Storage battery state coupling fault detection method for direct current bus power loss

Abstract

The invention discloses a storage battery state coupling fault detection method for DC bus power loss, and belongs to the technical field of power system fault monitoring, and the method comprises the steps: carrying out the real-time monitoring of a storage battery through a sensing module; performing standard state deviation detection on the storage battery according to the storage battery state data; performing fault prediction on the storage battery state deviation vector according to an ARIMA-LSTM combination model, and determining a risk fault prediction result corresponding to the fault storage battery; carrying out topological correlation analysis on the DC bus according to the fault storage battery; performing power-losing coupling fault prediction on the associated direct-current bus group according to a risk fault prediction result; and performing fault operation and maintenance on the power system according to the risk fault prediction result and the bus power-loss coupling fault prediction result. The technical problem that the bus power-loss risk is difficult to accurately predict in advance due to the lack of correlation analysis of the storage battery state and the DC bus power-loss fault in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault monitoring, and particularly to a method for detecting the coupled faults of battery states facing the loss of power of a DC bus. Background Art

[0002] As a key node for power transmission, the operation safety of the DC power supply bus directly affects the overall power supply reliability and stability. If the DC bus loses power, it may cause the shutdown of critical loads and seriously affect the safe operation of the global power grid in severe cases.

[0003] In the prior art, the loss of power fault of the DC bus is often highly coupled with the abnormal state of the battery, but there is a lack of effective fault prediction methods, making it difficult to identify and handle risks in advance, resulting in passive fault diagnosis and low operation and maintenance efficiency of the power system. Therefore, there is an urgent need for a fault detection method that can integrate real-time state monitoring, advanced data-driven model prediction, and topological correlation analysis to improve the timeliness and accuracy of fault prediction for the batteries and the loss of power of the DC bus in the power system. Summary of the Invention

[0004] This application provides a method for detecting the coupled faults of battery states facing the loss of power of a DC bus, aiming to solve the technical problem in the prior art that there is a lack of correlation analysis between the battery state and the loss of power fault of the DC bus, resulting in difficulty in accurately predicting the risk of power loss of the bus in advance.

[0005] In view of the above problems, this application provides a method for detecting the coupled faults of battery states facing the loss of power of a DC bus.

[0006] This application provides a method for detecting the coupled faults of battery states facing the loss of power of a DC bus. The method includes: real-time monitoring of each battery in the power system through a pre-deployed sensing module to obtain a plurality of battery state data, where the power system includes a plurality of batteries and a plurality of DC buses; performing a standard state deviation detection on the plurality of batteries according to the plurality of battery state data to determine a plurality of battery state deviation vectors; performing a fault prediction on the plurality of battery state deviation vectors according to an ARIMA-LSTM combined model to determine a risk fault prediction result corresponding to the faulty battery; performing a topological correlation analysis on the plurality of DC buses according to the faulty battery to determine an associated DC bus group; performing a loss of power coupled fault prediction on the associated DC bus group according to the risk fault prediction result to determine a plurality of bus loss of power coupled fault prediction results; and performing fault operation and maintenance on the power system according to the risk fault prediction result and the plurality of bus loss of power coupled fault prediction results.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: Due to the technical solution based on real-time monitoring data of storage batteries, ARIMA-LSTM combined model prediction, and topological correlation analysis, the technical problem in the prior art of lacking the correlation analysis between the state of storage batteries and the DC bus power failure fault is solved, thus making it difficult to accurately predict the risk of bus power failure in advance. The technical effect of accurately predicting and actively preventing and controlling the risk of bus power failure is achieved.

[0008] The above description is only an overview of the technical solution of this application. In order to be able to more clearly understand the technical means of this application, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the specific embodiments of this application are specifically given below. Brief Description of the Drawings

[0009] Figure 1 This is a schematic flowchart of the method for detecting the coupled fault of the storage battery state facing DC bus power failure provided by the embodiment of this application.

[0010] Figure 2 This is a schematic flowchart of determining the risk fault prediction result corresponding to the faulty storage battery in the method for detecting the coupled fault of the storage battery state facing DC bus power failure provided by the embodiment of this application. Detailed Description of the Embodiments

[0011] The general idea of the technical solution provided by this application is as follows: The embodiment of this application provides a method for detecting the coupled fault of the storage battery state facing DC bus power failure. By real-time monitoring the state data of the storage battery, identifying the deviation of the storage battery from the standard state, using the ARIMA-LSTM model to predict the faulty storage battery, and then based on topological correlation analysis to identify the risk-associated bus group and predict the risk of bus power failure coupling, accurate fault prediction and operation and maintenance of the power system are realized.

[0012] After introducing the basic principle of this application, the various non-limiting embodiments of this application will be specifically introduced below in conjunction with the drawings of the specification.

[0013] Embodiment, as Figure 1 shown, the embodiment of this application provides a method for detecting the coupled fault of the storage battery state facing DC bus power failure, and the method includes: Step S100: Real-time monitor each storage battery in the power system through a pre-deployed sensing module to obtain a plurality of storage battery state data, and the power system includes a plurality of storage batteries and a plurality of DC buses.

[0014] Specifically, the sensing module refers to a monitoring device composed of multiple sensors with different functions, mainly used for real-time collection of the operating data of the device. Taking the battery monitoring as an example, the sensing module includes a voltage sensor, a current sensor, a temperature sensor, an internal resistance measurement sensor, etc. The power system specifically refers to a DC power supply network composed of multiple batteries and multiple DC buses. The battery state data refers to the data set that can reflect the operating conditions of the battery, typically including parameters such as voltage value, current value, temperature value, state of charge, state of health, and internal resistance.

[0015] First, deploy the sensing module around each battery pack in the power system. Specifically, the sensing module includes: Voltage sensor: such as a high-precision voltage acquisition chip, used to measure the real-time voltage level of the battery. Current sensor: such as a Hall sensor or a shunt, used to monitor the charging and discharging current of the battery. Temperature sensor: such as a thermistor, used to collect the temperature data of the battery during operation. Internal resistance detector: used to periodically detect the change trend of the internal resistance of the battery.

[0016] The sensing module transmits the collected data to the system background in real time through a data acquisition terminal, such as realizing the rapid upload of data through RS485, CAN bus or wireless Zigbee network. After receiving the data, the background data server obtains a complete and clear data set of the states of multiple batteries through data storage, cleaning and sorting.

[0017] By pre-deploying a multi-functional and high-precision sensing module in the power system, it is possible to efficiently and real-time monitor and record the key parameters of the battery operation, obtain an accurate data set of the battery state, thereby providing a solid data basis for subsequent state analysis, anomaly prediction, and fault diagnosis, significantly improving the accuracy and real-time performance of the operation and maintenance of the entire power system, and effectively avoiding the risk of DC bus power loss caused by abnormal battery states.

[0018] Step S200: Perform a standard state deviation detection on the multiple batteries according to the multiple battery state data, and determine multiple battery state deviation vectors.

[0019] Specifically, the standard state refers to the typical state of the battery under normal operating conditions without any anomalies or faults, usually manifested as a stable voltage range, current value, temperature range, internal resistance value, as well as a good state of charge and state of health. Deviation detection refers to the process of analyzing and identifying the degree of difference between the actual state of the current battery and the standard state, mainly used to discover abnormal situations and potential risks of the battery. The state deviation vector refers to the difference between the real-time state data of the battery and the standard state data, which is quantitatively represented in the form of a vector, such as a vector composed of voltage deviation values, current deviation values, temperature difference values, internal resistance change values, etc., used to intuitively show the degree of device state abnormality.

[0020] First, based on multiple battery state data obtained from real-time monitoring, data preprocessing techniques (such as Kalman filtering, wavelet transform, moving average filtering, etc.) are used for filtering and denoising to obtain more accurate and clear actual operation data of the battery.

[0021] Secondly, the preprocessed data is vectorized, that is, multiple different types of data (such as voltage, current, temperature, internal resistance) are combined into a complete data vector for unified subsequent standard state comparison and analysis.

[0022] Next, standard state mining needs to be carried out. Specifically, a large number of normal operation samples are extracted from the historical operation data of the battery, and through central tendency analysis methods (such as mean analysis, median analysis or clustering analysis), a standard state vector that can represent the stable operation status of the battery is formed.

[0023] After completing the standard state mining, the battery state vector obtained in real time is compared item by item with the corresponding battery standard state vector. For example, the difference between each battery real-time state vector and the standard state vector is calculated by methods such as Euclidean distance, cosine similarity or Mahalanobis distance, so as to generate the corresponding state deviation vector.

[0024] For example, the standard voltage range of a certain battery pack in a DC power supply system is 51 - 53V, and the temperature range is 25 - 35°C. When the real-time monitoring data shows that the voltage of the battery pack continuously drops below 49V and the temperature exceeds 40°C, it indicates that the current state significantly deviates from the standard state. The system can quickly mark the abnormal state and store the data of "voltage deviation - 4V" and "temperature deviation + 5°C" into the deviation vector for subsequent analysis.

[0025] Through the standard state deviation detection technology of this step, the abnormal operation trend and potential risks of the battery can be quickly and accurately identified, and the abnormal state causing the DC bus fault can be discovered in advance. By constructing a quantifiable deviation vector, the deficiency of human subjective judgment is effectively avoided, and the accuracy and timeliness of abnormal identification are improved.

[0026] Step S300: Perform fault prediction on the multiple battery state deviation vectors according to the ARIMA-LSTM combined model, and determine the risk fault prediction result corresponding to the faulty battery.

[0027] Specifically, the ARIMA model, whose full name is Autoregressive Integrated Moving Average model, is a classic time series prediction model that can capture the linear trend and periodic characteristics in the data series and is commonly used for modeling and predicting short-term data with stable or transformable stability. The LSTM model, whose full name is Long Short-Term Memory network, is a special type of Recurrent Neural Network (RNN) that has the ability to remember long-term dependencies in sequential data and is particularly suitable for predicting time series data with obvious non-linear and long-term historical dependence characteristics. The ARIMA-LSTM combined model is a hybrid prediction model that integrates the advantages of the ARIMA model and the LSTM model. First, the ARIMA model is used to capture the short-term linear trend, then the LSTM model is used to capture the long-term non-linear relationship, and finally, the prediction results of the two are combined to improve the accuracy and robustness of the prediction. The risk fault prediction result refers to the probability or risk assessment result of the future failure of the battery obtained according to the prediction model, which is quantified in the form of a probability value or a risk score.

[0028] First, the obtained battery state deviation vector is used as the prediction input data. Considering the characteristics of the battery operation data, which have both short-term linear features (such as short-term voltage fluctuation trends) and long-term non-linear features (such as battery aging and internal resistance growth trends), the ARIMA-LSTM combined model prediction method is adopted to improve the accuracy of fault prediction.

[0029] In the specific implementation process: In the data preparation stage, the battery state deviation vector sequence is processed in chronological order to generate a historical chronological data set for modeling; an ARIMA prediction model is constructed; an LSTM prediction model is constructed; a prediction fusion mechanism is designed, and the prediction results respectively output by the ARIMA and LSTM models (such as the change in the battery state deviation in the future time period) are input into the fusion module. Usually, methods such as weighted average, BP neural network, and SVR support vector regression machine are used to fuse the two prediction results to obtain a comprehensive prediction value; a reasonable risk threshold is set for the fusion prediction result to evaluate the predicted abnormal deviation degree, so as to identify high-risk potential faulty batteries and give a risk probability or score.

[0030] For example: Suppose the state deviation data of a certain battery in the past month shows an obvious trend, where the voltage deviation gradually increases and the short-term fluctuation amplitude increases. First, ARIMA is used to capture the linear trend change of the short-term voltage fluctuation; then LSTM is used to deeply learn the non-linear pattern of the long-term internal resistance increase and aging trend; finally, combining the prediction results of the two models (such as the increase in short-term fluctuations and the continuous decrease in voltage), through neural network fusion prediction, it is determined that the probability of this battery failing within the next week is 85%. Since this probability exceeds the set fault prediction threshold (such as 75%), the system automatically includes this battery in the list of risk batteries and records the risk prediction result for subsequent analysis and processing.

[0031] By adopting the advantages of the ARIMA-LSTM combined model to integrate short-term linear trends and long-term non-linear characteristics, the accuracy and reliability of battery fault prediction can be greatly improved, effectively reducing the phenomena of prediction false alarms and missed alarms. At the same time, the system can identify high-risk batteries in advance and accurately, significantly improving the safety and stability of the power system operation, reducing the risk of DC bus power loss caused by battery anomalies, and providing reliable data basis and decision support for system operation and maintenance.

[0032] Step S400: Perform topological association analysis on the multiple DC buses according to the faulty battery to determine the associated DC bus group.

[0033] Specifically, topological association analysis refers to using graph theory or network analysis techniques to construct and analyze the topological connection relationship between the battery and the DC bus, and identify the association between the battery and the bus. The associated DC bus group refers to the set of one or more DC buses that have a strong association relationship with a specific faulty battery, indicating that the power loss risk of this bus group will occur once the battery fails.

[0034] First, clarify that the object to be analyzed is the faulty battery obtained by prediction, that is, the battery unit with a high fault probability determined by the ARIMA-LSTM combined model mentioned above. Subsequently, based on the topological structure of the power system operation (usually in the form of a topological graph or a network graph), construct a topological network graph between the battery and the bus: regard the battery as the "parent node" in the graph and the DC bus as the "child node"; the connection between the battery and the DC bus forms a connection edge through physical connections (such as power lines, cables), and then establish a battery-bus topological network graph.

[0035] Then, perform topological association evaluation on the constructed battery-bus topological network graph. Specifically, through the association degree algorithm in graph theory, such as: adjacency matrix analysis, path length analysis, network flow algorithm or topological correlation algorithm (such as calculating topological correlation coefficients) for calculation, so as to quantitatively characterize the association degree between each faulty battery and different buses, and obtain multiple battery-bus topological correlation coefficients.

[0036] Next, set a topological association threshold (such as 0.75), compare each correlation coefficient with this threshold, and judge whether it reaches or exceeds the threshold to determine the bus set with an obvious topological relationship with the faulty battery, which is the associated DC bus group.

[0037] For example, assume that the No. 1 battery in a substation is predicted to be a high-risk faulty battery. First, in the DC system of the substation, establish a battery-bus topology diagram (e.g., the No. 1 battery is connected to DC buses A, B, and C, and the No. 2 battery is connected to buses B and D, etc.); then use the topology analysis method to calculate the topology correlation coefficients between the No. 1 battery and each bus. Suppose the calculation results are: the correlation coefficient of bus A is 0.85, that of bus B is 0.90, and that of bus C is 0.65. If the predetermined topology correlation threshold is 0.75, then the correlation coefficients of buses A and B exceed the threshold. Therefore, buses A and B are determined as the associated DC bus group of the No. 1 faulty battery, while bus C is not included in this set.

[0038] In this step, through topology correlation analysis, the set of buses at risk of power loss affected by a specific faulty battery is accurately discovered, effectively narrowing the scope of fault investigation and prevention and control, and improving the accuracy of system safety management. Through the screening of topology correlation coefficients and thresholds, the risk-associated bus group is objectively and quantitatively identified, greatly reducing the risk of fault spread, providing a reliable decision-making basis for the operation and maintenance of the power system in advance, and improving the safety and reliability of the DC power supply system.

[0039] Step S500: Perform a power-loss coupling fault prediction on the associated DC bus group according to the risk fault prediction result, and determine multiple power-loss coupling fault prediction results for the buses.

[0040] Specifically, the power-loss coupling fault prediction refers to further evaluating the possibility and risk of DC bus power loss based on the association relationship for the occurring battery fault, and predicting the bus power-loss event caused by the battery fault. The power-loss coupling fault prediction result for the buses refers to the power-loss probability or risk assessment value obtained after predicting each associated bus, clearly indicating the DC buses where power loss occurs and their probabilities.

[0041] First, according to the obtained risk fault prediction result, that is, the battery with a high-risk fault probability is clearly predicted, combined with the associated DC bus group determined through the previous topology correlation analysis, as the object for further power-loss coupling fault prediction. Next, perform a power-loss coupling fault prediction on each associated DC bus respectively: extract the real-time status parameter data of each associated DC bus; perform a power-loss coupling fault registration analysis according to the prediction result of the risk fault battery and the real-time status data of the bus. Based on the above analysis, obtain the "power-loss coupling fault registration space" that occurs, that is, the set of all historical power-loss fault events that highly match the current risk prediction scenario. For each power-loss fault event in the registration space, use a confidence evaluation method, such as Bayesian inference, D-S evidence theory, fuzzy comprehensive evaluation, etc., to calculate the confidence coefficient of the registered fault. Then set a registration fault confidence threshold (e.g., 0.8), perform optimization screening on the registration space, and leave the events with confidence coefficients exceeding the threshold to form a power-loss coupling fault confidence space.

[0042] Finally, the data in the power failure coupling fault confidence space is fused using data fusion methods (such as weighted average, neural network fusion, or fuzzy comprehensive evaluation) to determine the final power failure probability or risk prediction result of the associated bus.

[0043] In this step, by combining the risk fault prediction of the battery and the real-time operation data of the DC bus, supported by the historical fault database, through methods such as power failure coupling registration, confidence evaluation, and data fusion, the probability and risk of power failure of the DC bus caused by battery faults are accurately and effectively predicted, realizing a leapfrog correlation prediction from the risk of a single battery to the risk of the bus. This effectively improves the foresight and accuracy of fault warning, helps the operation and maintenance personnel deploy risk control measures in advance, avoids the system cascade faults and losses caused by the power failure of the DC bus, and greatly improves the safety and stability of the power system operation.

[0044] Step S600: Perform fault operation and maintenance on the power system according to the risk fault prediction result and the multiple bus power failure coupling fault prediction results.

[0045] Specifically, based on the risk fault prediction result and the multiple bus power failure coupling fault prediction results obtained in the previous steps, these results are automatically converted into clear risk warning information or alarm reports.

[0046] Then, the system takes corresponding hierarchical operation and maintenance measures for different predicted risk levels: High risk level (such as fault probability ≥ 85%): The system automatically generates an emergency alarm; automatically notifies the operation and maintenance personnel to conduct on-site inspections or immediately start emergency disposal measures, such as battery replacement or redundant power supply line switching. Medium risk level (such as fault probability 60% - 85%): Arrange on-site equipment detection in advance, increase the inspection frequency; conduct equipment maintenance or prepare spare parts as appropriate; adjust equipment operation parameters or load to relieve the risk. Low risk level (such as fault probability < 60%): Maintain daily monitoring and regular inspections, pay attention to the risk trend; adjust the equipment operation strategy according to the trend change.

[0047] The operation and maintenance personnel view the detailed prediction information based on these automatically generated risk prediction reports and use professional power operation and maintenance tools and management systems (such as power fault management platforms, SCADA systems, on-line monitoring terminals) to formulate corresponding maintenance measures.

[0048] In specific implementation, the following methods can be adopted: on-site inspection: arrange personnel to conduct on-site inspections on the high-risk batteries and busbar equipment whose states are predicted by risk, and confirm the actual situation of the predicted risks; automated remote monitoring system (SCADA): track the changes in equipment operation data in real time, further confirm the risk level, and timely feedback abnormal situations; emergency response plan: formulate and initiate an emergency replacement or backup plan in advance for the equipment predicted to be at high risk (such as the failure probability of the battery reaching 90%), to avoid actual failures. Operation and maintenance work order system: automatically dispatch maintenance or inspection work orders to operation and maintenance personnel according to the prediction results, and urge them to perform maintenance operations in a timely manner.

[0049] Through this step, operation and maintenance management has changed from the traditional post-failure response to proactive and predictive maintenance. It has achieved the accurate and timely discovery of potential risks, clarified early warning information, and taken measures in advance to avoid failures; accurately identified risk areas and faulty equipment, avoided traditional blind inspections, improved maintenance efficiency, and reduced operation and maintenance costs.

[0050] Furthermore, as Figure 2 shown, fault prediction is carried out on the multiple battery state deviation vectors according to the ARIMA-LSTM combined model to determine the risk fault prediction results corresponding to the faulty batteries, including: carrying out fault prediction on the multiple battery state deviation vectors according to the ARIMA-LSTM combined model to obtain multiple battery fault prediction results; judging whether the multiple battery fault probabilities in the multiple battery fault prediction results are greater than or equal to a predetermined fault probability; if any one of the multiple battery fault probabilities is greater than or equal to the predetermined fault probability, obtaining a risk fault probability; adding the battery corresponding to the risk fault probability to the faulty battery, and adding the battery fault prediction result corresponding to the faulty battery to the risk fault prediction result.

[0051] Specifically, the predetermined fault probability refers to the set threshold probability used to judge whether the battery is in a high-risk fault state. When the predicted fault probability reaches or exceeds this probability, the battery is regarded as having potential risks. The risk fault probability refers to the battery fault probability that exceeds the predetermined fault probability, indicating that the battery has a relatively high risk of future failure events and needs to be focused on or processed in advance.

[0052] First, use the ARIMA-LSTM combined model to conduct fault prediction based on the obtained state deviation vectors of each battery, and output corresponding battery fault prediction results for each battery (usually expressed as the fault probability within a specified future time period). For example, for 10 batteries in a certain substation, after prediction, the fault probability prediction values for each of them in the next week are obtained.

[0053] Then, compare these predicted battery failure probabilities with a threshold. Specifically, set a predetermined failure probability threshold (e.g., 75%) to judge the potential risk level of the storage battery. If the failure probability of a certain storage battery predicted is greater than or equal to this predetermined failure probability, it indicates that the storage battery has a high risk.

[0054] Next, screen out all storage batteries that meet the above conditions (i.e., failure probability ≥ threshold) from the prediction results, and define the corresponding prediction probability as the "risk failure probability". Mark these high-risk storage batteries and process them centrally, adding them to the "failed storage battery" set for subsequent analysis and management. Subsequently, record and save the battery failure prediction probability results corresponding to these risk storage batteries to form a risk failure prediction result: database.

[0055] For example, assume there are 5 groups of storage batteries, and the ARIMA-LSTM combined model predicts the future state deviation data of these 5 groups of batteries respectively. The obtained failure probabilities are: Battery 1: 82%, Battery 2: 46%, Battery 3: 79%, Battery 4: 65%, Battery 5: 91%; if the predetermined failure probability threshold is set at 75%, the system automatically determines that the prediction probabilities of Battery 1, Battery 3, and Battery 5 exceed the threshold and belong to risk failure storage batteries. The system immediately adds these 3 groups of batteries to the "failed storage battery set" and records their corresponding failure probabilities (82%, 79%, 91%) in the risk failure prediction result, providing a basis for further in-depth analysis in the next step.

[0056] This step accurately identifies high-risk storage batteries through the ARIMA-LSTM combined model and accurately screens the prediction results through the predetermined failure probability threshold, greatly improving the automation and accuracy of failure prediction and risk assessment. It can identify and handle in advance the risk hidden dangers of storage batteries that cause the loss of power of the DC bus, thus effectively avoiding the chain accidents of the power supply system caused by battery anomalies.

[0057] Further, perform fault prediction on the multiple battery state deviation vectors according to the ARIMA-LSTM combined model to obtain multiple battery fault prediction results, including: extracting the first battery state deviation vector corresponding to the first battery according to the multiple battery state deviation vectors; retrieving the battery state deviation vector sample set and the battery fault sample set corresponding to the first battery; performing supervised training on the ARIMA model and the LSTM model in the ARIMA-LSTM combined model according to the battery state deviation vector sample set and the battery fault sample set to obtain a first battery fault predictor and a second battery fault predictor; training a battery fault prediction fusion device with the output data sets of the first battery fault predictor and the second battery fault predictor as input information and the battery fault sample set as output information; building a first battery fault prediction channel with the first battery fault predictor and the second battery fault predictor as the first node of battery fault prediction and the battery fault prediction fusion device as the second node of battery fault prediction; inputting the first battery state deviation vector into the first battery fault prediction channel to obtain a first battery fault prediction result, and adding the first battery fault prediction result to the multiple battery fault prediction results.

[0058] Specifically, the battery state deviation vector sample set refers to a data set composed of historically collected battery state deviation data and is used for model training. The battery fault sample set refers to a data set of fault records corresponding to the state deviation data, which annotates the time or state when a historical fault event occurs and is used as the true value of the model output during supervised training. The battery fault prediction fusion device refers to a module that effectively integrates the prediction results of the ARIMA model and the LSTM model. The battery fault prediction channel refers to a complete prediction system path formed by combining predictors and fusion devices in a specific manner, covering the entire process from data input to result output.

[0059] First, according to the multiple battery state deviation vectors, select the state deviation vector of the specific first battery to be predicted as the subsequent prediction analysis object. For example, among 10 batteries in a certain substation, select the current deviation state data vector of the No. 1 battery.

[0060] Second, retrieve the historical state deviation data sample set and the corresponding historical fault event record sample set of the first battery. The sample sets are pre-stored in the historical database. For example, the monthly voltage deviation, internal resistance deviation data and the corresponding fault event records (such as the time of fault occurrence, fault type, etc.) collected in the past year. Then, based on the historical sample data set, use the supervised learning method to train the ARIMA model and the LSTM model respectively: Using the battery state deviation vector sample set and the battery fault sample set, through ARIMA model training, a "first battery fault predictor" for capturing short-term linear trends is obtained; at the same time, using the same historical data, through LSTM model training, long-term non-linear characteristics are captured, and a "second battery fault predictor" is obtained. Specifically, a time series is established using the historical battery state deviation vector sample data, and the ADF unit root test is used for stationarity analysis. If it is not stationary, differencing processing is performed. After determining the model order through ACF and PACF, the ARIMA model parameters are estimated by the maximum likelihood method, and the model accuracy is verified through model diagnosis (such as residual analysis). Finally, a short-term trend predictor is formed. The historical battery state deviation sample sequence is divided into training and validation sets. With the state deviation data as the input and the fault occurrence label as the output, an LSTM neural network structure (input layer, hidden layer, output layer) is designed, reasonable node numbers are set, and the model is trained using the Adam optimizer and the cross-entropy loss function. After repeated iterative optimization, a long-term trend predictor is finally generated.

[0061] Next, using the output prediction result data sets of the first and second battery fault predictors as new input information, and the historical true fault sample set as the target output information, a fusion model (such as a BP neural network or an SVM support vector machine, etc.) is trained to form a "battery fault prediction fusion device". The fusion device is responsible for learning the differences and advantages between the two predictors and comprehensively integrating the results.

[0062] Subsequently, the above-trained predictors and fusion devices are connected in series in sequence, that is, the first battery fault predictor and the second battery fault predictor are used as the first node of the battery fault prediction, and the prediction fusion device is used as the second node of the battery fault prediction to build a complete battery fault prediction channel. Specifically, a BP neural network or an SVM support vector machine is used for fusion training, and the parameter weights are optimized using backpropagation and cross-validation. After verification, the final fusion prediction model is obtained.

[0063] Finally, the current state deviation vector of the first battery is input into this fault prediction channel. After obtaining the short-term and long-term prediction results through ARIMA and LSTM respectively, they are comprehensively processed by the fusion device, and the final first battery fault prediction result is output, and this result (such as the fault risk probability in the next week) is added to multiple battery fault prediction result databases.

[0064] This step combines the advantages of ARIMA and LSTM models to accurately capture the short-term linear trend and long-term non-linear trend of the battery operation data. Then, a fusion device is used to integrate the advantages of the two prediction models, improving the accuracy and robustness of the prediction effect. This prediction channel can effectively identify the high-risk state of the battery in advance, avoid the spread of faults and the occurrence of DC bus power loss accidents, provide more accurate and efficient operation and maintenance decision support for the power system, and ensure the safe and stable operation of the power system.

[0065] Furthermore, based on the risk fault prediction result, a power loss coupling fault prediction is performed on the associated DC bus group to determine multiple bus power loss coupling fault prediction results, including: extracting the first associated DC bus according to the associated DC bus group; collecting the real-time state parameters of the first associated DC bus to obtain the first associated bus state data; performing a power loss coupling fault registration on the first associated DC bus according to the risk fault prediction result and the first associated bus state data to obtain a first bus power loss coupling fault registration space; performing a confidence evaluation on each registered power loss coupling fault in the first bus power loss coupling fault registration space to obtain multiple registered fault confidence coefficients; based on the multiple registered fault confidence coefficients, performing an optimization screening on the first bus power loss coupling fault registration space according to a registered fault confidence threshold to generate a first bus power loss coupling fault confidence space; performing data fusion according to the first bus power loss coupling fault confidence space to generate a first bus power loss coupling fault prediction result, and adding the first bus power loss coupling fault prediction result to the multiple bus power loss coupling fault prediction results.

[0066] Specifically, the real-time state parameters refer to the state parameters monitored during the current actual operation of the bus, such as real-time voltage, current, load rate, bus temperature, etc. The power loss coupling fault registration refers to using the historical fault database and based on the current risk battery prediction result and the real-time state of the bus to find similar historical events and achieve the matching of the current state and historical power loss events. The power loss coupling fault registration space refers to the set of historical fault events with a relatively high matching degree with the current prediction scenario obtained after the power loss coupling fault registration. The confidence evaluation refers to a quantitative calculation of the similarity or credibility between the historical event and the current scenario, generally realized through methods such as Bayesian inference, D-S evidence theory, and fuzzy evaluation, and the result is reflected as a confidence coefficient between 0 and 1. The power loss coupling fault confidence space refers to a set of fault events with high confidence selected based on the confidence evaluation of the historical fault events in the registration space, used to achieve more accurate power loss prediction. The data fusion refers to comprehensively analyzing the prediction data of multiple events in the confidence space using methods such as weighted average, BP neural network, and fuzzy comprehensive evaluation, and outputting a unified and accurate final prediction result.

[0067] First, select a specific bus (i.e., the first associated DC bus) from multiple associated DC bus groups determined through topological association analysis for separate power loss coupling fault prediction analysis. Then, use real-time monitoring devices or sensors to collect the current real-time operating state parameter data of the first associated DC bus to form a real-time state data set of the current bus.

[0068] Next, implement "power loss coupling fault registration". Specifically: First, construct a historical power loss fault event library, which records the detailed data of battery faults and bus power loss events that occurred in history (such as battery fault types, bus operating states, and power loss situations); then input the prediction results of the current risk fault battery (such as predicted fault types and probabilities) and the real-time state data of the first associated bus into the system, and use fault scenario matching algorithms, such as similarity matching, association analysis, or machine learning clustering methods, to automatically compare with the data in the historical fault event library to obtain power loss coupling fault events with a relatively high degree of matching, forming a power loss coupling fault registration space.

[0069] Next, for each historical power loss event in the above registration space, through confidence evaluation methods (such as Bayesian inference method, evidence theory inference, or fuzzy logic evaluation), quantitatively calculate the credibility of these registered events to generate multiple specific confidence coefficients (between 0 and 1), so as to objectively judge the matching credibility between historical events and the current actual scenario.

[0070] Next, according to the pre-set registration fault confidence threshold (for example, 0.8), screen the historical events in the registration space, and only retain the highly credible events with a confidence level exceeding the threshold, so as to generate an accurate and reliable power loss coupling fault confidence space.

[0071] Finally, perform data fusion processing on the historical power loss event data in this confidence space. Specific methods include: weighted average method (weighted according to confidence); fuzzy comprehensive evaluation method (considering the comprehensive effects of multiple factors); BP neural network fusion (automatically learning the weight relationship between data); through the above fusion methods, calculate a unified prediction result of the power loss coupling fault of the first bus, such as obtaining the specific power loss risk probability or score of the bus in a future specific period, and add this result to the overall power loss coupling fault prediction result library of the bus.

[0072] Through the above detailed power loss coupling fault prediction steps, using real-time data and historical event matching, confidence quantification evaluation, and data fusion, the power loss risk of the associated bus can be predicted earlier and more accurately. Identify the hidden dangers leading to bus power loss in advance, effectively reduce the risk of system cascade faults caused by battery faults, greatly improve the operation stability and safety of the power system, and provide a scientific and accurate support basis for operation and maintenance decision-making.

[0073] Further, perform power loss coupling fault registration on the first associated DC bus according to the risk fault prediction result and the first associated bus status data to obtain a first bus power loss coupling fault registration space, including: obtaining a power loss coupling fault event set of the first associated DC bus; according to the power loss coupling fault event set, extracting a first power loss coupling fault event, where the first power loss coupling fault event includes a first battery fault record, a first bus status record, and a first power loss coupling fault record corresponding to the first associated DC bus; performing a correlation degree evaluation on the first battery fault record and the first bus status record according to the risk fault prediction result and the first associated bus status data to obtain a first power loss coupling scenario correlation coefficient; if the first power loss coupling scenario correlation coefficient is greater than or equal to the power loss coupling scenario correlation threshold, record the first power loss coupling fault record as a first registered power loss coupling fault, and add the first registered power loss coupling fault to the first bus power loss coupling fault registration space.

[0074] Specifically, the power loss coupling fault event set refers to a set of event records of a series of actual battery failures causing power loss of the DC bus recorded in the power system historical database, which records information such as battery fault status, bus status, and fault process.

[0075] The first power loss coupling fault event refers to a specific historical event record separately extracted from the power loss coupling fault event set, including: a first battery fault record; a first bus status record; a first power loss coupling fault record. The power loss coupling scenario correlation coefficient refers to a quantitative index (between 0 and 1) obtained through correlation degree evaluation, which is used to describe the matching degree between the current fault prediction scenario and the historical power loss fault scenario. The power loss coupling scenario correlation threshold refers to a preset critical value of correlation degree evaluation, which is used to screen out historical events with a high enough correlation degree with the current scenario, and the typical range is generally between 0.7 and 0.9. The registered power loss coupling fault refers to a historical power loss fault event that is determined to be highly matched with the current scenario and selected after correlation degree evaluation and threshold screening.

[0076] First, according to the determined first associated DC bus, call the power system historical database to obtain the "power loss coupling fault event set" that occurred in the past of this bus, and this data set records the specific details of how battery failures caused power loss of this bus in history. Secondly, extract specific historical power loss coupling fault events one by one from the above event set, such as the first power loss coupling fault event, including: a first battery fault record (voltage drop, internal resistance mutation amplitude, temperature anomaly record, etc. during the fault); a first bus status record (such as the voltage drop value of the bus at that time, current anomaly, sudden increase in load); a first power loss coupling fault record (specific power loss time, duration, affected range, fault handling situation, etc.).

[0077] Then, the system performs a correlation evaluation on the current risk fault prediction results, the current real-time status data of the first bus, and the "first battery fault record" and "first bus status record" in historical events. Specifically, methods such as Pearson correlation coefficient, cosine similarity, Euclidean distance, or similarity method based on machine learning are used to quantitatively obtain the matching degree between the current prediction status and historical events, and generate specific "power loss coupling scenario correlation coefficients". Subsequently, the calculated correlation coefficient is compared with a pre-set power loss coupling scenario correlation threshold (such as 0.8) to determine whether the matching degree between the historical event and the current prediction scenario is high enough. If the first power loss coupling scenario correlation coefficient reaches or exceeds the threshold, it is determined that the historical event is a successfully registered historical fault event, and the corresponding first power loss coupling fault record is officially added to the first bus power loss coupling fault registration space as the first registered power loss coupling fault; if it does not reach the threshold, it is considered that the matching is unsuccessful and is not added to the registration space.

[0078] Through the accurate historical event registration and quantitative scenario correlation evaluation method in this step, a highly intelligent matching between the current prediction scenario and historical power loss fault experience is achieved, and a set of historical events with high reference value for the current prediction scenario is scientifically and accurately selected. This significantly improves the accuracy and credibility of subsequent fault risk prediction, helps power system operation and maintenance personnel more timely and effectively prevent or control the potential power loss risk of the DC bus, greatly improves the safe and reliable operation ability of the system, and reduces the probability of accidents.

[0079] Furthermore, topological association analysis is performed on the multiple DC buses according to the faulty battery to determine the associated DC bus group, including: constructing a battery-bus topology graph with the multiple batteries as multiple parent nodes, the multiple DC buses as multiple child nodes, and the multiple connection relationships between the multiple batteries and the multiple DC buses as multiple connection edges; performing topological association evaluation on the faulty battery and the multiple DC buses according to the battery-bus topology graph to obtain multiple battery-bus topology correlation coefficients; judging whether the multiple battery-bus topology correlation coefficients are greater than or equal to a predetermined topology correlation coefficient to obtain multiple topology association judgment results; and screening the multiple DC buses according to the multiple topology association judgment results to generate the associated DC bus group.

[0080] Specifically, the battery node (parent node) represents a battery or a battery bank in the power system, which is used to provide DC electrical energy and is the superior node (parent node) in the topological network diagram. The DC bus node (child node) represents the bus line in the power system responsible for collecting and distributing DC electrical energy, which is the inferior node (child node) in the topological network diagram and directly supplies power to the load. The connecting edge refers to the actual connection line between the battery and the DC bus, which is an edge line reflecting the power flow or logical connection relationship in the topological diagram. The battery-bus topological diagram refers to a network diagram formed by multiple batteries as parent nodes and multiple DC buses as child nodes through actual connection relationships, reflecting the topological connection relationship between the battery and the bus. The battery-bus topological correlation coefficient refers to the quantitative index output by the topological correlation evaluation method, indicating the strength of the correlation between a specific battery and the bus node, usually represented by a value between 0 and 1. The topological correlation judgment result refers to the Boolean result obtained by comparing the topological correlation coefficient with a predetermined threshold, indicating whether a certain correlation relationship is valid or strongly correlated.

[0081] First, take all the batteries in the power system as parent nodes and all the DC buses as child nodes. Based on the actual physical connection relationships (such as cable connection relationships, power supply line configurations, etc.), use graph theory methods to establish a complete battery-bus topological diagram, clearly showing the connection structure and logical relationship between all the batteries and the buses. Secondly, for the predicted faulty batteries (i.e., the batteries with a high risk of failure probability predicted), use the topological correlation evaluation method for analysis, specifically including: Adjacency matrix method: Convert the topological network relationship into a matrix form and calculate the relationship strength between the battery and the bus; Path length analysis method: Calculate the path distance or path weight from the battery node to the bus node; Network correlation algorithms: Such as PageRank, Degree Centrality, etc., calculate the node correlation strength in the network.

[0082] Calculate multiple battery-bus topological correlation coefficients through the above methods to clearly represent the strength of the relationship between each faulty battery and each bus. For example, a correlation coefficient close to 1 indicates a strong correlation, and close to 0 indicates a weak correlation or no correlation. Then, set a predetermined topological correlation coefficient threshold (such as 0.8), and compare each calculated topological correlation coefficient with this threshold one by one to determine whether it reaches or exceeds the preset standard: If the correlation coefficient ≥ threshold, the corresponding battery-bus relationship is determined to be a valid correlation; If the correlation coefficient < threshold, it is regarded as having a non-significant correlation and is not included in the correlation relationship set. Finally, according to the above topological correlation judgment results, automatically screen out all the bus nodes that reach or exceed the threshold to form the final associated DC bus group.

[0083] In this step, by constructing a scientific battery - bus topology diagram and using a quantitative topological correlation evaluation method, accurate and automated analysis and identification of the coupling relationship between the risks of storage batteries and the safety risks of DC buses are achieved. After screening by the correlation threshold, the bus set highly correlated with high - risk storage batteries is objectively determined, enabling operation and maintenance personnel to clarify the scope of fault impact and more effectively conduct risk early warning and prevention and control deployments in advance. Overall, the safety, stability, and operation and maintenance efficiency of the DC power supply of the power system are significantly improved, avoiding serious consequences caused by bus power loss due to battery failures.

[0084] Furthermore, based on the multiple storage battery state data, standard - state deviation detection is performed on the multiple storage batteries to determine multiple storage battery state deviation vectors, including: filtering and denoising the multiple storage battery state data to obtain multiple battery state denoised data; performing vectorization processing on the multiple battery state denoised data to generate multiple storage battery state vectors; respectively performing standard - state mining on the multiple storage batteries to generate multiple storage battery standard - state vectors; and respectively performing deviation identification on the multiple storage battery state vectors according to the multiple storage battery standard - state vectors to generate the multiple storage battery state deviation vectors.

[0085] Specifically, filtering and denoising refers to removing random interference signals or noise in the original measurement data through digital filtering methods (such as Kalman filtering, wavelet transform, mean filtering, etc.) to make the data closer to the true state. Vectorization processing means combining multiple independent storage battery state parameters (such as voltage, current, internal resistance, temperature) into a unified data structure (vector form) for convenient unified analysis and processing. Standard - state mining refers to using data mining techniques (such as clustering analysis, statistical mean analysis) to extract the typical stable operation states of storage batteries from historical normal operation data of storage batteries, representing the typical states during the good operation of the equipment. The storage battery standard - state vector refers to a vector of typical state characteristic data obtained through standard - state mining, which characterizes the normal and stable operation of the storage battery and serves as a reference benchmark for subsequent abnormal analysis and fault identification.

[0086] First, multiple original state data of the storage battery are obtained by real - time collection through pre - deployed sensors. Since there will be environmental interference and measurement noise in the actual collection process, digital filtering methods (such as Kalman filtering, mean filtering, or wavelet filtering, etc.) are first used for filtering and denoising to eliminate interference and obtain more accurate and smoother storage battery state denoised data. Next, the above - mentioned denoised multiple state parameters are subjected to vectorization processing, that is, they are uniformly combined into a single storage battery state vector data.

[0087] Subsequently, the system conducts standard state mining on each storage battery from the historical long-term operating state data of the storage batteries, and uses data mining methods such as clustering analysis (such as K-Means clustering), central tendency analysis (such as statistical averaging method) or other statistical methods to extract the typical states of the storage batteries under stable and normal operating conditions, and form the standard state vectors (standard state feature vectors) of the storage batteries.

[0088] After that, based on the standard state vectors, deviation identification is performed on the current real-time state vectors of the storage batteries. By using difference calculation (such as Euclidean distance, Manhattan distance or vector cosine similarity method), the specific deviation magnitude and deviation trend between the real-time state and the standard state are calculated quantitatively, and the state deviation vectors of each storage battery at the current moment are generated.

[0089] This step significantly improves the accuracy of the storage battery data through filtering and denoising technology. The vectorization process unifies the data format for subsequent analysis. The standard state mining scientifically and accurately determines the normal operating benchmark of the storage battery, and the state deviation identification accurately captures abnormal conditions and potential faults. It realizes a high degree of automation, refinement and intelligence from data collection to abnormal state identification, significantly improves the timeliness and accuracy of storage battery fault diagnosis, can early warn of the abnormal operating trend of the storage battery, reduce the risk of DC bus power loss caused by storage battery faults, and ensure the safe and stable operation of the power system.

[0090] Furthermore, standard state mining is respectively performed on the multiple storage batteries to generate multiple standard state vectors of the storage batteries, including: retrieving normal samples for monitoring according to the multiple storage batteries to obtain multiple battery standard state records; respectively performing central tendency analysis on the multiple battery standard state records to generate the multiple standard state vectors of the storage batteries.

[0091] Specifically, the battery standard state record refers to the parameter data record collected during the normal operation of the storage battery in the historical monitoring process. Central tendency analysis is a statistical analysis method that determines the typical representative features of a data set by analyzing the central distribution trend of the data in the data set, usually including mean, median, and mode analysis methods.

[0092] First, access the historical monitoring database, and through the method of retrieving normal samples for monitoring (such as setting the normal operation threshold range, abnormal data filtering method or clustering analysis method), screen out the data records collected when a large number of storage batteries maintained normal and stable states during the past operation, which are the battery standard state records.

[0093] Specifically, the following methods can be used to obtain normal sample data records: the threshold range method, which sets the normal range thresholds for parameters such as voltage, internal resistance, temperature, and current, and automatically screens the data records within these ranges from historical data; the clustering analysis method, such as using K-means clustering to classify historical data and automatically identify and extract the clusters with stable data distribution and far from abnormal data as the standard state data set.

[0094] Secondly, for the standard state records of each battery selected, perform central tendency analysis (such as the mean method, median method, etc.) respectively to determine the typical and stable states of battery operation: mean analysis, take the average of all normal state data to obtain the central tendency of the data; median analysis, select the median after sorting the data, which can more effectively avoid the interference of outliers and has better stability.

[0095] After the central tendency analysis and processing, the standard state vector corresponding to each battery is automatically obtained. The standard state vector is the typical data representation of the normal state of the battery, and will be used as the basis for subsequent real-time state detection and deviation analysis.

[0096] Through this step, the automated and scientific excavation of the battery standard state is realized, and the standard reference state of the normal operation of the battery is accurately extracted from the vast amount of historical data. The central tendency analysis method effectively reduces the interference of abnormal data on the determination of the standard state, and improves the robustness and representativeness of the standard state.

[0097] In summary, the battery state coupling fault detection method provided by the embodiment of the present application has the following technical effects: 1. Through the real-time monitoring of the battery state and the accurate prediction of the ARIMA-LSTM combined model, the rapid identification and accurate early warning of the potential power loss risk of the DC bus are realized. It significantly improves the accuracy and timeliness of the fault correlation analysis between the battery and the bus, effectively prevents the chain risk of power loss of the bus caused by the abnormal state of the battery, and ensures the stable and safe operation of the power system.

[0098] 2. Construct an ARIMA-LSTM fusion prediction channel, fully combine the advantages of short-term linear and long-term non-linear trend analysis, and improve the accuracy and robustness of fault prediction. Make the identification of potential risks of the battery more detailed and reliable, ensure more accurate risk assessment, effectively discover potential fault hazards in advance, and reduce the risk of safety accidents caused by abnormal battery operation.

[0099] 3. Through the power loss coupling fault registration, confidence evaluation and data fusion, accurately predict the power loss risk of the associated bus. Identify the potential power loss risk range and probability of the DC bus in advance, provide detailed and reliable risk analysis results, greatly improve the bus fault risk prediction ability, enable the operation and maintenance personnel to better deploy targeted preventive measures, and ensure the safe operation of the system.

[0100] Any step of the method described above can be stored as computer instructions or programs in a computer memory without limitation and can be called and recognized by a computer processor without limitation to implement any of the methods in the embodiments of the present application, and no redundant limitation is made here.

[0101] Furthermore, the first or second described above does not only represent an order relationship, but also represents a specific concept, and / or means that multiple elements can be selected individually or in whole. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and variations.

Claims

1. A method for detecting the coupled faults of battery states facing the loss of power of the DC bus, characterized in that, The method includes: Real-time monitoring of each storage battery in the power system through a pre-deployed sensing module to obtain a plurality of storage battery state data, where the power system includes a plurality of storage batteries and a plurality of DC buses; Performing standard state deviation detection on the plurality of storage batteries according to the plurality of storage battery state data to determine a plurality of storage battery state deviation vectors; Performing fault prediction on the plurality of storage battery state deviation vectors according to the ARIMA-LSTM combined model to determine a risk fault prediction result corresponding to the faulty storage battery; Performing topological association analysis on the plurality of DC buses according to the faulty storage battery to determine an associated DC bus group; Performing power loss coupling fault prediction on the associated DC bus group according to the risk fault prediction result to determine a plurality of bus power loss coupling fault prediction results; Performing fault operation and maintenance on the power system according to the risk fault prediction result and the plurality of bus power loss coupling fault prediction results.

2. The method for detecting the coupling fault of the battery state facing the loss of power of the DC bus according to claim 1, wherein Performing fault prediction on the plurality of storage battery state deviation vectors according to the ARIMA-LSTM combined model to determine a risk fault prediction result corresponding to the faulty storage battery, including: Performing fault prediction on the plurality of storage battery state deviation vectors according to the ARIMA-LSTM combined model to obtain a plurality of battery fault prediction results; Judging whether a plurality of battery fault probabilities in the plurality of battery fault prediction results are greater than or equal to a predetermined fault probability; If any one of the plurality of battery fault probabilities is greater than or equal to the predetermined fault probability, obtaining a risk fault probability; Adding the storage battery corresponding to the risk fault probability to the faulty storage battery, and adding the battery fault prediction result corresponding to the faulty storage battery to the risk fault prediction result.

3. The battery state coupling fault detection method for DC bus power loss according to claim 2, characterized in that, Performing fault prediction on the plurality of storage battery state deviation vectors according to the ARIMA-LSTM combined model to obtain a plurality of battery fault prediction results, including: Extracting a first storage battery state deviation vector corresponding to a first storage battery according to the plurality of storage battery state deviation vectors; Retrieving a storage battery state deviation vector sample set and a storage battery fault sample set corresponding to the first storage battery; Supervising and training the ARIMA model and the LSTM model in the ARIMA-LSTM combined model according to the storage battery state deviation vector sample set and the storage battery fault sample set to obtain a first storage battery fault predictor and a second storage battery fault predictor; Training a storage battery fault prediction fusion device with the output data sets of the first storage battery fault predictor and the second storage battery fault predictor as input information and the storage battery fault sample set as output information; Taking the first storage battery fault predictor and the second storage battery fault predictor as the first node of the storage battery fault prediction, and taking the storage battery fault prediction fusion device as the second node of the storage battery fault prediction to build a first storage battery fault prediction channel; Inputting the first storage battery state deviation vector into the first storage battery fault prediction channel to obtain a first battery fault prediction result, and adding the first battery fault prediction result to the plurality of battery fault prediction results.

4. The battery state coupling fault detection method for DC bus power loss as described in claim 1, wherein, Perform a power loss coupling fault prediction on the associated DC bus group according to the risk fault prediction result, and determine multiple bus power loss coupling fault prediction results, including: Extract the first associated DC bus according to the associated DC bus group; Collect the real-time state parameters of the first associated DC bus to obtain the first associated bus state data; Perform a power loss coupling fault registration on the first associated DC bus according to the risk fault prediction result and the first associated bus state data to obtain a first bus power loss coupling fault registration space; Perform a confidence evaluation on each registered power loss coupling fault in the first bus power loss coupling fault registration space to obtain multiple registered fault confidence coefficients; Based on the multiple registered fault confidence coefficients, perform an optimization screening on the first bus power loss coupling fault registration space according to the registered fault confidence threshold to generate a first bus power loss coupling fault confidence space; Perform data fusion according to the first bus power loss coupling fault confidence space to generate a first bus power loss coupling fault prediction result, and add the first bus power loss coupling fault prediction result to the multiple bus power loss coupling fault prediction results.

5. The battery state coupling fault detection method for DC bus power loss as claimed in claim 4, characterized in that, Performing a power loss coupling fault registration on the first associated DC bus according to the risk fault prediction result and the first associated bus state data to obtain a first bus power loss coupling fault registration space, including: Obtain the power loss coupling fault event set of the first associated DC bus; According to the power loss coupling fault event set, extract the first power loss coupling fault event, and the first power loss coupling fault event includes the first battery fault record, the first bus state record and the first power loss coupling fault record corresponding to the first associated DC bus; Perform a correlation evaluation on the first battery fault record and the first bus state record according to the risk fault prediction result and the first associated bus state data to obtain a first power loss coupling scenario correlation coefficient; If the first power loss coupling scenario correlation coefficient is greater than or equal to the power loss coupling scenario correlation threshold, record the first power loss coupling fault record as the first registered power loss coupling fault, and add the first registered power loss coupling fault to the first bus power loss coupling fault registration space.

6. The battery state coupling fault detection method for DC bus power loss as described in claim 1, characterized in that Perform a topological association analysis on the multiple DC buses according to the faulty battery to determine the associated DC bus group, including: Construct a battery-bus topology graph with the multiple batteries as multiple parent nodes, the multiple DC buses as multiple child nodes, and the multiple connection relationships between the multiple batteries and the multiple DC buses as multiple connection edges; Perform a topological association evaluation on the faulty battery and the multiple DC buses according to the battery-bus topology graph to obtain multiple battery-bus topological correlation coefficients; Judge whether the multiple battery-bus topological correlation coefficients are greater than or equal to a predetermined topological correlation coefficient to obtain multiple topological association judgment results; According to the multiple topological association judgment results, screen the multiple DC buses to generate the associated DC bus group.

7. The battery state coupling fault detection method for DC bus power loss according to claim 1, wherein Perform standard state deviation detection on the multiple storage batteries according to the multiple storage battery state data, and determine multiple storage battery state deviation vectors, including: Perform filtering and denoising on the multiple storage battery state data to obtain multiple battery state denoised data; Perform vectorization processing on the multiple battery state denoised data to generate multiple storage battery state vectors; Perform standard state mining on the multiple storage batteries respectively to generate multiple storage battery standard state vectors; According to the multiple storage battery standard state vectors, perform deviation identification on the multiple storage battery state vectors respectively to generate the multiple storage battery state deviation vectors.

8. The battery state coupling fault detection method for DC bus power loss as described in claim 7, wherein Perform standard state mining on the multiple storage batteries respectively to generate multiple storage battery standard state vectors, including: Retrieve normal monitoring samples according to the multiple storage batteries to obtain multiple battery standard state records; Perform central tendency analysis on the multiple battery standard state records respectively to generate the multiple storage battery standard state vectors.

Images