Method and System for Fault Diagnosis and Early Warning of Battery Modules in Dynamically Reconfigurable Energy Storage Systems

Abstract

This invention relates to a method and system for fault diagnosis and early warning of battery modules in dynamically reconfigurable energy storage systems, specifically in the field of energy storage for new energy power systems. To address the shortcomings of existing battery module-level monitoring methods for new energy power storage systems, such as reliance on single indicators, poor dynamic topology adaptability, insufficient collaborative fault detection capabilities, low computational efficiency, high false negative rates for latent faults, and high maintenance costs, this invention constructs a voltage matrix; weighted and fused similarity matrices to generate complementary similarity matrices; constructs a weighted undirected graph; uses vertex consistency indicators as the module health benchmark; identifies connected components in the weighted undirected graph; and performs diagnosis by combining average edge weights and the number of modules with the smallest connected component using a hierarchical threshold; renders vertex and edge colors according to the graph model weights and overlays the diagnostic results. This invention is primarily used for fault diagnosis and early warning of battery modules in dynamically reconfigurable energy storage systems.

Description

Technical Field

[0001] This invention relates to the field of energy storage in new energy power systems, and in particular to a method and system for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system. Background Technology

[0002] With the continuous expansion of installed capacity of new energy power generation, the proportion of new energy power generation in the power grid is increasing. However, due to the small capacity, large number, and dispersed distribution of new energy power generation units, as well as their significant intermittency, volatility, and randomness, the high proportion of new energy grid connection inevitably brings unprecedented challenges to the power system's supply and demand balance, safety, and stability control. Energy storage systems are a key component in regulating the imbalance between new energy power generation and the power system's supply and demand, and in achieving energy management and optimization. Unlike the traditional fixed series battery structure of energy storage, dynamic reconfigurable battery networks consist of several modules combined through series and parallel structures, switching topologies in real time according to grid demand to achieve second-level power response and coordinated charging and discharging of modules within the cluster. However, this dynamic characteristic also brings new challenges:

[0003] 1. Increased risk of fault propagation: A single module fault may propagate to adjacent modules through electrical coupling during the reconfiguration process;

[0004] 2. Hidden anomalies are difficult to trace: Frequent topology switching causes dynamic changes in the module association relationship, and traditional diagnostic models based on fixed groups cannot locate the fault source;

[0005] 3. Upgraded system reliability requirements: Energy storage systems need to undertake key tasks such as grid frequency regulation and peak shaving and valley filling. A single failure outage may lead to local grid instability.

[0006] Currently, battery module-level monitoring technology is widely used in the energy storage field of new energy power systems, but traditional methods have significant limitations:

[0007] 1. Dependence on a single indicator:

[0008] Current solutions rely on voltage and temperature thresholds to determine faults, ignoring the dynamic correlation between modules. When a module's voltage curve shows local outliers but does not exceed the threshold, traditional methods cannot effectively identify early anomalies.

[0009] 2. Poor adaptability to dynamic topology:

[0010] The module switching of a dynamically reconfigurable system causes real-time changes in the topology. The fault diagnosis model of the traditional fixed series-parallel structure cannot adaptively update the correlation relationship and may mistakenly identify the reconfiguration action as a fault.

[0011] 3. Blind spots in collaborative fault detection capabilities:

[0012] There is a lack of quantitative assessment methods for multi-module collaborative anomalies, which can only identify single-point failures and cannot provide early warning of systemic consistency degradation.

[0013] 4. Low computational efficiency:

[0014] Similarity methods based on raw data from all time periods have high computational complexity and are difficult to meet the needs of real-time diagnosis.

[0015] 5. High rate of missed detection for latent faults:

[0016] Traditional clustering anomaly detection algorithms are sensitive to outliers. When the number of faulty modules increases, they are prone to misclassifying abnormal clusters as normal subgroups, leading to missed detections.

[0017] 6. High maintenance costs:

[0018] Traditional methods rely on regular manual inspections and full-capacity testing, with an average fault location delay of more than 24 hours and a lack of early warning capabilities for early-stage hidden faults.

[0019] Therefore, there is a need for a method and system for fault diagnosis and early warning of battery modules in dynamic reconfigurable energy storage systems that can improve early anomaly identification capabilities, overcome the limitations of single indicators, adapt to dynamic topology changes, avoid misdiagnosis, accurately locate synergistic degradation, and have high computational efficiency. Summary of the Invention

[0020] This invention addresses the shortcomings of existing battery module-level monitoring methods for new energy power storage systems, which rely on single indicators, have poor dynamic topology adaptability, insufficient collaborative fault detection capabilities, low computational efficiency, high rate of missed reports of latent faults, and high maintenance costs. It provides a method and system for fault diagnosis and early warning of battery modules in dynamic reconfigurable energy storage systems that can improve early anomaly identification capabilities, overcome the limitations of single indicators, adapt to dynamic topology changes, avoid misdiagnosis, accurately locate collaborative degradation, and have high computational efficiency.

[0021] The present invention provides a method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system, comprising the following steps:

[0022] S1. Data Acquisition and Data Preprocessing:

[0023] Collect time-series voltage data of all battery modules under the same battery cluster and construct a voltage matrix V;

[0024] S2, Multidimensional Similarity Fusion:

[0025] The Pearson similarity matrix P, cosine similarity matrix S, and Euclidean distance similarity matrix D between the voltage matrices of any two battery modules are calculated in parallel; the three similarity matrices are weighted and fused to generate a complementary fused similarity matrix R.

[0026] S3. Battery State Diagram Model Construction:

[0027] A weighted undirected graph G is constructed with battery modules as vertices and similarity as edges; and the consistency index of the vertices is calculated, which is used as the health benchmark of the battery modules.

[0028] S4. Fault Diagnosis and Early Warning: Based on the depth-first search method, the weighted undirected graph G is identified to obtain the connected components of the weighted undirected graph G. The average edge weight and the number of battery modules with the minimum connected component are combined to perform diagnosis through hierarchical thresholds.

[0029] S5. Visualization Output:

[0030] The vertex and edge colors are rendered based on the weights of the battery state graph model, and the diagnostic results are overlaid.

[0031] Furthermore: In S1, the data acquisition and data preprocessing specifically include:

[0032] Using a digital energy network card, the operating voltage data of all battery modules in the same cluster of a dynamically reconfigurable energy storage system are collected to obtain voltage and time series, where the elements of the voltage series correspond to the elements of the time series.

[0033] Based on the time series, sort the voltage series in ascending order of time;

[0034] Based on the time series, duplicate values ​​and their corresponding voltage sequence elements are removed to obtain the voltage matrix V. V is an N-row, T-column matrix, where N represents the number of battery modules in a cluster, and T represents the number of voltage data acquisition times. The element in the i-th row and j-th column of the voltage matrix V is... This represents the operating voltage value of the i-th module at sampling time j.

[0035] Furthermore: In S1, linear interpolation is performed in the column direction for the empty values ​​in the voltage matrix V, as shown in the following formula:

[0036] ;

[0037] in, This represents the element in the (i-1)th row of the j-th column of matrix V. This represents the element in the j-th column and i-th row of matrix V. This represents the element in the (i+1)th row of the j-th column of matrix V.

[0038] Furthermore: In S2, the formula for the Pearson similarity matrix P is as follows:

[0039] ;

[0040] in, This represents the a-th column of the obtained voltage matrix V. This represents the b-th column of the obtained voltage matrix V. Represents the voltage matrix V's first... The covariance of the column, This represents the standard deviation of the b-th column of the voltage matrix V. Representing the voltage matrix The standard deviation of column b, It is the element in the a-th row and b-th column of the Pearson similarity matrix P;

[0041] Normalizing the obtained Pearson similarity matrix P, we get the normalized Pearson similarity formula as follows:

[0042] ;

[0043] in, It is the element in the a-th row and b-th column of the normalized Pearson similarity matrix P'. These represent the maximum values ​​of the Pearson similarity matrix P, respectively. This represents the minimum value of the Pearson similarity matrix P;

[0044] Calculate the cosine similarity between the operating voltages of any two battery modules to obtain the cosine similarity matrix S. The calculation formula is as follows:

[0045] ;

[0046] in, This yields the element in the a-th row and b-th column of the cosine similarity matrix S. This represents the magnitude of the a-th column of the voltage matrix V. This represents the magnitude of the b-th column of the voltage matrix V. The adjustment coefficients for adjusting the vector cosine distance and cosine phase velocity;

[0047] Calculate the Euclidean distance similarity between the operating voltages of any two battery modules to obtain the Euclidean distance similarity matrix D. The calculation formula is as follows:

[0048] ;

[0049] in, is the element in the a-th row and b-th column of the Euclidean distance similarity matrix D, where e is the natural logarithm. This represents the Euclidean distance between the a-th and b-th columns of the voltage matrix V. It is a coefficient used to adjust the similarity between Euclidean distances.

[0050] Furthermore: In S2, the obtained Pearson similarity matrix P, cosine similarity matrix S, and Euclidean distance similarity matrix D are weighted and averaged to obtain the fused complementary similarity matrix R, as shown in the following formula:

[0051] ;

[0052] in, , , These are weighting coefficients, satisfying... .

[0053] Furthermore: In S3, the dynamically reconfigurable battery system is transformed into a weighted undirected graph. Where M is the set of vertices, representing each battery module under the dynamically reconfigurable battery cluster; E is the set of edges, representing the behavioral associations between battery modules; and W is the set of edge weights, used to quantify the association strength between battery modules.

[0054] The steps for constructing a battery state diagram model are as follows:

[0055] Construct a vertex set M. Each vertex Map a physical battery module, and calculate the consistency index of battery module i based on the calculated similarity matrix R. As a vertex attribute, the formula is as follows:

[0056] ;

[0057] in, Represents the calculated similarity matrix The element in the i-th row and j-th column;

[0058] Construct the edge set for i, ,if Then at the vertex and Add edges between ,in, To establish a threshold for determining the edge between two vertices;

[0059] for The similarity between the two modules is used as the weight of the edge, i.e. , For the edge The weights are then determined; this completes the construction of the weighted undirected graph G.

[0060] Further: In S4, the specific process of identifying the connected components of the weighted undirected graph G based on the depth-first search method is as follows:

[0061] From a weighted undirected graph Starting from any unvisited vertex, visit its unvisited adjacent vertices that are directly connected by an edge. When it is impossible to continue to the next level, return to the previous level vertex and try to visit its other unvisited adjacent vertices. Repeat this process until all vertices connected to the starting vertex by a path have been traversed. The visited vertices and their connecting edges form a connected component, denoted as . ;

[0062] Select a weighted undirected graph Using the next unvisited vertex as a new starting point, repeat the depth-first search method described above until the weighted undirected graph is reached. All vertices in the graph have been visited; the resulting subgraph, consisting of all independent, disconnected sets of vertices, is the weighted undirected graph. All connected components are denoted as ,in, The total number of connected components;

[0063] Compute a weighted undirected graph The average weight of all edges is calculated using the following formula:

[0064] ;

[0065] in, Represents a weighted undirected graph The number of edge set elements;

[0066] Compute a weighted undirected graph Minimum number of vertices in all connected components The calculation formula is as follows:

[0067] ;

[0068] in, This indicates that for all the connected components obtained The number of vertices;

[0069] Set the average edge weight threshold Based on the size of the connected component K and the average edge weight Fault diagnosis and early warning are performed by matching the preset diagnostic and early warning rules.

[0070] Furthermore: the preset diagnostic warning rules include:

[0071] when When the system is deemed to be in good condition, the strategy to be implemented is to maintain the current operation and maintenance plan. The diagnostic criteria are: if all batteries constitute a single system and the overall consistency level is high, then the anomaly level is determined to be level 0.

[0072] when When an abnormal battery module is detected, it is determined to be an abnormal single battery module. The abnormal battery module is then isolated for testing. If an independent battery module is found to be significantly out of the group, the abnormality level is determined to be Level 1.

[0073] when When the battery module batch is identified as abnormal, maintenance is performed by grouping it into subsystems. If the system is split into K subsystems, the abnormality level is determined to be level 2.

[0074] when When a system is deemed to be deteriorating in consistency, a comprehensive battery diagnostic is initiated. If the system's average similarity is below the average edge weight threshold... If so, the anomaly level is determined to be level 3.

[0075] Furthermore: In S5, the specific process of rendering vertex and edge colors according to the battery state graph model weights and overlaying the diagnostic results is as follows:

[0076] According to the weighted undirected graph Perform image rendering, filling vertices and edges with different depths of color according to their weights;

[0077] The results of fault diagnosis and early warning are displayed in the image.

[0078] The fault diagnosis and early warning system for implementing the fault diagnosis and early warning method for battery modules of the dynamically reconfigurable energy storage system, as described in this invention, includes a data acquisition and data preprocessing module, a multi-dimensional similarity fusion module, a battery state diagram modeling module, a fault diagnosis and early warning module, and a visualization module.

[0079] The data acquisition and data preprocessing module is used to acquire time-series voltage data of all battery modules under the same battery cluster, and to perform linear interpolation on the column direction for the empty values ​​in the matrix.

[0080] The multidimensional similarity fusion module is used to calculate in parallel the Pearson similarity matrix P, cosine similarity matrix S, and Euclidean distance similarity matrix D between the working voltages of any two battery modules; the three similarity matrices are weighted and fused to generate a complementary fusion similarity matrix R;

[0081] The battery state graph modeling module is used to construct a weighted undirected graph G with battery modules as vertices and similarity as edges; and to calculate the consistency index of the vertices, using the vertex consistency index as the benchmark for module health.

[0082] The fault diagnosis and early warning module identifies the weighted undirected graph G based on the depth-first search method, obtains the connected components of the weighted undirected graph G, and performs diagnosis by combining the average edge weight and the number of battery modules with the minimum connected component through a hierarchical threshold.

[0083] The visualization module renders vertex and edge colors according to the weights of the battery state graph model and overlays the diagnostic results.

[0084] The beneficial effects of this invention are:

[0085] 1. Overcome the limitations of single indicators and improve the ability to identify early anomalies.

[0086] By integrating Pearson correlation coefficient, cosine similarity, and Euclidean distance similarity, the numerical outliers, phase alignment, and morphological differences of voltage curves between modules are comprehensively quantified, effectively identifying local voltage outliers that are missed by traditional methods and significantly reducing the false alarm rate of latent faults.

[0087] 2. Adapts to dynamic topology changes, avoiding misdiagnosis.

[0088] A weighted undirected graph model is constructed based on a real-time similarity matrix, replacing the fixed grouping structure with dynamically linked edges. During topology switching, the graph model updates the module relationships in real time through edge weights, eliminating the interference of reconstruction actions on diagnosis.

[0089] 3. Accurately pinpoint synergistic decline and provide early warning of systemic risks.

[0090] By analyzing connectivity components and calculating consistency indices, anomalies in multi-module collaboration are quantified.

[0091] 4. Meet real-time diagnostic needs and optimize computational efficiency.

[0092] A lightweight graph traversal algorithm is adopted to replace full-time data clustering, which greatly reduces computational complexity; parallel computation of weighted similarity matrix ensures second-level response and adapts to high-frequency diagnostic scenarios in dynamic reconstruction systems.

[0093] 5. Significantly reduces operation and maintenance costs and shortens fault location time.

[0094] The tiered diagnostic strategy automatically outputs fault levels and handling suggestions, reducing fault location delays from 24 hours to minutes and minimizing reliance on manual inspections.

[0095] 6. Visualization assists decision-making, improving operation and maintenance efficiency.

[0096] The graph model weighted rendering topology status, overlaid with diagnostic results, intuitively displays the location of faulty modules, associated paths, and system health, supporting rapid operation and maintenance response. Attached Figure Description

[0097] Figure 1This is a topology diagram of a dynamically reconfigurable energy storage system's battery cluster.

[0098] Figure 2 It is a voltage curve of each battery module under a cluster when a fault occurs in a dynamically reconfigurable battery system.

[0099] Figure 3 This is a visual diagram of the diagnostic and early warning results. Detailed Implementation

[0100] The following are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. The embodiments described below are only for explaining the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention should be determined by the scope of the claims. The embodiments of the present invention are described in detail below. In order to facilitate the description of the present invention and simplify the description, the technical terms used in the specification of the present invention should be interpreted broadly, including but not limited to conventional alternatives not mentioned in this application, and including both direct and indirect implementation methods.

[0101] Example 1

[0102] Combination Figures 1-3 This embodiment describes a method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system, specifically including:

[0103] S1. Data acquisition and data preprocessing;

[0104] Using a digital energy network card, the operating voltage data of all battery modules in the same cluster of a dynamically reconfigurable energy storage system are collected to obtain voltage and time series, where the elements of the voltage series correspond to the elements of the time series.

[0105] Figure 1 The digital energy switch in the system is a comprehensive control unit for energy scheduling, advanced strategy execution, and communication management of a dynamically reconfigurable energy storage system. It integrates numerous self-diagnostic, self-healing, and self-optimizing management algorithms. It is responsible for collecting battery and adapter status data measured by each hub, analyzing and calculating data from the batteries and other system components under its management, and formulating corresponding scheduling strategies. Simultaneously, it communicates with the digital energy gateway to coordinate the scheduling of related systems and executes corresponding actions based on system status and superior scheduling instructions.

[0106] Digital power hub: Responsible for device control, data acquisition, status judgment, and data forwarding of energy units at the equipment layer, and performs online dynamic reconstruction of the series and parallel topology of energy units, as well as detection of abnormal conditions of energy units and millisecond-level rapid protection actions for fault states.

[0107] Digital energy network card: This is a high-speed control and data acquisition device for energy units. It controls the on / off switching of the battery module's charge / discharge switch, provides a digital interface for the digital energy hub, and responds to the hub's control commands and configurations. It also serves as an energy transmission channel and is a primary and secondary energy interface device. The energy unit is the battery module.

[0108] Based on the time series, sort the voltage series in ascending order of time;

[0109] Based on the time series, duplicate values ​​and their corresponding voltage sequence elements are removed to obtain the voltage matrix V. V is an N-row, T-column matrix, where N represents the number of battery modules within the same battery cluster, and T represents the number of voltage data acquisition times. The element in the i-th row and j-th column of matrix V... This represents the operating voltage value of the i-th module at sampling time j.

[0110] For the null values ​​in matrix V, linear interpolation is performed along the column direction, using the following formula:

[0111] ;

[0112] in, This represents the element in the (i-1)th row of the j-th column of matrix V. This represents the element in the j-th column and i-th row of matrix V. This represents the element in the (i+1)th row of the j-th column of matrix V.

[0113] S2, Calculation of multidimensional similarity matrix;

[0114] Calculate the Pearson correlation coefficient between the operating voltages of any two battery modules to obtain the Pearson similarity matrix P, as shown in the following formula:

[0115] ;

[0116] in, This represents the a-th column of the obtained voltage matrix V. This represents the b-th column of the obtained voltage matrix V. Represents the voltage matrix V's first... The covariance of the column, This represents the standard deviation of the b-th column of the voltage matrix V. Representing the voltage matrix The standard deviation of column b, It is the element in the a-th row and b-th column of the Pearson similarity matrix P;

[0117] Normalizing the obtained Pearson similarity matrix P, we get the normalized Pearson similarity formula as follows:

[0118] ;

[0119] in, It is the element in the a-th row and b-th column of the normalized Pearson similarity matrix P'. These represent the maximum values ​​of the Pearson similarity matrix P, respectively. This represents the minimum value of the Pearson similarity matrix P;

[0120] like Figure 2 As shown, the normalized Pearson similarity matrix can measure the phase alignment of the voltage curves of the battery module during charging and discharging, and can effectively identify outliers in the module voltage values.

[0121] Calculate the cosine similarity between the operating voltages of any two battery modules to obtain the cosine similarity matrix S. The calculation formula is as follows:

[0122] ;

[0123] in, This yields the element in the a-th row and b-th column of the cosine similarity matrix S. This represents the magnitude of the a-th column of the voltage matrix V. This represents the magnitude of the b-th column of the voltage matrix V. The adjustment coefficients for the cosine distance and cosine phase velocity of the vector are used in this embodiment. .

[0124] Calculate the Euclidean distance similarity between the operating voltages of any two battery modules to obtain the Euclidean distance similarity matrix D. The calculation formula is as follows:

[0125] ;

[0126] in, is the element in the a-th row and b-th column of the Euclidean distance similarity matrix D, where e is the natural logarithm. This represents the Euclidean distance between the a-th and b-th columns of the voltage matrix V. This is a coefficient used to adjust the similarity between Euclidean distances. In this embodiment... .

[0127] The weighted average of the obtained Pearson similarity, cosine similarity, and Euclidean distance similarity matrices yields the fused complementary similarity matrix R, as shown in the following formula:

[0128] ;

[0129] in, , , These are weighting coefficients, satisfying... In this embodiment .

[0130] S3. Construct a battery state diagram model;

[0131] like Figure 3 As shown, the dynamically reconfigurable battery system is transformed into a weighted undirected graph. Where M is the set of vertices, representing each module under the dynamically reconfigurable battery cluster; E is the set of edges, representing the behavioral associations between modules; and W is the set of edge weights, used to quantify the association strength between modules.

[0132] The steps for constructing a battery state diagram model are as follows:

[0133] Construct a vertex set M. Each vertex Map a physical battery module, and calculate the consistency index of battery module i based on the calculated similarity matrix R. As a vertex attribute, using modules as vertices is a consistency indicator for modules from the perspective of entities; from the perspective of the constructed graph model, it is a consistency indicator for vertices. The formula is as follows:

[0134] ;

[0135] in, Represents the calculated similarity matrix The element in the i-th row and j-th column;

[0136] Construct the edge set for i, ,if Then at the vertex and Add edges between ,in, To establish a threshold for determining the edge between two vertices; in this embodiment... The value is set to 0.5, which is used for subsequent fault diagnosis.

[0137] for The similarity between the two modules is used as the weight of the edge, i.e. , For the edge The weights are then determined; this completes the construction of the weighted undirected graph G.

[0138] S4. Fault diagnosis and early warning;

[0139] The weighted undirected graph G is identified using a depth-first search (DFS) method to obtain its connected components. Starting from any unvisited vertex in G, the system visits as many unvisited adjacent vertices as possible that are directly connected by edges. When no further depth is possible, the system returns to the previous level and attempts to visit its other unvisited adjacent vertices. This process is repeated until all vertices connected to the starting vertex by a path have been traversed. These visited vertices and their connecting edges constitute a connected component, denoted as . Select the next unvisited vertex in the graph as the new starting point and repeat the DFS process until all vertices in the weighted undirected graph G have been visited. The subgraph formed by all the independent, disconnected vertex sets is the complete connected component of the weighted undirected graph G, denoted as . , where k is the total number of connected components.

[0140] The average weight of all edges in a weighted undirected graph G is calculated using the following formula:

[0141] ;

[0142] in, This represents the number of edge set elements in the weighted undirected graph G.

[0143] Compute a weighted undirected graph Minimum number of vertices in all connected components The calculation formula is as follows:

[0144] ;

[0145] in, This indicates that for all the connected components obtained The number of vertices;

[0146] Set the average edge weight threshold Based on the size of the connected component K and the average edge weight Fault diagnosis and early warning are performed, and the methods for diagnosis and early warning are shown in Table 1:

[0147] Table 1: Diagnostic and Early Warning Methods

[0148]

[0149] In this embodiment, =0.7.

[0150] S5. Results visualization;

[0151] The weighted undirected graph obtained from the construction Perform image rendering, filling vertices and edges with different depths of color according to their weights;

[0152] The results of fault diagnosis and early warning are displayed in the image.

[0153] Example 2

[0154] Combination Figures 1-3 This embodiment describes a fault diagnosis and early warning system for implementing the aforementioned method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system. The system includes a data acquisition and data preprocessing module, a multi-dimensional similarity fusion module, a battery state diagram modeling module, a fault diagnosis and early warning module, and a visualization module.

[0155] The data acquisition and data preprocessing module is used to acquire time-series voltage data of all battery modules under the same battery cluster, and to perform linear interpolation on the column direction for the empty values ​​in the matrix.

[0156] The multidimensional similarity fusion module is used to calculate in parallel the Pearson similarity matrix P, cosine similarity matrix S, and Euclidean distance similarity matrix D between the working voltages of any two battery modules; the three similarity matrices are weighted and fused to generate a complementary fusion similarity matrix R;

[0157] The battery state graph modeling module is used to construct a weighted undirected graph G with battery modules as vertices and similarity as edges; and to calculate the consistency index of the vertices, using the vertex consistency index as the benchmark for module health.

[0158] The fault diagnosis and early warning module identifies the weighted undirected graph G based on the depth-first search method, obtains the connected components of the weighted undirected graph G, and performs diagnosis by combining the average edge weight and the number of battery modules with the minimum connected component through a hierarchical threshold.

[0159] The visualization module renders vertex and edge colors according to the weights of the battery state graph model and overlays the diagnostic results.

Claims

1. A method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Collect time-series voltage data of all battery modules under the same battery cluster and construct a voltage matrix V; S2, Multidimensional Similarity Fusion: Parallel calculation of the Pearson similarity matrix P, cosine similarity matrix S, and Euclidean distance similarity matrix D between the voltage matrices of any two battery modules; The three similarity matrices are weighted and fused to generate a complementary fused similarity matrix R; S3. Battery State Diagram Model Construction: A weighted undirected graph G is constructed with battery modules as vertices and similarity as edges; and the consistency index of the vertices is calculated, which is used as the health benchmark of the battery modules. In S3, the dynamically reconfigurable battery system is transformed into a weighted undirected graph. Where M is the set of vertices, representing each battery module under the dynamically reconfigurable battery cluster; E is the set of edges, representing the behavioral associations between battery modules; and W is the set of edge weights, used to quantify the association strength between battery modules. The steps for constructing a battery state diagram model are as follows: Construct a vertex set M. Each vertex Map a physical battery module, and calculate the consistency index of battery module i based on the calculated similarity matrix R. As a vertex attribute, the formula is as follows: ； in, Represents the calculated similarity matrix The element in the i-th row and j-th column; Construct the edge set, for i, ,if Then at the vertex and Add edges between ,in, To establish the threshold for determining the edge between two vertices; N represents the number of battery modules in a cluster; for The similarity between the two modules is used as the weight of the edge, i.e. , For the edge The weights are then determined; this completes the construction of the weighted undirected graph G. S4. Fault Diagnosis and Early Warning: Based on the depth-first search method, the weighted undirected graph G is identified to obtain the connected components of the weighted undirected graph G. The average edge weight and the number of battery modules with the minimum connected component are combined to perform diagnosis through hierarchical thresholds. In S4, the specific process of identifying the connected components of the weighted undirected graph G based on the depth-first search method is as follows: From a weighted undirected graph Starting from any unvisited vertex, visit its unvisited adjacent vertices that are directly connected by an edge. When it is impossible to continue to the next level, return to the previous level vertex and try to visit its other unvisited adjacent vertices. Repeat this process until all vertices connected to the starting vertex by a path have been traversed. The visited vertices and their connecting edges form a connected component, denoted as . ; Select a weighted undirected graph Using the next unvisited vertex as a new starting point, repeat the depth-first search method described above until the weighted undirected graph is reached. All vertices in the graph have been visited; the resulting subgraph, consisting of all independent, disconnected sets of vertices, is the weighted undirected graph. All connected components are denoted as ,in, The total number of connected components; Compute a weighted undirected graph The average weight of all edges is calculated using the following formula: ； in, Represents a weighted undirected graph The number of edge set elements; Compute a weighted undirected graph Minimum number of vertices in all connected components The calculation formula is as follows: ； in, This indicates that for all the connected components obtained The number of vertices; Set the average edge weight threshold Based on the size of the connected component K and the average edge weight Match preset diagnostic and early warning rules to perform fault diagnosis and early warning; S5. Visualization Output: The vertex and edge colors are rendered based on the weights of the battery state graph model, and the diagnostic results are overlaid.

2. The method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system according to claim 1, characterized in that, In S1, the data acquisition and data preprocessing specifically include: Using a digital energy network card, the operating voltage data of all battery modules in the same cluster of a dynamically reconfigurable energy storage system are collected to obtain voltage and time series, where the elements of the voltage series correspond to the elements of the time series. Based on the time series, sort the voltage series in ascending order of time; Based on the time series, duplicate values ​​and their corresponding voltage sequence elements are removed to obtain the voltage matrix V. V is an N-row, T-column matrix, where N represents the number of battery modules in a cluster, and T represents the number of voltage data acquisition times. The element in the i-th row and j-th column of the voltage matrix V is... This represents the operating voltage value of the i-th module at sampling time j.

3. The method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system according to claim 2, characterized in that, In S1, linear interpolation is performed in the column direction for the empty values ​​in the voltage matrix V, as shown in the following formula: ； in, This represents the element in the (i-1)th row of the j-th column of matrix V. This represents the element in the j-th column and i-th row of matrix V. This represents the element in the (i+1)th row of the j-th column of matrix V.

4. The method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system according to claim 1, characterized in that, In S2, the formula for the Pearson similarity matrix P is as follows: ； in, This represents the a-th column of the obtained voltage matrix V. This represents the b-th column of the obtained voltage matrix V. Represents the voltage matrix V's first... The covariance of the column, This represents the standard deviation of the a-th column of the voltage matrix V. Representing the voltage matrix The standard deviation of column b, It is the element in the a-th row and b-th column of the Pearson similarity matrix P; Calculate the cosine similarity between the operating voltages of any two battery modules to obtain the cosine similarity matrix S. The calculation formula is as follows: ； in, This yields the element in the a-th row and b-th column of the cosine similarity matrix S. This represents the magnitude of the a-th column of the voltage matrix V. This represents the magnitude of the b-th column of the voltage matrix V. The adjustment coefficients for adjusting the vector cosine distance and cosine phase velocity; Calculate the Euclidean distance similarity between the operating voltages of any two battery modules to obtain the Euclidean distance similarity matrix D. The calculation formula is as follows: ； in, is the element in the a-th row and b-th column of the Euclidean distance similarity matrix D, where e is the natural logarithm. This represents the Euclidean distance between the a-th and b-th columns of the voltage matrix V. It is a coefficient used to adjust the similarity between Euclidean distances.

5. The method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system according to claim 1, characterized in that, In S2, the weighted average of the obtained Pearson similarity matrix P, cosine similarity matrix S, and Euclidean distance similarity matrix D is used to obtain the fused complementary similarity matrix R, as shown in the following formula: ； in, , , These are weighting coefficients, satisfying... .

6. The method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system according to claim 1, characterized in that, The preset diagnostic warning rules include: when When the system is deemed to be in good condition, the strategy to be implemented is to maintain the current operation and maintenance plan. The diagnostic criteria are: if all batteries constitute a single system and the overall consistency level is high, then the anomaly level is determined to be level 0. when When an abnormal battery module is detected, it is determined to be an abnormal single battery module. The abnormal battery module is then isolated for testing. If an independent battery module is found to be significantly out of the group, the abnormality level is determined to be Level 1. when When the battery module batch is identified as abnormal, maintenance is performed by grouping it into subsystems. If the system is split into K subsystems, the abnormality level is determined to be level 2. when When a system is deemed to be deteriorating in consistency, a comprehensive battery diagnostic is initiated. If the system's average similarity is below the average edge weight threshold... If so, the anomaly level is determined to be level 3.

7. The method for fault diagnosis and early warning of battery modules in a dynamically reconfigurable energy storage system according to claim 1, characterized in that, In S5, the specific process of rendering vertex and edge colors based on the battery state graph model weights and overlaying the diagnostic results is as follows: According to the weighted undirected graph Perform image rendering, filling vertices and edges with different depths of color according to their weights; The results of fault diagnosis and early warning are displayed in the image.

8. A fault diagnosis and early warning system for implementing the fault diagnosis and early warning method for battery modules of a dynamically reconfigurable energy storage system as described in any one of claims 1-7, characterized in that, It includes a data acquisition and preprocessing module, a multi-dimensional similarity fusion module, a battery state diagram modeling module, a fault diagnosis and early warning module, and a visualization module; The data acquisition and data preprocessing module is used to acquire time-series voltage data of all battery modules under the same battery cluster, and to perform linear interpolation on the column direction for the empty values ​​in the matrix. The multidimensional similarity fusion module is used to calculate in parallel the Pearson similarity matrix P, cosine similarity matrix S, and Euclidean distance similarity matrix D between the working voltages of any two battery modules; The three similarity matrices are weighted and fused to generate a complementary fused similarity matrix R; The battery state graph modeling module is used to construct a weighted undirected graph G with battery modules as vertices and similarity as edges; and to calculate the consistency index of the vertices, using the vertex consistency index as the benchmark for module health. The fault diagnosis and early warning module identifies the weighted undirected graph G based on the depth-first search method, obtains the connected components of the weighted undirected graph G, and performs diagnosis by combining the average edge weight and the number of battery modules with the minimum connected component through a hierarchical threshold. The visualization module renders vertex and edge colors according to the weights of the battery state graph model and overlays the diagnostic results.

Images