Intelligent testing method for substation equipment
By constructing multi-source datasets and intelligent evaluation models, automated substation equipment testing has been achieved, solving the problems of low efficiency and insufficient early warning in traditional methods, and realizing efficient and automatic equipment status assessment and fault early warning.
Patent Information
- Application Number
- CN202511591650.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-03
- Publication Date
- 2026-03-03
AI Technical Summary
Traditional substation equipment testing methods are inefficient, rely on manual labor, and the test results are greatly affected by the skills of the personnel. They also lack the ability to comprehensively analyze data, making it difficult to achieve early warning and intelligent diagnosis of faults.
A multi-source test dataset is constructed, multi-dimensional feature vectors are extracted using feature engineering, an intelligent equipment status evaluation model is trained, test tasks are automatically generated, and test instruments are automatically controlled through edge computing to form a closed-loop optimization.
It enables automatic generation of test tasks and automatic control of instruments, improving testing efficiency, reducing labor costs and errors, capturing early signs of equipment failure, providing proactive fault warnings, and adapting to equipment aging and new defects.
Smart Images
Figure CN121597994A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of substation equipment testing technology, and specifically to an intelligent testing method for substation equipment. Background Technology
[0002] Substations are a crucial component of power systems, and the reliable operation of their internal equipment (such as transformers, circuit breakers, disconnectors, and current / voltage transformers) directly impacts the safety and stability of the power grid. Traditional substation equipment testing methods primarily rely on periodic preventative testing and maintenance. Testing personnel must be on-site to conduct item-by-item tests on the equipment using specialized instruments, following predetermined testing procedures, such as insulation resistance testing, dielectric loss factor testing, and loop resistance testing.
[0003] Traditional methods have several drawbacks: 1. Low efficiency: The testing process relies on manual labor, which is time-consuming and labor-intensive, and the test results are greatly affected by the skills and experience of the personnel; 2. Isolated data: Test data, online monitoring data, historical ledger data, etc., are independent of each other, making it difficult to form an effective comprehensive analysis and accurately assess and predict the equipment status; 3. Lack of early warning capability: Traditional testing methods are mainly used to diagnose defects that have occurred or developed to a certain extent, and it is difficult to achieve early warning and intelligent diagnosis of faults. To solve the above problems, this invention provides an intelligent testing method for substation equipment. Summary of the Invention
[0004] To address the aforementioned technical shortcomings, the present invention aims to provide an intelligent testing method for substation equipment. This method achieves automatic generation of test tasks and automatic control of test instruments, significantly reducing manual intervention, improving testing efficiency, and lowering labor costs and human error. The model can capture subtle changes in equipment status and early fault symptoms, enabling early warning of faults and providing forward-looking basis for operation and maintenance decisions. Through a closed-loop optimization mechanism, the system can continuously improve itself using new test data, enabling the evaluation model to continuously evolve and adapt to equipment aging and new defect patterns. This invention solves the technical problems mentioned in the background section.
[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides an intelligent testing method for substation equipment, comprising: A1: Construct a multi-source test dataset for substation equipment; The multi-source test dataset includes historical test data of the equipment, real-time / near real-time online monitoring data of the equipment, equipment ledger parameter data, and equipment defect / fault record data; A2: Based on the multi-source test dataset, feature engineering methods are used to extract multi-dimensional feature vectors related to the health status of the equipment; A3: Construct and train an intelligent equipment status assessment model, wherein the intelligent equipment status assessment model takes the multidimensional feature vector as input and outputs the equipment status score and / or failure probability. A4: Based on the output of the intelligent equipment status assessment model, test tasks and strategies are automatically generated; A5: Execute the test task and automatically control the test instruments to complete the test through the edge computing gateway or station monitoring system; A6: Collect and analyze the test results, update the multi-source test dataset and the intelligent evaluation model of the device status, and form a closed-loop optimization.
[0006] Preferably, the historical test data of the equipment includes preventive test data, handover and acceptance test data, and special test data; the preventive test data are the results of standardized tests performed during each periodic maintenance; the handover and acceptance test data are all initial test data before the equipment is put into operation; the special test data are transformer short-circuit impedance test data and partial discharge location test data. The real-time / near real-time online monitoring data of the equipment includes electrical quantity monitoring data, status quantity monitoring data, physical quantity monitoring data, and video image data; The equipment ledger parameter data includes identification, technical parameters, and commissioning information; The equipment defect / fault record data includes defect records, fault waveform data, protection action information, and fault analysis conclusions.
[0007] Preferably, the multidimensional feature vector in A2 includes: Static features: equipment model, commissioning date, and manufacturer, extracted from the equipment ledger parameter data; Dynamic characteristics: Real-time monitored current, voltage, temperature, vibration, partial discharge, oil chromatography data and their changing trends are extracted from the real-time / quasi-real-time data; Historical characteristics: longitudinal comparison data of the results of previous preventive tests and historical defect records, extracted from the historical test data of the equipment and the real-time / quasi-real-time online monitoring data of the equipment; Derived features: Feature quantities calculated based on the above features, such as data change rate, horizontal comparison difference with similar devices, and feature frequency components.
[0008] Preferably, the construction and training of the intelligent evaluation model for device status in A3 includes model construction and model training; The model construction includes: using a Long Short-Term Memory (LSTM) network to process pure time series data and extract time series features; concatenating the time series features with static features, dynamic features, historical features, and derived features, and then inputting them into a Gradient Boosting Decision Tree (GBDT) model for state evaluation; The model training includes the following: Training data preparation and label definition: Select data samples at historical time points from the multi-source test dataset, define a health score for the device at each historical time point, and define a time window; Model training and tuning: Data division: Divide the dataset into a training set, a validation set, and a test set in chronological order; Training loop: Input the feature vectors and corresponding labels of the training set into the model; calculate the difference between the predicted value and the true label; adjust the internal parameters of the model through the backpropagation algorithm optimizer to minimize the difference; Hyperparameter tuning: Use the validation set to optimize the hyperparameters of the model and find the best configuration.
[0009] Preferably, A4 includes a series of preset decision thresholds and rules, and the index sequence is as follows: model output status, status score (HS), failure probability (FP), generated test task level, core strategy; The following are the decisions and rules corresponding to the index sequence: Healthy state, HS>85, FP<0.05, routine test, execute according to the original cycle, strategy optimization; Warning state, 70<HS≤85, 0.05≤FP<0.15, warning test, shorten the cycle, targeted test; Abnormal state, 60<HS≤70, 0.15≤FP<0.30, targeted special test, execute immediately, focus on risk points; Critical state, HS≤60, FP≥0.30, emergency diagnostic test, execute immediately, comprehensive diagnosis, link dispatching if necessary.
[0010] Preferably, the process of automatically generating test tasks and strategies in A4 is as follows: A41: Task priority and trigger mechanism: When the model output enters the "critical state" or multiple failure probabilities soar simultaneously, it is automatically and immediately triggered, and an alarm text message / APP push is sent to the operation and maintenance personnel at the same time; generate a work order with the highest priority; A42: Test item recommendation and optimization: Recommendation based on the type of failure. If the model output has a high "discharge failure probability", the following will be preferentially recommended: high-frequency / ultra-high-frequency partial discharge test, ultrasonic detection, and review of dissolved gas analysis in insulating oil; if the "overheating failure probability" is high, the following will be preferentially recommended: infrared precise temperature measurement, loop resistance test, and oil chromatography analysis; Based on feature importance, if the two features "acetylene production rate" and "vibration spectrum entropy" rank the most important in the evaluation, then the test items most relevant to these two features will be generated first. A43: Dynamic adjustment of the testing cycle: Health equipment: Maintain the original schedule; Warning status equipment: Automatically shortens the test cycle to 1 / 2 or 1 / 3 of the original cycle; For equipment in abnormal or above-level condition: follow up with a retest within one month after this targeted test to verify the effectiveness of the treatment or monitor the development of defects; A44: Intelligent scheduling of test resources: When generating a task, the system automatically lists the intelligent test equipment required for the recommended test items, marks the personnel skills required to perform the task, and clearly marks whether the test requires a power outage.
[0011] Preferably, the process of automatically controlling the testing instruments to complete the test in A5 through an edge computing gateway or station-side monitoring system is as follows: A51: Task Issuance and Parsing: The station-side monitoring system issues the generated test task work order to the edge computing gateway of the substation; the gateway receives the task and parses out the key instructions: call the "Mechanical Characteristics Test" driver script, control the robot to connect to the 102 circuit breaker, and start the circuit breaker analyzer; A52: Automatic test environment preparation: The edge gateway first communicates with the substation's "five-prevention system" to confirm that the 102 circuit breaker is under maintenance and obtains a "test allowed" safety signal, and then automatically connects the wires. A53: Automatic Instrument Parameter Configuration and Test Execution: The edge gateway establishes a communication connection with the "Circuit Breaker Analyzer" through the IoT gateway. The gateway executes the built-in "Mechanical Characteristic Test Procedure" driver script, automatically setting all parameters for the test instrument. The edge gateway sends a "Start Test" command to the test instrument. The gateway sends a short pulse signal to the circuit breaker's control circuit to simulate an opening / closing command, triggering the circuit breaker to perform an actual operation. A54: Automatic Data Acquisition and Upload: Automatically generates test reports and raw waveform data, and automatically uploads them to the station monitoring system via the edge gateway; A55: Automatic on-site recovery: The edge gateway instructs the wiring robot to disconnect the test cable and straighten it back into place; the gateway controls the test instrument to shut down or enter standby mode.
[0012] Preferably, A6 includes the following: Update the multi-source test dataset: Treat all test results, waveform files, and expert diagnostic conclusions as a new data packet, and accurately associate it with the online monitoring data and historical data that were previously used to trigger this test through the device ID and timestamp; store this data packet as a new record in the multi-source test dataset; Update the intelligent assessment model for device status: trigger an update every fixed period, or when the number of newly accumulated samples reaches a certain threshold; use the updated, larger multi-source test dataset to comprehensively retrain the model.
[0013] The beneficial effects of this invention are as follows: This invention enables the automatic generation of test tasks and the automatic control of test instruments, significantly reducing manual intervention, improving test efficiency, and lowering labor costs and human error. The model can capture subtle changes in equipment status and early signs of failure, enabling early warning of faults and providing forward-looking basis for operation and maintenance decisions. Through a closed-loop optimization mechanism, the system can continuously improve itself using new test data, enabling the evaluation model to continuously evolve and adapt to equipment aging and new defect patterns. Attached Figure Description
[0014] 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.
[0015] Figure 1 A flowchart of an intelligent testing method for substation equipment provided in an embodiment of the present invention.
[0016] Figure 2 This is a flowchart of the automatic generation of test tasks and strategies in A4 of the present invention.
[0017] Figure 3 This is a flowchart illustrating how the testing instruments are automatically controlled by an edge computing gateway or a station-side monitoring system to complete the testing process in A5 of this invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Example: Figures 1 to 3As shown, this invention provides an intelligent testing method for substation equipment, including: A1: Construct a multi-source test dataset for substation equipment; The multi-source test dataset includes historical test data of the equipment, real-time / near real-time online monitoring data of the equipment, equipment ledger parameter data, and equipment defect / fault record data; A2: Based on the multi-source test dataset, feature engineering methods are used to extract multi-dimensional feature vectors related to the health status of the equipment; A3: Construct and train an intelligent equipment status assessment model, wherein the intelligent equipment status assessment model takes the multidimensional feature vector as input and outputs the equipment status score and / or failure probability. A4: Based on the output of the intelligent equipment status assessment model, test tasks and strategies are automatically generated; A5: Execute the test task and automatically control the test instruments to complete the test through the edge computing gateway or station monitoring system; A6: Collect and analyze the test results, update the multi-source test dataset and the intelligent evaluation model of the device status, and form a closed-loop optimization.
[0020] This embodiment realizes the automatic generation of test tasks and the automatic control of test instruments, which greatly reduces manual intervention, improves test efficiency, and reduces labor costs and human error. The model can capture subtle changes in equipment status and early fault symptoms, realize early warning of faults, and provide forward-looking basis for operation and maintenance decisions. Through a closed-loop optimization mechanism, the system can continuously improve itself using new test data, enabling the evaluation model to continuously evolve, adapt to equipment aging and new defect patterns, and have long-term vitality.
[0021] Furthermore, the historical test data of the equipment includes preventive test data, handover and acceptance test data, and special test data; the preventive test data consists of standardized test results performed during each periodic maintenance; the handover and acceptance test data consists of all initial test data before the equipment is put into operation; the special test data consists of transformer short-circuit impedance test data and partial discharge location test data. The real-time / quasi-real-time online monitoring data of the equipment includes electrical quantity monitoring data, status quantity monitoring data, physical quantity monitoring data, and video image data; the equipment ledger parameter data includes identification, technical parameters, and commissioning information; the equipment defect / fault record data includes defect records, fault waveform data, protection action information, and fault analysis conclusions.
[0022] Furthermore, the multidimensional feature vectors described in A2 include: Static features: equipment model, commissioning date, and manufacturer, extracted from the equipment ledger parameter data; Dynamic characteristics: Real-time monitored current, voltage, temperature, vibration, partial discharge, oil chromatography data and their changing trends are extracted from the real-time / quasi-real-time data; Historical characteristics: longitudinal comparison data of the results of previous preventive tests and historical defect records, extracted from the historical test data of the equipment and the real-time / quasi-real-time online monitoring data of the equipment; Derived features: Feature quantities calculated based on the above features, such as data change rate, horizontal comparison difference with similar devices, and feature frequency components.
[0023] Suppose we want to extract the feature vector of a transformer at the time of 2023-11-01: Static characteristics: [Model: S11-M-10000 / 110, Years in operation: 8, Manufacturer: Manufacturer A] -> Code is [0.85, 8, 1]; Dynamic characteristics: [Load rate: 0.75, Oil temperature: 68, H2: 120] -> Normalized to [0.75, 1.2, 1.5]; Historical trend characteristics: [Annual change rate of insulation resistance: -0.08, gas production rate in the past 7 days: 0.15] -> [-0.08, 0.15]; Derived characteristics: [Difference between H2 content and isotype mean: +20, family defect marker: 1] -> [20, 1]; Finally, the multidimensional feature vector is [0.85, 8, 1, 0.75, 1.2, 1.5, -0.08, 0.15, 20, 1].
[0024] Furthermore, the construction and training of the intelligent assessment model for device status in A3 includes model construction and model training; The model construction includes: using a Long Short-Term Memory (LSTM) network to process pure time-series data and extract time-series features; concatenating the time-series features with static features, dynamic features, historical features, and derived features, and then inputting them into a Gradient Boosting Decision Tree (GBDT) model for state evaluation; the GBDT model trains multiple decision trees sequentially, with each tree learning to correct the errors of the previous tree, ultimately forming a powerful ensemble model; the LSTM network is a special type of recurrent neural network designed specifically for processing time-series data and has a "memory" function; The model training includes the following: Training Data Preparation and Label Definition: Select data samples at historical time points from the multi-source test dataset, and define a health score for the device at each historical time point; for example: when the device is operating normally, the label is 100 points; one month before a fault occurs, the label is 30 points; when a fault occurs, the label is 0 points; assign scores to intermediate states through smooth interpolation. Define a time window; if a device fails within 30 days after a certain historical data point, the label of this data point is 1, otherwise it is 0; Model Training and Tuning: Data Partitioning: Partition the dataset into a training set, a validation set, and a test set in chronological order; Training Loop: Input the feature vectors of the training set and their corresponding labels into the model; calculate the difference between the predicted value and the true label; adjust the internal parameters of the model through the backpropagation algorithm optimizer to minimize the difference; Hyperparameter Tuning: Use the validation set to optimize the hyperparameters of the model and find the best configuration.
[0025] When a new data point arrives, for example, the feature vector generated after feature engineering on the real-time data of the main transformer today: Feature Vector Input: Input the new feature vector V(new) into the trained model; Model Inference: For GBDT: V(new) will sequentially pass through each decision tree, and each tree gives a preliminary judgment, and finally the outputs of all trees are weighted and aggregated; For LSTM: V(new) will be calculated together with the previous historical sequence data in the memory unit of LSTM to obtain a comprehensive representation of the current state, and then the result is obtained through the output layer; Result Output: The final output layer of the model passes through a Sigmoid or Softmax activation function to map the original calculated value to the interval [0, 1] or the interval [0, 100]; Final Output: State Score = 72, Discharge Fault Probability = 0.18.
[0026] Furthermore, A4 includes a series of preset decision thresholds and rules, where the index sequence is as follows: model output state, state score (HS), fault probability (FP), generated test task level, core strategy; Below are the decisions and rules corresponding to the index sequence: Healthy State, HS > 85, FP < 0.05, Routine Test, Execute according to the original cycle, Strategy Optimization; Warning State, 70 < HS ≤ 85, 0.05 ≤ FP < 0.15, Warning Test, Shorten the cycle, Targeted Test; Abnormal state, 60 < HS ≤ 70, 0.15 ≤ FP < 0.30, targeted special test, execute immediately, focus on risk points; Critical state, HS ≤ 60, FP ≥ 0.30, emergency diagnostic test, execute immediately, comprehensive diagnosis, linkage dispatching if necessary.
[0027] Furthermore, the process of automatically generating test tasks and strategies in A4 is as follows: A41: Task priority and triggering mechanism: When the model output enters the "critical state" or multiple failure probabilities soar simultaneously, it is triggered automatically and immediately, and an alarm text message / APP push is sent to the operation and maintenance personnel at the same time; generate a work order with the highest priority; A42: Recommended and optimized test items: Recommendation based on fault type: If the model output has a high "discharge fault probability", the following tests will be recommended first: high-frequency / ultra-high frequency partial discharge test, ultrasonic detection, and review of dissolved gas analysis in insulating oil; if the "overheating fault probability" is high, the following tests will be recommended first: infrared precise temperature measurement, loop resistance test, and oil chromatographic analysis; Recommendation based on feature importance: If in the evaluation, the importance of the two features "acetylene gas production rate" and "vibration spectrum entropy" ranks the highest, the test items most relevant to these two features will be generated preferentially; A43: Dynamic adjustment of test cycle: Healthy equipment: Keep the original cycle; Equipment in warning state: Automatically shorten the test cycle to 1 / 2 or 1 / 3 of the original cycle; Equipment in abnormal and above states: Conduct a follow-up re-test within 1 month after this targeted test to verify the treatment effect or monitor the development of defects; A44: Intelligent scheduling of test resources: When the system generates a task, it automatically lists the intelligent test equipment required for the recommended test items, marks the personnel skills required to execute the task, and clearly marks whether the test requires power outage.
[0028] Furthermore, the process of automatically controlling the test instrument to complete the test through the edge computing gateway or the substation terminal monitoring system in A5 is as follows: A51: Task distribution and parsing: The substation terminal monitoring system distributes the generated test task work order to the edge computing gateway of this substation; the gateway receives the task and parses out the key instructions: call the "mechanical characteristic test" driver script, control the robot to connect to the 102 circuit breaker, and start the circuit breaker analyzer; A52: Automatic preparation of test environment: The edge gateway first communicates with the "five-prevention system" of the substation to confirm that the 102 circuit breaker is in the maintenance state, obtains the safety signal of "allow testing", and then automatically wires; A53: Automatic Instrument Parameter Configuration and Test Execution: The edge gateway establishes a communication connection with the "Circuit Breaker Analyzer" through the IoT gateway. The gateway executes the built-in "Mechanical Characteristic Test Procedure" driver script, automatically setting all parameters for the test instrument. The edge gateway sends a "Start Test" command to the test instrument. The gateway sends a short pulse signal to the circuit breaker's control circuit to simulate an opening / closing command, triggering the circuit breaker to perform an actual operation. A54: Automatic Data Acquisition and Upload: Automatically generates test reports and raw waveform data, and automatically uploads them to the station monitoring system via the edge gateway; A55: Automatic on-site recovery: The edge gateway instructs the wiring robot to disconnect the test cable and straighten it back into place; the gateway controls the test instrument to shut down or enter standby mode.
[0029] Furthermore, A6 includes the following: Update the multi-source test dataset: Treat all test results, waveform files, and expert diagnostic conclusions as a new data packet, and accurately associate it with the online monitoring data and historical data that were previously used to trigger this test through the device ID and timestamp; store this data packet as a new record in the multi-source test dataset; Update the intelligent assessment model for device status: trigger an update every fixed period, or when the number of newly accumulated samples reaches a certain threshold; use the updated, larger multi-source test dataset to comprehensively retrain the model.
[0030] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for intelligent testing of substation equipment, characterized in that, include: A1: Construct a multi-source test dataset for substation equipment; The multi-source test dataset includes historical test data of the equipment, real-time / near real-time online monitoring data of the equipment, equipment ledger parameter data, and equipment defect / fault record data; A2: Based on the multi-source test dataset, feature engineering methods are used to extract multi-dimensional feature vectors related to the health status of the equipment; A3: Construct and train an intelligent equipment status assessment model, wherein the intelligent equipment status assessment model takes the multidimensional feature vector as input and outputs the equipment status score and / or failure probability. A4: Based on the output of the intelligent equipment status assessment model, test tasks and strategies are automatically generated; A5: Execute the test task and automatically control the test instruments to complete the test through the edge computing gateway or station monitoring system; A6: Collect and analyze the test results, update the multi-source test dataset and the intelligent evaluation model of the device status, and form a closed-loop optimization.
2. The intelligent testing method for substation equipment as described in claim 1, characterized in that, The historical test data of the equipment includes preventive test data, handover and acceptance test data, and special test data; the preventive test data are the results of standardized tests performed during each periodic maintenance; the handover and acceptance test data are all initial test data before the equipment is put into operation; the special test data are transformer short-circuit impedance test data and partial discharge location test data. The real-time / near real-time online monitoring data of the equipment includes electrical quantity monitoring data, status quantity monitoring data, physical quantity monitoring data, and video image data; The equipment ledger parameter data includes identification, technical parameters, and commissioning information; The equipment defect / fault record data includes defect records, fault waveform data, protection action information, and fault analysis conclusions.
3. The intelligent testing method for substation equipment as described in claim 1, characterized in that, The multidimensional feature vectors mentioned in A2 include: Static features: equipment model, commissioning date, and manufacturer, extracted from the equipment ledger parameter data; Dynamic characteristics: Real-time monitored current, voltage, temperature, vibration, partial discharge, oil chromatography data and their changing trends are extracted from the real-time / quasi-real-time data; Historical characteristics: longitudinal comparison data of the results of previous preventive tests and historical defect records, extracted from the historical test data of the equipment and the real-time / quasi-real-time online monitoring data of the equipment; Derived features: Feature quantities calculated based on the above features, such as data change rate, horizontal comparison difference with similar devices, and feature frequency components.
4. The intelligent testing method for substation equipment as described in claim 1, characterized in that, Building and training an intelligent equipment status assessment model in A3 includes model building and model training; The model construction includes: using a Long Short-Term Memory (LSTM) network to process pure time series data and extract time series features; concatenating the time series features with static features, dynamic features, historical features, and derived features, and then inputting them into a Gradient Boosting Decision Tree (GBDT) model for state evaluation; The model training includes the following: Training data preparation and label definition: Select data samples from historical time points from the multi-source test dataset, define a health score for the device at each historical time point, and define a time window; Model training and tuning: Data partitioning: The dataset is divided into training set, validation set and test set in chronological order; Training loop: Input the feature vectors and corresponding labels from the training set into the model; calculate the difference between the predicted values and the true labels; adjust the model's internal parameters using the backpropagation algorithm optimizer to minimize the difference; Hyperparameter tuning: Optimize the model's hyperparameters using the validation set to find the optimal configuration.
5. The intelligent testing method for substation equipment as described in claim 1, characterized in that, A4 includes a series of preset decision thresholds and rules, among which the indicator sequence is as follows: model output state, state score (HS), failure probability (FP), generated test task level, and core strategy; Below are the decisions and rules corresponding to the indicator sequences: Health status, HS > 85, FP < 0.05, routine testing, execution according to the original cycle, strategy optimization; Warning status, 70 < HS≤85, 0.05≤FP < 0.15, warning test, shortened cycle, targeted test; Abnormal conditions, 60 < HS ≤ 70, 0.15 ≤ FP < 0.30, targeted testing, immediate execution, focusing on risk points; In critical situations, HS≤60, FP≥0.30, emergency diagnostic testing, immediate execution, comprehensive diagnosis, and coordinated dispatch when necessary.
6. The intelligent testing method for substation equipment as described in claim 5, characterized in that, The process of automatically generating test tasks and strategies in A4 is as follows: A41: Task Priority and Triggering Mechanism: Automatically and immediately trigger when the model output enters a "critical state" or when the probability of multiple faults spikes simultaneously, and simultaneously send alarm SMS / APP push to the operation and maintenance personnel; generate the highest priority work order; A42: Test Item Recommendations and Optimization: Based on the fault type, if the model outputs a high probability of "discharge fault", it will prioritize recommending: high frequency / ultra-high frequency partial discharge testing, ultrasonic testing, and dissolved gas analysis verification in insulating oil; if the model outputs a high probability of "overheating fault", it will prioritize recommending: infrared precise temperature measurement, loop resistance testing, and oil chromatography analysis. Based on feature importance recommendations, if the two features "acetylene production rate" and "vibration spectrum entropy" rank the most important in the evaluation, then the test items most relevant to these two features will be generated first. A43: Dynamic adjustment of the testing cycle: Health equipment: Maintain the original schedule; Warning status equipment: Automatically shortens the test cycle to 1 / 2 or 1 / 3 of the original cycle; For equipment in abnormal or above-level condition: follow up with a retest within one month after this targeted test to verify the effectiveness of the treatment or monitor the development of defects; A44: Intelligent scheduling of test resources: When generating a task, the system automatically lists the intelligent test equipment required for the recommended test items, marks the personnel skills required to perform the task, and clearly marks whether the test requires a power outage.
7. The intelligent testing method for substation equipment as described in claim 1, characterized in that, The process of automatically controlling the testing instruments to complete the test in A5 via an edge computing gateway or station-side monitoring system is as follows: A51: Task Issuance and Parsing: The station-side monitoring system issues the generated test task work order to the edge computing gateway of the substation; the gateway receives the task and parses out the key instructions: call the "Mechanical Characteristics Test" driver script, control the robot to connect to the 102 circuit breaker, and start the circuit breaker analyzer; A52: Automatic test environment preparation: The edge gateway first communicates with the substation's "five-prevention system" to confirm that the 102 circuit breaker is under maintenance and obtains a "test allowed" safety signal, and then automatically connects the wires. A53: Automatic Instrument Parameter Configuration and Test Execution: The edge gateway establishes a communication connection with the "Circuit Breaker Analyzer" through the IoT gateway. The gateway executes the built-in "Mechanical Characteristic Test Procedure" driver script, automatically setting all parameters for the test instrument. The edge gateway sends a "Start Test" command to the test instrument. The gateway sends a short pulse signal to the control circuit of the circuit breaker to simulate an opening / closing command, triggering the circuit breaker to perform an actual operation. A54: Automatic Data Acquisition and Upload: Automatically generates test reports and raw waveform data, and automatically uploads them to the station monitoring system via the edge gateway; A55: Automatic on-site recovery: The edge gateway instructs the wiring robot to disconnect the test cable and straighten it back into place; The gateway controls the test instrument to shut down or enter standby mode.
8. The intelligent testing method for substation equipment as described in claim 1, characterized in that, A6 includes the following: Update the multi-source test dataset: Treat all test results, waveform files, and expert diagnostic conclusions as a new data package, and accurately associate it with the online monitoring data and historical data that were previously used to trigger this test using the device ID and timestamp; Store this data packet as a new record in the multi-source test dataset; Update the intelligent assessment model for device status: trigger an update every fixed period, or when the number of newly accumulated samples reaches a certain threshold; use the updated, larger multi-source test dataset to comprehensively retrain the model.