Method, system and device for quickly positioning energy storage fault high-voltage box and storage medium

Abstract

The invention discloses a method, system and device for rapidly positioning an energy storage fault high-voltage box and a storage medium, and relates to the technical field of fault diagnosis of a battery management system, and the method comprises the steps: obtaining a battery cluster current parameter in real time, and carrying out the dual-threshold judgment, and obtaining a state identifier; acquiring a current sensor parameter, and calculating a dynamic threshold parameter; constructing and adjusting a sensor-fault incidence matrix based on the state identifier in combination with historical fault data; based on the state identifier and the battery cluster current parameter, extracting spatial features and time features, and fusing to obtain a joint feature vector; calculating the probability distribution of the fault type based on the joint feature vector, and obtaining a fault diagnosis result in combination with the state identifier and a preset judgment condition; performing visual display based on the dynamic threshold parameter and the fault diagnosis result; the system state can be accurately identified, different working conditions are adapted, the diagnosis efficiency, accuracy and reliability are improved, and stable operation of the battery management system is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of battery management system fault diagnosis technology, and in particular to a method, system, device and storage medium for quickly locating faulty high-voltage boxes in energy storage systems. Background Technology

[0002] The main faults currently found in high-voltage energy storage boxes include: insulation failure, fuse blown, contactor failure, BCU (Battery Control Unit) failure, BMS (Battery Management System) communication failure, and control power module malfunction. The main symptoms are: insulation failure: the system reports a "low insulation resistance" alarm; leakage current exceeds the safety threshold; fuse blown: the system suddenly loses power and cannot recover; fuse surface is burned or the indicator pops out; contactor failure: the high-voltage box cannot open or close normally; contactor coil overheats or burns out; the system reports a "contactor status abnormality" alarm; BMS communication failure: communication between the BMS and BMU (Battery Management Unit) or BAU (Battery Array Controller). Unit) communication failure, resulting in the inability to report battery status or receive control commands; BMS battery control unit failure: intermittent single cell voltage or two consecutive single cell voltage abnormalities (such as displaying 0V or jumping); control power module failure: unstable output voltage or voltage directly returning to zero, the system reports "power supply abnormality" or "power supply damage".

[0003] Currently, the main methods for fault location are as follows: current analysis, insulation detection loop method, real-time temperature and humidity monitoring, segmented insulation inspection, and high-voltage interlock (HVIL) detection. Current difference analysis primarily monitors the current difference between adjacent battery modules within the same battery cluster, combining this with a preset threshold to determine the fault node. If the difference exceeds the threshold, the fault location is quickly determined based on the identification information, making it suitable for short-circuit fault detection. The insulation detection loop method involves arranging multiple detection loops inside the high-voltage box, accurately locating insulation failure points by measuring changes in the conductivity between each network node and the grounding terminal. For example, a time-division multiplexing detection branch method is used, combined with processor analysis of abnormal insulation resistance locations. Real-time temperature and humidity monitoring mainly utilizes temperature sensors (such as those monitoring high-voltage connectors and acquisition board temperatures) and humidity sensors, combined with algorithms to determine the risk of thermal runaway or insulation failure. When temperature and humidity are abnormal, a protection mechanism is triggered and the fault area is marked. The segmented insulation inspection method involves segmenting the high-voltage circuit for testing. Workers use an insulation resistance tester to measure the resistance between the positive and negative poles and ground segment by segment. If the resistance of a segment is lower than the standard, that segment is identified as the fault point. High-voltage interlock detection monitors the integrity of high-voltage connectors through a low-voltage signal circuit. If the interlock circuit breaks, the system immediately cuts off the high voltage and reports the fault location, commonly used in scenarios involving loose plugs or abnormal contactors. As the capacity of energy storage power stations gradually increases, segmented insulation checks are not only costly and inefficient but also pose a risk of personal injury. With increasing equipment capacity, a single fault location method is insufficient to meet the safety and stability requirements of large-capacity energy storage power stations. As the scale and application scope of independent energy storage power stations expand, the number of daily charge / discharge cycles increases, leading to a decline in the electrical, thermal, and mechanical performance of the high-voltage box. A high-voltage box failure can prevent individual energy storage units from participating in charging and discharging, severely impacting normal power station operation. Real-time tracking of high-voltage box parameters allows for the detection of potential defects and proactive intervention, effectively preventing revenue declines due to equipment failures. Therefore, timely and effective prediction, location, and tracing of high-voltage box faults in energy storage power stations are crucial. However, the current traditional high-voltage box fault location methods for energy storage power stations have the following problems: the diagnostic methods are prone to false alarms when a certain parameter exceeds the threshold due to the selection of a single parameter; the correlation between parameters is not effectively utilized, resulting in insufficient fault location accuracy, and maintenance personnel cannot find the fault point in time, affecting the efficiency of fault troubleshooting. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is: in the context of the gradual increase in the capacity of energy storage power stations, the complexity and diversity of fault types, and the difficulty in meeting the needs of a single fault location method, how to achieve accurate and efficient diagnosis and location of multiple types of faults such as insulation failure of energy storage high-voltage boxes and fuse blowing, reduce labor costs and fault misjudgment and omission rate, improve system reliability and adaptability, and achieve timely early warning and intuitive visualization display.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for rapidly locating a faulty high-voltage box in energy storage, comprising:

[0008] As a preferred method for quickly locating faulty high-voltage boxes in energy storage systems, the following is provided:

[0009] The real-time acquisition of battery cluster current parameters and the performance of dual threshold determination to obtain status indicators include:

[0010] Establish a current intensity time series window to monitor the current parameters of the battery cluster in real time;

[0011] The system determines whether the absolute value of the current is continuously lower than the silent threshold. When the current value is less than the set value for a set number of consecutive sampling periods, the system is determined to be in standby mode.

[0012] The beneficial effects of this preferred technical solution are as follows: by establishing a time series window to monitor current parameters in real time, and by comparing the absolute value of the current with the silent threshold to determine the standby state, the standby status of the system can be accurately identified, providing a basic status judgment for subsequent fault diagnosis, avoiding unnecessary fault diagnosis operations in the standby state, and improving diagnostic efficiency and accuracy.

[0013] As a preferred method for quickly locating faulty high-voltage boxes in energy storage systems, the following is provided:

[0014] The process of acquiring battery cluster current parameters in real time and performing dual threshold determination to obtain a status identifier also includes:

[0015] The direction and polarity of the current are used to determine whether it is in a discharging state or a charging state.

[0016] The beneficial effects of this preferred technical solution are as follows: using the polarity of the current direction to determine the charging and discharging state is simple, direct and effective, enabling the system to clearly distinguish the different working states of the battery cluster, thereby providing a basis for setting fault diagnosis and judgment conditions based on different working states, and enhancing the pertinence and accuracy of fault diagnosis.

[0017] As a preferred method for quickly locating faulty high-voltage boxes in energy storage systems, the following is provided:

[0018] The acquisition of current sensor parameters and calculation of dynamic threshold parameters include:

[0019] Establish a sliding data window to continuously record current sensor parameters and calculate the moving average and standard deviation of each parameter in real time;

[0020] When the sliding data window is not full, an initial threshold is generated through a three-layer perceptron network; when the sliding data window is full, a dynamic threshold range is generated based on the mean and standard deviation.

[0021] The beneficial effects of this preferred technical solution are as follows: by using a sliding data window to record parameters and calculate the moving average and standard deviation, and combining it with a three-layer perceptron network to generate an initial threshold and a dynamic threshold range, the threshold can be dynamically adjusted according to the real-time operating data of the system to adapt to different operating conditions, thereby improving the sensitivity and accuracy of fault detection and avoiding misjudgment or missed judgment that may occur under complex operating conditions due to fixed thresholds.

[0022] As a preferred method for quickly locating faulty high-voltage boxes in energy storage systems, the following is provided:

[0023] The process of constructing and adjusting the sensor-fault correlation matrix based on status identifiers and historical fault data includes:

[0024] A sensor-fault correlation matrix with a specific dimension is preset. The rows of the matrix correspond to the sensors, and the columns correspond to the fault types. The initial weights of the matrix elements are set based on historical fault case data.

[0025] For specific fault detection, an enhanced detection module is set up to continuously track specific sensor parameters; when the enhanced detection module detects that specific parameter conditions are met, the correlation weight between the fault and related sensors is enhanced by a product factor.

[0026] By utilizing an attention mechanism network, combining current sensor readings and the characteristics of the correlation matrix, dynamic weight coefficients with state adaptability are generated, enabling dynamic adjustment of the sensor-fault correlation matrix.

[0027] The beneficial effects of this preferred technical solution are as follows: By presetting the correlation matrix and setting initial weights based on historical fault data, basic correlation information is provided for fault diagnosis. The enhanced detection module settings and the enhanced correlation weight mechanism can highlight the correlation between relevant sensors and faults under specific conditions, improving the detection capability of specific faults. The dynamic weight coefficients generated by the attention mechanism network can adaptively adjust the correlation matrix according to the real-time status of the system, improving the accuracy and flexibility of fault diagnosis and better coping with complex and ever-changing fault situations.

[0028] As a preferred method for quickly locating faulty high-voltage boxes in energy storage systems, the following is provided:

[0029] The method of extracting spatial and temporal features based on state identifiers and battery cluster current parameters, and fusing them to obtain a joint feature vector, includes:

[0030] A fully connected topology is constructed using a specific network structure, with battery cluster current parameters as node features and state identifiers as edge attribute features.

[0031] By utilizing the multi-layer structure of the network, spatial features of a certain dimension are extracted in the first layer and multi-head attention is performed. The features are then aggregated in the second layer to obtain spatial features of a specific dimension. A time series processing network with a specific structure is used to process the time series of sensor data through a gating mechanism to extract time features of a specific dimension.

[0032] The extracted spatial and temporal features are merged, and feature interaction is achieved through a multilayer perceptron to output a joint feature vector of a specific dimension.

[0033] As a preferred method for quickly locating faulty high-voltage boxes in energy storage systems, the following is provided:

[0034] The probability distribution of fault types calculated based on joint feature vectors, combined with status identifiers and preset judgment conditions, yields the following fault diagnosis results:

[0035] Based on the fused joint feature vector, the feature vector is processed by a dual-channel activation function: the first layer compresses the features to eliminate feature redundancy, and the last layer performs a linear transformation to output the fault probability distribution corresponding to multiple fault types, and normalizes the probability distribution.

[0036] Based on the status identifier, dual judgment conditions are set. In one state, the fault type that meets certain probability conditions and has the highest probability is selected as the diagnosis result; in the other state, all fault probabilities are forcibly reduced to zero and the state is judged as normal.

[0037] To address conflicts caused by multiple concurrent faults, a probability difference threshold is set; when the difference between the highest probability and the second highest probability is less than the probability difference threshold, a review warning is triggered.

[0038] Secondly, the present invention provides a system for rapidly locating faulty high-voltage boxes in energy storage systems, comprising:

[0039] The status identifier acquisition module is used to acquire battery cluster current parameters in real time and perform dual threshold judgment to obtain status identifiers.

[0040] The dynamic threshold calculation module is used to acquire current sensor parameters and calculate dynamic threshold parameters.

[0041] The correlation matrix construction and adjustment module is used to construct and adjust the sensor-fault correlation matrix based on the status identifier and combined with historical fault data.

[0042] The joint feature vector acquisition module is used to extract spatial and temporal features based on state identifiers and battery cluster current parameters, and fuse them to obtain a joint feature vector;

[0043] The fault diagnosis module is used to calculate the probability distribution of fault types based on joint feature vectors, and obtain fault diagnosis results by combining status identifiers and preset judgment conditions.

[0044] The visualization module is used to visualize the results based on dynamic threshold parameters and fault diagnosis.

[0045] Thirdly, the present invention provides an electronic device, comprising:

[0046] Memory and processor;

[0047] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the method for quickly locating faulty high-voltage boxes in energy storage as described in this invention.

[0048] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method for rapidly locating faulty high-voltage boxes in energy storage systems.

[0049] The beneficial effects of this invention are as follows: By analyzing potential faults in the high-voltage box of an energy storage power station, this invention establishes the influence relationship between major fault types and sensor parameters, and introduces physical prior-guided fault reasoning. The weight of the fault-sensor association is dynamically adjusted by physical priors and an attention mechanism, the latter of which can dynamically adjust the prior weights of the former. It emphasizes a multi-modal feature fusion architecture, utilizing joint spatial-temporal modeling; it employs a dynamic threshold generation mechanism, setting an adaptive statistical threshold model; and it adopts a state-adaptive diagnostic strategy: emphasizing computational resource optimization, executing the complete diagnostic process only in active states (charging / discharging), disabling high-order feature calculations in standby mode, and setting fault probability and dynamic thresholds to zero; a noise suppression mechanism automatically shields sensor noise in standby mode to avoid false alarms; and it designs an abnormal situation response plan, adopting a data visualization scheme to achieve health status assessment of key equipment in the energy storage power station, improving risk warning, fault tracing, and intelligent decision-making capabilities, and enhancing the operation and maintenance technical support level and system safety and stability level of the energy storage power station. In actual energy storage power station operation and maintenance scenarios, the above improvements of this invention can be directly translated into significant economic and social benefits. By improving the accuracy and timeliness of fault diagnosis, downtime and maintenance costs caused by faults are reduced, minimizing potential losses from safety accidents. Optimized resource utilization and lower operation and maintenance costs make energy storage power station operations more efficient and economical. Enhanced fault tracing capabilities and improved operation and maintenance efficiency shorten fault repair time and ensure the continuity of power supply. Visualized management and intelligent decision-making improve the overall management level of the power station, contributing to the refined operation of energy storage power stations. Adapting to complex environments improves system stability, ensuring reliable operation of energy storage power stations under various harsh conditions, providing solid technical support for the large-scale storage and utilization of new energy sources, and promoting the healthy development of the energy storage industry. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is an overall flowchart of the method for quickly locating faulty high-voltage boxes in energy storage provided by the present invention. Detailed Implementation

[0052] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0053] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for rapidly locating a faulty high-voltage box in energy storage, comprising:

[0054] S1: Real-time acquisition of battery cluster current parameters and dual threshold determination to obtain status identifier;

[0055] S2: Obtain current sensor parameters and calculate dynamic threshold parameters;

[0056] S3: Based on status identifiers and combined with historical fault data, construct and adjust the sensor-fault correlation matrix;

[0057] S4: Based on the state identifier and battery cluster current parameters, extract spatial and temporal features and fuse them to obtain a joint feature vector;

[0058] S5: Calculate the probability distribution of fault types based on joint feature vectors, and obtain fault diagnosis results by combining status identifiers and preset judgment conditions;

[0059] S6: Visualize the results based on dynamic threshold parameters and fault diagnosis.

[0060] It should be noted that through steps S1-S6, a complete and efficient fault diagnosis and display system for battery clusters in energy storage power stations was constructed. Starting with real-time data acquisition, this system comprehensively utilizes advanced methods such as dynamic threshold generation, correlation matrix construction, and multi-feature fusion to accurately diagnose battery cluster faults and present the results intuitively through visualization. This system significantly improves the accuracy, timeliness, and intelligence of fault diagnosis in energy storage power stations, providing strong technical support for the safe and stable operation, efficient maintenance, and risk prevention of energy storage power stations, and powerfully promoting the development of energy storage power station operation and maintenance management towards a more scientific, intelligent, and reliable direction.

[0061] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the previous embodiment, a method for rapidly locating a faulty high-voltage box in energy storage is provided, comprising:

[0062] In this embodiment, the real-time acquisition of battery cluster current parameters and the performance of dual threshold determination in step S1 above, resulting in a status identifier, include:

[0063] A current intensity time series window is established to monitor battery cluster current parameters in real time. A dual threshold determination mechanism is set: the system is determined to be in standby mode if the absolute value of the current is continuously lower than the quiescent threshold and the current value is less than 0.1 amperes for 10 consecutive sampling periods; secondly, based on the polarity of the current direction, positive current is determined to be in the discharging state and negative current is determined to be in the charging state.

[0064] It should be noted that the obtained status identifier will serve as the condition judgment benchmark for all subsequent diagnostic steps, directly affecting the feature extraction strategy and the selection of diagnostic logic branches. That is, the complete diagnostic process is started in the charging / discharging state, while only basic monitoring is performed in the standby state.

[0065] In this embodiment, the acquisition of current sensor parameters and calculation of dynamic threshold parameters in step S2 above include:

[0066] A sliding data window with 100 sampling periods was established to continuously record six types of sensor parameters (insulation resistance, relay temperature, cluster current, cluster voltage, control power supply voltage, and BMS temperature). The moving average and standard deviation of each parameter were calculated in real time to provide basic statistics for dynamic threshold generation. An exponential decay method was used to address the initial data insufficiency problem before the sliding data window was filled.

[0067] Furthermore, the dynamic threshold calculation adopts a dual-mode approach: in the early stage of data accumulation (when the window is not full), an initial threshold is generated through a three-layer perceptron network; when the data is sufficient (when the statistical window is full), a dynamic threshold range is generated based on the mean ± 2.5 times the standard deviation.

[0068] It should be noted that the sensor's dynamic threshold parameter is adaptively adjusted according to the system's operating status, serving as a benchmark reference value for subsequent anomaly detection.

[0069] In this embodiment, step S3 above, which involves constructing and adjusting the sensor-fault correlation matrix based on the status identifier and historical fault data, includes:

[0070] A preset 6×6 dimensional sensor-fault correlation matrix is ​​used, where rows correspond to six sensor types and columns correspond to six fault types. The matrix element values ​​(initial weights) are set based on historical fault case data.

[0071] For example, the initial correlation between insulation failure and insulation resistance is set to 0.92, and the initial correlations between fuse failure and cluster current sensor and relay temperature sensor are 0.75 and 0.15, respectively.

[0072] To detect BMS communication faults, an enhanced detection module is implemented to continuously track the cluster voltage standard deviation and cluster current intensity. When a voltage fluctuation rate below 0.05 volts and a current consistently above 0.1 amperes are detected, the association weights between the BMS communication fault and the current and voltage sensors are amplified by a factor of 3. An attention mechanism network integrates current sensor readings with the characteristics of the association matrix to generate dynamic weight coefficients with state adaptability.

[0073] In another possible implementation, reinforcement learning algorithms can be introduced. Using the system's operating state and fault diagnosis results as environmental feedback, the reinforcement learning agent continuously adjusts the weights of the sensor-fault association matrix to adapt to the system's dynamic changes. For example, when the system frequently experiences a certain type of fault, the reinforcement learning agent can automatically increase the association weights between relevant sensors and that fault type.

[0074] In this embodiment, step S4 above, based on the state identifier and battery cluster current parameters, extracts spatial and temporal features and fuses them to obtain a joint feature vector, including:

[0075] A fully connected topology with 36 edges is constructed using a graph attention network (GAT), with battery cluster current parameters as node features and state identifiers as edge attribute features.

[0076] The first layer of the network extracts 32-dimensional spatial features and performs four-head attention calculation. The second layer of the network aggregates the features and finally outputs 128-dimensional spatial features.

[0077] The sensor data time series is processed using a bidirectional recursive structure via a Long Short-Term Memory (LSTM) network, and 128-dimensional temporal features are captured through a gating mechanism. The feature fusion layer merges the 128-dimensional spatial features with the 128-dimensional temporal features, and the feature interaction is realized through a multilayer perceptron, finally outputting a 256-dimensional joint feature vector.

[0078] In another possible implementation, a Graph Convolutional Network (GCN) can be used instead of a Graph Attention Network (GAT) for spatial feature extraction. GCNs can process graph-structured data more efficiently, automatically extracting spatial relationship features between nodes through convolution operations. Furthermore, the number of layers and the size of the convolutional kernels can be adjusted to optimize the spatial feature extraction performance.

[0079] In another possible implementation, the Transformer architecture can be used instead of the Long Short-Term Temporal Memory (LSTM) network for temporal feature extraction. The Transformer architecture has stronger parallel computing capabilities and long sequence processing capabilities, enabling it to better capture temporal dependencies in sensor data. Through a multi-head self-attention mechanism, the Transformer can weighted aggregate information from different time steps to extract richer temporal features.

[0080] In another possible implementation, a weighted fusion approach can be used during feature fusion. Different weights are assigned to spatial and temporal features based on their importance, and then the features are summed using weighted methods. The weights can be optimized using cross-validation to find the best fusion result.

[0081] In this embodiment, the probability distribution of fault types calculated based on joint feature vectors in step S5 above, combined with status identifiers and preset judgment conditions, yields the following fault diagnosis results:

[0082] The classifier network receives the fused feature vectors and processes them using a dual-channel activation function: the first layer uses the hyperbolic tangent function for feature compression, outputting 64 dimensions and eliminating feature redundancy; the last layer uses linear transformation to output a six-dimensional fault probability distribution, and uses the softmax function to achieve probability normalization.

[0083] Based on status indicators, a dual judgment condition is set: in the charging / discharging state, the fault type with the highest probability value exceeding 0.5 is selected as the diagnostic result; in the standby state, all fault probabilities are forcibly reduced to zero and the system is judged as normal. For situations where multiple faults cause conflicts, a probability difference threshold is set. When the difference between the highest probability and the second highest probability is less than 0.15, a review warning is triggered. The system provides two options: one is to use manual review to avoid misjudgment; the other is for the system to automatically combine the diagnostic results of the previous 5 cycles to perform trend analysis and select the most matching fault type.

[0084] In another possible implementation, in addition to the dual decision criteria, a prior probability of the fault type can be introduced. By statistically analyzing the occurrence probability of various fault types based on historical data, the prior probability can be combined with the fault probability distribution during diagnosis for a more accurate judgment. For example, for fault types with a low occurrence probability, the diagnostic threshold can be appropriately increased.

[0085] In this embodiment, the visualization based on dynamic threshold parameters and fault diagnosis results in step S6 includes:

[0086] The data visualization interface features a three-view diagnostic panel: the left-hand heatmap displays the dynamically adjusted sensor-fault correlation matrix, using a blue-yellow gradient to visually indicate weight intensity; the central line graph shows the deviation between sensor readings and dynamic thresholds, with error bands indicating the allowable fluctuation range; the right-hand bar chart annotates the probability values ​​of each fault type, with bars exceeding 0.5 automatically switching to red as a warning. The status indicator module dynamically displays the current system status (charging / discharging / standby) at the top of the panel, and the timeline navigation function supports retrospective viewing of diagnostic results from the past 100 sampling periods.

[0087] In another possible implementation, the confidence level (maximum failure probability) and false alarm events for each diagnosis are recorded. When the average confidence level of 100 consecutive diagnoses is below 0.85 or the false alarm rate (false alarms are defined as situations where the system is determined to have a fault by manual judgment, the model fails to identify the fault, or the system is determined to have no fault by manual judgment, or the model reports an error) exceeds 5%, a parameter optimization mechanism is triggered. The sliding window module automatically expands the window capacity to 150 sampling periods, and the dynamic threshold generation module automatically switches to a more stringent 2.8 times standard deviation to calculate the threshold range. The weights of the correlation matrix are recalibrated using historical fault case data. The optimization process employs a gradual adjustment strategy, with each update not exceeding 30% of the original value to ensure system stability.

[0088] Example 3: The above is an illustrative scheme of the method for quickly locating a faulty high-voltage box in energy storage according to this embodiment. It should be noted that the technical solution of the system for quickly locating a faulty high-voltage box in energy storage and the technical solution of the method for quickly locating a faulty high-voltage box in energy storage belong to the same concept. Details not described in detail in the technical solution of the system for quickly locating a faulty high-voltage box in energy storage in this embodiment can be found in the description of the technical solution of the method for quickly locating a faulty high-voltage box in energy storage described above.

[0089] This embodiment also provides a system for quickly locating faulty high-voltage boxes in energy storage systems, including:

[0090] The status identifier acquisition module is used to acquire battery cluster current parameters in real time and perform dual threshold judgment to obtain status identifiers.

[0091] The dynamic threshold calculation module is used to acquire current sensor parameters and calculate dynamic threshold parameters.

[0092] The correlation matrix construction and adjustment module is used to construct and adjust the sensor-fault correlation matrix based on the status identifier and combined with historical fault data.

[0093] The joint feature vector acquisition module is used to extract spatial and temporal features based on state identifiers and battery cluster current parameters, and fuse them to obtain a joint feature vector;

[0094] The fault diagnosis module is used to calculate the probability distribution of fault types based on joint feature vectors, and obtain fault diagnosis results by combining status identifiers and preset judgment conditions.

[0095] The visualization module is used to visualize the results based on dynamic threshold parameters and fault diagnosis.

[0096] This embodiment also provides an electronic device applicable to the method of quickly locating faulty high-voltage boxes in energy storage systems, including:

[0097] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the method for quickly locating faulty high-voltage boxes in energy storage systems, as proposed in the above embodiments.

[0098] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for quickly locating faulty high-voltage boxes in energy storage as proposed in the above embodiments.

[0099] The storage medium proposed in this embodiment and the method for quickly locating faulty high-voltage boxes in energy storage proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0100] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for rapidly locating faulty high-voltage boxes in energy storage systems, characterized in that, include: Real-time acquisition of battery cluster current parameters and dual threshold determination are performed to obtain status indicators; Obtain current sensor parameters and calculate dynamic threshold parameters; Based on status identifiers and combined with historical fault data, a sensor-fault correlation matrix is ​​constructed and adjusted; Based on the state identifier and battery cluster current parameters, spatial and temporal features are extracted and fused to obtain a joint feature vector. The probability distribution of fault types is calculated based on joint feature vectors, and the fault diagnosis results are obtained by combining status identifiers and preset judgment conditions. Based on dynamic threshold parameters and fault diagnosis results, a visual representation is provided.

2. The method for rapidly locating a faulty high-voltage box in energy storage as described in claim 1, characterized in that, The real-time acquisition of battery cluster current parameters and the performance of dual threshold determination to obtain status indicators include: Establish a current intensity time series window to monitor the current parameters of the battery cluster in real time; The system determines whether the absolute value of the current is continuously lower than the silent threshold. When the current value is less than the set value for a set number of consecutive sampling periods, the system is determined to be in standby mode.

3. The method for rapidly locating a faulty high-voltage box in energy storage as described in claim 2, characterized in that, The process of acquiring battery cluster current parameters in real time and performing dual threshold determination to obtain a status identifier also includes: The direction and polarity of the current are used to determine whether it is in a discharging state or a charging state.

4. The method for rapidly locating a faulty high-voltage box in energy storage as described in claim 3, characterized in that, The acquisition of current sensor parameters and calculation of dynamic threshold parameters include: Establish a sliding data window to continuously record current sensor parameters and calculate the moving average and standard deviation of each parameter in real time; When the sliding data window is not full, an initial threshold is generated through a three-layer perceptron network; when the sliding data window is full, a dynamic threshold range is generated based on the mean and standard deviation.

5. The method for rapidly locating a faulty high-voltage box in energy storage as described in claim 4, characterized in that, The process of constructing and adjusting the sensor-fault correlation matrix based on status identifiers and historical fault data includes: A sensor-fault correlation matrix with a specific dimension is preset. The rows of the matrix correspond to the sensors, and the columns correspond to the fault types. The initial weights of the matrix elements are set based on historical fault case data. For specific fault detection, an enhanced detection module is set up to continuously track specific sensor parameters; when the enhanced detection module detects that specific parameter conditions are met, the correlation weight between the fault and related sensors is enhanced by a product factor. By utilizing an attention mechanism network, combining current sensor readings and the characteristics of the correlation matrix, dynamic weight coefficients with state adaptability are generated, enabling dynamic adjustment of the sensor-fault correlation matrix.

6. The method for rapidly locating a faulty high-voltage box in energy storage as described in claim 5, characterized in that, The method of extracting spatial and temporal features based on state identifiers and battery cluster current parameters, and fusing them to obtain a joint feature vector, includes: A fully connected topology is constructed using a specific network structure, with battery cluster current parameters as node features and state identifiers as edge attribute features. By utilizing the multi-layer structure of the network, spatial features of a certain dimension are extracted in the first layer and multi-head attention is performed. The features are then aggregated in the second layer to obtain spatial features of a specific dimension. A time series processing network with a specific structure is used to process the time series of sensor data through a gating mechanism to extract time features of a specific dimension. The extracted spatial and temporal features are merged, and feature interaction is achieved through a multilayer perceptron to output a joint feature vector of a specific dimension.

7. The method for rapidly locating a faulty high-voltage box in energy storage as described in claim 6, characterized in that, The probability distribution of fault types calculated based on joint feature vectors, combined with status identifiers and preset judgment conditions, yields the following fault diagnosis results: Based on the fused joint feature vector, the feature vector is processed by a dual-channel activation function: the first layer compresses the features to eliminate feature redundancy, and the last layer performs a linear transformation to output the fault probability distribution corresponding to multiple fault types, and normalizes the probability distribution. Based on the status identifier, dual judgment conditions are set. In one state, the fault type that meets certain probability conditions and has the highest probability is selected as the diagnosis result; in the other state, all fault probabilities are forcibly reduced to zero and the state is judged as normal. To address conflicts caused by multiple concurrent faults, a probability difference threshold is set; when the difference between the highest probability and the second highest probability is less than the probability difference threshold, a review warning is triggered.

8. A system for rapidly locating faulty high-voltage boxes in energy storage systems, using the method described in any one of claims 1 to 7, characterized in that, include: The status identifier acquisition module is used to acquire battery cluster current parameters in real time and perform dual threshold judgment to obtain status identifiers. The dynamic threshold calculation module is used to acquire current sensor parameters and calculate dynamic threshold parameters. The correlation matrix construction and adjustment module is used to construct and adjust the sensor-fault correlation matrix based on the status identifier and combined with historical fault data. The joint feature vector acquisition module is used to extract spatial and temporal features based on state identifiers and battery cluster current parameters, and fuse them to obtain a joint feature vector; The fault diagnosis module is used to calculate the probability distribution of fault types based on joint feature vectors, and obtain fault diagnosis results by combining status identifiers and preset judgment conditions. The visualization module is used to visualize the results based on dynamic threshold parameters and fault diagnosis.

9. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images