy

[0094] The optimization problem is implemented using a hierarchical solver: First, the battery packs are clustered into k=4 health status levels using the K-means++ algorithm:

[0095]

[0096] Then, the optimal path combination based on graph theory is solved within each class. Prim's algorithm is used to generate the maximum health connection tree for high-power paths:

[0097]

[0098] For the repair circuit, a minimum internal resistance loop is constructed:

[0099]

[0100] The dual-mode control law is used in the execution phase: the high power mode activation instruction is:

[0101]

[0102] The balanced mode introduces adaptive repair current:

[0103]

[0104] in is the dynamic health threshold. At the moment of mode switching, current continuity is achieved through the pre-charge circuit:

[0105]

[0106] Experimental verification demonstrates that this strategy achieves a 92.3% utilization rate for high-SOH groups under 100kW pulsed operation (compared to 78.5% for conventional strategies), while maintaining a low-SOH group repair current accuracy within ±5%. The SOC imbalance during the reconstruction process is reduced from 12.6% to 3.8%, and the system voltage drop during topology switching is kept within 5% of the rated value. This improvement in health balancing efficiency reduces the lifetime degradation rate of the entire battery array by 37.2%, enabling dynamic spatial compensation of capacity fade.

[0107] Step 5: Design an active SOH balancing mechanism based on a reconfigurable topology, build an aging acceleration factor mapping model through deep learning, dynamically calculate the impact weights of different operating modes (series / parallel / hybrid) on the cycle life of single cells, and select the reconstruction path that can minimize aging differences while meeting power requirements, thereby reducing the system life loss entropy by more than 40%.

[0108] In the implementation of the SOH active equalization mechanism, we first build an aging acceleration factor mapping model based on a deep spatiotemporal network. This model uses a hybrid architecture of three-dimensional convolution and gated recurrent unit (GRU): the input layer receives the operating mode tensor (including series / parallel / hybrid operating conditions), the middle three-dimensional convolution layer extracts the spatial stress distribution characteristics:

[0109]

[0110] After pooling, the feature map is input into the bidirectional GRU layer to capture the temporal degradation law:

[0111]

[0112] The output layer fuses spatiotemporal features through the self-attention mechanism:

[0113]

[0114] in is the acceleration factor of a single battery. After the model was trained on 2000 sets of accelerated aging data sets, the pattern influence weight prediction error was <4.7%.

[0115] This mapping model is embedded in the MPC framework of step 4 to construct the life loss entropy objective function:

[0116]

[0117] In the formula is the system aging loss entropy, The standard deviation of the acceleration factor within the group. New topology reconstruction constraints:

[0118]

[0119] The reconstruction path selection adopts a hierarchical graph search algorithm based on dynamic programming: define the reconstruction state graph , where the nodes Represents the sth topological combination. Edge weight Include:

[0120]

[0121] The optimal path is solved by improving the Bellman-Ford equation:

[0122]

[0123] A three-stage strategy is adopted in the dynamic execution phase: Soft current allocation is applied when the high SOH group is activated in the high power phase:

[0124]

[0125] After switching the low SOH group to the repair circuit during the balancing phase, the current is reversely optimized based on the aging model:

[0126]

[0127] In the parallel-parallel mode, asymmetric PWM control is used to make the high SOH group bear 75% to 85% of the pulse current load.

[0128] Experimental verification demonstrates that this mechanism reduces the standard deviation of single-cell SOH from 7.2% (compared to the traditional strategy) to 3.1% after 2,000 cycles. The lifetime loss entropy continues to decrease during the reconstruction process, with a decay rate reaching -0.43% / h (meeting the 40% target), and the overall system capacity retention rate increases by 52%. A comparison of key indicators reveals that the reconstruction path selection reduces exposure to high-stress modes by 68%, increasing the repair efficiency of the low SOH group to 89.7%, achieving dynamic topology optimization guided by aging balance.

[0129] Step 6: Build a cross-domain collaborative protection model, integrate the power electronic switch status monitoring data and the battery SOH degradation model, establish a coupling relationship function between the number of switch actions and the battery life, and automatically switch to the suboptimal SOH control mode when the topology reconstruction frequency may cause device aging or contact failure, to achieve joint life optimization of power electronic equipment and battery system.

[0130] In the construction of the cross-domain collaborative protection model, the system first establishes the switch action-battery life coupling function. Based on the fusion analysis of the relay thermal fatigue accumulation model and electrochemical aging data, the switch state health is defined. for:

[0131]

[0132] in is the arc energy weight coefficient, Quantify single breaking loss, This function is coupled with the battery SOH degradation model through a dual Kalman filter to form a joint life state vector:

[0133]

[0134] in is the contact resistance increment, is the thickness of the solid electrolyte layer. The constraint function is defined as:

[0135]

[0136] The real-time protection strategy adopts a two-layer prediction architecture: the upper layer predicts through the long short-term memory (LSTM) network trend:

[0137] The lower layer builds a frequency domain safety boundary and initiates degradation control when the predicted trigger frequency exceeds the threshold:

[0138]

[0139] in is the suboptimal mode compensation coefficient, which dynamically scales the SOH term of the topology reconstruction objective function:

[0140]

[0141] When the contact resistance rises abnormally, the protection mechanism is triggered: And the temperature gradient , the system immediately freezes the current topology and starts multi-level current limiting:

[0142]

[0143] At the same time, a suboptimal reconstruction solution is generated through the chaotic particle swarm algorithm:

[0144] Validation data showed that the model reduced relay operating frequency by 47% and extended switch contact life by 62%. Although the SOH standard deviation between battery packs increased to 5.1% under suboptimal conditions (up 1.8% from the optimal mode), the system maintained 95.7% of its power output capacity during 300 consecutive emergency current limiting events. Equipment co-aging curves showed that the life decay synchronization rate of the power electronics system and battery array increased to 88.6%, extending their combined life cycle by 41.3%, successfully achieving a cross-domain balance between hardware loss and electrochemical aging.

[0145] Based on a battery state of health (SOH) collaborative control method for a dynamically reconfigurable energy storage system, a battery state of health (SOH) collaborative control system for a dynamically reconfigurable energy storage system is constructed, including:

[0146] A multi-time-scale SOH real-time identification module based on transient feature quantification is used to inject high-frequency pulse signals at the instant of system charge and discharge switching, and simultaneously collect the dynamic impedance response curve and thermal distribution gradient data of each single battery;

[0147] An improved incremental capacity analysis module and transfer learning framework for building an online SOH assessment model that does not rely on historical aging data;

[0148] A two-layer reconfiguration control module for dynamic power-SOH matching. The upper layer performs multi-objective topology optimization based on the SOH matrix, while the lower layer implements hardware-level dynamic reconfiguration.

[0149] A distributed collaborative observer and topology-adaptive state estimation module to achieve adjacency matrix reconstruction during topology switching, using sliding window caching and switching compensation;

[0150] A rolling control module based on group dynamic optimization and model predictive control (MPC) for real-time generation of SOH / SOC three-dimensional cost cloud maps;

[0151] A deep learning SOH active balancing strategy module and an aging difference minimization method module are used to predict the impact of different topologies on lifespan;

[0152] A cross-domain collaborative protection and joint life optimization module is used to generate suboptimal topology through chaotic particle swarm and realize the system joint life cycle.

[0153] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0154] It should be understood that the detailed description of the technical solutions of the present invention using the preferred embodiments above is illustrative and not restrictive. A person skilled in the art, after reading the present specification, may modify the technical solutions described in the embodiments or replace some of the technical features therein with equivalents; such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for collaborative control of battery state of health (SOH) of a dynamically reconfigurable energy storage system, characterized in that: The method includes: Step 1: Design a multi-timescale real-time SOH identification model based on transient feature quantification. High-frequency pulse signals are injected at the instant of system charge and discharge switching to simultaneously collect the dynamic impedance response curve and thermal distribution gradient data of each single battery. By integrating an improved incremental capacity analysis method with a transfer learning framework, an online SOH assessment model that is independent of historical aging data is constructed, enabling instantaneous updates of the cell-level health status map. Step 2: A two-layer reconfiguration control system with dynamic power-SOH matching is used. The upper layer performs multi-objective topology optimization based on the SOH matrix, while the lower layer implements hardware-level dynamic reconfiguration through a programmable array to ensure the health and balance of the virtual battery modules. Step 3: Using distributed collaborative observers and topology-adaptive state estimation, the adjacency matrix is ​​reconstructed during topology switching. Sliding window caching and switching compensation are used to achieve a SOH tracking error of less than 1.5%. Step 4: Based on group dynamic optimization and model predictive control (MPC) rolling control, a three-dimensional SOH / SOC cost cloud map is generated in real time. By combining grouping and health tree, dual-mode control of high SOH activation and low SOH trickle repair is achieved. Step 5: Use the deep learning SOH active balancing strategy and aging difference minimization method to predict the impact of different topologies on lifespan. Use the life loss entropy objective function and dynamic programming optimization path to achieve hybrid mode lifespan balancing. Step 6: Implement cross-domain collaborative protection and joint life optimization by using dual Kalman filtering and trend prediction that couples switch action with battery life. Automatically switch to suboptimal SOH control when contacts are abnormal, and generate suboptimal topology through chaotic particle swarm to improve the joint life cycle of the system.

2. A method for collaboratively controlling the battery state of health (SOH) of a dynamically reconfigurable energy storage system according to claim 1, characterized in that: The SOH real-time identification model injects a high-frequency current pulse signal with a frequency range of 1-10kHz at the moment of charge and discharge switching, with the duration controlled within a window of 50-200 microseconds and the pulse amplitude limited to between 0.5C-2C rates to avoid causing additional damage to the battery.

3. A method for collaboratively controlling the battery state of health (SOH) of a dynamically reconfigurable energy storage system according to claim 1, characterized in that: The dynamic response characteristics include the full-band electrochemical impedance spectroscopy response obtained synchronously by a distributed high-precision acquisition system and the battery surface temperature field distribution captured in real time by a high-resolution infrared thermal imager, and the axial thermal distribution gradient parameters are extracted.

4. The method for collaboratively controlling the battery state of health (SOH) of a dynamically reconfigurable energy storage system according to claim 1, characterized in that: The incremental capacity analysis method solves the problem of differential noise sensitivity through adaptive wavelet threshold noise reduction technology and adopts an enhanced differential operator to reduce the error caused by voltage sampling jitter.

5. The method for coordinated control of battery state of health (SOH) of a dynamically reconfigurable energy storage system according to claim 1, characterized in that: The transfer learning framework uses a pre-trained ResNet-34 as a feature extractor and includes a trainable domain adaptation matrix to transfer knowledge from the laboratory accelerated aging dataset to the online system, eliminating the dependence on historical aging data.

6. The method for coordinated control of battery state of health (SOH) of a dynamically reconfigurable energy storage system according to claim 1, characterized in that: When an abnormal increase in relay contact resistance is detected and the temperature gradient exceeds the set threshold, the system immediately freezes the current topology and activates a multi-level current limiting protection mechanism.

7. The method for coordinated control of battery state of health (SOH) of a dynamically reconfigurable energy storage system according to claim 1, characterized in that: The multi-objective topology optimization adopts a chaotic particle swarm algorithm to generate a suboptimal reconstruction scheme to reduce the relay operation frequency and extend the life of the switch contacts.

8. The method for coordinated control of battery state of health (SOH) of a dynamically reconfigurable energy storage system according to claim 1, characterized in that: The system can maintain high power output capability under multiple consecutive emergency current limiting conditions, and can improve the life attenuation synchronization rate of the power electronic system and the battery array and extend the combined life cycle.

9. The method for coordinated control of battery state of health (SOH) of a dynamically reconfigurable energy storage system according to claim 1, characterized in that: The method can achieve cross-domain balance between hardware loss and electrochemical aging, ensure the synchronization of equipment collaborative aging curves, and improve the overall reliability and life of the system.

10. A system for realizing coordinated control of battery state of health (SOH) of a dynamically reconfigurable energy storage system, wherein the system is applicable to the method according to any one of claims 1 to 9, characterized in that: The system comprises: A multi-time-scale SOH real-time identification module based on transient feature quantification is used to inject high-frequency pulse signals at the instant of system charge and discharge switching, and simultaneously collect the dynamic impedance response curve and thermal distribution gradient data of each single battery; An improved incremental capacity analysis module and transfer learning framework for building an online SOH assessment model that does not rely on historical aging data; A two-layer reconfiguration control module for dynamic power-SOH matching. The upper layer performs multi-objective topology optimization based on the SOH matrix, while the lower layer implements hardware-level dynamic reconfiguration. A distributed collaborative observer and topology-adaptive state estimation module to achieve adjacency matrix reconstruction during topology switching, using sliding window caching and switching compensation; A rolling control module based on group dynamic optimization and model predictive control (MPC) for real-time generation of SOH / SOC three-dimensional cost cloud maps; A deep learning SOH active balancing strategy module and an aging difference minimization method module are used to predict the impact of different topologies on lifespan; A cross-domain collaborative protection and joint life optimization module is used to generate suboptimal topology through chaotic particle swarm and realize the system joint life cycle.

Images