A substation intelligent operation and maintenance method and system for cloud large model fault diagnosis and prediction

Abstract

The application discloses a substation intelligent operation and maintenance method and system for cloud large model fault diagnosis and prediction, and belongs to the technical field of substation operation and maintenance. The method comprises the following steps: analyzing a to-be-built large model, obtaining diagnosis collection data and prediction collection data required for fault diagnosis and fault prediction of the to-be-built large model, generating a data optimization scheme for an edge end according to the diagnosis collection data and the prediction collection data, and configuring a corresponding data optimization model for the edge end according to the data optimization scheme; collecting real-time data of a substation to obtain collection data of the corresponding substation; processing the preprocessed collection data through a preset data optimization model to obtain corresponding diagnosis collection data and prediction collection data, and sending the diagnosis collection data and the prediction collection data to a cloud end; and analyzing the diagnosis collection data and the prediction collection data through a preset cloud large model to obtain corresponding fault diagnosis result data and fault prediction result data.

Description

Technical Field

[0001] This invention belongs to the field of substation operation and maintenance technology, specifically a cloud-based large-scale model fault diagnosis and prediction method and system for intelligent operation and maintenance of substations. Background Technology

[0002] With the rapid development of technologies such as the Internet of Things, big data, and artificial intelligence, intelligent methods are gradually being introduced into the substation operation and maintenance field to improve management. Early intelligent operation and maintenance systems deployed various sensors to collect equipment operating data in real time and used simple data analysis methods (such as threshold comparison and trend analysis) to conduct preliminary assessments of equipment status. However, these systems have significant shortcomings: on the one hand, the data analysis methods are relatively crude, only able to identify abnormal changes in single parameters, making it difficult to capture the multi-parameter correlation characteristics under complex equipment fault modes, resulting in low fault diagnosis accuracy; on the other hand, the systems lack in-depth mining and learning of historical equipment operating data, making it impossible to predict future fault development trends, still relying mainly on post-incident handling, and failing to fundamentally change the passive situation of traditional operation and maintenance models.

[0003] In recent years, large-scale modeling technology in the field of artificial intelligence has made groundbreaking progress, demonstrating powerful capabilities in multiple areas such as natural language processing, image recognition, and time series prediction. Trained on massive amounts of data, large-scale models can automatically learn complex patterns and inherent laws within the data, possessing powerful feature extraction and generalization capabilities. Applying large-scale modeling technology to the field of substation fault diagnosis and prediction provides a new approach to solving the challenges of traditional operation and maintenance models; based on this, this invention provides a cloud-based large-scale model-based intelligent operation and maintenance method and system for substation fault diagnosis and prediction. Summary of the Invention

[0004] To address the problems of the above solutions, this invention provides a smart operation and maintenance method and system for substations based on cloud-based large-scale model fault diagnosis and prediction.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A smart operation and maintenance system for substations with cloud-based large-scale model fault diagnosis and prediction, including cloud, edge, and device terminals;

[0007] The cloud-based system includes a large model analysis module, a fault diagnosis module, and a fault prediction module.

[0008] The large model analysis module is used to analyze the large model to be built, obtain the diagnostic and predictive data required for fault diagnosis and fault prediction of the large model to be built, generate a data optimization scheme for the edge end based on the diagnostic and predictive data, and configure the corresponding data optimization model for the edge end based on the data optimization scheme.

[0009] Furthermore, the model to be built is analyzed, including:

[0010] Define intermediate data, which is the intermediate result generated by the edge device after optimizing the preprocessed collected data for use in large-scale cloud model analysis;

[0011] Based on the definition of intermediate data, various intermediate data are identified for fault diagnosis and fault prediction in substations.

[0012] Each intermediate data is prioritized, and the intermediate data with the highest priority is marked as the target intermediate data. Based on the target intermediate data, the diagnostic and predictive data required for fault diagnosis and fault prediction of the large model to be built are determined.

[0013] Furthermore, based on the definition of intermediate data, various intermediate data for fault diagnosis and fault prediction in substations are identified, including:

[0014] The process involves identifying various large-scale model building schemes for fault diagnosis and fault prediction in substations, identifying the large-scale models to be built based on the schemes, and processing the initial diagnostic data and initial prediction data separately for fault diagnosis and fault prediction. This results in the formation of diagnostic data change chains and prediction data change chains corresponding to fault diagnosis and fault prediction, respectively. The initial diagnostic data and initial prediction data refer to the input data obtained after preprocessing the collected data for fault diagnosis and fault prediction.

[0015] By combining the change chains of diagnostic data and the change chains of predicted data, various intermediate data are obtained.

[0016] Furthermore, the data from each intermediate component is prioritized, including:

[0017] Identify resource data at the edge, filter the data of each intermediate entity based on the resource data, and remove intermediate data that does not meet the resource data restriction requirements;

[0018] The fault diagnosis efficiency, fault diagnosis accuracy, fault prediction efficiency, and fault prediction accuracy are estimated based on the corresponding intermediate data. The fault diagnosis efficiency, fault diagnosis accuracy, fault prediction efficiency, and fault prediction accuracy are then substituted into the preset priority value formula for calculation to determine the priority value corresponding to each intermediate data.

[0019] The intermediate data are sorted in descending order of priority, and the intermediate data ranked first is selected as the target intermediate data.

[0020] Furthermore, the intermediate data is filtered based on the resource data, including:

[0021] Build a resource calibration model, and the expression of the resource calibration model is:

[0022]

[0023] In the formula: (s, V) is the input data, s is the intermediate data and the collected data, and V is the resource data; s→V means that the corresponding intermediate data meets the limitation requirements of the resource data; the output data is the resource calibration value PL(s, V), and the resource calibration value is 1 or 0;

[0024] Integrate each intermediate data and the corresponding collected data into the corresponding input data and input it into the resource calibration model for analysis to obtain the resource calibration value of the corresponding intermediate data;

[0025] When the resource calibration value is 1, no corresponding operation is performed;

[0026] When the resource calibration value is 0,剔除 the corresponding intermediate data.

[0027] Further, the priority formula is:

[0028] QW = λ1×(b1×GL + b2×LD) + λ2×(b1×GY + b2×YD);

[0029] In the formula: QW is the priority, b1 and b2 are both proportionality coefficients, and the value range is 0 < b1 ≤ 1, 0 < b2 ≤ 1; λ1 and λ2 are the weight coefficients corresponding to the corresponding fault diagnosis and fault prediction respectively, and λ1 + λ2 = 1; GL is the fault diagnosis efficiency; LD is the fault diagnosis accuracy; GY is the fault prediction efficiency; YD is the fault prediction accuracy.

[0030] The fault diagnosis module is used to perform fault diagnosis on the substation, real-time identify the received diagnostic acquisition data, and call a preset large model for analysis according to the diagnostic acquisition data to obtain the corresponding fault diagnosis result data.

[0031] The fault prediction module is used to perform fault prediction on the substation, real-time identify the received prediction acquisition data, and call a preset large model for analysis according to the prediction acquisition data to obtain the corresponding fault prediction result data.

[0032] The device end includes a collection module;

[0033] The collection module is used to perform real-time data collection on the substation, obtain the collection data of the corresponding substation, and send the collection data to the edge end.

[0034] The edge end includes a data processing module;

[0035] The data processing module is used to process data, receive the collected data sent by the device, preprocess the collected data, process the preprocessed collected data through a preset data optimization model, obtain corresponding diagnostic collected data and predictive collected data, and send the diagnostic collected data and predictive collected data to the cloud.

[0036] Furthermore, the edge terminal also includes an emergency processing module, which is used to perform emergency processing, acquire fault prediction result data and scene status in real time, analyze the scene status in real time, and determine whether the emergency standards are met.

[0037] If the emergency criteria are not met, no corresponding action will be taken.

[0038] When the emergency criteria are met, the safe period is determined in real time based on the fault prediction results. If the time is within the safe period, no action is taken. If the time exceeds the safe period, an early warning is issued to the staff, and the pre-processed collected data is analyzed using a preset simplified fault diagnosis model to obtain emergency fault diagnosis results.

[0039] A smart operation and maintenance method for substations based on cloud-based large-scale model fault diagnosis and prediction includes:

[0040] The large-scale model to be built is analyzed to obtain the diagnostic and predictive data required for fault diagnosis and prediction. Based on the diagnostic and predictive data, a data optimization scheme for the edge is generated, and a corresponding data optimization model is configured for the edge based on the data optimization scheme.

[0041] Real-time data acquisition is performed on substations to obtain the corresponding substation data, and the acquired data is sent to the edge terminal.

[0042] The edge device preprocesses the collected data and uses a preset data optimization model to process the preprocessed data to obtain corresponding diagnostic and predictive collected data, which are then sent to the cloud.

[0043] The diagnostic and predictive data are analyzed using a pre-set cloud-based big data model to obtain corresponding fault diagnosis and prediction results.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] Traditional intelligent substation operation and maintenance systems employ simple data analysis methods, such as threshold comparison and trend analysis, which can only identify abnormal changes in single parameters. They struggle to capture the multi-parameter correlation characteristics of equipment under complex fault modes, resulting in low fault diagnosis accuracy. This invention, however, introduces cloud-based large-scale model technology. These models, trained on massive amounts of data, possess powerful feature extraction and generalization capabilities, automatically learning complex patterns and inherent laws in equipment operating data to deeply explore the correlations between multiple parameters. Through comprehensive analysis of these complex features, the fault types and states of equipment can be identified more accurately, significantly improving fault diagnosis accuracy, reducing false positives and false negatives, and providing a more reliable guarantee for the stable operation of substation equipment. Simultaneously, edge-based data optimization processing accelerates fault diagnosis and prediction efficiency while reducing cloud-based pressure. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0047] Figure 1 This is a block diagram illustrating the principle of the present invention. Detailed Implementation

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

[0049] like Figure 1 As shown, a cloud-based large-scale model fault diagnosis and prediction system for substations is presented, comprising cloud, edge, and device terminals.

[0050] The cloud-based system includes a large model analysis module, a fault diagnosis module, and a fault prediction module.

[0051] The large model analysis module is used to analyze the large model to be built, obtain the diagnostic and predictive data required for fault diagnosis and fault prediction of the large model to be built, generate a data optimization scheme for the edge end based on the diagnostic and predictive data, and deploy the corresponding data optimization model for the edge end according to the data optimization scheme.

[0052] The large-scale model to be built refers to the large-scale model yet to be established for fault diagnosis and prediction in substations. Depending on the scenario requirements, the large-scale model can adopt a single large-scale model or a dual-model architecture. A single large-scale model means that the single large-scale model performs dual functions; a dual-model architecture refers to situations where the diagnostic and prediction tasks differ significantly (e.g., diagnosis requires high-precision classification, while prediction requires long-term time-series modeling), and the independent model can have its architecture optimized for its respective task. For example: the diagnostic model uses a lightweight CNN to quickly process real-time data streams; the prediction model uses a deep Transformer to capture long-term dependencies. Once the diagnostic and prediction data collection processes are clear, the large-scale model will be built based on existing large-scale model technologies.

[0053] In one embodiment, the established large model can be analyzed using diagnostic and predictive data, or it can be analyzed using data collected after initial preprocessing. Redundancy design ensures continuous operation and maintenance.

[0054] In one embodiment, the analysis of the large model to be built includes:

[0055] Intermediate data is defined as the intermediate result generated by the edge after feature extraction of preprocessed collected data, which is used for large-scale cloud model analysis. Its core objective is to retain the most critical information for cloud tasks (such as fault diagnosis and prediction) while reducing the amount of data transmission and improving the analysis efficiency of large-scale cloud models.

[0056] Based on the definition of intermediate data and existing large model technology, the intermediate data that can be used to perform fault diagnosis and fault prediction in substations is determined. This is the integrated data of diagnostic data and prediction data. The simplest intermediate data is the pre-processed data. Then, depending on the analysis requirements of the large model, the data is gradually processed to form process data of different degrees, which are marked as intermediate data.

[0057] Each intermediate data is prioritized, and the intermediate data with the highest priority is marked as the target intermediate data. Based on the target intermediate data, the diagnostic and predictive data required for fault diagnosis and fault prediction of the large model to be built are determined.

[0058] In one embodiment, the intermediate data that can be used to perform fault diagnosis and fault prediction in substations is determined based on the definition of intermediate data and existing large model technology. The intermediate data is determined according to the above definition and existing technology, such as building an intelligent model based on deep learning algorithms, machine learning and other intelligent algorithms.

[0059] In one embodiment, various intermediate data for fault diagnosis and fault prediction of substations are determined based on intermediate data definitions, including:

[0060] Based on existing large model technology, various large model establishment schemes for realizing fault diagnosis and fault prediction in substations are identified. According to the large model establishment scheme, the process of processing the initial diagnostic data and the initial prediction data for fault diagnosis and fault prediction is carried out separately to identify the large model to be built. The process forms the diagnostic data change chain and the prediction data change chain corresponding to fault diagnosis and fault prediction, respectively. The initial diagnostic data and the initial prediction data refer to the input data for fault diagnosis and fault prediction obtained after only preprocessing the collected data.

[0061] By combining and analyzing the diagnostic data change chain and the predicted data change chain, various intermediate data are obtained. For example, if the diagnostic data change chain is AB and the predicted data change chain is CD, then there are AC, AD, BC, and BD data combinations, with each combination corresponding to one intermediate data.

[0062] The diagnostic data change chain refers to the gradual change process of the initial fault diagnosis data. For example, the initial diagnostic data is processed into data A, data A is processed into data B, and so on, until the fault diagnosis result data is obtained (not in the change chain).

[0063] For example, edge preprocessing, electrical quantity dimensionality reduction:

[0064] Time-frequency analysis: Perform short-time Fourier transform on the current waveform to extract the energy distribution in the 0-500Hz frequency band; Feature extraction: Calculate the fundamental amplitude, total harmonic distortion (THD), and number of transient overvoltages;

[0065] Standardization of non-electrical quantities:

[0066] Gas data: Encoded using the IEC three-ratio method (e.g., C2H2 / C2H4 = 0.1 → code 0, CH4 / H2 = 0.8 → code 1); Ultrasonic signals: Extracting features such as peak energy, pulse interval time, and frequency band energy ratio.

[0067] Time alignment: Synchronize multi-source data according to timestamps to generate comprehensive data packets with a 1-minute granularity.

[0068] Multimodal data fusion:

[0069] Spatiotemporal correlation: cross-analysis of gas concentration changes with load curves and temperature trends; Fault mode mapping: gas code [0, 1, 2] → corresponds to "medium and low temperature overheating" fault mode; sudden increase in partial discharge pulse frequency → corresponds to "insulation aging" characteristic;

[0070] Dynamic threshold adjustment:

[0071] Based on historical equipment operating data, the normal range threshold is adaptively updated (e.g., the normal oil temperature range is dynamically adjusted from 65-75℃ to 62-78℃).

[0072] Feature compression:

[0073] PCA dimensionality reduction was applied to compress the original 30-dimensional features into an 8-dimensional key feature vector;

[0074] Input data format:

[0075] Time-series characteristics: 8-dimensional time series of the most recent hour; Static characteristics: equipment model, years of operation, historical fault records; Real-time alarms: list of alarm signals that have not yet been reset;

[0076] Model processing procedure:

[0077] Phase 1: The Transformer encoder captures temporal dependencies and generates contextual feature representations; Phase 2: The graph neural network fuses device topology relationships (such as the connection relationship between transformers and circuit breakers); Phase 3: The multi-task learning head simultaneously outputs: fault type probability distribution (e.g., [winding deformation: 0.72, core grounding: 0.18, normal: 0.10]); fault severity classification (levels 1-5); remaining service life prediction (e.g., "fault is expected within 3 months");

[0078] Quantification of uncertainty:

[0079] Output confidence scores: 92% confidence for primary fault types and 68% confidence for secondary fault types.

[0080] In one embodiment, prioritizing the data from each intermediate component includes:

[0081] Identifying edge resource data refers to resource data related to the subsequent data optimization model setup. This avoids disruptions to normal operations due to insufficient data. It primarily considers core factors such as hardware resource limitations, real-time requirements, environmental adaptability, and security and reliability of edge devices, including computing resources, energy and power consumption, and real-time and latency requirements. Based on the resource data, intermediate data is filtered, eliminating intermediate data that exceeds the resource data limits after processing the collected data at the edge.

[0082] The efficiency, accuracy, and prediction efficiency of fault diagnosis are estimated based on the analysis of the corresponding intermediate data. This involves estimating the efficiency, accuracy, and prediction accuracy of fault diagnosis by analyzing the intermediate data using a large model. This can be achieved through simulation evaluation using a similar large model built in the same way, or by evaluating historical data from similar large models. Alternatively, it can be determined using common knowledge in the field, specifically by employing various existing methods to determine the efficiency, accuracy, and prediction accuracy of fault diagnosis.

[0083] The fault diagnosis efficiency, fault diagnosis accuracy, fault prediction efficiency, and fault prediction accuracy are substituted into the preset priority value formula for calculation to determine the priority value corresponding to each intermediate data.

[0084] The intermediate data are sorted in descending order of priority, and the intermediate data ranked first is selected as the target intermediate data.

[0085] In one embodiment, the intermediate data is filtered based on resource data. This can be done using existing filtering and evaluation algorithms, such as estimating whether the resources required to obtain the intermediate data from the collected data exceed the limits of the resource data.

[0086] In one embodiment, filtering the intermediate data based on resource data includes:

[0087] A resource calibration model is established to assess whether processing the collected data into corresponding intermediate data using existing technologies exceeds the limits of resource data. The expression for the resource calibration model is:

[0088]

[0089] In the formula: (s, V) are the input data, s are the intermediate data and the collected data, and V are the resource data; s→V means that the corresponding intermediate data meets the resource data constraint requirements; the corresponding training set is set up using the historical training data of the existing large model for training; the output data is the resource calibration value PL(s, V), and the resource calibration value is 1 or 0;

[0090] The intermediate data and the corresponding collected data are integrated into the corresponding input data and input into the resource calibration model for analysis to obtain the resource calibration value of the corresponding intermediate data;

[0091] When the resource calibration value is 1, no corresponding operation is performed;

[0092] When the resource calibration value is 0, the corresponding intermediate data is discarded.

[0093] In one embodiment, the priority value formula is:

[0094] QW = λ1 × (b1 × GL + b2 × LD) + λ2 × (b1 × GY + b2 × YD);

[0095] In the formula: QW is the priority value, b1 and b2 are both proportionality coefficients, and the value range is 0 < b1 ≤ 1, 0 < b2 ≤ 1; λ1 and λ2 are the weight coefficients corresponding to the respective fault diagnosis and fault prediction, and λ1 + λ2 = 1; GL is the fault diagnosis efficiency; LD is the fault diagnosis accuracy; GY is the fault prediction efficiency; YD is the fault prediction accuracy.

[0096] In one embodiment, the priority value formula can be the existing calculation formula for the priority value. For example, if the above priority value formula does not consider the proportionality coefficient and the weight coefficient, the efficiency change weight can also be increased according to the exponential relationship.

[0097] In one embodiment, an edge - side data optimization scheme is generated based on the diagnostic acquisition data and the prediction acquisition data, that is, the target intermediate data is clarified. Taking the target intermediate data as the goal, the processing method for processing the pre - processed acquisition data into the target intermediate data is determined and marked as the data optimization scheme. The existing data processing methods can be screened according to the use of edge - side resources, data processing efficiency, etc.; subsequently, a data optimization model for automatically processing the pre - processed acquisition data is established according to the data optimization method.

[0098] The fault diagnosis module is used to perform fault diagnosis on the substation, identify the received diagnostic acquisition data in real - time, and analyze it by calling the corresponding preset large - model according to the diagnostic acquisition data to obtain the corresponding fault diagnosis result data.

[0099] The fault prediction module is used to perform fault prediction on the substation, identify the received prediction acquisition data in real - time, and analyze it by calling the corresponding preset large - model according to the prediction acquisition data to obtain the corresponding fault prediction result data.

[0100] The device - side includes a collection module;

[0101] The collection module is used to perform real - time data collection on the substation, obtain the acquisition data of the corresponding substation, and send the acquisition data to the edge - side.

[0102] The device - side is equipped with various types of sensors to collect the electrical parameters (such as voltage, current, power, etc.), mechanical parameters (such as vibration, displacement, etc.), temperature parameters, and environmental parameters (such as temperature and humidity, gas concentration, etc.) of the substation equipment in real - time, summarize them into acquisition data, and transmit the acquisition data to the edge - side quickly and stably through wired or wireless communication methods.

[0103] The edge - side includes a data processing module and an emergency processing module;

[0104] The data processing module is used to process data, receive the collected data sent by the device, preprocess the collected data, process the preprocessed collected data through a preset data optimization model, obtain corresponding diagnostic collected data and predictive collected data, and send the diagnostic collected data and predictive collected data to the cloud.

[0105] The emergency processing module is used for emergency handling, acquiring fault prediction result data and scenario status in real time. Scenario status refers to the status of related backgrounds such as data acquisition and transmission at the acquisition end, data processing and transmission at the edge end, and fault diagnosis and prediction in the cloud. It determines whether there are any abnormal situations that prevent the timely output of substation fault diagnosis result data. Therefore, it is necessary to acquire the corresponding scenario status in real time. In other words, scenario status refers to data related to the generation of fault diagnosis result data. The module performs real-time analysis of the scenario status to determine whether it meets the emergency criteria. Emergency criteria are situations that lead to the generation of fault diagnosis result data.

[0106] If the emergency criteria are not met, no corresponding action will be taken.

[0107] When the emergency criteria are met, the safe period is determined in real time based on the fault prediction results. That is, the time period within which there will be no fault is determined based on the fault prediction results. Alternatively, a period can be shifted forward for subsequent data analysis. Generally, the safe period is determined based on the final fault prediction results. If the fault prediction results can be continuously updated, the safe period will be continuously updated. When the time is within the safe period, no corresponding processing is performed. When the time exceeds the safe period, an early warning is issued to the staff, and the pre-processed collected data is analyzed using a preset simplified fault diagnosis model to obtain emergency fault diagnosis results.

[0108] In one embodiment, the simplified fault diagnosis model is a simplified fault diagnosis model established using existing fault diagnosis technology, used to conduct preliminary fault analysis during emergency periods and provide auxiliary reference for staff.

[0109] In one embodiment, a simplified fault model may not be set up, that is, the process ends after issuing an early warning.

[0110] In one embodiment, the scenario status is analyzed in real time to determine whether it meets the emergency criteria. The judgment is made based on existing methods, and various scenario statuses that meet the emergency criteria are listed for subsequent matching and judgment. Simulation analysis and evaluation can also be used to determine whether the emergency criteria are met.

[0111] A cloud-based large-scale model-based intelligent operation and maintenance method for substations includes:

[0112] The large-scale model to be built is analyzed to obtain the diagnostic and predictive data required for fault diagnosis and prediction. Based on the diagnostic and predictive data, a data optimization scheme for the edge is generated, and a corresponding data optimization model is configured for the edge based on the data optimization scheme.

[0113] Real-time data acquisition is performed on substations to obtain the corresponding substation data, and the acquired data is sent to the edge terminal.

[0114] The edge device preprocesses the collected data and uses a preset data optimization model to process the preprocessed data to obtain corresponding diagnostic and predictive collected data, which are then sent to the cloud.

[0115] The diagnostic and predictive data are analyzed using a pre-set cloud-based big data model to obtain corresponding fault diagnosis and prediction results.

[0116] The above formulas are all numerical calculations after removing dimensions. The formulas are obtained by software simulation based on a large amount of data and are closest to the real situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by simulation based on a large amount of data.

[0117] The above embodiments are only used to illustrate the technical methods 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 methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A cloud-based large-scale model-based intelligent operation and maintenance system for substation fault diagnosis and prediction, characterized in that, Including cloud, edge, and device; The cloud-based system includes a large model analysis module, a fault diagnosis module, and a fault prediction module. The large model analysis module is used to analyze the large model to be built, obtain the diagnostic and predictive data required for fault diagnosis and fault prediction of the large model to be built, generate a data optimization scheme for the edge end based on the diagnostic and predictive data, and configure the corresponding data optimization model for the edge end based on the data optimization scheme. The fault diagnosis module is used to diagnose faults in substations, identify the received diagnostic data in real time, and call a preset large model to analyze the data to obtain the corresponding fault diagnosis results. The fault prediction module is used to predict faults in substations, identify the received prediction data in real time, and call a preset large model to analyze the prediction data to obtain the corresponding fault prediction results. The device includes a data acquisition module; The acquisition module is used to acquire data from the substation in real time, obtain the acquisition data of the corresponding substation, and send the acquisition data to the edge terminal. The edge terminal includes a data processing module; The data processing module is used to process data, receive the collected data sent by the device, preprocess the collected data, process the preprocessed collected data through a preset data optimization model, obtain the corresponding diagnostic collected data and predictive collected data, and send the diagnostic collected data and predictive collected data to the cloud. The analysis of the model to be built includes: Define intermediate data, which is the intermediate result generated by the edge device after optimizing the preprocessed collected data for use in large-scale cloud model analysis; Based on the definition of intermediate data, various intermediate data are identified for fault diagnosis and fault prediction in substations. Prioritize each intermediate data, mark the intermediate data with the highest priority as the target intermediate data, and determine the diagnostic and predictive data required for fault diagnosis and fault prediction of the large model to be built based on the target intermediate data. Based on the definition of intermediate data, various intermediate data are identified for fault diagnosis and fault prediction in substations, including: The process involves identifying various large-scale model building schemes for fault diagnosis and fault prediction in substations, identifying the large-scale models to be built based on the schemes, and processing the initial diagnostic data and initial prediction data separately for fault diagnosis and fault prediction. This results in the formation of diagnostic data change chains and prediction data change chains corresponding to fault diagnosis and fault prediction, respectively. The initial diagnostic data and initial prediction data refer to the input data obtained after preprocessing the collected data for fault diagnosis and fault prediction. By combining the change chains of diagnostic data and the change chains of predicted data, various intermediate data are obtained. Prioritize the data from each intermediate component, including: Identify resource data at the edge, filter the data of each intermediate entity based on the resource data, and remove intermediate data that does not meet the resource data restriction requirements; Estimate the fault diagnosis efficiency, fault diagnosis accuracy, fault prediction efficiency, and fault prediction accuracy for analysis according to the corresponding intermediate data. Substitute the fault diagnosis efficiency, fault diagnosis accuracy, fault prediction efficiency, and fault prediction accuracy into the preset priority value formula for calculation to determine the priority value corresponding to each intermediate data; Sort the intermediate data in descending order of the priority value, and select the intermediate data ranked first as the target intermediate data.

2. The intelligent operation and maintenance system for substation fault diagnosis and prediction based on a cloud-based large-scale model, as described in claim 1, is characterized in that... Screen each intermediate data according to the resource data, including: Establish a resource calibration model, and the expression of the resource calibration model is: ； In the formula: (s, V) is the input data, s is the intermediate data and the collected data, V is the resource data; s→V means that the corresponding intermediate data meets the limitation requirements of the resource data; the output data is the resource calibration value PL(s, V), and the resource calibration value is 1 or 0; Integrate each intermediate data with the corresponding collected data into the corresponding input data and input it into the resource calibration model for analysis to obtain the resource calibration value of the corresponding intermediate data; When the resource calibration value is 1, no corresponding operation is performed; When the resource calibration value is 0, the corresponding intermediate data is excluded.

3. The intelligent operation and maintenance system for substation fault diagnosis and prediction based on a cloud-based large model, as described in claim 2, is characterized in that... The priority value formula is: QW = λ1×(b1×GL + b2×LD) + λ2×(b1×GY + b2×YD); In the formula: QW is the priority value, b1 and b2 are both proportionality coefficients, and the value range is 0 < b1 ≤ 1, 0 < b2 ≤ 1; λ1 and λ2 are the weight coefficients corresponding to the corresponding fault diagnosis and fault prediction respectively, and λ1 + λ2 = 1; GL is the fault diagnosis efficiency; LD is the fault diagnosis accuracy; GY is the fault prediction efficiency; YD is the fault prediction accuracy.

4. The intelligent operation and maintenance system for substation fault diagnosis and prediction based on a cloud-based large-scale model as described in claim 1, characterized in that, The edge side further includes an emergency handling module, and the emergency handling module is used for emergency handling, obtaining the fault prediction result data and the scene status in real time, analyzing the scene status in real time, and judging whether it meets the emergency standard; When it is judged that the emergency standard is not met, no corresponding operation is performed; When it is judged that the emergency standard is met, determine the safe time period in real time according to the fault prediction result data. When the time is within the safe time period, no corresponding treatment is performed; When the time exceeds the safe time period, give a warning to the staff, and analyze the preprocessed collected data through a preset simple fault diagnosis model to obtain the emergency fault diagnosis result data.

5. A cloud-based large-scale model-based intelligent operation and maintenance method for substation fault diagnosis and prediction, characterized in that, Applied to a substation intelligent operation and maintenance system for cloud large model fault diagnosis and prediction as described in any one of claims 1 to 4, the method includes: Analyze the to-be-built large model, obtain the diagnostic collection data and prediction collection data required for the to-be-built large model to perform fault diagnosis and fault prediction, generate a data optimization plan for the edge side according to the diagnostic collection data and prediction collection data, and configure the corresponding data optimization model for the edge side according to the data optimization plan; Collect real-time data from the substation, obtain the collection data of the corresponding substation, and send the collection data to the edge side; The edge side preprocesses the collection data, processes the preprocessed collection data through a preset data optimization model, obtains the corresponding diagnostic collection data and prediction collection data, and sends the diagnostic collection data and prediction collection data to the cloud; The diagnostic and predictive data are analyzed using a pre-set cloud-based big data model to obtain corresponding fault diagnosis and prediction results.

Images