Mobile energy storage device fault diagnosis and early warning method and system

Abstract

This invention relates to the field of fault data processing technology and discloses a method and system for fault diagnosis and early warning of mobile energy storage devices. The method includes: acquiring the operating data of the target energy storage object; analyzing the pre-processed real-time operating data to obtain a multi-dimensional situation field; the target energy storage object is an outdoor energy storage power source; calculating the multi-dimensional situation field to obtain the synaptic traces of abnormal data and identifying antibody modes; analyzing the stability of each antibody mode to obtain the mode confidence. This solution accurately captures the dynamic evolution characteristics of multi-parameter coupling of faults by constructing a dynamic cellular network and a multi-dimensional situation field, and achieves deep causal inference based on synaptic traces, antibody modes, and mode confidence. It locks the root cause of the fault by the root cause contribution and generates a fault early warning command by combining the fault outbreak potential energy. This can help non-professional outdoor personnel quickly locate faults and improve the accuracy of diagnosis and the timeliness of early warning.

Description

Technical Field

[0001] This invention relates to the field of fault data processing technology, specifically to a method and system for fault diagnosis and early warning of mobile energy storage devices. Background Technology

[0002] Currently, fault diagnosis of mobile energy storage devices often involves performing isolated single-variable analysis on data collected by sensors such as current, voltage, and temperature during data processing, and setting a static safety threshold for each parameter. After simple processing, the data is independently compared with its respective threshold. Once a data point exceeds the threshold, the corresponding alarm signal is triggered, such as overvoltage or overtemperature.

[0003] However, when the above data processing methods are applied to outdoor energy storage power supplies, the following shortcomings still exist: Most faults of outdoor energy storage power supplies are not isolated events with instantaneous changes, but rather a dynamic process involving multiple parameters and evolving along a specific path. Currently, data analysis is often done in a single-factor manner, lacking multivariate correlation analysis and causal inference capabilities. This results in diagnoses that are often based on a single fault, failing to determine the root cause of the fault. Consequently, non-professional outdoor maintenance personnel are unable to locate faults in a timely manner, reducing the effectiveness of diagnosis and early warning. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for fault diagnosis and early warning of mobile energy storage devices, thus solving the aforementioned problems.

[0005] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0006] Methods for fault diagnosis and early warning of mobile energy storage devices include:

[0007] Step S1: Obtain the operating data of the target energy storage object, analyze the preprocessed real-time operating data to obtain a multi-dimensional situation field, and the target energy storage object is an outdoor energy storage power source.

[0008] Step S2: Calculate the multidimensional situation field, analyze the synaptic weight traces of abnormal data, identify antibody modes, analyze the stability of each antibody mode, and obtain the mode confidence.

[0009] Step S3: Based on the antibody modality and modality confidence, reverse reasoning is performed to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, and the root cause contribution is obtained.

[0010] Step S4: Based on the diagnostic points where the root cause contribution exceeds the threshold, construct a key root cause set, calculate the key root cause set and the root cause contribution, and generate a fault warning instruction.

[0011] Furthermore, the preprocessed real-time operational data is analyzed to obtain a multi-dimensional situation field, including:

[0012] Each parameter in the real-time running data is regarded as a dynamic cell, and the physical causal and temporal correlation between each dynamic cell is analyzed to construct a dynamic cell network. The dynamic cell network is then analyzed to generate cell synergy.

[0013] Based on dynamic cellular networks, the real-time data of each dynamic cell is mapped to a situation vector in a high-dimensional space. The initial situation field formed by all situation vectors is analyzed to obtain the field chaos index.

[0014] Furthermore, analysis of the preprocessed real-time operational data yields a multi-dimensional situation field, which also includes:

[0015] The initial state field is adjusted based on the field chaos index to generate a field mode sequence;

[0016] A multidimensional situation field is generated by superimposing modal sequences based on field mode sequences.

[0017] Furthermore, the multidimensional situation field is calculated, the synaptic weight traces of anomalous data are analyzed, and antibody modalities are identified. The stability of each antibody modality is analyzed to obtain modality confidence, including:

[0018] The multidimensional situation field is calculated, the continuous change of field strength is converted, and the situation pulse sequence is obtained by combining the cell coordination degree.

[0019] The situation pulse sequence is analyzed to construct a pulse neural cluster and trigger a co-activation signal;

[0020] Based on co-activation signals, the abnormal propagation path of abnormal data is traced in reverse within the spiking neural cluster to obtain synaptic traces.

[0021] Furthermore, the multidimensional situation field is calculated, the synaptic weight traces of anomalous data are analyzed, and antibody modalities are identified. The stability of each antibody modality is analyzed to obtain modality confidence. This also includes:

[0022] Based on synaptic traces, each stable abnormal propagation path is identified, and antibody modalities are generated.

[0023] Modal stability of antibody modalities was analyzed to obtain modal maturity.

[0024] Based on modality maturity, antibody modalities are evaluated, and modality confidence scores are generated.

[0025] Furthermore, based on antibody modalities and modal confidence levels, reverse reasoning is performed to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, thus obtaining the root cause contribution, including:

[0026] Based on antibody modes and field chaos index, the diagnostic points in the target energy storage object are analyzed, and a fault propagation graph representing the abnormal propagation path is constructed to generate propagation topology coefficients.

[0027] Based on modal confidence, the weight of each antibody modality is calculated to generate modal influence.

[0028] Starting from the current antibody modality, we trace back along the fault propagation graph and combine the propagation topology coefficients to analyze the effect of antibody modalities on the path, thereby generating the source tracing strength.

[0029] Furthermore, based on antibody modalities and modal confidence, reverse reasoning is performed to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, thus obtaining the root cause contribution. This also includes:

[0030] Based on the source tracing strength of all antibody modalities, modality influence is used for fusion to generate fusion evidence values;

[0031] Based on the fused evidence value, the probability of each diagnostic point being the root cause of the current abnormal state is calculated, and the root cause contribution is generated.

[0032] Furthermore, based on diagnostic points where the root cause contribution exceeds a threshold, a key root cause set is constructed. The key root cause set and root cause contribution are calculated to generate fault warning instructions, including:

[0033] Diagnostic points whose root cause contribution exceeds the threshold are combined to obtain the key root cause set.

[0034] Based on the key root cause set, root cause contribution, and modality maturity, the degree of abnormal concentration at each diagnostic point is calculated to generate the root cause abnormal cohesion.

[0035] Furthermore, based on diagnostic points where the root cause contribution exceeds a threshold, a key root cause set is constructed. The key root cause set and root cause contribution are calculated to generate fault warning instructions, which also include:

[0036] Based on the root cause anomaly cohesion, the potential burst capability of each diagnostic point is calculated to generate the fault burst potential energy.

[0037] Based on the potential energy of the fault outbreak, the diagnostic points in the key root cause set are sorted and classified to generate fault warning instructions.

[0038] Furthermore, the mobile energy storage device fault diagnosis and early warning system, applied to the aforementioned mobile energy storage device fault diagnosis and early warning method, includes:

[0039] The data analysis unit is used to acquire the operating data of the target energy storage object, analyze the pre-processed real-time operating data, and obtain a multi-dimensional situation field. The target energy storage object is an outdoor energy storage power source.

[0040] The data anomaly calculation unit is used to calculate the multidimensional situation field, analyze the synaptic weight traces of the abnormal data, identify antibody modes, analyze the stability of each antibody mode, and obtain the mode confidence.

[0041] The data contribution calculation unit is used to perform reverse reasoning based on antibody modality and modality confidence, analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, and obtain the root cause contribution.

[0042] The fault early warning unit is used to construct a key root cause set based on diagnostic points where the root cause contribution exceeds a threshold, calculate the key root cause set and the root cause contribution, and generate a fault early warning command.

[0043] In summary, the present invention has the following main beneficial effects:

[0044] Step S1 transforms the operating parameters into dynamic cells, constructing a dynamic cellular network containing physical causality and temporal correlation. Combining cell synergy, field chaos index, and modal superposition, a multidimensional situation field is generated, enabling in-depth mining of multi-parameter correlations. This effectively captures the dynamic characteristics of fault evolution and solves the problem of traditional analysis lacking multivariate correlation capabilities. Step S2 uses the generated synaptic traces to identify stable antibody modes. Combining mode maturity and maturity volatility, mode confidence is calculated, which can filter out interference false signals, reduce false alarm rate, and clearly restore the path of abnormal propagation.

[0045] The root cause contribution is calculated by combining the fault propagation map constructed in step S3 with the propagation topology coefficient, modal influence degree and source tracing strength, so as to realize the root cause probability assessment of each diagnostic point. Step S4 filters the key root cause set, calculates the root cause anomaly cohesion degree and fault outbreak potential energy, and divides the warning level into three levels. It generates a fault warning instruction containing diagnostic points, warning level and fault outbreak potential energy, intuitively presenting the core and severity of the fault, and helping non-professional outdoor personnel to quickly locate the fault. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the fault diagnosis and early warning method for mobile energy storage devices according to the present invention.

[0047] Figure 2 This is a schematic diagram of the mobile energy storage device fault diagnosis and early warning system of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] refer to Figure 1 and Figure 2 Methods for fault diagnosis and early warning of mobile energy storage devices include:

[0050] Step S1: Obtain the operating data of the target energy storage object, analyze the preprocessed real-time operating data to obtain a multi-dimensional situation field, and the target energy storage object is an outdoor energy storage power source.

[0051] Operating data includes: voltage, current, power, charge / discharge efficiency, energy conversion efficiency, battery capacity, battery health status, battery internal resistance, battery cycle count, battery temperature, etc.

[0052] Step S2: Calculate the multidimensional situation field, analyze the synaptic weight traces of abnormal data, identify antibody modes, analyze the stability of each antibody mode, and obtain the mode confidence.

[0053] Step S3: Based on the antibody modality and modality confidence, reverse reasoning is performed to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, and the root cause contribution is obtained.

[0054] Step S4: Based on the diagnostic points where the root cause contribution exceeds the threshold, construct a key root cause set, calculate the key root cause set and the root cause contribution, and generate a fault warning instruction.

[0055] In one embodiment, the preprocessed real-time running data is analyzed to obtain a multi-dimensional situation field, including:

[0056] Each parameter in the real-time running data is regarded as a dynamic cell, and the physical causal and temporal correlation between the dynamic cells is analyzed to construct a dynamic cell network. The dynamic cell network is analyzed to generate cell synergy. Specifically, the following steps are taken: an initial network framework is established with each running parameter in the real-time running data as a node, where each parameter such as voltage, current, temperature, and internal resistance is a dynamic cell. Based on all cell pairs that may have physical causal relationships, a candidate edge set is formed. For each candidate edge in the set, the following analysis is performed: a sliding time window with a length of N consecutive sampling periods is extracted, the state sequences of dynamic cell X and dynamic cell Y within this window are obtained, and the maximum information coefficient between the two sequences is calculated. The maximum information coefficient ranges from 0 to 1, which represents the causal strength of this candidate edge.

[0057] Within the same sliding time window, the state sequence of cell X is shifted relative to the state sequence of cell Y by 0 to K (K=10) sampling periods, and the cross-correlation coefficient between the two sequences is calculated after each shift. The shift corresponding to the maximum cross-correlation coefficient is the specific delay time of this candidate edge. Thus, a dynamic cellular network containing nodes, edges and edge attributes is constructed, where nodes represent dynamic cells, edges represent the association between dynamic cells, and edge attributes represent causal strength and specific delay time.

[0058] Set an analysis period. For each sampling time within the period, calculate the difference between the real-time state value of each dynamic cell and the average real-time state value of the previous five sampling times to obtain the state deviation sequence of each cell.

[0059] Using all causal intensities as elements, construct an edge weight matrix, calculate the covariance between each pair of cells in the state deviation sequence during the analysis period, and form a state deviation covariance matrix; multiply the corresponding elements of the edge weight matrix and the state deviation covariance matrix to obtain the weighted covariance matrix.

[0060] The cellular covariance is calculated by dividing the maximum value of all eigenvalues ​​in the weighted covariance matrix by the sum of all eigenvalues ​​and normalizing the result to the 0-1 interval.

[0061] Based on a dynamic cellular network, the real-time data of each dynamic cell is mapped to a situation vector in a high-dimensional space. The initial situation field formed by all situation vectors is analyzed to obtain the field chaos index. Specifically, this includes: for each dynamic cell in the dynamic cellular network, its real-time state value at the current sampling moment is taken as the first core component of the situation vector; the causal intensity of all outgoing edges of the dynamic cell is arranged in a fixed order according to the target dynamic cell type (dynamic cell type such as current, temperature, etc.) to form the second component of the situation vector, namely the causal intensity sequence, where the outgoing edge is the edge pointing from this dynamic cell to other dynamic cells; the specific delay times of these outgoing edges are arranged in the same order to form the third component of the situation vector, namely the delay time sequence; the real-time state value, the causal intensity sequence, and the delay time sequence are concatenated to form the situation vector representing the multidimensionality of the dynamic cell.

[0062] During the analysis period, the set of situation vectors of all dynamic cells at each sampling moment is regarded as an instantaneous situation point in a high-dimensional space, and the situation points at all sampling moments together constitute the initial situation field.

[0063] Calculate the weighted Euclidean distance between any two situation vectors in the initial situation field, where the weights are the causal strengths on the corresponding edges in the dynamic cellular network, to obtain a distance matrix; divide the variance of the rate of change of the eigenvalues ​​of the distance matrix over five consecutive sampling periods by the mean of all variances, and normalize the calculation result to the 0-1 interval, which is the field chaos index.

[0064] In one embodiment, analyzing the preprocessed real-time running data to obtain a multi-dimensional situation field further includes:

[0065] The initial situation field is adjusted based on the field chaos index to generate a field mode sequence. Specifically, the field chaos index is used as an adjustment factor to perform weighted correction on the causal intensity sequence and the delayed time sequence of each situation vector in the initial situation field. The weight of the causal intensity sequence is 1 minus the field chaos index, and the weight of the delayed time sequence is the square root of the field chaos index.

[0066] All corrected situation vectors are arranged in chronological order. Each five consecutive sampling periods are divided into a time window. The mean of all corrected situation vectors within each time window is calculated. The vector formed by multiple means is a field mode. These field modes are arranged in the order of their corresponding time windows, and the resulting sequence is the field mode sequence.

[0067] Modal superposition based on field mode sequences generates a multidimensional situation field, specifically including: calculating the Euclidean distance between any two field modes in the field mode sequence to form a modal distance matrix;

[0068] Extract all non-repeating non-zero distance values ​​from the modal distance matrix to form a distance set. Calculate the mean and standard deviation of all distance values ​​in the distance set. Subtract the field chaos index + (squared field chaos index divided by 2) from 1 to obtain the adjustment coefficient. Multiply the adjustment coefficient by the standard deviation and then add it to the mean to obtain the similarity threshold.

[0069] For each distance value in the modal distance matrix, if it is less than or equal to the similarity threshold, the corresponding two field modes are considered to be similar and related, and are marked as 1 in the matrix; if it is greater than the similarity threshold, they are considered to be dissimilar and unrelated, and are marked as 0 in the matrix, thus obtaining a modal adjacency matrix composed of 0 and 1.

[0070] The adjacency matrix is ​​decomposed into eigenvalues, and the eigenvector corresponding to its largest eigenvalue is used as the superposition weight of each field mode. All field modes in the field mode sequence are multiplied by their corresponding superposition weights and then summed to obtain a weighted composite vector.

[0071] The weighted composite vector and the cell synergy degree are concatenated to form a new vector. This new vector is then processed by a hyperbolic tangent function, which smoothly compresses the value of each element in the new vector to the range of -1 to +1. The processed vector is the multidimensional situation field.

[0072] By transforming operating parameters into dynamic cells, a dynamic cellular network containing causal strength and specific delay time is constructed. Combined with cell synergy, multi-parameter correlation analysis is achieved, overcoming the shortcomings of univariate isolated analysis and generating a multidimensional situation field. This effectively uncovers the root cause of faults rather than just identifying a single fault, improving the accuracy of fault location and helping non-professional outdoor personnel to handle faults in a timely manner. At the same time, it enhances the comprehensiveness of diagnosis and the timeliness of early warning.

[0073] In one embodiment, the multidimensional situation field is calculated, the synaptic weight traces of anomalous data are analyzed, and antibody modalities are identified. The stability of each antibody modality is analyzed to obtain modality confidence, including:

[0074] The multidimensional situation field is calculated, the continuous change of field strength is converted, and the situation pulse sequence is obtained by combining the cell coordination degree. Specifically, the variance of the rate of change of each vector component in the multidimensional situation field within ten consecutive sampling periods is calculated, the variance is used as the instantaneous field strength, the historical instantaneous field strength of any 20 sampling periods is obtained, and the mean and standard deviation of the historical instantaneous field strength are calculated.

[0075] Multiply the cell coherence degree by the field chaos exponent, then multiply by the standard deviation of the historical instantaneous field strength, and then add it to the mean of the historical instantaneous field strength to obtain the dynamic field strength threshold.

[0076] When the instantaneous field strength exceeds the dynamic field strength threshold, a pulse signal is generated. The absolute value of the difference between the instantaneous field strength and the cell coordination degree is calculated, which is the amplitude of the pulse signal. All pulse signals are arranged in time sequence to form a situation pulse sequence.

[0077] The situation pulse sequence is analyzed to construct a pulse neural cluster and trigger a cooperative activation signal. Specifically, a sliding time window with a duration of five sampling periods is set. When at least three pulse signals appear within the sliding time window, the variance of the amplitude of these pulses is calculated. If the variance is less than 0.1, a pulse neural cluster is formed; otherwise, it is not. The mean of the amplitude of all pulses within the pulse neural cluster is calculated as the cluster strength. The cellular cooperativeness is added to the field chaos index and then divided by two to obtain the cooperative activation threshold.

[0078] When the cluster strength is greater than the co-activation threshold, the co-activation signal is triggered. The cluster strength is divided by the co-activation threshold to obtain the strength value of the co-activation signal.

[0079] Based on the co-activation signal, the abnormal propagation path of abnormal data is traced in reverse within the spiking neural cluster to obtain synaptic traces. Specifically, starting from the spiking neural cluster that triggered the co-activation signal, the sampling times corresponding to the three pulses with the highest cluster intensity are taken as key time points.

[0080] In dynamic cellular networks, starting from the abnormal dynamic cells corresponding to these sampling times, the source is traced along the opposite direction of the edges. For each reverse path, the causal strength of each edge on the path is multiplied by the strength value of the co-activation signal. The path segments with a product result > 0.6 are connected in reverse time order, and the sequence of cells passed through and their corresponding specific delay times are recorded. This sequence is the synaptic trace that characterizes the abnormal propagation path.

[0081] In one embodiment, the multidimensional situation field is calculated, the synaptic weight traces of anomalous data are analyzed, and antibody modalities are identified. The stability of each antibody modality is analyzed to obtain modality confidence. The method also includes:

[0082] Based on synaptic traces, each stable abnormal propagation path is identified and an antibody modality is generated. Specifically, this includes: collecting all synaptic traces over ten consecutive sampling periods, constructing a propagation path set from all cell sequences, calculating the standard deviation of a specific delay time between adjacent dynamic cells in each path, and marking paths with a standard deviation less than 0.05 as stable paths.

[0083] The longest common subsequence algorithm is used to calculate the sequence overlap between any two stable paths. Stable paths with a sequence overlap of more than 80% are grouped into the same category. In each category, the cell sequence with the highest frequency is taken as the typical path. The median of all specific delay times on the typical path is calculated as the standard delay. The combination of the typical path and the standard delay is taken as the antibody modality.

[0084] Modal stability analysis of antibody modalities is performed to obtain modal maturity. Specifically, this includes: counting the number of times the typical path corresponding to the antibody modality appears in all synaptic traces over the past ten sampling periods, and using this as the activation count; calculating the absolute value of the difference between the specific delay time of each segment on the typical path and the current measured delay time, and calculating the average of multiple absolute differences as the path delay deviation; dividing the activation count by the path delay deviation to obtain the initial maturity index, and normalizing the initial maturity index to the 0-1 interval, which is the modal maturity.

[0085] Based on modality maturity, the antibody modality is evaluated and modality confidence is generated. Specifically, this includes: calculating the standard deviation and mean of the modality maturity sequence of the antibody modality in the most recent five sampling periods, and dividing the standard deviation by the mean to obtain the maturity volatility.

[0086] Obtain the last cell coherence degree calculated within the analysis period, multiply the modal maturity by the last cell coherence degree, and then multiply by (1 - maturity volatility) to obtain the initial confidence index;

[0087] If the initial confidence index is less than 0.5, multiply the initial confidence index by 0.7, and then normalize the product to the 0-1 interval to obtain the modal confidence level; if the initial confidence index is greater than or equal to 0.5, multiply the initial confidence index by 1.3, and then normalize the product to the 0-1 interval to obtain the modal confidence level.

[0088] Specifically, when the initial confidence index is <0.5, it indicates that the antibody modality is likely a transient, atypical interference, or a false signal generated by noise. Therefore, the initial confidence index is reduced to 70% of its original value to filter out suspicious modalities with low credibility, prevent them from affecting subsequent root cause inference, and reduce the false alarm rate of faults. When the initial confidence index is ≥0.5, it indicates that the antibody modality has shown high maturity, good system synergy, and low volatility, which may correspond to a potential fault. Therefore, the initial confidence index is increased to 130% of its original value to strengthen it. The purpose is to increase the weight of these high-value, high-reliability modalities in subsequent analysis, ensure the capture of true fault precursors, and thus improve the accuracy and timeliness of fault warnings.

[0089] By generating a sequence of situational pulses, combining cellular synergy and field chaos index to construct pulsed neural clusters and trigger co-activation signals, dynamic anomalies involving multi-parameter coupling are accurately captured. Simultaneously, synaptic traces are generated through reverse tracing to clearly reconstruct the path of anomaly propagation. Stable antibody modalities are identified based on these traces, and modal confidence is calculated using modal maturity and maturity volatility, effectively filtering out interference and false signals and reducing false alarm rates. This allows for precise identification of the root cause of faults, helping non-professional outdoor personnel quickly locate faults and improving diagnostic accuracy and timely warnings.

[0090] In one embodiment, reverse reasoning is performed based on antibody modalities and modal confidence levels to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, thus obtaining the root cause contribution, including:

[0091] Based on antibody modes and field chaos indexes, diagnostic points in the target energy storage object are analyzed, and a fault propagation graph representing the abnormal propagation path is constructed to generate propagation topology coefficients. Specifically, based on dynamic cellular networks, nodes representing physical components such as battery packs, inverters, and BMS in the target energy storage object are used as diagnostic points, and connection edges are established from nodes to relevant cells: battery pack nodes are connected to dynamic cells such as voltage, current, and temperature; inverter nodes are connected to dynamic cells such as voltage, power, and energy conversion efficiency; and BMS nodes are connected to dynamic cells such as health status, internal resistance, and cycle count.

[0092] Multiply the causal strength of each edge in the typical path of each antibody modality by its corresponding modality confidence to obtain the weighted causal strength of the typical path. Using the dynamic cellular network as the basis, while maintaining all the original nodes and connections, integrate the weighted causal strength of each antibody modality onto the corresponding edge to form a fault propagation graph.

[0093] Based on each node in the fault propagation graph, a topological influence value is assigned to each node. The initial value of the topological influence value is 1, and the topological influence value is updated through an iterative process. Then, the propagation of the node in the fault propagation graph is analyzed. In each iteration, the topological influence value of each node is updated to the sum of the current topological influence values ​​of all its neighboring nodes. The contribution value of each neighboring node is equal to the current topological influence value of the neighboring node multiplied by the weighted causal strength of the edge connected to the current node, and then multiplied by (1 minus the field chaos exponent). The above iterative process is repeated until the change in the topological influence value of all nodes is less than one-thousandth. At this time, the topological influence value is considered to have reached stability.

[0094] After stabilization, the topological influence values ​​of each node are normalized to the range of 0-1, which are then used as the propagation topological coefficients for each diagnostic point.

[0095] Based on modal confidence, the weight of each antibody modality is calculated to generate modal influence. Specifically, this includes: squaring the modal confidence of all antibody modalities to obtain the squared confidence value of each modality, and calculating the arithmetic sum of these squared confidence values; then dividing the squared confidence value of each antibody modality by the arithmetic sum to obtain the initial influence weight of that modality; and multiplying the initial influence weight of each modality by its corresponding modal confidence to obtain the modal influence.

[0096] Starting from the current antibody modality, the system traces backward along the fault propagation graph and analyzes the effect of the antibody modality on the path using the propagation topology coefficient to generate the source tracing strength. Specifically, this involves: taking the typical path of the antibody modality as the starting point, traversing backward along the fault propagation graph to each node; for each node, calculating the product of its propagation topology coefficient and the modality influence of the current antibody modality as the path contribution value of that node; when the propagation topology coefficient > modality influence, multiplying the contribution value by 1.2 for reinforcement; when the propagation topology coefficient ≤ modality influence, multiplying the contribution value by 0.8 for attenuation; summing the adjusted contribution values ​​of all nodes on the path and normalizing the result to the 0-1 interval to obtain the source tracing strength.

[0097] In one embodiment, reverse reasoning is performed based on antibody modalities and modal confidence levels to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, thus obtaining the root cause contribution. This also includes:

[0098] Based on the source tracing strength of all antibody modalities, modality influence is used for fusion to generate fusion evidence values. Specifically, this includes: multiplying the source tracing strength of each antibody modality by its corresponding modality influence to obtain the weighted source tracing strength of each modality; calculating the mean of all weighted source tracing strengths as the benchmark value; squaring the difference between each weighted source tracing strength and the benchmark value, and normalizing the calculation result to the 0-1 interval, which is the fusion evidence value.

[0099] Based on the fusion evidence value, the probability of each diagnostic point being the root cause of the current abnormal state is calculated, and the root cause contribution is generated. Specifically, this includes: constructing a diagnostic point-antibody modality association matrix, where each element in the diagnostic point-antibody modality association matrix is ​​the number of times the diagnostic point appears on the reverse tracing path of the typical path of the corresponding antibody modality multiplied by the propagation topology coefficient.

[0100] For each antibody modality, its fusion evidence value is multiplied by the modality influence to obtain the total evidence quantity of that antibody modality. This total evidence quantity is then allocated to each diagnostic point according to the proportion of each element value in the correlation matrix. The evidence quantities allocated to all antibody modalities at each diagnostic point are summed to obtain the initial contribution value. The initial contribution values ​​of all diagnostic points are normalized to the 0-1 interval, which is the root cause contribution of each diagnostic point.

[0101] By constructing a fault propagation graph of associated diagnostic points and dynamic cells, integrating weighted causal strength and field chaos index to iteratively generate propagation topology coefficients, and combining modal confidence to calculate modal influence, the path contribution value is adjusted by tracing back along the fault propagation graph and the propagation topology coefficients to generate source strength and fusion evidence value. Finally, the root cause contribution is accurately derived, breaking through the limitations of traditional univariate isolated analysis, realizing multi-parameter coupling correlation and deep causal inference, and solving the problem that traditional methods can only identify a single fault and cannot locate the root cause. Among them, the root cause contribution intuitively quantifies the root cause probability of each diagnostic point, helping non-professional outdoor personnel to quickly locate the core fault and improve the accuracy of fault diagnosis.

[0102] In one embodiment, based on diagnostic points where the root cause contribution exceeds a threshold, a key root cause set is constructed. The key root cause set and root cause contribution are calculated to generate a fault warning instruction, including:

[0103] Diagnostic points whose root cause contribution exceeds a threshold are grouped to obtain a key root cause set. Specifically, this involves: adding the average root cause contribution of all diagnostic points to the standard deviation of the root cause contribution to obtain the contribution threshold; using diagnostic points whose root cause contribution exceeds the contribution threshold as candidate root causes, calculating the interaction strength between any two candidate root causes in the fault propagation graph: dividing the product of the propagation topology coefficients of the two diagnostic points by their average distance in the antibody modality tracing path to obtain the interaction strength; when the interaction strength is greater than 0.6, the corresponding diagnostic points are merged into a root cause combination; multiplying the sum of the root cause contributions of all diagnostic points within the root cause combination by the average interaction strength to obtain the density index of each root cause combination; and forming a key root cause set by the top three root cause combinations with the highest density index. If there are fewer than three root cause combinations, all root cause combinations are sorted by density index and included in the key root cause set.

[0104] Based on the key root cause set, root cause contribution, and modality maturity, the degree of anomalous concentration for each diagnostic point is calculated to generate root cause anomalous cohesion. Specifically, this includes: calculating the average interaction strength between each diagnostic point and all other diagnostic points in the key root cause set, multiplying the average interaction strength by the root cause contribution of the diagnostic point to obtain an initial cohesion value; obtaining the modality maturity of all antibody modalities involved in the diagnostic point, and calculating the geometric mean of these modality maturity values ​​as a maturity factor; multiplying the initial cohesion value by the maturity factor, and then multiplying by the frequency of the diagnostic point in the key root cause set to obtain the original cohesion; and normalizing the original cohesion of all diagnostic points to the 0-1 interval to obtain the root cause anomalous cohesion of each diagnostic point.

[0105] In one embodiment, based on diagnostic points where the root cause contribution exceeds a threshold, a key root cause set is constructed. The key root cause set and root cause contribution are calculated to generate a fault warning instruction. The method also includes:

[0106] Based on the root cause anomaly cohesion, the potential burst capability of each diagnostic point is calculated to generate the fault burst potential energy. Specifically, this includes: multiplying the root cause anomaly cohesion of the diagnostic point by its propagation topology coefficient, and then multiplying by the square of the field chaos exponent to obtain the basic potential energy value; using the weighted causal intensity sum of all edges of the diagnostic point in the fault propagation graph as the influence range coefficient; and multiplying the basic potential energy value by the influence range coefficient and then by the reciprocal of the cell synergy degree to obtain the fault burst potential energy.

[0107] Based on the fault outbreak potential energy, the diagnostic points in the key root cause set are sorted and classified to generate fault early warning instructions. Specifically, this includes: sorting all diagnostic points in the key root cause set according to their fault outbreak potential energy from high to low, and then dividing the diagnostic points into three early warning levels according to the fault outbreak potential energy: when the fault outbreak potential energy is >0.8, it is a level 1 early warning; when the fault outbreak potential energy is between 0.5 and 0.8, it is a level 2 early warning; and when the fault outbreak potential energy is <0.5, it is a level 3 early warning.

[0108] This generates a triplet sequence containing diagnostic points, warning levels, and fault outbreak potential energy, forming a complete fault warning instruction. The diagnostic point here is the final fault point.

[0109] By screening candidate root causes and constructing a key root cause set based on interaction intensity, and then integrating root cause contribution and modal maturity to generate root cause anomaly cohesion, the potential energy of fault outbreak is calculated based on propagation topology coefficients and field chaos indexes, and three-level early warning levels are divided to generate fault early warning instructions. This approach breaks through the limitations of single-variable static threshold analysis, realizes the aggregation and in-depth mining of fault root causes under multi-parameter coupling, accurately quantifies the severity and risk level of faults, and can then accurately locate the core root cause of the fault and display diagnostic points, allowing non-professional outdoor personnel to determine the direction of repair, thus improving the efficiency of diagnosis and the accuracy of early warning.

[0110] In one embodiment, the mobile energy storage device fault diagnosis and early warning system, applied to the aforementioned mobile energy storage device fault diagnosis and early warning method, includes:

[0111] The data analysis unit is used to acquire the operating data of the target energy storage object, analyze the pre-processed real-time operating data, and obtain a multi-dimensional situation field. The target energy storage object is an outdoor energy storage power source.

[0112] The data anomaly calculation unit is used to calculate the multidimensional situation field, analyze the synaptic weight traces of the abnormal data, identify antibody modes, analyze the stability of each antibody mode, and obtain the mode confidence.

[0113] The data contribution calculation unit is used to perform reverse reasoning based on antibody modality and modality confidence, analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, and obtain the root cause contribution.

[0114] The fault early warning unit is used to construct a key root cause set based on diagnostic points where the root cause contribution exceeds a threshold, calculate the key root cause set and the root cause contribution, and generate a fault early warning command.

[0115] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for fault diagnosis and early warning of mobile energy storage devices, characterized in that, include: Step S1: Obtain the operating data of the target energy storage object, regard each parameter in the real-time operating data as a dynamic cell, analyze the physical causal and temporal correlation between each dynamic cell, construct a dynamic cell network, analyze the dynamic cell network, and generate cell synergy degree. Based on a dynamic cellular network, the real-time data of each dynamic cell is mapped to a situation vector in a high-dimensional space. The initial situation field formed by all situation vectors is analyzed to obtain the field chaos index. The initial situation field is adjusted based on the field chaos index to generate a field mode sequence. The field mode sequence is then superimposed to generate a multidimensional situation field. The target energy storage object is an outdoor energy storage power source. Step S2 involves calculating the multidimensional situation field, converting the continuous changes in field strength, and analyzing them in conjunction with cell coordination degree to obtain a situation pulse sequence. This includes: calculating the variance of the rate of change of each vector component in the multidimensional situation field within a continuous sampling period as the instantaneous field strength; obtaining historical instantaneous field strengths and calculating their mean and standard deviation; co-calculating the cell coordination degree, field chaos index, historical instantaneous field strength standard deviation, and historical instantaneous field strength mean to obtain the dynamic field strength threshold; then co-analyzing the instantaneous field strength, dynamic field strength threshold, and cell coordination degree to obtain the amplitude of the pulse signal; and arranging all pulse signals in time sequence to form a situation pulse sequence. The situation pulse sequence is analyzed to construct a pulse neural cluster and trigger a co-activation signal; Based on co-activation signals, the abnormal propagation path of abnormal data is traced in reverse within a spike neural cluster to obtain synaptic traces. Based on synaptic traces, each stable anomalous propagation path is identified, and an antibody modality is generated. This includes: analyzing synaptic traces within a continuous sampling period to obtain a set of propagation paths; marking paths with a standard deviation of less than 0.05 for a specific delay time between adjacent dynamic cells as stable anomalous propagation paths; calculating the sequence overlap between any two stable anomalous propagation paths; classifying stable anomalous propagation paths based on the sequence overlap; identifying typical paths in each category; calculating the median of all specific delay times on the typical path as the standard delay; and using the combination of the typical path and the standard delay as the antibody modality. Modal stability analysis of antibody modalities yields modal maturity, including: counting the number of times the typical path corresponding to the antibody modality appears in all synaptic traces within historical sampling periods to obtain activation counts; calculating the path delay deviation by comparing the specific delay time of each segment on the typical path with the current delay time; dividing the activation counts by the path delay deviation and normalizing the result to obtain modal maturity; and evaluating the antibody modality based on the modal maturity to generate modal confidence. Step S3: Based on the antibody modality and modality confidence, perform reverse reasoning to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, and obtain the root cause contribution. Step S4: Based on the diagnostic points where the root cause contribution exceeds the threshold, construct a key root cause set, calculate the key root cause set and the root cause contribution, and generate a fault warning instruction.

2. The method for fault diagnosis and early warning of mobile energy storage devices according to claim 1, characterized in that, Based on antibody modalities and modal confidence, reverse reasoning is performed to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, and the root cause contribution is obtained, including: Based on antibody modes and field chaos index, the diagnostic points in the target energy storage object are analyzed, and a fault propagation graph representing the abnormal propagation path is constructed to generate propagation topology coefficients. Based on modal confidence, the weight of each antibody modality is calculated to generate modal influence. Starting from the current antibody modality, we trace back along the fault propagation graph and combine the propagation topology coefficients to analyze the effect of antibody modalities on the path, thereby generating the source tracing strength.

3. The method for fault diagnosis and early warning of mobile energy storage devices according to claim 2, characterized in that, Based on antibody modalities and modal confidence, reverse reasoning is performed to analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, thus obtaining the root cause contribution. This also includes: Based on the source tracing strength of all antibody modalities, modality influence is used for fusion to generate fusion evidence values; Based on the fused evidence value, the probability of each diagnostic point being the root cause of the current abnormal state is calculated, and the root cause contribution is generated.

4. The method for fault diagnosis and early warning of mobile energy storage devices according to claim 3, characterized in that, Based on diagnostic points where the root cause contribution exceeds a threshold, a key root cause set is constructed. The key root cause set and root cause contribution are calculated to generate fault warning instructions, including: Diagnostic points whose root cause contribution exceeds the threshold are combined to obtain the key root cause set. Based on the key root cause set, root cause contribution, and modality maturity, the degree of abnormal concentration at each diagnostic point is calculated to generate the root cause abnormal cohesion.

5. The method for fault diagnosis and early warning of mobile energy storage devices according to claim 4, characterized in that, Based on diagnostic points where the root cause contribution exceeds a threshold, a key root cause set is constructed. The key root cause set and root cause contribution are calculated to generate fault warning instructions. This also includes: Based on the root cause anomaly cohesion, the potential burst capability of each diagnostic point is calculated to generate the fault burst potential energy. Based on the potential energy of the fault outbreak, the diagnostic points in the key root cause set are sorted and classified to generate fault warning instructions.

6. A mobile energy storage device fault diagnosis and early warning system, applied to the mobile energy storage device fault diagnosis and early warning method according to any one of claims 1-5, characterized in that, include: The data analysis unit is used to acquire the operating data of the target energy storage object, analyze the pre-processed real-time operating data, and obtain a multi-dimensional situation field. The target energy storage object is an outdoor energy storage power source. The data anomaly calculation unit is used to calculate the multidimensional situation field, analyze the synaptic weight traces of anomalous data, identify antibody modes, analyze the stability of each antibody mode to obtain mode maturity, and evaluate the antibody modes based on the mode maturity to obtain mode confidence. The data contribution calculation unit is used to perform reverse reasoning based on antibody modality and modality confidence, analyze the probability that each diagnostic point in the target energy storage object is the root cause of the current abnormal state, and obtain the root cause contribution. The fault early warning unit is used to construct a key root cause set based on diagnostic points where the root cause contribution exceeds a threshold, calculate the key root cause set and the root cause contribution, and generate a fault early warning command.

Images