Intelligent energy distribution system and method for multiple battery modules

Abstract

The invention belongs to the technical field of battery distribution management, and provides an intelligent energy distribution system and method for a multi-battery module, and the method comprises the steps: constructing a battery cell network topological graph according to the internal battery cells of a battery pack and the connection relation between the battery cells; according to the cell network topological graph, real-time voltage, current and temperature signals of part of cells in each battery module are obtained, and part of sampling observation data is formed; on the basis of a graph signal processing method, mapping part of sampling observation data into graph signals, estimating the state of an unsampled battery cell by adopting a graph Laplacian regularization method, detecting abnormal nodes in a graph on the basis of a state estimation value of the battery cell, and correspondingly identifying an abnormal battery cell; constructing an energy distribution model according to the state estimation value data of each battery cell, which is estimated by graph signal processing; risks caused by reduction of sampling wire harnesses and reduction of monitoring precision under a CTP architecture are effectively relieved, the cell state estimation precision and the energy distribution stability are improved, and the overall safe life of the battery pack is prolonged.

Description

Technical Field

[0001] This invention belongs to the field of battery distribution management technology, specifically an intelligent energy distribution system and method for multiple battery modules. Background Technology

[0002] With the rapid development of new energy vehicles and energy storage systems, the safety, lifespan consistency and energy utilization efficiency of lithium-ion batteries have become key factors limiting the improvement of system performance. In order to improve energy density and reduce system costs, the industry has gradually adopted highly integrated battery structures represented by CTP (Cell to Pack), which directly integrates the cells into the battery pack, eliminating the traditional module level, thereby significantly improving space utilization and manufacturing efficiency. However, while the CTP structure brings high energy density, it also brings new management and control problems. Traditional battery management systems (BMS) usually monitor and control on a module-by-module basis, and can directly obtain the voltage, current and temperature information of individual cells in each module. When the module level is eliminated, the system can often only obtain comprehensive measurements of several "groups" or "regions", such as group voltage, bus current or average temperature. This means that it is no longer possible to identify the state of each cell individually, and can only obtain aggregated information of a certain region. In this situation, during the intelligent energy distribution process of multiple battery modules, since the observed data is a regional composite quantity, it is impossible to directly determine which cell has an abnormal charge or health status. Furthermore, when there are aging differences or uneven internal resistance among cells, the group average voltage or SOH index may still remain within the normal range, masking the severe degradation of individual cells. This average smoothing effect will cause the system to misjudge battery health, which in turn will cause the intelligent energy distribution to lose its accurate data foundation, resulting in distorted energy distribution decisions. Therefore, the present invention provides an intelligent energy distribution system and method for multi-battery modules. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0004] The technical solution adopted by this invention to solve its technical problem is: a smart energy distribution method for multi-battery modules, comprising: Construct a cell network topology diagram based on the cells inside the battery pack and the connections between the cells; Based on the cell network topology diagram, real-time voltage, current and temperature signals of some cells in each battery module are obtained to form partial sampling observation data; Based on graph signal processing, some sampled observation data are mapped to graph signals. The graph Laplace regularization method is used to estimate the state of unsampled cells, and abnormal nodes in the graph are detected based on the state estimates of the cells, thus identifying the presence of abnormal cells. An energy allocation model is constructed based on the estimated state values ​​of each cell obtained through signal processing. The energy allocation model takes maximizing the energy utilization rate of the entire battery pack as the objective function and constrains each cell node in the cell network topology diagram to obtain the optimal energy allocation strategy for each battery module. Based on the output-based energy allocation strategy, a reverse verification mechanism is introduced in the energy allocation stage to dynamically evaluate the reliability of the sampled data according to the actual battery response and adaptively correct low reliability observation nodes. A smart energy distribution system for multi-battery modules, the system comprising: Cell diagram construction module: Constructs a cell network topology diagram based on the cells inside the battery pack and the connections between the cells; Observation data acquisition module: Based on the cell network topology diagram, it acquires the real-time voltage, current and temperature signals of some cells in each battery module to form partial sampling observation data; Graph signal processing module: Based on graph signal processing methods, it maps some sampled observation data into graph signals, uses graph Laplace regularization method to estimate the state of unsampled cells, and detects abnormal nodes in the graph based on the state estimate of the cells, and identifies the existence of abnormal cells. The energy allocation strategy output module constructs an energy allocation model based on the estimated state values ​​of each cell obtained from graph signal processing. The energy allocation model takes maximizing the energy utilization rate of the entire battery pack as the objective function and constrains each cell node in the cell network topology diagram to obtain the optimal energy allocation strategy for each battery module. Reverse verification module: Based on the output energy allocation strategy, a reverse verification mechanism is introduced in the energy allocation stage. The reliability of the sampled data is dynamically evaluated based on the actual battery response, and low reliability observation nodes are adaptively corrected.

[0005] The beneficial effects of this invention are as follows: This invention models the battery cell and its electrical, thermal, and aging coupling relationships as a network topology graph. By using graph Laplace regularization to reconstruct signals and detect abnormal nodes in some sampled data, it is possible to accurately estimate the state of unsampled battery cells under sampling constraints. This invention introduces node confidence into the energy allocation strategy and dynamically adjusts the power allocation constraints based on the uncertainty of cell state estimation, thereby achieving risk-adaptive energy allocation optimization. This invention introduces reverse verification in the energy allocation execution phase. By comparing the deviation between the predicted value of the allocation strategy and the actual response, the sampling observation data is dynamically evaluated and corrected. When sensor drift or observation anomaly is detected, the system automatically adjusts the weight of the graph signal model and re-estimates the state, realizing closed-loop self-correction of data sampling, state estimation and control scheduling. This invention effectively mitigates the risks caused by the reduction in sampling harnesses and decreased monitoring accuracy under the CTP architecture, and improves the accuracy of cell status estimation, energy distribution stability and overall safe life of the battery pack. Attached Figure Description

[0006] The invention will now be further described with reference to the accompanying drawings.

[0007] Figure 1 This is a flowchart of the steps of an intelligent energy distribution method for multi-battery modules according to the present invention; Figure 2 This is a partial flowchart of the intelligent energy distribution method for multi-battery modules according to the present invention. Figure 3 This is a module architecture diagram of an intelligent energy distribution system for multi-battery modules according to the present invention. Detailed Implementation

[0008] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0009] Example 1: Please see Figure 1 and Figure 2 As shown in the embodiments of the present invention, an intelligent energy distribution method for multi-battery modules addresses the problems of decreased cell monitoring accuracy, incomplete sampling data, and difficulty in anomaly identification in the intelligent energy distribution of multi-battery modules under the CTP architecture. It proposes a cell state estimation and energy distribution method based on graph signal processing. By modeling the cell and its electrical, thermal, and aging coupling relationships as a network topology graph, graph Laplace regularization is used to reconstruct signals and detect abnormal nodes in partially sampled observation data, compensating for the state information of unsampled cells. Then, an energy distribution optimization model is established by combining the cell node confidence, and a reverse verification mechanism is introduced to assess the reliability of the sampled observation data. This maximizes the overall energy utilization of the battery pack and provides adaptive power constraints for abnormal cells, improving the accuracy of cell state estimation and the stability of energy distribution under limited sampling conditions, and optimizing the overall safety and lifespan of the CTP battery pack. The method includes the following steps: Step S10: Construct a cell network topology diagram based on the cells inside the battery pack and the connection relationships between the cells; Specifically, obtain all the cells inside the battery pack and treat each cell in the battery pack as a node; Based on the electrical connections, thermal coupling relationships, and aging-related factors between battery cells, the edges between nodes in the graph are defined, specifically including: Electrical connection edge: If two cells form a current path through a direct physical conductive structure (such as a copper busbar, connecting piece, or solder joint), regardless of whether they are in series, parallel, or mixed connection, an electrical connection edge is established between the corresponding nodes. The weight of the edge can be based on the reciprocal of the connection resistance. Thermal coupling edge: Based on the physical structure of the battery pack and thermal simulation or measured data, if the geometric center distance between two cells is less than the set thermal coupling threshold, or if they are placed together in an independent liquid cooling plate channel, air channel or heat insulation zone, then it is considered that there is a significant thermal conduction relationship between them, and a thermal coupling edge is established. The weight of the edge can be assigned according to the reciprocal of the cell spacing. Aging-related edge: Based on the historical operating data of the battery cell (such as cycle count, cumulative throughput, historical temperature distribution) or initial performance parameters (such as initial internal resistance, capacity), calculate the similarity of aging behavior between any two battery cells (for example, use Pearson correlation coefficient to analyze capacity decay trajectory). If the similarity exceeds the preset correlation coefficient threshold, an aging-related edge is established between the corresponding nodes, and the weight can be directly assigned to the correlation coefficient. Based on the edge relationships between nodes in the graph defined above, construct the adjacency matrix A of the graph, where the matrix elements are... Represents a node With nodes The connection strength between them, i and j represent the cell number index, which can be set to a binary value (0 or 1) or a weighted value according to the actual system characteristics; Based on the adjacency matrix A, the degree matrix D of the graph is calculated (a diagonal matrix, where the diagonal elements are the degrees of the corresponding nodes), and then the graph Laplacian matrix L is constructed: L=DA; the Laplacian matrix is ​​set up for regularization and smoothing constraints in subsequent graph signal processing; The mathematical representation of the battery cell network topology is output, including the adjacency matrix A or the Laplace matrix L. The constructed battery cell network topology structure data (including node set, edge set, adjacency matrix, and Laplace matrix) is stored in system memory or database for subsequent graph signal processing and state estimation modules to call.

[0010] The construction of the cell network topology in step S10 ensures the local smoothness and global correlation of the cell state, facilitates signal reconstruction under partial sampling conditions, captures complex interactions within the battery pack, and provides a data foundation for state estimation and energy allocation optimization based on graph signals.

[0011] Step S20: Based on the cell network topology diagram, obtain the real-time voltage, current and temperature signals of some cells in each battery module to form partial sampling observation data; Specifically, based on the cell network topology, representative sampling nodes covered by sensors are set up, and a set of cells is identified as the target cells for data acquisition. Preferably, for sampling nodes, the following nodes are selected, but not limited to: Structurally critical nodes (based on network centrality): Select cells that occupy structurally critical positions in the topology graph. Specifically, this can be identified by calculating and ranking the following indicators: Nodes with high degree centrality: Cells with a large number of connected edges, whose state changes will directly affect more neighbors; Nodes with high eigenvector centrality are not only highly connected to themselves, but also have many highly interconnected neighbors. These nodes are the core of the network and can reflect the overall behavior of the cluster. Performance vulnerability nodes (based on historical data): cells with abnormal historical records and faster performance degradation, to enable predictive maintenance and risk control; Historical parameter anomalies: Prioritize cells that have historically experienced voltage jumps, persistently high temperatures, or internal resistance growth rates higher than average. Aging trajectory deviation: Based on aging-related edges, prioritize cells with low confidence in the estimated health status or whose aging behavior deviates (is inconsistent) from that of connected cells. Spatial representative nodes (based on physical layout): To ensure that the sampling can capture the environmental gradient inside the battery pack, the sampling points should be evenly distributed in the physical space to cover key areas; Thermal boundary areas: These include cells located at the corners and periphery of the battery pack. These locations are usually more susceptible to ambient temperature and are weak points in thermal management. Core area: This includes the battery cells at the geometric center of the battery pack. Heat tends to accumulate in this area, forming high-temperature hotspots. For example, if the battery module has 16 cells arranged in a 4x4 grid; By calculating centrality, it may be found that the degree centrality and eigenvector centrality of several cells located in the middle position (such as cells 6, 7, 10, and 11) are the highest because they are connected to the most neighbors. A review of historical databases revealed that the internal resistance of cell number 1 in the corner was too high, and the capacity decay curve of cell number 16 deviated from that of similar cells. To cover the space, you need to select: Corner: No. 1, 4, 13, 16; Center: A portion of numbers 6, 7, 10, and 11; The final sample set could be: 1, 4, 6, 7, 10, 11, 13, 16; Deploy voltage sensors, current sensors, and temperature sensors on or near the identified target battery cell for data acquisition. Optionally, the voltage sensor can be an integrated battery monitoring chip, the current sensor can be a Hall sensor or a shunt, and the temperature sensor can be a thermocouple or an NTC thermistor. Synchronous data acquisition: Real-time signals of the target battery cell are collected, including voltage, current, and temperature signals. Voltage signal: Measure the terminal voltage of each target battery cell for data acquisition; Current signal: Since the total bus current is usually measured in CTP architecture, the current flowing through the critical path can be estimated by using a shunt or by combining the cell connection relationship; Temperature signal: Measures the surface temperature of each target battery cell for data acquisition; It should be noted that when acquiring current signals, the critical path in the associated path refers to those paths in the electrical connection topology of the battery pack that can represent or determine the main current flow direction. Specifically: For series paths: In a branch consisting of multiple cells connected in series, since the current is the same everywhere, any path containing any cell can be regarded as a critical path. For parallel paths: When multiple (series) branches are connected in parallel, the total current is distributed among the different branches. The critical path refers to a specific branch that is selected to characterize the current of the branch in which it is located. Therefore, the critical path is a representative current path selected based on the electrical topology and measurement strategy to indirectly extrapolate the global current distribution; The acquired real-time signals are preprocessed, specifically including filtering and normalization. The preprocessed data is then organized into structured partial sampled observation data, such as vectors or matrices, where each element corresponds to the voltage, current, and temperature values ​​of a sampled battery cell.

[0012] Step S20 selectively selects sampling nodes to cover structurally critical, performance-vulnerable, and spatially representative cells, ensuring that the data reflects the overall state of the battery pack and the characteristics of key cells. With limited hardware resources, it efficiently acquires data that reflects the overall operating state of the battery pack, providing data support for subsequent state estimation and anomaly detection based on graph signal processing.

[0013] Step S30: Based on the graph signal processing method, the sampled observation data is mapped into a graph signal. The graph Laplace regularization method is used to estimate the state of the unsampled cells. Based on the state estimate of the cells, abnormal nodes in the graph are detected, and abnormal cells are identified accordingly. It is worth explaining why graph signal processing was chosen: Since the battery pack under the CTP architecture is a natural graph structure system, each cell forms a complex coupling relationship through electrical connection, thermal conduction, and aging characteristics. The cell is not an independent individual, but an energy network with topological connections. In this case, the traditional cell-by-cell sampling and independent estimation methods cannot work effectively because the observation data is incomplete and the signals have strong correlations. By introducing the Graph Signal Processing (GSP) method, the cell state can be regarded as a "signal on a graph". The sampled cells are used as observation nodes, and the unsampled cell states are interpolated and reconstructed through graph Laplace regularization or graph filtering, thereby recovering the state distribution of the entire package under limited sampling conditions. At the same time, GSP can use the local smoothness feature of the graph to identify abnormal nodes, that is, to discover cells with abnormal behavior. This not only improves the accuracy of cell state estimation, but also provides a complete and reliable input basis for subsequent intelligent energy allocation, realizing adaptive power constraints and risk suppression for abnormal cells. Specifically, some sampled observation data are mapped onto the cell network topology to form a graph signal; Specifically, a graphical signal vector is constructed for each physical quantity (voltage, current, temperature), where: For cell nodes where sensors have been installed and sampling has been successfully completed, the signal value is the measured value; For unsampled cell nodes, the signal value is initialized to the missing value; The state estimation based on graph Laplacian regularization is as follows: The main approach utilizes the assumption of smoothness in the signal on the graph to fill in the state values ​​of all unsampled nodes through optimization methods. The physical quantity to be estimated (e.g., voltage) is defined as a graphical signal, represented as an N-dimensional column vector x. Where N is the total number of battery cells, and x is the estimated state vector; In N battery cells, M cells are covered by the sensor and successfully sampled. The M observations form an M-dimensional column vector y, where y is the observation signal vector. Define an M×N sampling matrix C, where each row of the matrix has only one element that is 1 and the rest are 0, to select the M sampled cell states from the full signal x; For example, if there are a total of 4 cells (N=4), and only the 1st and 3rd cells are sampled (M=2), then: , ; Construct the cost function: ,in, Here, Cx-y represents the data fitting term, and Cx-y is the residual vector between the estimated and measured values. The graph smoothness regularization term quantitatively describes the total smoothness (or total variation) of the graph signal x across the entire cell network. The regularization parameter was determined experimentally based on historical data and system performance using techniques such as cross-validation. This is the optimal estimation vector for the cell state; It should be noted that the core idea of ​​the cost function is to find the "best guess" for the state of all cells in the entire battery pack. This conjecture must be able to explain the data we actually measured well (as guaranteed by the data fitting term) and also conform to the inherent physical correlation of the cell network, that is, the connected cells are in similar states (as guaranteed by the graph smoothness regularization term). The aforementioned cost function is a convex quadratic function with respect to X. The global optimal solution can be obtained by setting its derivative to zero. By taking the derivative of the cost function and setting it to zero, a deterministic linear equation system is obtained. ; ; ; in, It is an N×N diagonal matrix, whose diagonal elements are 1 (corresponding to the sampled cell) or 0 (corresponding to the unsampled cell). The linear equations are well-state and can be solved efficiently using standard numerical methods (such as the conjugate gradient method and Cholesky decomposition) to obtain state estimates for all N cells. Anomaly detection is performed based on the calculated state estimates of N cells. For a cell node that is successfully sampled, the residual between the observed value and the state estimate is calculated. If the residual of a successfully sampled cell node is greater than a preset residual threshold, the observation data of that node is considered to be unreliable, and the corresponding cell node is marked as an abnormal observation node. If the residual of a successfully sampled cell node is less than or equal to a preset residual threshold, the observations of that node are considered reliable, and the corresponding cell node is marked as a normal observation node. Among them, the residual threshold is set by statistically analyzing a large amount of historical residual data under the known normal operating conditions of the battery system, and selecting the high quantile of the statistical distribution (such as the 99.5th percentile) as the residual threshold. The residual threshold represents the maximum reasonable deviation between the observed value and the model estimate under fault-free conditions. Any residual exceeding this threshold is judged as a statistically significant anomaly. For all cell nodes, calculate the difference between the state estimate and the mean of the state estimates of all adjacent cell nodes, and take the absolute value as the local consistency value; If the local consistency value is greater than the local consistency threshold, the corresponding cell is considered to be in an abnormal state and is marked as an abnormal node, i.e., an abnormal cell. If the local consistency value is less than or equal to the local consistency threshold, it is marked as a normal node; The setting of the local consistency threshold is mainly based on the statistical distribution of the differences between the states (such as voltage and temperature) of all cells and their neighboring cells under normal historical operating conditions of the battery pack. Specifically, it is achieved by analyzing the long-term records of the difference between the state value of each cell and the average state value of all its directly connected neighbors in the historical data, calculating the average value μ and standard deviation σ of all differences, and setting the threshold to μ+nσ (e.g., n=2), thereby ensuring that under the assumption of normal distribution, the vast majority (e.g., 95%) of normal fluctuations will not be misjudged as abnormal.

[0014] Step S30 maps limited partial sampled observation data onto the cell network topology graph and estimates the complete state (voltage, temperature, etc.) of all cells (including unsampled cells) by solving an optimization problem based on graph Laplace regularization, thereby optimizing the insufficient hardware sampling capability. Anomalies are identified through a dual detection mechanism. By comparing the residuals between the observed and estimated values ​​at sampling points, potential faulty sensors or abnormal data can be identified. By checking the local consistency between the state of each cell and the states of its neighbors, abnormal cells with severely deviated performance can be discovered, reducing the masking effect of the herd average.

[0015] Step S40: Construct an energy allocation model based on the estimated state values ​​of each cell obtained from graph signal processing; The energy allocation model takes maximizing the energy utilization rate of the entire battery pack as the objective function and constrains each cell node in the cell network topology diagram to obtain the optimal energy allocation strategy for each battery module. Specifically, the power allocation of each battery module is defined as the optimization variable P. Where K is the total number of battery modules; With the goal of maximizing the total energy utilization of the entire battery pack, an energy allocation model (objective function) is constructed. A typical implementation maximizes the total available energy while minimizing heat loss due to internal resistance and connection losses. The objective function can be: ; in, It is the overall efficiency coefficient of the k-th module. It is the equivalent internal resistance of the k-th module. These are weighting coefficients used to balance energy output and loss; in, It is a weighting coefficient, and the basis for setting it is: the initial value of γ can be set as the reciprocal of the rated power of the system. Based on the initial value, offline optimization is carried out in combination with the historical operating data of the battery pack (especially the temperature rise data under different power). It is usually transformed into a multi-objective optimization problem to find an optimal γ so that under typical operating conditions, after power allocation, the energy efficiency and the maximum temperature / temperature difference of the battery pack can reach an ideal balance. It is usually accomplished through techniques such as cross-validation. Set constraints, including system power requirement conventions, module physical limit constraints, and adaptive constraints based on cell node confidence. System power requirement constraint: The sum of the power allocated to each module must meet the total power requirement of the entire vehicle or load; Module physical limit constraints: The power allocated to each module cannot exceed its own charging and discharging power limit range; Adaptive constraints based on cell node confidence: Based on the confidence of abnormal nodes and state estimation output in the aforementioned step S30, more stringent power constraints are applied to modules from different modules or modules containing abnormal cells. For modules containing faulty cells (i.e., cells that are faulty themselves), reduce the power limit (especially during discharge), or directly constrain the power to near 0 (i.e., isolate it) to prevent the fault from spreading. For modules composed of cells with low confidence in state estimation (e.g., most of which are not sampled and are located at the network edge), a relatively conservative power limit is adopted to avoid the risks caused by uncertainty. Conversely, modules composed of cells with high confidence in state estimation are allowed to operate over a wider power range, fully utilizing their performance. It should be noted that the confidence level for state estimation is used to represent the level of confidence in each cell state value estimated through graph signal processing. High confidence level: This means there is a strong degree of confidence that the estimated value of the battery cell is close to the actual state. Low confidence level: This means that the estimated value of the battery cell has a high degree of uncertainty and may deviate significantly from the actual state. The confidence level of the state estimate is calculated by multiplying the proximity factor and the reliability factor. The proximity factor is set as follows: if a node is marked as a normal observation node, the proximity factor is set to 1.0; if a node is marked as an abnormal observation node, the proximity factor is set to 0.1. For unsampled cell nodes, in the cell network topology graph, find the three cells that are closest to the unsampled node and have been sampled and marked as normal observation nodes. Calculate the average topological distance between the unsampled node and these three normal observation nodes, and use 1 and the reciprocal of the sum of the average topological distances as the proximity factor. The reliability factor is as follows: if a node is marked as an abnormal observation node, the reliability factor is set to 0.1; if a node is marked as a normal observation node, the difference between the maximum permissible residual and the residual of the normal observation node is calculated, and then the ratio of the maximum permissible residual is used to calculate the reliability factor. For unsampled cell nodes, calculate the average residual of all directly adjacent cells marked as normal observation nodes. Then calculate the difference between the maximum allowable residual and the average residual of the cells, and use the ratio of the maximum allowable residual to the reliability factor. If there are no normal observation nodes as direct neighbors, the reliability factor is set to 0.1. Efficient numerical optimization algorithms (such as interior point method and quadratic programming solver) are used to solve the energy allocation model, and the power allocation vector of each module that satisfies all constraints and optimizes the objective function is calculated.

[0016] Step S40 establishes a mathematical optimization model with the goal of maximizing the overall energy utilization rate of the battery pack. This elevates the scheduling objective from the local performance of a single module or cell to the system-level performance of the entire battery pack, thereby improving the overall energy efficiency of the system. The key perception results, such as the list of abnormal cells and the confidence level of the state estimation, output in step S30 are directly used as adaptive constraints in the optimization model. This enables the energy allocation strategy to proactively avoid risks, achieve power limitation or isolation of abnormal cells, and conservative scheduling of low-confidence modules, thus improving the safety and reliability of the system.

[0017] Step S50: Based on the output energy allocation strategy, a reverse verification mechanism is introduced in the energy allocation stage to dynamically evaluate the reliability of the sampled data according to the actual battery response and adaptively correct the low reliability observation nodes. Specifically, based on the energy distribution strategy, the system performs charge and discharge control on the battery pack. Within a fixed time window after the strategy is executed, the system synchronously collects real-time data from all sampled cells to form an actual response value dataset. The actual response data includes: actual voltage response value, actual temperature response value, and actual current response value. The real-time response value is compared with the predicted value of the energy distribution model, and the prediction error and residual of each battery module are calculated. If the actual response value of a battery module deviates from the predicted value by more than the set tolerance threshold, it is determined that the sampling observation nodes contained in the module may have data anomalies or sensor drift, that is, they contain observation nodes with low reliability. The response tolerance threshold is set based on the statistical distribution of historical response residual data of the battery pack under normal operating conditions (e.g., taking the 99th percentile). When a low-reliability observation node is detected, adaptive correction of the low-reliability observation node is triggered, including the following steps: Set the confidence level of low-reliability observation nodes directly to the lowest level (for example, set both the proximity factor and the reliability factor to 0.1). In step S30 of the next scheduling cycle, a sampling matrix and observation vector are constructed, and low-reliability observation nodes are deweighted (e.g., their weight coefficient in the data fitting term is reduced from 1.0 to 0.1), or they are directly treated as unsampled nodes for multiple consecutive cycles, and their status is estimated from the information of adjacent cell nodes. The state estimation process based on graph Laplace regularization is re-executed to re-estimate the state of the local cell network region centered on low-reliability nodes, thereby obtaining a more reliable and corrected state of the entire cell. Based on the revised cell state estimation results and the updated node confidence, adjust the power constraints in the energy allocation model (e.g., reduce the power upper limit).

[0018] Step S50 establishes a reverse verification mechanism based on the output of the energy allocation strategy and the actual operating response, and performs real-time verification of the observed data and model self-correction. This enables the system to automatically adjust the graph signal estimation results and energy allocation strategy in the event of distorted sampling data or sensor failure. This allows the system to maintain the accuracy of state estimation and the stability of energy scheduling under limited sampling conditions, thereby achieving intelligent energy allocation for multiple battery modules and suppressing the risk of abnormal cells, further improving the safety, reliability and service life of the CTP battery pack.

[0019] Example 2: Based on the same inventive concept as the intelligent energy distribution method for multi-battery modules in the foregoing embodiments, such as Figure 3 As shown, this application provides an intelligent energy distribution system for multi-battery modules, wherein the system specifically includes: Cell diagram construction module: Constructs a cell network topology diagram based on the cells inside the battery pack and the connections between the cells; Observation data acquisition module: Based on the cell network topology diagram, it acquires the real-time voltage, current and temperature signals of some cells in each battery module to form partial sampling observation data; Graph signal processing module: Based on graph signal processing methods, it maps some sampled observation data into graph signals, uses graph Laplace regularization method to estimate the state of unsampled cells, and detects abnormal nodes in the graph based on the state estimate of the cells, and identifies the existence of abnormal cells. The energy allocation strategy output module constructs an energy allocation model based on the estimated state values ​​of each cell obtained from graph signal processing. The energy allocation model takes maximizing the energy utilization rate of the entire battery pack as the objective function and constrains each cell node in the cell network topology diagram to obtain the optimal energy allocation strategy for each battery module. Reverse verification module: Based on the output energy allocation strategy, a reverse verification mechanism is introduced in the energy allocation stage. The reliability of the sampled data is dynamically evaluated based on the actual battery response, and low reliability observation nodes are adaptively corrected.

[0020] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A smart energy distribution method for multi-battery modules, characterized in that: include: Construct a cell network topology diagram based on the cells inside the battery pack and the connections between the cells; Based on the cell network topology diagram, real-time voltage, current and temperature signals of some cells in each battery module are obtained to form partial sampling observation data; Based on graph signal processing, some sampled observation data are mapped to graph signals. The graph Laplace regularization method is used to estimate the state of unsampled cells, and abnormal nodes in the graph are detected based on the state estimates of the cells, thus identifying the presence of abnormal cells. An energy allocation model is constructed based on the estimated state values ​​of each cell obtained through signal processing. The energy allocation model takes maximizing the energy utilization rate of the entire battery pack as the objective function and constrains each cell node in the cell network topology diagram to obtain the optimal energy allocation strategy for each battery module. Based on the output-based energy allocation strategy, a reverse verification mechanism is introduced in the energy allocation stage to dynamically evaluate the reliability of the sampled data according to the actual battery response and adaptively correct low-reliability observation nodes.

2. The intelligent energy distribution method for multi-battery modules according to claim 1, characterized in that: Construction of the battery cell network topology: Each battery cell is treated as a node. Based on the electrical connection, thermal coupling and aging correlation between the cells, the edges between the nodes in the graph are defined, the adjacency matrix of the graph is constructed, and the degree matrix of the graph is calculated based on the adjacency matrix. Output the mathematical representation of the battery cell network topology, including the adjacency matrix or Laplace matrix, which will construct the battery cell network topology structure data.

3. The intelligent energy distribution method for multi-battery modules according to claim 1, characterized in that: Acquisition of partial sampled observation data: Based on the cell network topology, a subset of cells is identified as the target cells for data acquisition. The system simultaneously acquires the voltage, current, and temperature signals of the target battery cell. The acquired real-time signals are preprocessed, and the preprocessed data is organized into structured partial sampled observation data.

4. The intelligent energy distribution method for multi-battery modules according to claim 1, characterized in that: The process of estimating the state of unsampled cells using the Graph Laplace regularization method: The voltage, current, and temperature that need to be estimated are defined as graph signals; In N battery cells, M cells are covered by sensors and successfully sampled, and the M observations form an M-dimensional column vector; Define a sampling matrix such that each row of the matrix has only one element that is 1 and the rest are 0; Construct a cost function. The global optimal solution can be obtained by setting the derivative of the cost function to zero. By taking the derivative of the cost function and setting it to zero, a deterministic linear equation system is obtained. The state estimates of all battery cells were obtained by solving the problem using standard numerical methods.

5. The intelligent energy distribution method for multi-battery modules according to claim 4, characterized in that: Obtaining abnormal battery cells: For all cell nodes, calculate the difference between the state estimate and the mean of the state estimates of all adjacent cell nodes, and take the absolute value as the local consistency value; If the local consistency value is greater than the local consistency threshold, it is marked as an abnormal node, which is an abnormal battery cell.

6. The intelligent energy distribution method for multi-battery modules according to claim 1, characterized in that: The process of solving the optimal energy allocation strategy: The power allocation of each battery module is defined as an optimization variable, and an energy allocation model is constructed with the goal of maximizing the total energy utilization of the entire battery pack. The constraints include: adaptive constraints based on cell node confidence; System power requirement constraint: The sum of the power allocated to each module must meet the total power requirement of the load; Module physical limit constraints: The power allocated to each module cannot exceed its own charging and discharging power limit range; A numerical optimization algorithm is used to solve the energy allocation model, and the power allocation vector of each module that satisfies all constraints and optimizes the objective function is calculated.

7. The intelligent energy distribution method for multi-battery modules according to claim 6, characterized in that: The adaptive constraint based on cell node confidence: Based on the confidence level of abnormal nodes and state estimation, power constraints are set for different modules or modules containing abnormal cells. For modules containing faulty battery cells, reduce the power limit or directly constrain the power to 0; For modules composed of cells with low confidence in state estimation, a conservative power limit is adopted; For modules composed of cells with high confidence in state estimation, a wide power limit is applied.

8. The intelligent energy distribution method for multi-battery modules according to claim 7, characterized in that: The confidence level of the state estimation is calculated by multiplying the proximity factor and the reliability factor; Proximity factor: 1.0 for normal observation nodes and 0.1 for abnormal observation nodes; For an unsampled cell node, find the three cells that are closest to the unsampled node and have been sampled and marked as normal observation nodes. Calculate the average topological distance between the unsampled node and these three normal observation nodes, and use 1 and the reciprocal of the sum of the average topological distances as the proximity factor. Reliability factor: For abnormal observation nodes, the value is set to 0.

1. For normal observation nodes, the difference between the maximum permissible residual and the residual of the normal observation node is calculated, and then the ratio of the difference to the maximum permissible residual is used as the reliability factor. For unsampled cell nodes, calculate the average residual of all directly adjacent cells marked as normal observation nodes. Then calculate the difference between the maximum allowable residual and the average residual of the cells, and use the ratio of the maximum allowable residual to the reliability factor. If there are no directly adjacent normal observation nodes, set it to 0.

1.

9. The intelligent energy distribution method for multi-battery modules according to claim 1, characterized in that: The process of assessing the reliability of sampling data: Based on the energy distribution strategy, charge and discharge control of the battery pack is performed. Within a fixed time window after execution, real-time data of all sampled cells are collected synchronously. The actual response data includes: actual voltage response value, actual temperature response value, and actual current response value. If the actual response value of a battery module deviates from the predicted value of the allocation model by more than the set tolerance threshold, it is determined that it contains an observation node with low reliability.

10. A smart energy distribution system for multi-battery modules, characterized in that, The system is used to perform the method according to any one of claims 1-9, the system comprising: Cell diagram construction module: Constructs a cell network topology diagram based on the cells inside the battery pack and the connections between the cells; Observation data acquisition module: Based on the cell network topology diagram, it acquires the real-time voltage, current and temperature signals of some cells in each battery module to form partial sampling observation data; Graph signal processing module: Based on graph signal processing methods, it maps some sampled observation data into graph signals, uses graph Laplace regularization method to estimate the state of unsampled cells, and detects abnormal nodes in the graph based on the state estimate of the cells, and identifies the existence of abnormal cells. The energy allocation strategy output module constructs an energy allocation model based on the estimated state values ​​of each cell obtained from graph signal processing. The energy allocation model takes maximizing the energy utilization rate of the entire battery pack as the objective function and constrains each cell node in the cell network topology diagram to obtain the optimal energy allocation strategy for each battery module. Reverse verification module: Based on the output energy allocation strategy, a reverse verification mechanism is introduced in the energy allocation stage. The reliability of the sampled data is dynamically evaluated based on the actual battery response, and low reliability observation nodes are adaptively corrected.

Images