Battery module adaptive equalization method and system based on intelligent algorithm

Abstract

This invention relates to the field of battery management technology, and more particularly to an adaptive equalization method and system for battery modules based on intelligent algorithms. The method constructs a state trajectory matrix by collecting voltage and temperature data of individual battery cells and calculates mutual information entropy. Based on this, a coupled topology graph is constructed to identify dominant nodes and propagate to generate a set of predicted trajectories. The endpoint voltage is extracted and combined with topology weights to construct an optimization function to solve for the target voltage distribution. Then, an inverse search is performed to generate an intermediate voltage sequence, and finally, the conduction time and current of each equalization path are calculated to control the equalization circuit. This invention achieves efficient and accurate equalization of battery modules, improving consistency and lifespan.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and in particular to an adaptive balancing method and system for battery modules based on intelligent algorithms. Background Technology

[0002] In the field of battery module management, equalization control is a core technology for ensuring battery pack safety, extending cycle life, and improving energy utilization. Current conventional practices mainly rely on collecting single parameters such as voltage or state of charge of each individual battery cell, and determining the degree of imbalance between cells by setting fixed or dynamic thresholds. When the deviation of a cell from the average value exceeds a preset threshold, the system triggers passive equalization (e.g., consuming excess energy through parallel resistors) or active equalization (e.g., transferring energy between cells through bidirectional converters) until the deviation falls back within the allowable range. This type of method constructs a simple closed loop from measurement to calculation and has long dominated industrial practice.

[0003] However, conventional practices typically treat battery modules as a collection of independent cells, focusing only on the current voltage or state of charge differences, neglecting the complex coupling relationships between cells through electrochemical characteristics, thermal field distribution, and connection resistance. This isolated perspective leads to a lack of globality in balancing decisions—directly comparing instantaneous differences in voltage or state of charge fails to reflect the evolution trend of cell states during the dynamic charging and discharging process, easily causing oscillating balancing or over-balancing. Especially under non-steady-state conditions (such as variable rate charging and discharging or significant temperature gradients), the applicability of a single threshold drops sharply, and the balancing system frequently starts and stops ineffectively, wasting energy and accelerating connector aging. Existing balancing methods are essentially reactive: intervention only occurs after deviations exceed limits, rather than predicting the direction of imbalance development based on the overall state trajectory of the battery module. This causes balancing actions to lag behind system changes, and the endpoint state of each balancing cycle is entirely determined by a preset threshold, failing to adaptively adjust the target voltage distribution based on the intrinsic coupling topology of the module (such as current path dependence and heat transfer relationships). Therefore, the balancing process often has slow convergence speed, redundant energy transfer paths, and may even induce new sources of imbalance due to local over-balancing. For example, traditional active balancing only focuses on energy transfer between adjacent cells, but fails to utilize global coupling information to optimize energy transfer priorities and pathways, resulting in low balancing efficiency, which is particularly prominent in modules with a large number of cells connected in series. In summary, existing technologies lack in-depth modeling of the inherent dynamic coupling relationships and state evolution paths of battery modules, making it difficult to achieve truly adaptive and highly efficient balancing control. Summary of the Invention

[0004] This invention provides a battery module adaptive balancing method and system based on intelligent algorithms, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a battery module adaptive balancing method based on intelligent algorithms, comprising: Collect voltage and temperature data of each individual cell in the battery module, construct a state trajectory matrix from the voltage and temperature data, and calculate the mutual information entropy between each individual cell in the state trajectory matrix. A coupled topology graph is constructed based on mutual information entropy. Dominant nodes are identified in the coupled topology graph. The state trajectories corresponding to the rows of dominant nodes are extracted from the state trajectory matrix. The extracted state trajectories are propagated to other nodes along the edges of the coupled topology graph to generate a set of predicted trajectories. Extract the endpoint voltage value of each predicted trajectory from the predicted trajectory set, construct an optimization function with topological constraints by combining the adjacency weights of each node in the coupled topology graph, and solve the optimization function to obtain the target voltage distribution; The target voltage distribution is taken as the termination state, and the voltage data is taken as the initial state. The priority is determined according to the weight of the edges in the coupled topology graph. The intermediate voltage sequence is generated by searching backward from the termination state to the initial state according to the priority. The conduction time and conduction current of each equalization path are calculated based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology diagram, thereby controlling the equalization circuit in the battery module.

[0006] Voltage and temperature data are used to construct a state trajectory matrix. The mutual information entropy between individual cells in the state trajectory matrix is ​​calculated, including: The voltage and temperature data collected from each individual cell at continuous time are used to form coordinate pairs. The coordinate pairs of each individual cell are arranged in chronological order to form a state trajectory. The state trajectories of each individual battery cell are stacked row by row to construct a state trajectory matrix; Divide the voltage-temperature coordinate plane into grid regions, map the coordinate pairs of each row of state trajectories in the state trajectory matrix to the grid regions, record the grid region number to which each coordinate pair belongs, and arrange the grid region numbers of each row of state trajectories in chronological order to form a grid number sequence. Extract the grid number sequence corresponding to any two rows from the state trajectory matrix, count the number combinations that appear at the same time for the two grid number sequences, and calculate the mutual information entropy between each individual cell based on the joint distribution characteristics of the number combinations and the marginal distribution characteristics of the individual number sequences.

[0007] A coupled topology graph is constructed based on mutual information entropy. Dominant nodes are identified within the coupled topology graph, and the state trajectories corresponding to these dominant nodes are propagated along the edges of the coupled topology graph to other nodes, generating a set of predicted trajectories, including: Each individual battery cell is mapped to a node, and the mutual information entropy is used as the weight of the directed edge to construct a coupled topology graph; Traverse the nodes in the coupled topology graph, count the sum of the weights of the incoming edges and the sum of the weights of the outgoing edges of each node, calculate the difference between the sum of the weights of the incoming edges and the sum of the weights of the outgoing edges as an asymmetry index, select the node with the largest absolute value of the asymmetry index as the dominant node, and extract the state trajectory of the row corresponding to the dominant node from the state trajectory matrix as the seed trajectory. Extract the weights of directed edges from the dominant node to the adjacent node in the coupled topology graph. Perform a decay transformation on the seed trajectory based on the weights of the directed edges to generate the propagation trajectory. Extract the state trajectory of the corresponding row of the adjacent node from the state trajectory matrix. Calculate the spatial distance deviation between the propagation trajectory and the state trajectory of the adjacent node. Correct the propagation trajectory based on the spatial distance deviation to generate the corrected trajectory. Calculate the cumulative deviation between the corrected trajectory and the state trajectories of adjacent nodes. Based on the cumulative deviation, update the weights of the directed edges from the dominant node to the adjacent nodes in the coupled topology graph. Use the updated directed edge weights to perform a decay transformation on the corrected trajectory to generate the predicted trajectory. Summarize the predicted trajectories of each adjacent node to form a set of predicted trajectories.

[0008] Extract the directed edge weights from the dominant node to its adjacent nodes in the coupled topology graph. Perform a decay transformation on the seed trajectory based on these weights to generate the propagation trajectory. Extract the state trajectories corresponding to the adjacent nodes from the state trajectory matrix. Calculate the spatial distance deviation between the propagation trajectory and the adjacent node state trajectories. Correct the propagation trajectory based on this spatial distance deviation to generate the corrected trajectory, including: Search for multiple propagation paths from the dominant node to each neighboring node in the coupled topology graph, extract the weights of the directed edges connected in series on each propagation path and multiply them to generate a path propagation factor, and apply a decay transformation to the seed trajectory based on the path propagation factor to generate multiple candidate propagation trajectories. Extract the state trajectory of each adjacent node from the state trajectory matrix, match the state trajectory of each adjacent node with the corresponding multiple candidate propagation trajectories in the voltage-temperature coordinate space, calculate the trajectory similarity between each candidate propagation trajectory and the state trajectory of the adjacent node, select the candidate propagation trajectory with the highest trajectory similarity as the propagation trajectory, and record the path propagation factor corresponding to the propagation trajectory. Calculate the Euclidean distance between the propagation trajectory and the state trajectory of adjacent nodes at each time point in the voltage-temperature coordinate space, accumulate the Euclidean distances at each time point to obtain the spatial distance deviation, and calculate the correction weight based on the spatial distance deviation and the path propagation factor. Based on the spatial distance deviation, a time-by-time deviation vector is constructed from the propagation trajectory to the state trajectory of the adjacent node. The deviation vector is then scaled according to the correction weight to generate a correction vector. The propagation trajectory is then superimposed along the direction of the correction vector to generate the correction trajectory.

[0009] The endpoint voltage values ​​of each predicted trajectory in the predicted trajectory set are extracted. An optimization function with topological constraints is constructed by combining the adjacency weights of each node in the coupled topology graph. Solving the optimization function yields the target voltage distribution, including: Extract the endpoint voltage value corresponding to the end of the time series for each predicted trajectory from the predicted trajectory set, and construct an initial voltage vector by arranging the endpoint voltage values ​​according to the node index. The adjacency matrix formed by the adjacency weights of each node in the coupled topology graph is decomposed into a topological eigenspace. The initial voltage vector is then mapped to the topological eigenspace to obtain the modal characterization vector. The degenerate adjacency matrix is ​​obtained by filtering and removing edges whose adjacency weight values ​​are less than a preset weight threshold from the adjacency matrix. The degenerate eigenspace is obtained by performing spectral decomposition on the degenerate adjacency matrix. The degenerate mode vector is then mapped to the degenerate eigenspace to obtain the degenerate mode vector. The modal characterization vector is inversely mapped to the node space to obtain the reconstructed voltage vector. The deviation vector between the reconstructed voltage vector and the initial voltage vector is calculated. The conservation deviation vector of the voltage values ​​of adjacent nodes in the reconstructed voltage vector is calculated according to the adjacency weight. The magnitude change of the degenerate mode vector is extracted. The deviation vector, the conserved deviation vector, and the magnitude change are used as constraints to construct an optimization function with topological constraints. The optimization function is iteratively solved to update the modal representation vector until convergence is obtained to obtain the optimal modal representation vector. The optimal modal representation vector is then inversely mapped to the node space to obtain the target voltage distribution.

[0010] The priority is determined based on the edge weights in the coupled topology graph. The intermediate voltage sequence is generated by reverse searching from the termination state to the initial state according to the priority, including: Calculate the voltage difference between each node in the final state and the initial state. Construct a backpropagation weight matrix based on the voltage difference and the weight of the edge in the coupled topology graph. Sort the backpropagation weight matrix by row to generate a back transfer priority sequence. Predecessor nodes are selected from the termination state according to the reverse transition priority sequence. The weight values ​​corresponding to the predecessor nodes and the termination state in the reverse propagation weight matrix are used as reverse coefficients. The reverse voltage values ​​of the predecessor nodes are calculated based on the reverse coefficients and the node voltage values ​​in the termination state. The reverse voltage values ​​of each predecessor node are used as the current search layer state. Calculate the deviation between the current search layer state and the voltage values ​​of each node in the initial state, correct the corresponding weight values ​​in the backpropagation weight matrix based on the deviation, and update the backpropagation priority sequence by sorting the corrected backpropagation weight matrix by row. Using the updated reverse transition priority sequence, the current search layer state is used as the starting point to continue selecting predecessor nodes and calculating the reverse voltage value. This process is repeated until the deviation converges. The intermediate voltage sequence is generated by arranging the voltage values ​​of each node in the state of each search level in the order of search level.

[0011] The conduction time and conduction current of each equalization path are calculated based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology graph. The equalization circuit in the battery module includes: Extract the voltage values ​​corresponding to each time step from the intermediate voltage sequence and calculate the voltage change rate at each node; The energy transfer intensity is obtained by multiplying the adjacency weight in the coupled topology graph by the voltage change rate. Node pairs with energy transfer intensity exceeding the preset intensity threshold are selected as node pairs to be balanced. The connection edges between the node pairs to be balanced are extracted as balancing paths. The adjacency weights corresponding to each balancing path are extracted as path weights. Construct a path admittance matrix based on the path weight of each equalization path, input the voltage change rate into the path admittance matrix to solve the current distribution coefficient of each equalization path, and sort the equalization paths according to the current distribution coefficient to generate an equalization path sequence. Extract the voltage values ​​of the nodes to be balanced corresponding to each equalization path in the intermediate voltage sequence at each time step according to the equalization path sequence, calculate the voltage difference at each time step, and calculate the instantaneous equalization current based on the voltage difference and the current distribution coefficient. The instantaneous equalization current of each equalization path is integrated over time until the charge transfer saturation is reached. The integration time is recorded as the conduction time, and the integral mean of the instantaneous equalization current is calculated as the conduction current. The equalization control command is generated based on the conduction time and conduction current of each equalization path to control the equalization circuit in the battery module.

[0012] A second aspect of this invention provides a battery module adaptive balancing system based on intelligent algorithms, comprising: The data acquisition unit is used to collect voltage and temperature data of each individual cell in the battery module, construct a state trajectory matrix from the voltage and temperature data, and calculate the mutual information entropy between each individual cell in the state trajectory matrix. The topology building unit is used to construct a coupled topology graph based on mutual information entropy, identify the dominant node in the coupled topology graph, extract the state trajectory corresponding to the row of the dominant node from the state trajectory matrix, and propagate the extracted state trajectory along the edge of the coupled topology graph to other nodes to generate a set of predicted trajectories. An optimization unit is used to extract the endpoint voltage value of each predicted trajectory in the predicted trajectory set. It combines the adjacency weights of each node in the coupled topology graph to construct an optimization function with topological constraints, and solves the optimization function to obtain the target voltage distribution. The reverse search unit is used to take the target voltage distribution as the termination state and the voltage data as the initial state, determine the priority according to the weight of the edges in the coupled topology graph, and reverse search from the termination state to the initial state according to the priority to generate an intermediate voltage sequence. The equalization control unit is used to calculate the conduction time and conduction current of each equalization path based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology graph, and to control the equalization circuit in the battery module.

[0013] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] This method calculates the nonlinear correlation between batteries using mutual information entropy. The constructed coupled topology graph can accurately capture the dynamic coupling relationship of each individual battery under complex operating conditions, significantly improving the ability to perceive differences in the internal state of the battery pack. Based on the state trajectory propagation mechanism of dominant node identification, it effectively avoids the inefficiency of traditional methods that traverse the entire dataset, significantly reducing the computational complexity of prediction while preserving temporal evolution characteristics. Combined with the target voltage distribution obtained by solving the optimization function of topological constraints, it fundamentally overcomes the defects of over- or under-balancing caused by traditional fixed threshold or averaging strategies, achieving differentiated convergence of each battery voltage towards the optimal target. When generating intermediate voltage sequences through reverse search, priority is dynamically determined based on edge weights, causing the balancing path to automatically favor key difference units, avoiding unnecessary energy migration, thereby minimizing balancing energy consumption and simultaneously reducing the balancing cycle. The conduction duration and current of each balancing path are accurately calculated based on real-time voltage difference and topological adjacency weights, ensuring a high degree of matching between control commands and the current state of the battery. It can adapt to complex scenarios such as battery aging and uneven temperature, so that the battery module always maintains voltage consistency, effectively delays capacity decay, extends the overall service life, and improves energy utilization efficiency and system operation stability. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the adaptive balancing method for battery modules based on intelligent algorithms in this embodiment. Figure 2 This is a flowchart illustrating the execution of the battery module equalization control scheme in this embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0019] Figure 1 This is a flowchart illustrating the adaptive balancing method for battery modules based on intelligent algorithms, as described in an embodiment of the present invention.

[0020] Battery module adaptive balancing methods based on intelligent algorithms include: Collect voltage and temperature data of each individual cell in the battery module, construct a state trajectory matrix from the voltage and temperature data, and calculate the mutual information entropy between each individual cell in the state trajectory matrix. A coupled topology graph is constructed based on mutual information entropy. Dominant nodes are identified in the coupled topology graph. The state trajectories corresponding to the rows of dominant nodes are extracted from the state trajectory matrix. The extracted state trajectories are propagated to other nodes along the edges of the coupled topology graph to generate a set of predicted trajectories. Extract the endpoint voltage value of each predicted trajectory from the predicted trajectory set, construct an optimization function with topological constraints by combining the adjacency weights of each node in the coupled topology graph, and solve the optimization function to obtain the target voltage distribution; The target voltage distribution is taken as the termination state, and the voltage data is taken as the initial state. The priority is determined according to the weight of the edges in the coupled topology graph. The intermediate voltage sequence is generated by searching backward from the termination state to the initial state according to the priority. The conduction time and conduction current of each equalization path are calculated based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology diagram, thereby controlling the equalization circuit in the battery module.

[0021] Voltage and temperature data are used to construct a state trajectory matrix. The mutual information entropy between individual cells in the state trajectory matrix is ​​calculated, including: The voltage and temperature data collected from each individual cell at continuous time are used to form coordinate pairs. The coordinate pairs of each individual cell are arranged in chronological order to form a state trajectory. The state trajectories of each individual battery cell are stacked row by row to construct a state trajectory matrix; Divide the voltage-temperature coordinate plane into grid regions, map the coordinate pairs of each row of state trajectories in the state trajectory matrix to the grid regions, record the grid region number to which each coordinate pair belongs, and arrange the grid region numbers of each row of state trajectories in chronological order to form a grid number sequence. Extract the grid number sequence corresponding to any two rows from the state trajectory matrix, count the number combinations that appear at the same time for the two grid number sequences, and calculate the mutual information entropy between each individual cell based on the joint distribution characteristics of the number combinations and the marginal distribution characteristics of the individual number sequences.

[0022] During the operation of the battery module, voltage and temperature sensors are used to monitor each individual cell in real time. The voltage sensor collects the terminal voltage of each individual cell at a fixed sampling frequency (e.g., once per second or once every 100 milliseconds), while the temperature sensor simultaneously collects the temperature value of the surface or interior of each individual cell. Assuming the battery module contains several individual cells, each individual cell accumulates a series of voltage and temperature measurements over a continuous time window (e.g., 10 to 30 minutes).

[0023] For any single battery cell, the voltage and temperature values ​​collected at a given moment constitute a coordinate pair. For example, at the first sampling moment, the voltage of a single battery cell is 3.65V and the temperature is 25.3℃, so the coordinate pair at that moment is (3.65, 25.3). As time progresses, the voltage of the same battery cell at the second sampling moment may change to 3.66V and the temperature to 25.5℃, corresponding to the coordinate pair (3.66, 25.5). Arranging the coordinate pairs of the single battery cell at all sampling moments in chronological order forms a trajectory curve in the voltage-temperature two-dimensional plane. This curve is the state trajectory of the single battery cell. The state trajectory reflects the dynamic evolution of the voltage and temperature of the single battery cell within the observation time window, and can reflect the battery's charge-discharge characteristics, thermal effects, and interactions with other batteries.

[0024] The above process is repeated for each individual cell in the battery module, generating a state trajectory for each cell. The state trajectories of all individual cells are then stacked row-wise to construct a matrix structure. Each row of this matrix corresponds to the state trajectory of one individual cell, and each column corresponds to the voltage-temperature coordinates of all individual cells at a given sampling time. This matrix is ​​the state trajectory matrix, with the number of rows equal to the number of individual cells and the number of columns equal to the number of sampling times. The state trajectory matrix stores all the state information of the battery module within the observation time window in a compact form, providing a data foundation for subsequent coupling relationship analysis.

[0025] To quantify the coupling strength between individual battery cells, it is necessary to calculate the mutual information entropy between the state trajectories of any two individual cells. Mutual information entropy measures the degree of statistical dependence between two random variables; a higher value indicates a stronger correlation between the two variables. In the battery module scenario, if the state trajectories of two individual cells have high mutual information entropy, it indicates that they have significant synergistic or coupling effects during voltage and temperature evolution, possibly due to factors such as physical proximity, shared current paths, or heat conduction.

[0026] To calculate the mutual information entropy, a grid region is first defined within the voltage-temperature coordinate plane. Specifically, based on the battery module's voltage operating range (e.g., 3.0V to 4.2V) and temperature operating range (e.g., 15℃ to 45℃), both the voltage and temperature axes are divided into several equally spaced intervals. For example, the voltage axis is divided into 24 intervals, each with a width of 0.05V; the temperature axis is divided into 30 intervals, each with a width of 1℃. The intervals of the voltage and temperature axes are combined to form a grid region, resulting in a total of 720 grid cells. Each grid cell corresponds to a rectangular region within the voltage-temperature plane and is assigned a unique number (e.g., from 1 to 720).

[0027] The state trajectory in each row of the state trajectory matrix is ​​mapped to a grid region. For each coordinate pair in a given row of the state trajectory, the grid cell into which the coordinate pair falls is determined based on its voltage and temperature values, and the grid cell number is recorded. For example, if the coordinate pair (3.65, 25.3) falls into grid cell number 158, then the grid cell number corresponding to that moment is 158. All coordinate pairs in that row of the state trajectory are mapped to grid numbers sequentially, forming a grid number sequence in chronological order. The length of this sequence is equal to the number of sampling moments, and each element in the sequence is an integer between 1 and 720. This mapping process is repeated for each row of the state trajectory matrix, generating a grid number sequence for each individual cell.

[0028] Extract the grid number sequences corresponding to any two rows from the state trajectory matrix, denoted as Sequence A and Sequence B respectively. Sequence A and Sequence B have the same length, equal to the number of sampling times. At the same time, Sequence A and Sequence B each have one grid number, forming a number combination. For example, at the 5th sampling time, Sequence A has the number 158, and Sequence B has the number 162, so the number combination at that time is (158, 162). Iterate through all sampling times, counting all occurrences of number combinations and their frequency. The total number of number combinations is at most the square of the number of grid cells (i.e., 720 × 720), but the actual number of number combinations is usually much smaller than this theoretical upper limit because the battery state trajectory only covers a finite region within the voltage-temperature plane.

[0029] Based on the statistical results of the numbered combinations, the joint distribution characteristics are calculated. The joint probability of each numbered combination is obtained by dividing the number of occurrences of each combination by the total number of sampling times. For example, if the numbered combination (158, 162) appears 3 times in 100 sampling times, its joint probability is 0.03. The joint probabilities of all numbered combinations constitute the joint probability distribution of sequence A and sequence B.

[0030] Simultaneously, the marginal distribution characteristics of a single number sequence are calculated. For sequence A, the number of times each grid number appears in sequence A is counted, and this number is divided by the total number of sampling times to obtain the marginal probability of that number in sequence A. For example, number 158 appears 5 times in sequence A, so its marginal probability is 0.05. The same statistical analysis is performed on sequence B to obtain the marginal probability distribution of sequence B.

[0031] The mutual information entropy between sequence A and sequence B is calculated using the joint probability distribution and marginal probability distribution. The formula for calculating mutual information entropy is: ,in, Indicates number combination The joint probability, Indicates number Marginal probabilities in sequence A Indicates number Marginal probabilities in sequence B. Summation is performed by iterating through all actually occurring number combinations. Mutual information entropy. The value reflects the statistical dependence between sequence A and sequence B. The larger the value, the more similar the evolution patterns of the two sequences are, and the stronger the coupling between the two individual cells.

[0032] By pairwise combining all rows in the state trajectory matrix, the above calculation process is repeated to obtain the mutual information entropy between any two individual cells. All mutual information entropy values ​​are then organized into a symmetric matrix, where the number of rows and columns equals the number of individual cells. Line 1 The elements of the column represent the first... The single cell and the first The mutual information entropy between individual battery cells. This matrix provides quantified coupling strength information for subsequent construction of the coupled topology graph, enabling the equalization strategy to adaptively adjust according to the actual coupling relationship between the cells, thereby improving equalization efficiency and the overall performance of the battery module.

[0033] Using the above method, the original voltage and temperature data are transformed into a state trajectory matrix, and the coupling relationships between individual battery cells are extracted through gridded mapping and mutual information entropy calculation. This process does not require pre-setting the connection topology or physical model between batteries; it is entirely based on measured data for adaptive analysis, and can adapt to structural differences and operating condition changes in different battery modules, providing reliable data support for subsequent equalization control.

[0034] A coupled topology graph is constructed based on mutual information entropy. Dominant nodes are identified within the coupled topology graph, and the state trajectories corresponding to these dominant nodes are propagated along the edges of the coupled topology graph to other nodes, generating a set of predicted trajectories, including: Each individual battery cell is mapped to a node, and the mutual information entropy is used as the weight of the directed edge to construct a coupled topology graph; Traverse the nodes in the coupled topology graph, count the sum of the weights of the incoming edges and the sum of the weights of the outgoing edges of each node, calculate the difference between the sum of the weights of the incoming edges and the sum of the weights of the outgoing edges as an asymmetry index, select the node with the largest absolute value of the asymmetry index as the dominant node, and extract the state trajectory of the row corresponding to the dominant node from the state trajectory matrix as the seed trajectory. Extract the weights of directed edges from the dominant node to the adjacent node in the coupled topology graph. Perform a decay transformation on the seed trajectory based on the weights of the directed edges to generate the propagation trajectory. Extract the state trajectory of the corresponding row of the adjacent node from the state trajectory matrix. Calculate the spatial distance deviation between the propagation trajectory and the state trajectory of the adjacent node. Correct the propagation trajectory based on the spatial distance deviation to generate the corrected trajectory. Calculate the cumulative deviation between the corrected trajectory and the state trajectories of adjacent nodes. Based on the cumulative deviation, update the weights of the directed edges from the dominant node to the adjacent nodes in the coupled topology graph. Use the updated directed edge weights to perform a decay transformation on the corrected trajectory to generate the predicted trajectory. Summarize the predicted trajectories of each adjacent node to form a set of predicted trajectories.

[0035] After obtaining the state trajectory matrix and the mutual information entropy between individual cells, it is necessary to transform these numerical correlation strengths into an operable network structure. Each individual cell in the battery module is abstracted as a node, and the connection relationships between nodes are determined by the mutual information entropy. Specifically, if an individual cell... With single cell The mutual information entropy between them is Then, slave nodes are established in the coupled topology graph. Pointing to node A directed edge, the weight of which is... Because mutual information entropy itself is asymmetric, that is... Not necessarily equal to Therefore, the constructed coupled topology graph is a directed graph. This directed graph can characterize the direction and intensity of influence between individual cells, providing a topological basis for subsequent dominant node identification and state propagation.

[0036] After constructing the coupled topology graph, it is necessary to identify the nodes that play a dominant role in the state evolution of the entire battery module. This involves traversing all nodes in the coupled topology graph and processing each node... The sum of the weights of all directed edges pointing to this node is denoted as the sum of the weights of the incoming edges. Simultaneously, the sum of the weights of the directed edges pointing from this node to other nodes is calculated and denoted as the sum of the outgoing edge weights. Calculate the difference between the two. This difference reflects the node Asymmetry in information dissemination. If A positive value and a large value indicate that the node... The node whose influence on other nodes is stronger than the influence of other nodes on it holds a dominant position in information output within the topology. (Selection) The largest node is designated as the dominant node, and the individual cell corresponding to this node possesses the strongest state guidance capability within the battery module. The state trajectory corresponding to the dominant node is extracted from the state trajectory matrix. This trajectory contains the voltage and temperature evolution sequence of the individual cell throughout the entire acquisition cycle and is used as the seed trajectory for subsequent propagation.

[0037] The propagation of the seed trajectory is based on the directed edges in the coupled topological graph. The weights of the directed edges from the dominant node to each adjacent node are extracted, and the dominant node is denoted as node A. One of its adjacent nodes is a node. The corresponding directed edge weight is The seed trajectory is attenuated based on the directed edge weights to generate the propagation trajectory.

[0038] Specifically, if the seed trajectory is at time step The voltage value is Temperature value Then it propagates to the node. The transmission trajectory in time steps The voltage value is Temperature value ,in and The attenuation coefficient controls the degree of information loss during propagation. The attenuation transformation reflects the intensity attenuation law of information propagation in the topology; the larger the edge weight, the smaller the attenuation, and the closer the propagation trajectory is to the seed trajectory.

[0039] After the propagation trajectory is generated, it needs to be compared and corrected with the actual state trajectories of adjacent nodes. Nodes are extracted from the state trajectory matrix. The state trajectory of the corresponding row is denoted as and Calculate the spatial distance deviation between the propagation trajectory and the state trajectories of adjacent nodes.

[0040] Calculate the deviation at each time step along the voltage dimension. In terms of temperature, the deviation at each time step is calculated. After normalizing the voltage and temperature deviations, the overall spatial distance deviation is calculated. ,in and As a reference scale, it is used to unify the dimensions of different physical quantities. The propagation trajectory is corrected based on spatial distance deviation; the corrected trajectory is then updated at the time step. The voltage value is Temperature value ,in This is a correction coefficient used to control the pull strength of the actual state trajectory on the propagation trajectory. The corrected trajectory retains the propagation characteristics of the seed trajectory while incorporating the actual state information of adjacent nodes, thus reflecting the node's position more accurately. The state evolution trend under the influence of the dominant node.

[0041] To further improve prediction accuracy, the directed edge weights in the coupled topology graph need to be updated in reverse based on the cumulative deviation between the corrected trajectory and the actual state trajectories of adjacent nodes. The cumulative deviation between the corrected trajectory and the state trajectories of adjacent nodes over the entire time series is calculated. This cumulative deviation reflects the current directed edge weights. For nodes The accuracy of state prediction. If the accumulated deviation is large, it indicates that the current edge weights fail to accurately depict the influence of the dominant node on its adjacent nodes, and the edge weights need to be adjusted. A gradient descent strategy is used to update the edge weights, and the updated edge weights are... ,in The learning rate controls the update step size. Updating the edge weights allows the coupled topology graph to dynamically adapt to the actual operating state of the battery module, improving the accuracy of subsequent propagation processes.

[0042] Use the updated directed edge weights The corrected trajectory is then subjected to another attenuation transformation to generate the final predicted trajectory.

[0043] Predicted trajectory at time step The voltage value is Temperature value This predicted trajectory integrates the propagation characteristics of the seed trajectory, the actual state information of adjacent nodes, and the updated topological weights, enabling accurate prediction of the future state evolution of adjacent nodes under the influence of the dominant node. The above propagation, correction, and prediction process is repeated for all adjacent nodes of the dominant node, and the predicted trajectories of each adjacent node are aggregated to form a predicted trajectory set. This set contains the predicted state evolution of each individual battery cell in the battery module under the guidance of the dominant node, providing crucial input for subsequent target voltage distribution construction and equalization strategy formulation.

[0044] In practical applications, attenuation coefficient and The value of needs to be adjusted based on the specific characteristics of the battery module. For battery modules with low internal resistance and good consistency, a smaller attenuation coefficient can be selected to make the propagation trajectory closer to the seed trajectory, making full use of the state information of the dominant node. For battery modules with high internal resistance and poor consistency, a larger attenuation coefficient can be selected to enhance the influence of the actual state of adjacent nodes on the predicted trajectory, avoiding prediction bias caused by over-reliance on the dominant node. Correction coefficient The value of also needs to be considered in relation to the degree of integration between propagation characteristics and the actual state, and is typically between 0.3 and 0.7. Learning rate The choice of learning rate needs to consider the balance between convergence speed and stability. An excessively large learning rate may cause oscillations in edge weight updates, while an excessively small learning rate will reduce the efficiency of adaptive adjustment. By properly configuring these parameters, the predicted trajectory set can achieve an optimal balance between accuracy and robustness, providing a reliable state prediction basis for the adaptive equilibrium control of the battery module.

[0045] Extract the directed edge weights from the dominant node to its adjacent nodes in the coupled topology graph. Perform a decay transformation on the seed trajectory based on these weights to generate the propagation trajectory. Extract the state trajectories corresponding to the adjacent nodes from the state trajectory matrix. Calculate the spatial distance deviation between the propagation trajectory and the adjacent node state trajectories. Correct the propagation trajectory based on this spatial distance deviation to generate the corrected trajectory, including: Search for multiple propagation paths from the dominant node to each neighboring node in the coupled topology graph, extract the weights of the directed edges connected in series on each propagation path and multiply them to generate a path propagation factor, and apply a decay transformation to the seed trajectory based on the path propagation factor to generate multiple candidate propagation trajectories. Extract the state trajectory of each adjacent node from the state trajectory matrix, match the state trajectory of each adjacent node with the corresponding multiple candidate propagation trajectories in the voltage-temperature coordinate space, calculate the trajectory similarity between each candidate propagation trajectory and the state trajectory of the adjacent node, select the candidate propagation trajectory with the highest trajectory similarity as the propagation trajectory, and record the path propagation factor corresponding to the propagation trajectory. Calculate the Euclidean distance between the propagation trajectory and the state trajectory of adjacent nodes at each time point in the voltage-temperature coordinate space, accumulate the Euclidean distances at each time point to obtain the spatial distance deviation, and calculate the correction weight based on the spatial distance deviation and the path propagation factor. Based on the spatial distance deviation, a time-by-time deviation vector is constructed from the propagation trajectory to the state trajectory of the adjacent node. The deviation vector is then scaled according to the correction weight to generate a correction vector. The propagation trajectory is then superimposed along the direction of the correction vector to generate the correction trajectory.

[0046] After obtaining the dominant node and its seed trajectory, the seed trajectory needs to be propagated to other nodes along the edges of the coupled topology graph. Due to the complex coupling relationships between individual cells in the battery module, the propagation process cannot simply proceed along a single path, but needs to consider multiple possible propagation paths from the dominant node to the target adjacent node.

[0047] For each adjacent node in the coupled topology graph, a breadth-first search algorithm is used to search for all reachable paths starting from the dominant node. Let the dominant node be... The target adjacent node is The search result is the first The path is denoted as The path consists of a sequence of nodes, represented as ,in This is an intermediate node on the path. This represents the path length. To avoid the path search getting stuck in a loop, the maximum search depth is set to 5, meaning that only paths that pass through no more than 5 intermediate nodes are considered.

[0048] For the Path Extract the weights of all connected directed edges along the path. Specifically, the weights of the directed edges between adjacent node pairs along the path are as follows: , ... The path propagation factor of this path is obtained by multiplying the weights of these directed edges. The calculation formula is: Path propagation factor This reflects the cumulative attenuation of information as it propagates along the path. Since the weight of each edge is usually less than 1, the longer the path, the smaller the cumulative product, indicating more severe propagation attenuation.

[0049] Based on the path propagation factor, a decay transformation is applied to the seed trajectory to generate candidate propagation trajectories. Let the seed trajectory be at time step... The voltage value is Temperature value . No. The candidate propagation trajectories corresponding to each path at time step voltage value and temperature value The calculation is as follows: , ,in, The voltage decay index, typically ranging from 0.8 to 1.2, is used to adjust the non-linearity of voltage decay. The temperature decay index is typically between 0.5 and 1.0. The ambient temperature is typically set to 25 degrees Celsius. The temperature decay formula reflects the physical characteristic of temperature reverting to the ambient temperature.

[0050] For the target adjacent node Extract the state trajectory of the corresponding row from the state trajectory matrix, denoted as . The state trajectory is then matched with all candidate propagation trajectories in a voltage-temperature two-dimensional coordinate space. Specifically, for the... Calculate the distance between each candidate propagation trajectory and the node. Trajectory similarity of state trajectories Trajectory similarity is calculated using the reciprocal of the dynamic time-warped distance, and the formula is as follows: ,in, For the first Candidate propagation trajectories and nodes The dynamic time warped distance between state trajectories. The dynamic time warped distance is calculated by constructing a cumulative distance matrix and finding the optimal alignment path, which can handle the nonlinear alignment problem of trajectories in the time dimension.

[0051] Among all candidate propagation trajectories, trajectory similarity is selected. The largest candidate propagation trajectory is taken as the final propagation trajectory, denoted as... Simultaneously record the path propagation factor corresponding to this propagation trajectory. This is used for subsequent weight correction calculations. After obtaining the propagation trajectory, it is necessary to calculate its relationship with the nodes. Spatial distance deviation between state trajectories. In the voltage-temperature two-dimensional coordinate space, the propagation trajectory at time step... The coordinates of the point are ,node State trajectory at time step The coordinates of the point are Calculate the Euclidean distance between two points. : ;in, This is a voltage normalization scale, typically taken as 10% of the battery's rated voltage; for example, 0.37 volts for a 3.7-volt lithium battery. This is a temperature normalization scale, typically taken as 10 degrees Celsius. Normalization ensures that physical quantities with different dimensions, voltage and temperature, are comparable in distance calculations.

[0052] The spatial distance deviation is obtained by summing the Euclidean distances of all time steps. Based on spatial distance deviation and path propagation factor, the corrected weights are calculated. The calculation of the correction weight needs to consider two factors: first, the greater the spatial distance deviation, the stronger the correction should be; second, the smaller the path propagation factor, the longer the propagation path or the weaker the coupling, and the lower the reliability of the correction should be. The formula for calculating the correction weight is: ,in, This is the distance threshold, typically set to 0.5, used to control the saturation characteristics of the correction weights. Hyperbolic tangent function. Ensure that the adjustment weights change smoothly between 0 and 1 to avoid over-adjustment.

[0053] Constructing a path from the propagation trajectory to the node The time-by-time deviation vector of the state trajectory. At time step... Deviation vector In voltage-temperature coordinate space, it is represented as: The direction of the deviation vector points to the actual trajectory, and the magnitude reflects the degree of deviation at that moment.

[0054] The deviation vector is scaled by the correction weights to generate the correction vector. : The correction vector maintains the direction of the deviation vector, but the magnitude is adjusted according to the correction weight, thus achieving adaptive correction force control.

[0055] The propagation trajectory is vector-superimposed along the direction of the correction vector to generate the corrected trajectory. The corrected trajectory is generated at time step [missing information]. voltage value and temperature value The calculation is as follows: , ,in, and These represent the components of the correction vector in the voltage and temperature dimensions, respectively. The corrected trajectory integrates the global trend of the propagation trajectory and the local characteristics of the actual state trajectory. It retains the guiding role of the dominant node while taking into account the individual differences of each node, providing a more accurate basis for subsequent prediction trajectory generation.

[0056] The endpoint voltage values ​​of each predicted trajectory in the predicted trajectory set are extracted. An optimization function with topological constraints is constructed by combining the adjacency weights of each node in the coupled topology graph. Solving the optimization function yields the target voltage distribution, including: Extract the endpoint voltage value corresponding to the end of the time series for each predicted trajectory from the predicted trajectory set, and construct an initial voltage vector by arranging the endpoint voltage values ​​according to the node index. The adjacency matrix formed by the adjacency weights of each node in the coupled topology graph is decomposed into a topological eigenspace. The initial voltage vector is then mapped to the topological eigenspace to obtain the modal characterization vector. The degenerate adjacency matrix is ​​obtained by filtering and removing edges whose adjacency weight values ​​are less than a preset weight threshold from the adjacency matrix. The degenerate eigenspace is obtained by performing spectral decomposition on the degenerate adjacency matrix. The degenerate mode vector is then mapped to the degenerate eigenspace to obtain the degenerate mode vector. The modal characterization vector is inversely mapped to the node space to obtain the reconstructed voltage vector. The deviation vector between the reconstructed voltage vector and the initial voltage vector is calculated. The conservation deviation vector of the voltage values ​​of adjacent nodes in the reconstructed voltage vector is calculated according to the adjacency weight. The magnitude change of the degenerate mode vector is extracted. The deviation vector, the conserved deviation vector, and the magnitude change are used as constraints to construct an optimization function with topological constraints. The optimization function is iteratively solved to update the modal representation vector until convergence is obtained to obtain the optimal modal representation vector. The optimal modal representation vector is then inversely mapped to the node space to obtain the target voltage distribution.

[0057] After obtaining the set of predicted trajectories, it is necessary to extract the endpoint voltage value corresponding to each predicted trajectory at the end of the time series. The set of predicted trajectories contains multiple trajectories generated from the dominant node to other nodes, and each trajectory records the voltage evolution process from the initial time to the prediction termination time. For nodes... The corresponding predicted trajectory, its endpoint voltage value is denoted as This represents the voltage value of the predicted trajectory at the last time step. The initial voltage vector is constructed by sequentially arranging the endpoint voltage values ​​of all nodes in ascending order of node index. ,in This represents the total number of individual cells in the battery module. The initial voltage vector reflects the voltage state of each individual cell at a future moment, predicted based on the propagation mechanism, but this state may have local inconsistencies or violate topological constraints.

[0058] To optimize the initial voltage vector using the structural information contained in the coupled topology graph, spectral decomposition of the adjacency matrix, which is composed of the adjacency weights of each node in the coupled topology graph, is required. (Adjacency matrix) It is A square matrix, in which elements Represents a node Pointing to node The directed edge weights. The adjacency matrix is ​​symmetric by performing a symmetric transformation on it to obtain a symmetric adjacency matrix. Then, calculate the eigenvalues ​​and eigenvectors of the symmetric adjacency matrix. The eigenvalues ​​are arranged in descending order and denoted as . The corresponding eigenvector is denoted as These eigenvectors form an orthogonal basis of the topological eigenspace, capable of characterizing the multi-scale structural features of the coupled topological graph. The initial voltage vector... Mapping to the topological eigenspace yields the modal representation vector. , of which The modal coefficients are obtained through projection calculation: The modal representation vector decomposes the voltage distribution into a superposition of different frequency components. Lower-order modes correspond to global smooth changes, while higher-order modes correspond to local detailed fluctuations.

[0059] In actual battery modules, the coupling relationships between some individual cells are relatively weak, resulting in small adjacency weights. These weakly coupled edges contribute little to the overall balancing process but introduce additional degrees of freedom and computational complexity during optimization. To simplify the optimization problem and highlight the constraint effect of strong coupling relationships, we use the adjacency matrix... The adjacent weight values ​​are selected when they are less than the preset weight threshold. Remove the edges. Specifically, iterate through all elements in the adjacency matrix; if... If the element is zero, then set the element to zero to obtain the degenerate adjacency matrix. The degenerate adjacency matrix preserves the main strong coupling connections, forming a sparse topology. Symmetric processing of the degenerate adjacency matrix yields... Then, spectral decomposition is performed to obtain eigenvalues. and eigenvectors These eigenvectors constitute the degenerate eigenspace. The modal representation vectors... Mapping to the degenerate eigenspace yields the degenerate mode vector. , of which The degradation mode coefficients are: The degenerate mode vector reflects the mode decomposition characteristics of the voltage distribution under the simplified topology.

[0060] To evaluate the validity of the current modal representation vector, it is inversely mapped to the node space to obtain the reconstructed voltage vector. The inverse mapping is achieved through a linear combination of the modal coefficients and eigenvectors: ,in The reconstructed voltage vector represents a reinterpretation of the initial voltage vector under topological eigenspace constraints. The deviation vector between the reconstructed and initial voltage vectors is calculated; each element of the deviation vector reflects the difference between the reconstructed voltage at a node and the predicted endpoint voltage. The norm of the deviation vector measures the magnitude of the reconstruction error, which should be minimized during optimization to maintain fidelity to the predicted trajectory information.

[0061] In addition to reconstruction error constraints, the conservation constraints of the topology on voltage distribution also need to be considered. In a coupled topology graph, there is an energy exchange relationship between adjacent nodes, and their voltage difference should be consistent with the adjacency weights. For nodes in the reconstructed voltage vector... and nodes If there is an edge connecting them and the adjacency weight is Then the conservation deviation is defined as: This metric characterizes the weighted square of the voltage difference between adjacent nodes; edges with larger weights should correspond to smaller voltage differences to satisfy strong coupling constraints. The conserved deviation is calculated for all node pairs with connected edges and summed to obtain the total conserved deviation vector. The total amount of the conserved deviation vector should be minimized during the optimization process to ensure that the voltage distribution conforms to the topological constraints.

[0062] The magnitude change of the degenerate mode vector reflects the distribution characteristics of the mode energy under the simplified topology. For the th Degenerate mode coefficients Its amplitude change is defined as the change in amplitude relative to the corresponding complete mode coefficient. The difference: The absolute value of the magnitude change This represents the energy loss or gain of the mode during the degradation process. The total amplitude change is obtained by weighted summation of the amplitude changes across all modes. ,in For the first The weight coefficients for each mode are typically set to be proportional to the eigenvalues. The amplitude should be proportional to the value of the low-order modes to emphasize their stability. The total amplitude variation should be constrained within a reasonable range during the optimization process to avoid introducing excessive modal distortion during degradation.

[0063] An optimization function with topological constraints is constructed using the deviation vector, the conserved deviation vector, and the magnitude change as constraints. The optimization function adopts a multi-objective weighted form: ,in , , These are weighting coefficients used to balance the relative importance of the three constraints. The first term... To ensure the reconstructed voltage vector is close to the initial predicted value, the second term... To ensure that the voltage distribution satisfies the topological conservation constraints, the third term... To ensure the mode decomposition remains stable during degradation, the independent variable of the optimization function is the mode representation vector. By adjusting the modal coefficients, the optimal balance between fidelity and topological consistency can be found.

[0064] The optimization function is iteratively solved, and the modal representation vector is updated until convergence. Iterative optimization is performed using gradient descent or a quasi-Newton method. In the next iteration, the modal representation vector is updated as follows: ,in The iteration step size, To optimize the gradient of the function at the current modal representation vector, gradient calculation involves taking partial derivatives with respect to three constraints and then summing them with weights. The iterative process continues until a convergence condition is met, which can be set to the change in the modal representation vector between two consecutive iterations being less than a preset threshold. Or, the change in the optimized function value is less than a preset threshold: After convergence, the optimal modal representation vector is obtained. .

[0065] The optimal modal representation vector The target voltage distribution is obtained by inverse mapping to the node space. Inverse mapping is achieved through a linear combination of the optimal mode coefficients and the eigenvectors of the topological eigenspace. ,in Elements in the target voltage distribution Represents a node The target voltage value corresponding to a single cell comprehensively considers predicted trajectory information, topological conservation constraints, and modal stability requirements, representing the ideal termination state for equalization control. Compared to the initial voltage vector, the target voltage distribution exhibits better topological consistency and global coordination, effectively guiding subsequent equalization path planning and control strategy generation.

[0066] The priority is determined based on the edge weights in the coupled topology graph. The intermediate voltage sequence is generated by reverse searching from the termination state to the initial state according to the priority, including: Calculate the voltage difference between each node in the final state and the initial state. Construct a backpropagation weight matrix based on the voltage difference and the weight of the edge in the coupled topology graph. Sort the backpropagation weight matrix by row to generate a back transfer priority sequence. Predecessor nodes are selected from the termination state according to the reverse transition priority sequence. The weight values ​​corresponding to the predecessor nodes and the termination state in the reverse propagation weight matrix are used as reverse coefficients. The reverse voltage values ​​of the predecessor nodes are calculated based on the reverse coefficients and the node voltage values ​​in the termination state. The reverse voltage values ​​of each predecessor node are used as the current search layer state. Calculate the deviation between the current search layer state and the voltage values ​​of each node in the initial state, correct the corresponding weight values ​​in the backpropagation weight matrix based on the deviation, and update the backpropagation priority sequence by sorting the corrected backpropagation weight matrix by row. Using the updated reverse transition priority sequence, the current search layer state is used as the starting point to continue selecting predecessor nodes and calculating the reverse voltage value. This process is repeated until the deviation converges. The intermediate voltage sequence is generated by arranging the voltage values ​​of each node in the state of each search level in the order of search level.

[0067] After obtaining the target voltage distribution and initial voltage data, it is necessary to construct a reverse search path from the target state to the initial state, generating intermediate voltage sequences at each time step to provide a precise voltage regulation trajectory for the subsequent equalization circuit control. The core of the reverse search lies in using the weight information of the edges in the coupled topology graph to determine the transition priority between nodes, and through iterative correction, ensuring that the generated intermediate voltage sequence not only conforms to the topology constraints but also smoothly transitions from the initial state to the target state.

[0068] First, calculate the voltage difference between each node in the final state and the initial state. The final state is the target voltage distribution obtained from the aforementioned optimization solution. The initial state is the actual voltage data collected. For any given node, the voltage difference reflects the voltage magnitude that needs to be adjusted from the initial state to the target state. The larger the difference, the more significant the equalization control required for that node.

[0069] After obtaining the voltage differences at each node, a backpropagation weight matrix is ​​constructed by combining the edge weights in the coupled topology graph. The construction of the backpropagation weight matrix needs to consider two factors: first, the original weights of the edges in the coupled topology graph; and second, the impact of the voltage differences on the propagation path. For each node... Pointing to node The directed edges, their backpropagation weights Calculated as ,in For nodes in the coupled topology graph Pointing to node The weight of the directed edge. Let be the voltage difference scale parameter. This formula shows that when the voltage differences between two nodes are similar, the backpropagation weight is larger, and these nodes are preferentially selected as the backpropagation search path; conversely, when the voltage differences are large, the backpropagation weight is smaller, and the selection priority is reduced.

[0070] The backpropagation weight matrix is ​​sorted row-wise to generate a back transition priority sequence. Specifically, for each node... Extract the first weight from the backpropagation weight matrix. Sort all non-zero elements in a row by their weight values ​​in descending order to obtain the node. Predecessor node priority list ,in For nodes The number of incoming edges, The predecessor node with the highest weight is selected. The priority list of predecessor nodes for all nodes is then compiled to form the reverse transition priority sequence.

[0071] The predecessor node is selected from the terminated state according to the reverse transfer priority sequence. The target voltage distribution is used as the basis for this selection. As the starting point for the reverse search, for each node Select the predecessor node with the highest weight from its predecessor node priority list. The weight values ​​corresponding to the predecessor node and the termination state in the backpropagation weight matrix are then used. As a backward coefficient The backward coefficient reflects the degree of decay during propagation from the terminal state to the predecessor node.

[0072] The reverse voltage value of the predecessor node is calculated based on the reverse coefficient and the node voltage value in the termination state. For node... Its reverse voltage value Calculated as This formula indicates that the reverse voltage value is obtained by weighting the voltage difference based on the target voltage value of the predecessor node using a reverse coefficient. When the reverse coefficient is large, the reverse voltage value is closer to the target voltage value of the predecessor node; when the reverse coefficient is small, the reverse voltage value is closer to the node's target voltage value. The target voltage value. The inverse voltage values ​​of each predecessor node are summarized to form the current search layer state. .

[0073] Calculate the deviation between the current search layer state and the voltage values ​​of each node in the initial state. For each node... Its deviation Calculated as The deviation reflects the distance between the current search layer state and the initial state; the smaller the deviation, the closer the reverse search is to the initial state.

[0074] The weight values ​​in the backpropagation weight matrix are adjusted based on the deviation. The purpose of this adjustment is to adjust the backpropagation path so that subsequent search layers can converge to the initial state more quickly. For nodes... predecessor node Its corrected backpropagation weights Calculated as ,in To correct the strength coefficient, This represents the maximum deviation of all nodes in the current search layer. This formula indicates that nodes with larger deviations receive a greater increase in their backpropagation weights, thus prioritizing the adjustment of their backpropagation voltage values ​​in subsequent searches.

[0075] The corrected backpropagation weight matrix is ​​sorted row-wise, and the backpropagation priority sequence is updated. The updated priority sequence reflects the deviation distribution between the current search layer state and the initial state, which can guide subsequent search layers to tilt towards regions with larger deviations, thus accelerating the convergence process.

[0076] Use the updated reverse transition priority sequence to change the current search layer state. Starting from this point, we continue to select predecessor nodes and calculate the reverse voltage values. For nodes... Select the predecessor node with the highest weight from the updated predecessor node priority list. Calculate the inverse coefficient And calculate the new reverse voltage value based on the reverse coefficient. The new reverse voltage values ​​of each node are summarized to form the state of the next search layer. .

[0077] Repeat the above iterative process, generating a new search layer state in each iteration and calculating the deviation between this search layer state and the initial state. The iteration terminates when the deviation converges, i.e., when the deviation of all nodes is less than a preset threshold. When, or when the rate of change of the deviation between two consecutive iterations is less than a preset threshold. Stop iteration when the time is reached. Assume iteration... If the convergence occurs subsequently, then co-generation will occur. Search layer state .

[0078] The voltage values ​​of each node in each search level state are arranged in search hierarchy order to generate an intermediate voltage sequence. The intermediate voltage sequence is a three-dimensional data structure: the first dimension is the node index, the second dimension is the search hierarchy index, and the third dimension is the voltage value. For each node... The intermediate voltage sequence is The sequence starts from the initial state, passes through each search layer state, and finally reaches the target state, forming a node. The complete control trajectory from the initial voltage to the target voltage.

[0079] To ensure the smoothness of the intermediate voltage sequence, interpolation can be performed on the voltage values ​​between adjacent search layers. For nodes... In the search layer and search layer Between, interpolated voltage values Calculated as ,in These are the interpolation parameters. By inserting several interpolation points between each search layer, a finer intermediate voltage sequence can be generated, providing higher time resolution for the precise control of the equalization circuit.

[0080] The generated intermediate voltage sequence not only contains the voltage regulation trajectory of each node, but also implicitly contains the transfer priority information between nodes. In the subsequent equalization circuit control, the conduction time and conduction current of each equalization path can be calculated based on the voltage difference between adjacent time steps in the intermediate voltage sequence and the weight of the edges in the coupled topology graph, thereby realizing adaptive equalization control of the battery module.

[0081] like Figure 2 The diagram shows the execution flowchart of the battery module equalization control scheme in this embodiment.

[0082] The conduction time and conduction current of each equalization path are calculated based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology graph. The equalization circuit in the battery module includes: Extract the voltage values ​​corresponding to each time step from the intermediate voltage sequence and calculate the voltage change rate at each node; The energy transfer intensity is obtained by multiplying the adjacency weight in the coupled topology graph by the voltage change rate. Node pairs with energy transfer intensity exceeding the preset intensity threshold are selected as node pairs to be balanced. The connection edges between the node pairs to be balanced are extracted as balancing paths. The adjacency weights corresponding to each balancing path are extracted as path weights. Construct a path admittance matrix based on the path weight of each equalization path, input the voltage change rate into the path admittance matrix to solve the current distribution coefficient of each equalization path, and sort the equalization paths according to the current distribution coefficient to generate an equalization path sequence. Extract the voltage values ​​of the nodes to be balanced corresponding to each equalization path in the intermediate voltage sequence at each time step according to the equalization path sequence, calculate the voltage difference at each time step, and calculate the instantaneous equalization current based on the voltage difference and the current distribution coefficient. The instantaneous equalization current of each equalization path is integrated over time until the charge transfer saturation is reached. The integration time is recorded as the conduction time, and the integral mean of the instantaneous equalization current is calculated as the conduction current. The equalization control command is generated based on the conduction time and conduction current of each equalization path to control the equalization circuit in the battery module.

[0083] After obtaining the intermediate voltage sequence, it needs to be converted into actual equalization control commands. The intermediate voltage sequence records the transition process from the initial state to the target voltage distribution, where each time step corresponds to a set of voltage values. To achieve precise equalization control, the voltage values ​​corresponding to each time step are first extracted from the intermediate voltage sequence. Assume the intermediate voltage sequence contains... Each time step, node In the The voltage value at each time step is denoted as ,in , This represents the index of a single cell in the battery module. The voltage change rate of each node can be obtained by calculating the voltage difference between adjacent time steps. Node At time step The voltage change rate is calculated as follows: This rate of voltage change reflects the speed and direction of voltage adjustment at the node during the equilibration process; a positive value indicates a voltage increase, and a negative value indicates a voltage decrease.

[0084] In a coupled topology graph, adjacency weights reflect the coupling strength between nodes, while voltage change rates reflect the dynamic adjustment needs of the nodes. Combining these two factors allows for the quantification of energy transfer intensity between node pairs. For node pairs with connecting edges in a coupled topology graph... Its energy transfer intensity is calculated as ,in Represents a node Pointing to node The adjacency weight, Represents nodes Adjacent node indices. A higher energy transfer intensity indicates a more drastic change in voltage difference between node pairs, requiring stronger equalization intervention. To avoid unnecessary equalization operations on minor voltage differences, a preset intensity threshold is set, and node pairs with energy transfer intensities exceeding this threshold are selected as nodes to be equalized. The connection edges between all node pairs to be equalized are extracted as equalization paths; these paths constitute the actual equalization circuit paths that need to be activated. Simultaneously, the adjacency weights corresponding to each equalization path are extracted as path weights, denoted as... .

[0085] After determining the balancing path, it is necessary to calculate the current distribution coefficient of each path to achieve a reasonable energy distribution. Assume there are a total of... The first balanced pathway, the first Equalized path connecting nodes and nodes Its pathway weight is ,in Construct the path admittance matrix based on the path weights of each equilibrium path. The diagonal elements of the path admittance matrix represent the first... The total coupling strength between the path and other paths, with off-diagonal elements representing the first path. The first pathway and the second The coupling relationship between the pathways, where This represents the path index. The voltage change rate of each node is constructed as a voltage change rate vector. , of which The elements are By solving the system of linear equations The current distribution coefficient vector can be obtained. , of which element Indicates the first The current allocation coefficients of each balancing path are determined. These coefficients reflect the proportion of current each path should handle during the balancing process; a larger value indicates that the path needs to transfer more charge. The balancing paths are then sorted in descending order according to their current allocation coefficients, generating a balancing path sequence. This sorted sequence ensures that paths with higher current demands are processed first, improving balancing efficiency.

[0086] After obtaining the equalization path sequence, it is necessary to calculate the instantaneous equalization current of each equalization path at each time step. Process each path sequentially according to the equalization path sequence. For the first... The equalization path is defined, and the voltage values ​​of the corresponding nodes to be equalized at each time step in the intermediate voltage sequence are extracted. and Calculate the voltage difference at each time step. The sign of the voltage difference determines the direction of current flow; a positive value indicates that current flows from the node. Flow to Node A negative value indicates reverse flow. The instantaneous equalization current is calculated based on the voltage difference and the current distribution coefficient. The formula for calculating the instantaneous equalization current is as follows: ,in This represents the equivalent resistance of the equalization circuit; this parameter is determined by the hardware characteristics of the equalization circuit. The instantaneous equalization current reflects the current at time step... Time The magnitude of the current that should be applied to the path.

[0087] To determine the conduction duration of each equalization path, the instantaneous equalization current needs to be integrated over time. The integration process starts from the time step... Initially, the cumulative charge transfer amount ,in This represents the time interval between adjacent time steps. This represents the time step index for the integral summation. When the charge transfer reaches saturation, i.e. Stop integrating and record the time step at that point. Charge transfer saturation value According to the node and nodes The initial voltage difference between the cells and the capacity of each individual cell are calculated, specifically as follows: ,in This indicates the rated capacity of a single battery cell. The integration time is the conduction time, calculated as follows: Simultaneously, the integral mean of the instantaneous equalization current over the conduction time is calculated as the conduction current, as shown in the formula: The conduction current represents the average current that the equalization circuit should apply over the entire conduction duration.

[0088] After calculating the conduction duration and conduction current of each balancing path, balancing control commands are generated. These commands include path identifiers, conduction durations, conduction currents, and current directions. The path identifier specifies the balancing circuit path to be activated, the conduction duration determines the operating time of the balancing circuit, the conduction current determines the operating intensity of the balancing circuit, and the current direction is determined by the sign of the voltage difference. These control commands are sent to the balancing circuit controller in the battery module. The controller activates the corresponding switching devices according to the commands, establishes the balancing path, and applies the specified current. During the balancing process, the voltage changes of each individual battery cell are monitored in real time. When the actual voltage approaches the target voltage distribution, the conduction current is gradually reduced or the conduction duration is shortened to avoid over-balancing leading to voltage overshoot. This balancing control method based on intermediate voltage sequences and coupled topology diagrams enables precise and efficient balancing of the battery module, extending battery life and improving system safety.

[0089] A second aspect of this invention provides a battery module adaptive balancing system based on intelligent algorithms, comprising: The data acquisition unit is used to collect voltage and temperature data of each individual cell in the battery module, construct a state trajectory matrix from the voltage and temperature data, and calculate the mutual information entropy between each individual cell in the state trajectory matrix. The topology building unit is used to construct a coupled topology graph based on mutual information entropy, identify the dominant node in the coupled topology graph, extract the state trajectory corresponding to the row of the dominant node from the state trajectory matrix, and propagate the extracted state trajectory along the edge of the coupled topology graph to other nodes to generate a set of predicted trajectories. An optimization unit is used to extract the endpoint voltage value of each predicted trajectory in the predicted trajectory set. It combines the adjacency weights of each node in the coupled topology graph to construct an optimization function with topological constraints, and solves the optimization function to obtain the target voltage distribution. The reverse search unit is used to take the target voltage distribution as the termination state and the voltage data as the initial state, determine the priority according to the weight of the edges in the coupled topology graph, and reverse search from the termination state to the initial state according to the priority to generate an intermediate voltage sequence. The equalization control unit is used to calculate the conduction time and conduction current of each equalization path based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology graph, and to control the equalization circuit in the battery module.

[0090] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0091] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0092] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A battery module adaptive equalization method based on intelligent algorithm, characterized in that, include: Collect voltage and temperature data of each individual cell in the battery module, construct a state trajectory matrix from the voltage and temperature data, and calculate the mutual information entropy between each individual cell in the state trajectory matrix. A coupled topology graph is constructed based on mutual information entropy. Dominant nodes are identified in the coupled topology graph. The state trajectories corresponding to the rows of dominant nodes are extracted from the state trajectory matrix. The extracted state trajectories are propagated to other nodes along the edges of the coupled topology graph to generate a set of predicted trajectories. Extract the endpoint voltage value of each predicted trajectory from the predicted trajectory set, construct an optimization function with topological constraints by combining the adjacency weights of each node in the coupled topology graph, and solve the optimization function to obtain the target voltage distribution; The target voltage distribution is taken as the termination state, and the voltage data is taken as the initial state. The priority is determined according to the weight of the edges in the coupled topology graph. The intermediate voltage sequence is generated by searching backward from the termination state to the initial state according to the priority. The conduction time and conduction current of each equalization path are calculated based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology diagram, thereby controlling the equalization circuit in the battery module.

2. The method according to claim 1, characterized in that, Voltage and temperature data are used to construct a state trajectory matrix. The mutual information entropy between individual cells in the state trajectory matrix is ​​calculated, including: The voltage and temperature data collected from each individual cell at continuous time are used to form coordinate pairs. The coordinate pairs of each individual cell are arranged in chronological order to form a state trajectory. The state trajectories of each individual battery cell are stacked row by row to construct a state trajectory matrix; Divide the voltage-temperature coordinate plane into grid regions, map the coordinate pairs of each row of state trajectories in the state trajectory matrix to the grid regions, record the grid region number to which each coordinate pair belongs, and arrange the grid region numbers of each row of state trajectories in chronological order to form a grid number sequence. Extract the grid number sequence corresponding to any two rows from the state trajectory matrix, count the number combinations that appear at the same time for the two grid number sequences, and calculate the mutual information entropy between each individual cell based on the joint distribution characteristics of the number combinations and the marginal distribution characteristics of the individual number sequences.

3. The method according to claim 1, characterized in that, A coupled topology graph is constructed based on mutual information entropy. Dominant nodes are identified within the coupled topology graph, and the state trajectories corresponding to these dominant nodes are propagated along the edges of the coupled topology graph to other nodes, generating a set of predicted trajectories, including: Each individual battery cell is mapped to a node, and the mutual information entropy is used as the weight of the directed edge to construct a coupled topology graph; Traverse the nodes in the coupled topology graph, count the sum of the weights of the incoming edges and the sum of the weights of the outgoing edges of each node, calculate the difference between the sum of the weights of the incoming edges and the sum of the weights of the outgoing edges as an asymmetry index, select the node with the largest absolute value of the asymmetry index as the dominant node, and extract the state trajectory of the row corresponding to the dominant node from the state trajectory matrix as the seed trajectory. Extract the weights of directed edges from the dominant node to the adjacent node in the coupled topology graph. Perform a decay transformation on the seed trajectory based on the weights of the directed edges to generate the propagation trajectory. Extract the state trajectory of the corresponding row of the adjacent node from the state trajectory matrix. Calculate the spatial distance deviation between the propagation trajectory and the state trajectory of the adjacent node. Correct the propagation trajectory based on the spatial distance deviation to generate the corrected trajectory. Calculate the cumulative deviation between the corrected trajectory and the state trajectories of adjacent nodes. Based on the cumulative deviation, update the weights of the directed edges from the dominant node to the adjacent nodes in the coupled topology graph. Use the updated directed edge weights to perform a decay transformation on the corrected trajectory to generate the predicted trajectory. Summarize the predicted trajectories of each adjacent node to form a set of predicted trajectories.

4. The method according to claim 3, characterized in that, Extract the directed edge weights from the dominant node to its adjacent nodes in the coupled topology graph. Perform a decay transformation on the seed trajectory based on these weights to generate the propagation trajectory. Extract the state trajectories corresponding to the adjacent nodes from the state trajectory matrix. Calculate the spatial distance deviation between the propagation trajectory and the adjacent node state trajectories. Correct the propagation trajectory based on this spatial distance deviation to generate the corrected trajectory, including: Search for multiple propagation paths from the dominant node to each neighboring node in the coupled topology graph, extract the weights of the directed edges connected in series on each propagation path and multiply them to generate a path propagation factor, and apply a decay transformation to the seed trajectory based on the path propagation factor to generate multiple candidate propagation trajectories. Extract the state trajectory of each adjacent node from the state trajectory matrix, match the state trajectory of each adjacent node with the corresponding multiple candidate propagation trajectories in the voltage-temperature coordinate space, calculate the trajectory similarity between each candidate propagation trajectory and the state trajectory of the adjacent node, select the candidate propagation trajectory with the highest trajectory similarity as the propagation trajectory, and record the path propagation factor corresponding to the propagation trajectory. Calculate the Euclidean distance between the propagation trajectory and the state trajectory of adjacent nodes at each time point in the voltage-temperature coordinate space, accumulate the Euclidean distances at each time point to obtain the spatial distance deviation, and calculate the correction weight based on the spatial distance deviation and the path propagation factor. Based on the spatial distance deviation, a time-by-time deviation vector is constructed from the propagation trajectory to the state trajectory of the adjacent node. The deviation vector is then scaled according to the correction weight to generate a correction vector. The propagation trajectory is then superimposed along the direction of the correction vector to generate the correction trajectory.

5. The method according to claim 1, characterized in that, The endpoint voltage values ​​of each predicted trajectory in the predicted trajectory set are extracted. An optimization function with topological constraints is constructed by combining the adjacency weights of each node in the coupled topology graph. Solving the optimization function yields the target voltage distribution, including: Extract the endpoint voltage value corresponding to the end of the time series for each predicted trajectory from the predicted trajectory set, and construct an initial voltage vector by arranging the endpoint voltage values ​​according to the node index. The adjacency matrix formed by the adjacency weights of each node in the coupled topology graph is decomposed into a topological eigenspace. The initial voltage vector is then mapped to the topological eigenspace to obtain the modal characterization vector. The degenerate adjacency matrix is ​​obtained by filtering and removing edges whose adjacency weight values ​​are less than a preset weight threshold from the adjacency matrix. The degenerate eigenspace is obtained by performing spectral decomposition on the degenerate adjacency matrix. The degenerate mode vector is then mapped to the degenerate eigenspace to obtain the degenerate mode vector. The modal characterization vector is inversely mapped to the node space to obtain the reconstructed voltage vector. The deviation vector between the reconstructed voltage vector and the initial voltage vector is calculated. The conservation deviation vector of the voltage values ​​of adjacent nodes in the reconstructed voltage vector is calculated according to the adjacency weight. The magnitude change of the degenerate mode vector is extracted. The deviation vector, the conserved deviation vector, and the magnitude change are used as constraints to construct an optimization function with topological constraints. The optimization function is iteratively solved to update the modal representation vector until convergence is obtained to obtain the optimal modal representation vector. The optimal modal representation vector is then inversely mapped to the node space to obtain the target voltage distribution.

6. The method according to claim 1, characterized in that, The priority is determined based on the edge weights in the coupled topology graph. The intermediate voltage sequence is generated by reverse searching from the termination state to the initial state according to the priority, including: Calculate the voltage difference between each node in the final state and the initial state. Construct a backpropagation weight matrix based on the voltage difference and the weight of the edge in the coupled topology graph. Sort the backpropagation weight matrix by row to generate a back transfer priority sequence. Predecessor nodes are selected from the termination state according to the reverse transition priority sequence. The weight values ​​corresponding to the predecessor nodes and the termination state in the reverse propagation weight matrix are used as reverse coefficients. The reverse voltage values ​​of the predecessor nodes are calculated based on the reverse coefficients and the node voltage values ​​in the termination state. The reverse voltage values ​​of each predecessor node are used as the current search layer state. Calculate the deviation between the current search layer state and the voltage values ​​of each node in the initial state, correct the corresponding weight values ​​in the backpropagation weight matrix based on the deviation, and update the backpropagation priority sequence by sorting the corrected backpropagation weight matrix by row. Using the updated reverse transition priority sequence, the current search layer state is used as the starting point to continue selecting predecessor nodes and calculating the reverse voltage value. This process is repeated until the deviation converges. The intermediate voltage sequence is generated by arranging the voltage values ​​of each node in the state of each search level in the order of search level.

7. The method according to claim 1, characterized in that, The conduction time and conduction current of each equalization path are calculated based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology graph. The equalization circuit in the battery module includes: Extract the voltage values ​​corresponding to each time step from the intermediate voltage sequence and calculate the voltage change rate at each node; The energy transfer intensity is obtained by multiplying the adjacency weight in the coupled topology graph by the voltage change rate. Node pairs with energy transfer intensity exceeding the preset intensity threshold are selected as node pairs to be balanced. The connection edges between the node pairs to be balanced are extracted as balancing paths. The adjacency weights corresponding to each balancing path are extracted as path weights. Construct a path admittance matrix based on the path weight of each equalization path, input the voltage change rate into the path admittance matrix to solve the current distribution coefficient of each equalization path, and sort the equalization paths according to the current distribution coefficient to generate an equalization path sequence. Extract the voltage values ​​of the nodes to be balanced corresponding to each equalization path in the intermediate voltage sequence at each time step according to the equalization path sequence, calculate the voltage difference at each time step, and calculate the instantaneous equalization current based on the voltage difference and the current distribution coefficient. The instantaneous equalization current of each equalization path is integrated over time until the charge transfer saturation is reached. The integration time is recorded as the conduction time, and the integral mean of the instantaneous equalization current is calculated as the conduction current. The equalization control command is generated based on the conduction time and conduction current of each equalization path to control the equalization circuit in the battery module.

8. A battery module adaptive balancing system based on intelligent algorithms, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition unit is used to collect voltage and temperature data of each individual cell in the battery module, construct a state trajectory matrix from the voltage and temperature data, and calculate the mutual information entropy between each individual cell in the state trajectory matrix. The topology building unit is used to construct a coupled topology graph based on mutual information entropy, identify the dominant node in the coupled topology graph, extract the state trajectory corresponding to the row of the dominant node from the state trajectory matrix, and propagate the extracted state trajectory along the edge of the coupled topology graph to other nodes to generate a set of predicted trajectories. An optimization unit is used to extract the endpoint voltage value of each predicted trajectory in the predicted trajectory set. It combines the adjacency weights of each node in the coupled topology graph to construct an optimization function with topological constraints, and solves the optimization function to obtain the target voltage distribution. The reverse search unit is used to take the target voltage distribution as the termination state and the voltage data as the initial state, determine the priority according to the weight of the edges in the coupled topology graph, and reverse search from the termination state to the initial state according to the priority to generate an intermediate voltage sequence. The equalization control unit is used to calculate the conduction time and conduction current of each equalization path based on the voltage value corresponding to each time step in the intermediate voltage sequence and the adjacency weight of the corresponding node in the coupled topology graph, and to control the equalization circuit in the battery module.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images