Monitoring and maintenance system and method for transformer substation

Abstract

The invention relates to the technical field of power system automation, in particular to a monitoring and maintenance system and method for a transformer substation. The system comprises a quantitative classification module, a difference analysis module and a grading adjustment module. Equipment in a transformer substation is classified through the quantitative classification module, correction parameters such as aging and load loss are introduced for single-equipment monitoring through the difference analysis module, misjudgment and missed judgment caused by equipment state changes are reduced by combining multi-level threshold values, associated equipment is modeled through a GNN model, neighbor states are aggregated, the equipment health degree is output, and the equipment quality is improved. The method comprises the following steps of: quantifying a fault propagation probability, pre-warning cascading fault risks in advance, and adjusting a threshold value of a single device in a linkage manner while a hierarchical adjustment module takes risk response measures, so that single device monitoring is adaptively matched with an associated device risk state, full-coverage accurate monitoring is realized, and local and global risks are considered.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, and more specifically, to a monitoring and maintenance system and method for substations. Background Technology

[0002] As the core hub of the power system's generation, transmission, transformation, and distribution links, the operating status of substations directly determines the reliability of the power grid and the economics of operation and maintenance. Traditional substation monitoring and maintenance systems generally rely on equipment status acquisition, abnormal alarms, and manual repair as their core logic. They collect electrical parameters (voltage, current, partial discharge) and physical conditions (temperature, vibration) of equipment such as transformers, switchgear, and busbars through sensors, transmit the data to the station control layer or cloud platform, and then maintenance personnel will conduct on-site troubleshooting, formulate and execute repair plans, and manually record the maintenance results after completion.

[0003] However, with the expansion of substation scale, the increase in equipment types, and the improvement of users' requirements for power supply reliability, the equipment in the substation is interconnected during monitoring and maintenance. Different equipment will have different impacts during maintenance, some with greater impact and some with less. Traditional substation monitoring and maintenance systems only monitor individual equipment and ignore the interrelationships between them. This can lead to misjudgments during monitoring and maintenance, resulting in incorrect maintenance measures and failure to maintain equipment in a timely manner, causing irreparable damage to the substation. Therefore, we propose a substation monitoring and maintenance system and method. Summary of the Invention

[0004] The purpose of this invention is to provide a monitoring and maintenance system and method for substations to solve any of the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a monitoring and maintenance system and method for substations, comprising: The quantitative classification module collects electrical connection data and parameter correlation data, calculates the electrical connectivity and data correlation of the equipment, quantifies the scope of the equipment's fault impact, obtains a comprehensive score, compares the comprehensive score with a preset threshold, and classifies the equipment type into single equipment type and associated equipment type. The difference analysis module uses a threshold comparison method to monitor the status of a single device. It collects the functional parameters of a single device in real time and compares them with the rated parameters of the single device to determine the abnormal situation of the single device. It uses a GNN global propagation model to monitor the status of related devices. It models the related devices and their connection relationships as a graph structure. With the node feature matrix and adjacency matrix as input, it uses a graph convolutional layer to aggregate information and output the node health and fault propagation probability. The tiered adjustment module classifies associated equipment anomalies into high-risk, medium-risk, and low-risk categories based on node health and fault propagation probability. For each anomaly level, adjustment measures are implemented with the goal of suppressing faults while minimizing their impact on normal grid operation.

[0006] As a further improvement to this technical solution, the quantitative classification module includes: The data acquisition and calculation unit extracts the static electrical connection relationships and dynamic switch status between equipment from the substation design drawings. It calculates the electrical connection degree by statistically analyzing the number of equipment that can be directly and indirectly affected by the equipment through electrical connections and the connection strength. It collects equipment operating parameters from sensors and smart meters, filters out strongly correlated parameter pairs, and calculates the data correlation degree by statistically analyzing the number of strongly correlated parameter pairs. The quantitative comparison unit quantifies the scope of equipment failure impact through weighted summation, calculates the equipment's comprehensive score, and compares the comprehensive score of the equipment with a preset threshold. If the comprehensive score is lower than the preset threshold, it is a single equipment type; if the comprehensive score exceeds the preset threshold, it is an associated equipment type.

[0007] As a further improvement to this technical solution, the quantitative comparison unit collects historical operating data of substation equipment, calculates the fault impact range of each piece of equipment, and uses data statistical analysis to draw a distribution histogram of the equipment fault impact, with the inflection point of the distribution curve as the initial preset threshold.

[0008] As a further improvement to this technical solution, the quantitative comparison unit collects the actual operating load of the equipment in the substation in real time, calculates the load rate of the equipment, classifies the operating conditions of the equipment into levels, determines the adjustment range of the initial threshold according to the magnitude of the impact of the operating condition level on the equipment failure, and establishes a rule mapping table between operating conditions and thresholds.

[0009] As a further improvement to this technical solution, the difference analysis module includes: The single-device monitoring unit determines the correction parameters based on the cumulative running time of the single device, corrects the rated parameters, compares the real-time functional parameters of the single device with the corrected rated parameters, outputs the abnormal status of the single device, and issues an early warning signal. The associated equipment monitoring unit treats each associated device as a node, the electrical connection relationships between devices as edges and assigns weights, and then transforms the nodes and edges into an adjacency matrix. The adjacency matrix and node feature matrix are input into the GNN global propagation model. Through graph convolutional layers, each node integrates its own and neighboring device state information, and outputs the node health and fault propagation probability of the associated devices.

[0010] As a further improvement to this technical solution, the single device monitoring unit acquires the cumulative running time of a single device, calculates the load loss coefficient and aging correction coefficient based on the cumulative full load time ratio, and determines the final correction parameters.

[0011] As a further improvement to this technical solution, the associated equipment monitoring unit classifies connections into hard connections, semi-hard connections, and soft connections according to the functional attributes of electrical connections, determines the basic weight of the edge between devices according to the connection type, and corrects the basic weight through the key parameters of different connections.

[0012] As a further improvement to this technical solution, the graded adjustment module aims to quickly isolate faults and minimize the scope of power outages for high-risk anomalies, to prevent fault escalation and fine-tune to maintain stability for medium-risk anomalies, and to minimize intervention and ensure normal operation for low-risk anomalies.

[0013] As a further improvement to this technical solution, after taking adjustment measures, the hierarchical adjustment module determines the adjustment direction of the threshold in a single device according to the abnormality level of the associated devices, determines the threshold adjustment range based on the node health and fault propagation probability, and adjusts the threshold of the single device.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The monitoring and maintenance system and method of the substation, the quantitative classification module classifies the equipment in the substation based on electrical connection and operating parameter quantification, avoids human subjective bias, sets the initial threshold by combining historical fault data, and then dynamically adjusts the threshold according to the equipment load rate to adapt to different operating scenarios of the substation and ensure that the classification is consistent with the actual impact of equipment faults; 2. The difference analysis module introduces correction parameters such as aging and load loss for single device monitoring, combined with multi-level thresholds, to reduce false and missed judgments caused by changes in device status. Associated devices are modeled using a GNN model, aggregating neighbor status, outputting both device health and quantifying the probability of fault propagation, providing early warning of cascading failure risks, achieving full-coverage and accurate monitoring, and taking into account both local and global risks. 3. The graded adjustment module quickly isolates high-risk faults to ensure core power supply, fine-tunes medium-risk faults for preventive upgrades, and reduces intervention for low-risk faults to avoid excessive operation and disturbance to the power grid. While taking risk response measures, it also adjusts the threshold of individual devices in a coordinated manner to make the monitoring of individual devices adaptively match the risk status of related devices.

[0015] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0016] Figure 1This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram illustrating the overall detailed process of the present invention; Figure 3 This is a schematic diagram of the method flow of the present invention.

[0017] The meanings of the labels in the diagram are as follows: 100. Quantitative Classification Module; 110. Data Acquisition and Calculation Unit; 120. Quantitative Comparison Unit; 200. Difference Analysis Module; 210. Single Device Monitoring Unit; 220. Associated Device Monitoring Unit; 300. Graded Adjustment Module. Detailed Implementation

[0018] The technical solutions in 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] Currently, with the expansion of substation scale, the increase in equipment types, and the improvement of users' requirements for power supply reliability, the equipment in substations is interconnected during monitoring and maintenance. Different equipment will have different impacts during maintenance, some with greater impact and some with less. Traditional substation monitoring and maintenance systems only monitor individual equipment and ignore the interrelationships between them. This can lead to misjudgments during equipment monitoring and maintenance, resulting in incorrect maintenance measures and failure to maintain equipment in a timely manner, causing irreparable damage to the substation.

[0020] Therefore, this invention proposes that the quantitative classification module divides equipment into single-equipment types and associated equipment types by calculating electrical connectivity, data correlation, and comprehensive score, directly providing the difference analysis module with the basis for classifying monitoring objects. The difference analysis module outputs core monitoring results for the two types of equipment, providing direct input for risk assessment and measure formulation to the graded adjustment module. The graded adjustment module adjusts the threshold of the single-equipment local threshold diagnosis model in the difference analysis module in reverse according to the risk level of the associated equipment.

[0021] Specifically as follows: Please see Figure 1 As shown, the present invention provides a monitoring and maintenance system and method for substations, including a quantitative classification module 100, which collects electrical connection data and parameter correlation data, calculates the electrical connection degree and data correlation degree of the equipment, quantifies the fault impact range of the equipment, obtains a comprehensive score, compares the comprehensive score with a preset threshold, and classifies the equipment type into single equipment type and associated equipment type. Electrical connection data refers to data reflecting the physical and functional connection relationships and real-time on / off status between substation equipment. Its core function is to define which equipment is interconnected, providing a basis for subsequent determination of whether a fault will propagate between equipment. Parameter correlation data refers to data reflecting the operating status of substation equipment itself and the strong correlation between different equipment and parameters. Its core function is to determine whether the equipment is operating normally and which parameter changes will affect each other, providing a basis for equipment anomaly diagnosis and risk warning.

[0022] like Figure 2 As shown, the quantitative classification module 100 includes: The data acquisition and calculation unit 110 extracts the static electrical connection relationships and dynamic switch states between equipment from the substation design drawings. It calculates the electrical connection strength by statistically analyzing the number of devices directly and indirectly affected by electrical connections and the connection strength, using the following formula: Electrical connectivity = (Number of directly connected devices × a + Number of indirectly connected devices × b) ÷ Maximum possible number of connections; Where 'a' is the weight of the number of directly effective connected devices. Direct effective connection is the core path of fault propagation. Fault current, voltage abnormalities, etc. can be transmitted instantly and without attenuation through direct connection (e.g., a bus short circuit will directly cause the connected main transformer to overload). Its impact on the target equipment is direct and inevitable. It is the main carrier of fault chain propagation and should be given the highest weight to reflect its decisive role. It is generally set to 1. b represents the weight of the number of effective indirectly connected devices. Indirect connections require relaying through directly connected devices. Fault propagation is delayed, attenuated, and can be blocked (e.g., a fault in an indirect device can be isolated by an intermediate switch without affecting the target device). The impact on the target device is indirect and accidental. The weight is determined proportionally based on the number of device intervals between the device and the target device (the more intermediate interval devices there are, the lower the weight of the number of effective indirectly connected devices). To determine the number of directly connected devices, we need to identify those that are physically connected to the target device (e.g., by conductors, busbars, or copper busbars) and whose switching equipment (circuit breakers, disconnectors) along the connection path is in a closed, active state. We obtain a list of all physically connected devices to the target device from the substation's CAD primary wiring diagram. We also obtain the real-time open / closed status of the switching equipment along the connection path from the SCADA system, excluding open connections. The number of devices in the connection list that are directly connected to the target device and whose switches are closed represents the number of directly connected devices. Indirectly connected devices refer to devices that are not physically connected to the target device but are indirectly connected through Level 1 directly connected devices, and all intermediate switches are in the closed and conducting state. The number of devices directly connected to the directly connected devices with closed switches in the connection list is the number of indirectly connected devices. The maximum possible number of connections refers to the maximum number of devices that can theoretically be directly connected to the target equipment during the substation design phase, based on the physical wiring topology and functional positioning. This information can be obtained directly from the substation design drawings and equipment technical manuals. The electrical connectivity is calculated using the above formula, which not only considers whether there is a physical connection, but also combines the real-time status of the switch (closed / open) to screen for valid connections, avoiding the inclusion of invalid connections that have been disconnected by the open switch, thus conforming to the actual operating scenario. The system collects operating parameters from sensors and smart meters, identifies strongly correlated parameter pairs, and calculates the data correlation degree by counting the number of strongly correlated parameter pairs. The formula is as follows: Data correlation degree = (number of core strongly correlated pairs × c + number of secondary strongly correlated pairs × d) ÷ total number of key parameters of the target equipment; Where c is the weight of the number of core strongly correlated pairs. The core parameters are the direct causes of equipment failure. For example, current exceeding the rated value directly leads to overload failure, and power abnormality directly reflects overload. Their correlation plays a decisive role in judging whether the equipment is abnormal. They need to be given the highest weight to ensure the accuracy of fault diagnosis. They are generally set to 1. d represents the weight of the secondary strongly correlated pair. Secondary parameters are indirect signals of equipment failure. For example, a temperature rise is a result of overload (not the cause), and abnormal vibration needs to be combined with current parameters to determine whether it is a fault (high vibration alone may be an installation problem). Its correlation only plays an auxiliary role and is determined according to the functional priority of the parameter (the greater the impact on the core parameter, the higher the weight d of the secondary strongly correlated pair). Among them, the core strongly correlated pairs refer to the vital parameter pairs for equipment operation. The parameters have extremely strong linear and nonlinear correlations and directly reflect the safe operating boundary of the equipment. Abnormalities in these pairs will directly lead to equipment failure or trigger protection actions. Real-time operating parameters of the target equipment (such as current, voltage, power, temperature, vibration, partial discharge, etc.) are obtained from smart instruments. The time granularity is uniformly set to 15 minutes. The correlation of parameter pairs is calculated using statistical methods, and the number of parameter pairs that are directly associated with the equipment is recorded. This number is the core strongly correlated pair count. Secondary strongly correlated pairs refer to auxiliary parameter pairs of equipment operation. The parameters are strongly correlated and only indirectly reflect the equipment status. Their abnormality is a signal of potential equipment problems. The number of parameter pairs that only indirectly reflect the equipment status is the number of secondary strongly correlated pairs. The total number of key parameters of the target equipment refers to the total number of all core and secondary operating parameters that need to be monitored from the perspective of ensuring the safe and stable operation of a specific target equipment (such as a main transformer, circuit breaker, or busbar). These parameters are obtained from the monitoring parameters specified by the manufacturer in the equipment's technical manual. The above formula is used to calculate the data correlation of the equipment, distinguish between core and secondary strong correlation pairs, avoid treating auxiliary parameters as equal, and ensure that the data correlation aligns with the priority requirements of fault diagnosis.

[0023] Among them, the quantitative comparison unit 120 quantifies the fault impact range of the equipment by weighted summation, calculates the comprehensive score of the equipment, compares the comprehensive score of the equipment with a preset threshold, and if the comprehensive score is lower than the preset threshold, it is a single equipment type; if the comprehensive score exceeds the preset threshold, it is an associated equipment type. Based on the calculated electrical connectivity and data correlation, in power systems, electrical connectivity has a stronger determinant effect on the scope of fault impact, while parameter correlation only plays a supplementary role. This is the core basis for weight allocation. We statistically analyze substation fault cases from the past 3-5 years, counting separately the number of cases where faults in related equipment were caused solely by electrical connectivity, and the number of cases where faults in related equipment were indirectly caused solely by parameter correlation. The results are then calculated using the formula: in, For electrical connectivity weights, For data relevance weights, This refers to the number of cases where related equipment malfunctions solely due to electrical connections. This represents the number of cases where related equipment failures are indirectly caused solely by parameter correlation.

[0024] The formula for calculating the weighted sum of the impact range of equipment failures is as follows: in, For the overall score, For electrical connectivity, For data correlation, For electrical connectivity weights, This represents the weight of data relevance.

[0025] Among them, the quantitative comparison unit 120 collects historical operating data of substation equipment, calculates the fault impact range of each piece of equipment, and uses data statistical analysis to draw a distribution histogram of equipment fault impact, using the inflection point of the distribution curve as the initial preset threshold. Based on the formula for the comprehensive score, calculate the impact range of each device's failure. Divide the data into intervals according to the comprehensive score range (0 to 1), with a group interval of 0.1 (balancing resolution and simplicity), for a total of 10 groups. Use the score interval as the X-axis and the device frequency as the Y-axis. Each interval corresponds to a rectangular bar, with the bar height being the frequency of that interval and the bar width being the group interval (0.1). Add a chart title (e.g., Histogram of Comprehensive Score Distribution of Substation Equipment Failure Impact) and axis labels (X-axis: Comprehensive score of failure impact, Y-axis: Number of devices). Based on the histogram, add a frequency distribution curve, observe the shape of the distribution curve, and find the point where the curve changes from a steep rise (representing a rapid increase in the number of devices) to a gentle rise and then begins to fall. The X-axis (comprehensive score) corresponding to this point is the inflection point. Use the comprehensive score corresponding to the located inflection point as the initial preset threshold. For example, in a 110kV substation with 30 pieces of equipment, traditionally, based on experience, the main transformer and busbar are classified as related equipment, while the rest are classified as individual equipment. In practice, a circuit breaker failure affecting an entire area (which should be considered related) is not monitored in a key area; a voltage transformer failure that only affects metering (which should be considered an individual equipment) is monitored globally; and the threshold of 0.5 cannot be explained in terms of rationality, and misjudgment still occurs after adjustment. The weights were determined through data statistics (electrical connectivity 0.6, data correlation 0.4). Based on the comprehensive score formula F = electrical connectivity × 0.6 + data correlation × 0.4, the comprehensive scores were calculated as follows: main transformer 0.76, fault circuit breaker 0.56, surge arrester 0.16, voltage transformer 0.26. The scores were divided into 10 groups (0-0.1 to 0.9-1.0, group interval 0.1), the number of devices in each section was counted, histograms were plotted, and distribution curves were added. Observing the curve shape, it was found that the curve rises gently in the 0-0.5 range (the number of devices increases slowly, mostly weakly affected devices such as surge arresters and voltage transformers); the curve rises sharply in the 0.5-0.6 range (the number of devices increases rapidly to 5, including medium-affected devices such as faulty circuit breakers); above 0.6, the curve rises gently again and then falls (the growth rate of the number of devices slows down, mostly strong-affected devices such as main transformers and busbars). The critical point where the curve changes from a gentle rise to a steep rise corresponds to an X-axis score of 0.5. This point is the dividing point, so 0.5 is set as the initial preset threshold. By combining the overall score with the inflection point threshold, faulty circuit breakers (0.56 > 0.5) are accurately classified as associated devices to avoid underestimating the impact range; voltage transformers (0.26 < 0.5) are classified as single devices to reduce invalid monitoring.

[0026] In order to adjust the initial threshold according to the actual operating conditions of the equipment, the quantification comparison unit 120 collects the actual operating load of the equipment in the substation in real time, calculates the load rate of the equipment, classifies the operating conditions of the equipment into levels, determines the adjustment range of the initial threshold according to the magnitude of the impact of the operating condition level on the equipment failure, and establishes a rule mapping table between operating conditions and thresholds. Equipment load rate is essentially the percentage of actual load to rated load, and the formula is: in, For equipment load rate, This refers to the actual operating load of the equipment within a certain time period. The rated load capacity specified by the manufacturer when the equipment leaves the factory.

[0027] Under different operating conditions, the equipment's load-bearing pressure, aging rate, and fault transmission efficiency vary, directly leading to differences in the magnitude of fault impact. Based on the equipment load rate, the equipment operating conditions are divided into light-load, normal, and heavy-load conditions. The differences in the magnitude of fault impact are quantified: light-load conditions reduce the impact by 20%-30%, normal conditions show no significant change, and heavy-load conditions increase the impact by 15%-40%. The adjustment range should be based on the percentage change in the impact of the fault, using a percentage adjustment method. The specific adjustment ratios are as follows: Based on four core dimensions—equipment type, operating condition level (light load, normal, heavy load), corresponding load rate range, and initial threshold—the adjusted threshold is calculated according to the logic of increasing by 5%-10% for light load, decreasing by 10%-20% for heavy load, and remaining unchanged for normal load. The application rules for determining whether the comprehensive score exceeds the threshold are supplemented to form a structured table. The table is then digitized and entered into the substation automation system. An index is created by equipment ID and operating condition level to facilitate the system to automatically match thresholds after real-time load rate collection. Finally, the table is calibrated and updated quarterly based on changes in equipment parameters and fault data to ensure it is adapted to actual operation and maintenance needs. The comprehensive score obtained from the above calculation is compared with the adjusted preset threshold. If the comprehensive score is lower than the preset threshold, it is a single device type; if the comprehensive score exceeds the preset threshold, it is an associated device type.

[0028] In addition, the difference analysis module 200 uses the threshold comparison method to monitor the status of a single device. By collecting the functional parameters of the single device in real time and comparing them with the rated parameters of the single device, it determines the abnormal situation of the single device. The core of the threshold comparison method is to take the device's own status as the core, and compare the real-time parameters of a single device with the corrected rated parameters in a binary manner. It does not rely on information from neighboring devices and is suitable for single device types whose fault impact is limited to themselves (such as surge arresters and voltage transformers). The rated parameters are corrected according to the correction parameters. By comparing the real-time collected functional parameters of a single device with the rated parameters of the single device, the abnormality of the single device is determined. If the real-time collected functional parameters are lower than the rated parameters, the single device is normal. If the real-time collected functional parameters are higher than the rated parameters, the single device is abnormal and an alarm signal is issued. For example, in a 110kV substation, three 10kV surge arresters are deployed (the fault impact is limited to themselves and meets the definition of a single equipment type). Their operating status needs to be monitored by the threshold comparison method to avoid the problem of missed judgments due to the aging of surge arrester insulation and load loss leading to a decrease in withstand capability, or the problem of misjudgment due to normal fluctuations accidentally triggering fixed thresholds. The operating voltage and leakage current (two key functional parameters) of each surge arrester are collected in real time by sensors. After filtering and noise reduction preprocessing (to eliminate parameter jumps caused by instantaneous interference from the power grid), for example, if the operating voltage of the surge arrester is 10.2kV and the leakage current is 0.35mA at a certain time, the original rated parameters are extracted from the nameplate of the surge arrester, including a rated operating voltage of 10kV (the upper limit of voltage for long-term safe operation) and a rated leakage current of 0.5mA (the maximum allowable leakage current under normal insulation conditions). Extract the basic data of surge arresters from the substation PMS system, the total cumulative operating time (excluding maintenance time) is 50,000 hours, the cumulative full load time is 30,000 hours, and the time for each load range is: 0-5kV (T1) 5,000 hours, 5-9.5kV (T2) 15,000 hours, ≥9.5kV (T3) 30,000 hours. Calculate the load loss coefficient according to the formula. Substitute the corrected parameters into the threshold comparison method to correct the original rated parameters. Compare the pre-processed real-time functional parameters with the corrected rated parameters. Operating voltage comparison: real-time 10.2kV is greater than the corrected rated 9.4kV, so the parameter is judged to be abnormal. Leakage current comparison: real-time 0.35mA is less than the corrected rated 0.47mA, so the parameter is judged to be normal. If one of the real-time parameters is higher than the corrected rated parameter, the surge arrester is determined to be abnormal. The system immediately issues an audible and visual alarm signal and displays a pop-up message on the background operation and maintenance interface indicating that the surge arrester's operating voltage is out of limit and the insulation status needs to be checked.

[0029] Considering that traditional single-device monitoring only compares real-time parameters with rated parameters, without taking into account the decrease in the adaptability of rated parameters due to equipment load loss and aging, misjudgment is prone to occur. Therefore, the difference analysis module 200 includes: The single device monitoring unit 210 determines the correction parameters based on the cumulative running time of the single device, corrects the rated parameters, compares the real-time functional parameters of the single device with the corrected rated parameters, outputs the abnormal status of the single device, and issues an early warning signal. The device's functional parameters (such as current, temperature, and pressure) and rated parameters (such as rated current and maximum allowable temperature, from the device nameplate) are collected in real time by sensors or smart instruments, and then filtered and noise-reducing preprocessing is performed on them. In order to better determine the correction parameters, the single device monitoring unit 210 acquires the cumulative running time of the single device, calculates the load loss coefficient based on the cumulative full load time ratio, and determines the final correction parameters. Extract basic data from the substation PMS or SCADA system: obtain the total cumulative operating time of the equipment (T_total, excluding downtime for maintenance), the total full load time (T_full, operating time with a load rate greater than 90%), and calculate the operating time of the equipment in different load ranges (such as 0-50%, 50%-80%, and greater than 90%) (T1, T2, T3). Calculate the load loss factor (K-loss): First, calculate the proportion of each load interval (e.g., T1 / Ttotal, T2 / Ttotal), then take the average load of each interval (e.g., 25% for 0-50%, 65% for 50%-80%), according to the formula: Calculate the degree of additional loss caused by load fluctuations. The larger the K loss, the more severe the loss. Determine the final correction parameter. If the correction is to load-related thresholds such as current and power, the impact of equipment aging needs to be considered (a fixed aging coefficient, such as 0.95, can be used for simplification). Then the final correction parameter = K loss × aging coefficient. If only load loss-related thresholds are corrected, K loss can be directly used as the final correction parameter to adjust the rated parameters. For example, the main transformer of a substation (operated for 8 years) has a rated current of 1000A. In traditional monitoring, the real-time current is 950A (lower than the rated value), which is considered normal. However, due to years of full-load operation (60% of the time is at full load), winding losses have caused the actual safe current to drop to 920A. Continuing to operate at 950A has already created an overload risk, but no warning was given. A newly commissioned surge arrester has a rated temperature of 80℃. In traditional monitoring, the real-time temperature is 78℃ (close to the rated value), which is mistakenly judged as abnormal. In reality, there is no load loss, and the safe temperature still meets the requirements, resulting in ineffective operation and maintenance. Extracting main transformer data from PMS / SCADA: T total is 65,000 hours, T full load (load rate > 90%) is 39,000 hours, and the duration of each load interval is T1 (0-50%), T2 (50%-80%), and T3 (>90%), which is 13,000 hours. Calculate the interval percentages (T1 / T total is 20%, T2 / T total is 20%, T3 / T total is 60%), take the interval averages (25%, 65%, 95%), and calculate the load loss coefficient (K loss) according to the above formula (quantifying additional losses, K loss is 1.08 here). The main transformer needs to correct the current threshold. Combined with the aging coefficient of 0.95, the final correction parameter = 1.08 × 0.95 ≈ 1.03; After correction, the rated current of the main transformer is approximately 970A (1000A ÷ 1.03). If the real-time current is 950A, which is less than 970A, it is considered normal (if it is not corrected and the original rated value of 1000A is used, there is a risk of missing the 950A value, which is close to the actual safe value). If the new surge arrester has no obvious loss, K loss is 1, the correction parameter is 1, the rated temperature is still 80℃, and the real-time temperature is 78℃, it is considered normal to avoid false alarms. By following the steps above, the problem of decreased compatibility of rated parameters caused by load loss and aging can be solved, ensuring that the rated parameters match the actual condition of the equipment and improving monitoring accuracy.

[0030] A Global Neural Network (GNN) model is used to monitor the status of associated devices. The associated devices and their connections are modeled as a graph structure. Using the node feature matrix and adjacency matrix as input, a graph convolutional layer is used to aggregate information, outputting node health (an indicator that quantifies the current operational health of associated devices; the closer the value is to 1, the more stable the device's operation) and fault propagation probability (an indicator that quantifies the likelihood of a fault in one associated device propagating to other associated devices (target nodes); the value range is also 0-1). The steps are as follows: Each associated device is treated as a node, and the electrical connections between devices are treated as edges with weights. The nodes and edges are then transformed into an adjacency matrix. The adjacency matrix and the node feature matrix are input into the GNN global propagation model. The graph convolutional layer allows each node to fuse its own state information with that of its neighboring devices. The fused node features output from the graph convolutional layer are input into a fully connected layer. A linear transformation maps the feature dimension to 1 (single-value output). The Sigmoid activation function then compresses the output value to the 0-1 range. This value represents the health of the corresponding node (device), i.e., the node health. For each source node (assuming it is a faulty source device), its fused features and the features of all target nodes (other related devices) are input into another fully connected layer. A probability mapping relationship from the source node to all target nodes is constructed through linear transformation. Then, the Softmax activation function is applied to make the sum of the probabilities of the source node transmitting to each target node equal to 1. The output value corresponding to each target node is the probability of the source node transmitting to that target node after the source node fails. This probability is the fault propagation probability. All devices identified as associated devices (such as main transformers, busbars, and circuit breakers) are defined as node objects. Each node is assigned a unique ID (such as main transformer T1 and busbar B2), and the device type is recorded. The node attributes are determined. If two associated devices have a valid electrical connection (hard connection, semi-hard connection, or soft connection), an edge is established between the two nodes, and the edge weight is assigned. Among them, a hard connection refers to a physical connection that can directly and uninterruptedly transmit electrical energy (or strong electrical signals), with no controllable switching equipment on the connection path (or the switching equipment is in the closed state for a long time and there is no need to disconnect), and faults can be transmitted instantly through the connection; Semi-hard connection refers to a power transfer connection achieved through controllable switching equipment (such as circuit breakers and disconnectors). When the switch is closed, it is equivalent to a hard connection (power can be transferred and faults can be conducted immediately). When the switch is opened, the connection is broken (no power is transferred and faults cannot be conducted). The effectiveness of the connection depends on the state of the switch. A soft connection refers to a connection used only for transmitting control signals and monitoring data (non-electrical power). It has no power transmission capability, and faults must be triggered indirectly through signals (which involves delays and are not necessarily propagated). The effectiveness of the connection depends on signal quality. The graph structure is transformed into an adjacency matrix. The adjacency matrix is ​​a numerical representation of the graph structure, used to describe the connection relationships and weights between nodes. Let matrix A be, where A[i][j] represents the edge weight from node i (device i) to node j (device j). Assuming there are 3 nodes: 1 is the main transformer T1, 2 is bus B2, and 3 is circuit breaker Q3, then the adjacency matrix is: The node feature matrix is ​​a collection of device status information used in GNN to fuse neighborhood features. Its dimension is "total number of associated devices (N) × feature dimension (D)". It contains the core status features of the devices. The following features are extracted from sensors, SCADA, and PMS systems: real-time operating features, historical status features, and corrected features. Assuming feature dimension D=5 (real-time load rate, real-time temperature, cumulative full load percentage, load loss coefficient, aging correction coefficient), the feature matrix is: Preprocessing the adjacency matrix: Add self-loops (A'=A+I, where I is the identity matrix) to allow nodes to retain their own features during fusion. Neighborhood feature fusion is achieved through 1-2 graph convolutional layers. The propagation formula is simplified to: in, The characteristics after fusion For activation function, Let A be the normalized matrix. For the node features of the l-th layer, For the trainable weights of the l-th layer, This is a bias term.

[0031] Using a fully connected layer and a sigmoid activation function, output the health score of each node (0-1, the closer to 1 the healthier). Using a fully connected layer and a softmax activation function, output the probability that the failure of each node will be passed on to all other nodes. For example, a substation has three core interconnected devices (main transformer T1, busbar B2, and circuit breaker Q3). Traditional monitoring only collects parameters for each device individually: main transformer T1 is monitored for current and temperature, busbar B2 for voltage and load, and circuit breaker Q3 for tripping status and operating mechanism temperature. Problems arise in actual operation: First, when the temperature of main transformer T1 is abnormal, it is not combined with voltage fluctuation data from busbar B2, and is mistakenly judged as a minor anomaly based solely on T1's own parameters, failing to promptly investigate potential overload risks. Second, the cascading effects of equipment failures cannot be predicted. In one instance, a short-circuit fault occurred in main transformer T1, and because the probability of it transmitting to busbar B2 and circuit breaker Q3 was not known in advance, the tripping of the busbar B2 protection device was delayed, expanding the power outage area. The edge weights are determined based on the connection type between devices. Main transformer T1 and bus B2 are hard connected (weight 0.9), bus B2 and circuit breaker Q3 are semi-hard connected (closed state, weight 0.8), and the weights between devices without direct connections are set to 0, forming a 3×3 adjacency matrix that clearly presents the connection relationships and strength between nodes (devices). Five-dimensional core features (real-time load rate, real-time temperature, cumulative full load percentage, load loss coefficient, and aging correction coefficient) of each device are extracted from the sensor, SCADA, and PMS systems to form a 3x5 node feature matrix, which centrally stores the real-time, historical, and corrected status information of the devices. A self-loop (i.e., superimposed identity matrix) is added to the adjacency matrix to obtain a new matrix, ensuring that each device can obtain the status of adjacent devices without losing its own status information during subsequent feature fusion (e.g., main transformer T1 retains its own temperature feature during fusion). A simplified graph convolution propagation formula is adopted, and the normalized adjacency matrix and the initial node feature matrix (i.e., the node feature matrix mentioned above) are substituted into the calculation. Neighborhood feature fusion is achieved through a single layer of graph convolution. The main transformer T1 fuses the features of bus B2, bus B2 fuses the features of main transformer T1 and circuit breaker Q3, and circuit breaker Q3 fuses the features of bus B2. This breaks the limitation of isolated monitoring of single devices. Through a fully connected layer combined with the Sigmoid activation function, the health score of each device is output (values ​​range from 0 to 1, with the closer to 1 indicating a healthier device). Through a fully connected layer combined with the Softmax activation function, the probability of each device's failure being transmitted to other devices is output, providing a quantitative basis for risk assessment and early warning. Through the above steps, after integrating the status of neighboring devices, the health of the devices no longer depends solely on their own parameters. For example, by combining the health of the main transformer T1 with the status of the bus B2, the risk level can be determined more accurately, avoiding the misjudgment problem of traditional isolated monitoring. The probability of fault propagation between output devices can be clearly defined (such as the probability of a fault in the main transformer T1 being transmitted to the bus B2). The protection strategies of related devices can be adjusted in advance (such as shortening the tripping delay of the bus B2) to avoid delays in protection actions.

[0032] Considering that traditional associated device monitoring only analyzes the status of a single device in isolation, does not quantify the connection weight between devices (such as the difference between hard connection and semi-hard connection), and does not integrate the status information of neighboring devices, the associated device monitoring unit 220 divides the connection into hard connection, semi-hard connection and soft connection according to the functional attributes of the electrical connection, determines the basic weight of the edge between devices according to the connection type, and corrects the basic weight through the key parameters of different connections; The basic weights are determined based on the fault propagation capability. The connection is graded and assigned a weight according to the directness, speed and intensity of the fault transmission. Hard connection can directly and instantly transmit fault current (such as short circuit), has the strongest transmission capability and is assigned a high weight. Semi-hard connection (closed state) can transmit faults equivalent to hard connection, with the second strongest transmission capability and the next strongest weight. Soft connection only transmits faults indirectly through signals (with delay and not necessarily), has the weakest transmission capability and is assigned a low weight. The weights are used to quantify the inherent influence of the connection in fault propagation. The more direct the transmission and the greater the impact, the higher the weight. For hard connections, contact resistance is used for correction (the higher the resistance, the lower the weight). For semi-hard connections, the switch status is considered first (the weight of opening is 0), and the weight is corrected according to the number of actions when closing (the more actions, the lower the weight). For soft connections, signal-to-noise ratio is used for correction (the lower the signal-to-noise ratio, the lower the weight). Key parameters are used to reflect the real-time status of the connection. The worse the status, the more the weight is reduced. Through the above corrections, the weight of the edge is assigned.

[0033] Historical fault cases are collected. Based on the real data that can be collected, synthetic minority class oversampling technology is used to directly generate new synthetic samples in the feature space of existing minority class (fault) samples through linear interpolation. Real equipment status data is collected from SCADA and PMS systems to form an initial training set. The samples of fault categories are oversampled until the data volume of each category is balanced, retaining high-fidelity feature data to form a complete synthetic training data sample. The above data samples are divided into training and validation sets. The actual health status and actual fault propagation path of each device are labeled to train the model. The adjacency matrix and node feature matrix are input into the GNN global propagation model. The graph convolutional layer allows each node to fuse its own and neighboring device's state information, and outputs the node health status and fault propagation probability of the associated devices.

[0034] In addition, the graded adjustment module 300, based on the node health and fault propagation probability, classifies associated equipment abnormalities into high-risk, medium-risk, and low-risk anomalies. Based on the anomaly level, it takes adjustment measures with the goal of suppressing faults while reducing the impact on the normal operation of the power grid.

[0035] From the above training samples, the minimum health value and the minimum propagation probability of faulty equipment are statistically analyzed and used as the initial threshold for high risk. The range below this threshold is the high-risk range. Then, the health value and propagation probability of faultless equipment but with potential risks are statistically analyzed and used as the medium-risk threshold. The range above the initial threshold but below the medium-risk threshold is the medium-risk range, and the range above the medium-risk threshold is the low-risk range. Among them, the graded adjustment module 300 is designed for high-risk anomalies, aiming to quickly isolate faults and minimize the scope of power outages; for medium-risk anomalies, it aims to prevent fault escalation and fine-tune to maintain stability; and for low-risk anomalies, it aims to minimize intervention and ensure normal operation. For high-risk anomalies, power supply is maintained while preventing the spread of faults through isolation, load reduction, and backup operation. If the equipment has backup (such as dual main transformers or backup lines), the backup equipment is put into operation first to carry the load, and then the high-risk equipment is shut down for maintenance. If there is no backup equipment, the equipment load is temporarily limited to reduce the probability of fault triggering, and emergency maintenance is arranged (within 24 hours). For the protection of related equipment, the protection thresholds of related equipment are adjusted in advance (such as shortening the circuit breaker tripping delay) to prevent the fault from spreading. For medium-risk anomalies, enhanced monitoring and preventative maintenance are adopted to eliminate potential hazards in advance. Monitoring frequency is increased: the interval for real-time parameter acquisition of equipment is reduced from 15 minutes to 5 minutes, with a focus on tracking key parameters such as temperature and current. Planned maintenance is carried out: preventative maintenance (such as replacing aging components of circuit breakers) is conducted during periods of low grid load (such as late at night) to avoid affecting peak electricity demand. Related equipment early warning is implemented: related equipment is marked in the SCADA system, and if the parameters of medium-risk equipment are abnormal, an early warning prompt is immediately triggered for the related equipment (without tripping). For low-risk anomalies, optimize operating parameters through parameter optimization and planned maintenance to avoid excessive intervention: fine-tune equipment operating thresholds, strengthen early warning, and incorporate them into routine maintenance: include equipment anomalies in the next quarter's planned maintenance list, requiring no emergency handling, and avoiding adjustments to related equipment: only record low-risk anomalies without changing the operating status and protection settings of related equipment.

[0036] Considering that the adjustment measures taken by associated devices will affect the threshold of a single device, the hierarchical adjustment module 300, after taking adjustment measures, determines the direction of threshold adjustment in a single device based on the abnormality level of the associated devices, determines the threshold adjustment range based on the node health and fault propagation probability, and adjusts the threshold of the single device. The higher the risk of associated devices, the more sensitive the local threshold for a single device needs to be (lowering it to trigger warnings earlier); the lower the risk, the more lenient the threshold can be (raising it to reduce false alarms). For high-risk associated devices, the threshold should be significantly lowered; for medium-risk devices, it should be slightly lowered; and for low-risk devices, it should be appropriately raised. The adjustment methods are as follows: First, determine the direction of adjustment: If the associated equipment is a high-risk anomaly, it means that the environment in which the equipment is located is easily affected by the fault, and the local threshold needs to be significantly lowered to make the early warning more sensitive and avoid missed detections; if the associated equipment is a medium-risk anomaly, there are certain hidden dangers in the environment, and the threshold only needs to be slightly lowered to balance the sensitivity of the early warning and the normal operation of the power grid; if the associated equipment is a low-risk anomaly, the environmental risk is low, and the threshold can be appropriately raised to reduce the interference of false alarms on the power grid. Next, the adjustment range is calculated: In high-risk scenarios, the adjustment range is obtained by weighted superposition of the node health inverse index and the fault propagation probability. First, the first part of the range is calculated based on the node health inverse value (i.e., the index reflecting the degree of equipment condition deterioration) with high weighting of node health; then, the second part of the range is calculated based on the fault propagation probability (the index reflecting the possibility of a fault affecting a single device) with high weighting of fault propagation probability; finally, the two parts are added together to obtain the overall reduction range. The worse the equipment condition and the easier it is for the fault to propagate, the larger the reduction range, ensuring that the threshold is sensitive enough. Among them, the high weight of node health and the high weight of fault propagation probability are determined by the proportion of the impact of the two types of high-risk indicators on the fault outcome in the statistical data sample, and the basic weight is allocated according to the proportion of impact. For example, since the driving effect of the deterioration of health on the fault is stronger, its weight is set to 10%, and the weight of propagation probability is set to 5% proportionally (the ratio of the two is about 2:1, which is close to the impact proportion of the case). In medium-risk scenarios, the adjustment magnitude is a weighted sum of the difference between node health and the upper limit of medium risk, and the difference between the fault propagation probability and the lower limit of medium risk. First, the difference between node health and the highest threshold of medium risk is calculated, and the first part of the magnitude is obtained by giving node health a lower weight. Then, the portion of the fault propagation probability that exceeds the lowest threshold of medium risk is calculated, and the second part of the magnitude is obtained by giving the fault propagation probability a lower weight. The sum of the two parts is the overall downward adjustment magnitude, which is smaller than that in high-risk scenarios. This covers potential risks while avoiding false alarms caused by over-adjustment. Among them, the lower weights of node health and fault propagation probability are determined by the proportion of the impact of the two types of indicators of medium risk on the fault outcome in the statistical data sample. The basic weights are allocated according to the proportion of the impact. For example, since the impact of medium-risk faults is only 1 / 2 to 2 / 3 of that of high-risk faults, the weights are reduced proportionally on the basis of the high-risk weights. The health weight is reduced from 10% to 5%, and the propagation probability weight is reduced from 5% to 3% (to maintain the same proportion of the impact as high-risk faults). In low-risk scenarios, the adjustment range is obtained by weighted subtraction of the portion of node health exceeding the low-risk lower limit and the portion of fault propagation probability falling below the low-risk upper limit. First, the portion of node health exceeding the low-risk minimum threshold is calculated, and the first part of the range is obtained with low weighting of node health. Then, the portion of fault propagation probability falling below the low-risk maximum threshold is calculated, and the second part of the range is obtained with low weighting of fault propagation probability. Finally, the first part of the result is subtracted from the second part of the result to obtain the overall upward adjustment range. The better the equipment's own condition and the more difficult it is for faults to propagate, the more reasonable the upward adjustment range is, reducing invalid false alarms during normal operation. Among them, the low weight of node health and the low weight of fault propagation probability are determined by the proportion of the impact of the two types of low-risk indicators on the fault results in the statistical data sample. The basic weights are allocated according to the proportion of the impact. For example, since health is more critical to the judgment of low risk, its weight is set to 8% (higher than the propagation probability weight), and the propagation probability weight is set to 3% (close to the contribution proportion). In actual operation, if it is found that there are still a few invalid warnings after the threshold is raised, the propagation probability weight is slightly reduced (e.g., from 3% to 2.5%) to ensure that the weight can effectively relax the threshold, reduce false alarms, and at the same time not miss extreme risks. Finally, verify the adjustment effect: After adjustment, observe the warning situation of single device. If there is a problem of missed judgment when actual abnormality is not triggered, further lower the threshold; if there are many false judgments when warning is frequently triggered in normal state, slightly increase the threshold. Subsequently, the threshold will be dynamically iterated and optimized every day based on the latest health status and fault propagation probability data of associated devices to ensure that it always adapts to the current risk. Considering that frequent threshold adjustments may lead to frequent false alarms and affect system stability, a minimum adjustment interval is set to limit the adjustment frequency. Based on the characteristics of real-time data acquisition by the SCADA system (time granularity of 15 minutes), it is stipulated that the minimum interval between two threshold adjustments for a single device is greater than 1 hour, and adjustments are only triggered when the risk level of the associated device changes across levels (such as from medium risk to high risk, or from low risk to medium risk); if the risk level fluctuates only within the same level, no adjustment is triggered to avoid adjusting the threshold for minor fluctuations. When the difference analysis module 200 performs filtering and noise reduction preprocessing on the data, the following requirements apply: when adjusting the threshold, the real-time instantaneous value of the associated device is not used, but the average value of the node health and fault propagation probability within 15 minutes is used to ensure that the adjustment is based on a stable risk state and to eliminate false risk signals caused by sensor instantaneous errors and short-term fluctuations in the power grid.

[0037] For example, in a certain substation, there is a hard connection between core related equipment and individual equipment: when the related equipment enters a high-risk state due to overload, the original local threshold of the individual equipment is not adjusted in time. Even if the operating parameters of the individual equipment are close to the safety threshold, no warning is triggered, which poses a risk of missed fault detection. When the related equipment returns to a low-risk state, the threshold of the individual equipment is not relaxed accordingly. The parameter fluctuations during normal operation frequently trigger invalid warnings, increasing the troubleshooting burden on maintenance personnel. Moreover, when the related equipment fluctuates slightly within the same risk level, the system still frequently adjusts the threshold of the individual equipment, causing the warning state to switch repeatedly, which affects the stability of the power grid operation. The direction of adjusting the local threshold for a single device is determined based on the risk level of the associated devices. If the associated device is in a high-risk anomaly, it indicates that the environment in which the device is located is susceptible to fault impact, and the local threshold needs to be significantly lowered to improve early warning sensitivity and avoid missed detections. If the associated device is in a medium-risk anomaly, there are certain hidden dangers in the environment, and the threshold is only slightly lowered to balance early warning sensitivity with normal grid operation. If the associated device is in a low-risk anomaly, the environmental risk is relatively low, and the threshold is appropriately raised to reduce the interference of false alarms on the grid. The adjustment range is calculated by combining the node health and fault propagation probability of the associated devices. In the case of a high-risk anomaly, the threshold adjustment is adjusted accordingly. The reverse indicator of node health and the probability of fault propagation are weighted and superimposed to determine the significant downward adjustment range. The lower the health of the associated equipment and the higher the probability of fault propagation, the greater the downward adjustment range. For medium-risk anomalies, a slight downward adjustment range is calculated based on the difference between the middle range of node health and the probability of fault propagation, with a lower weight, to avoid over-adjustment. For low-risk anomalies, a reasonable upward adjustment range is determined by weighting the portion of node health exceeding the low-risk threshold and the portion of fault propagation probability below the low-risk threshold. The higher the health of the associated equipment and the lower the probability of fault propagation, the more the upward adjustment range is adapted to the actual needs. After adjusting the threshold, observe the early warning situation of a single device. If there is a missed detection of actual abnormalities that do not trigger an early warning, further lower the threshold. If early warnings are frequently triggered under normal operating conditions, slightly reduce the threshold. Subsequently, the threshold will be dynamically iterated and optimized daily based on the latest health status and fault propagation probability data of the associated devices to ensure that the threshold always adapts to the current risk status. Through the above steps, when the associated equipment is at high risk, the single-device early warning is sensitive and there are no missed faults; when the risk is low, the early warning is lenient and there are no frequent false alarms. The accuracy of single-device anomaly judgment is significantly improved, the adjustment frequency is greatly reduced, the system oscillation caused by frequent threshold changes is avoided, the number of false alarms is reduced, and the power grid operation is not affected by invalid early warnings.

[0038] like Figure 3 As shown, a method for monitoring and maintaining a substation includes the following steps: S1. Collect electrical connection data and parameter correlation data, quantify the scope of equipment failure impact, obtain a comprehensive score, compare the comprehensive score with a preset threshold, and classify the equipment type into single equipment type and associated equipment type; S2. A local threshold diagnostic model is used to monitor the status of a single device and determine the abnormal situation of the single device. A GNN global propagation model is used to monitor the status of related devices. The node feature matrix and adjacency matrix are used as inputs, and information is aggregated using graph convolutional layers to output the node health and fault propagation probability. S3. Based on the node health and fault propagation probability, identify high-risk, medium-risk, and low-risk anomalies of associated equipment. Based on the anomaly level, take adjustment measures with the goal of suppressing faults while reducing the impact on the normal operation of the power grid.

[0039] In summary, the working principle of this solution is as follows: The monitoring and maintenance system and method of this substation, the quantitative classification module 100 classifies the equipment in the substation based on electrical connection and operating parameter quantification, avoids human subjective bias, sets the initial threshold by combining historical fault data, and then dynamically adjusts the threshold according to the equipment load rate to adapt to different operating scenarios of the substation and ensure that the classification is consistent with the actual impact of equipment faults. The difference analysis module 200 introduces correction parameters such as aging and load loss for single device monitoring. Combined with multi-level thresholds, it reduces false and missed judgments caused by changes in device status. Associated devices are modeled using a GNN model, which aggregates the status of neighbors. It outputs both device health and quantifies the probability of fault propagation, providing early warning of cascading fault risks and achieving full-coverage and accurate monitoring, taking into account both local and global risks. The graded adjustment module 300 is designed to quickly isolate high-risk faults to ensure core power supply, fine-tune and upgrade prevention for medium-risk faults, and reduce intervention for low-risk faults to avoid excessive operation and disturbance to the power grid. While taking risk response measures, it also adjusts the threshold of individual devices in a coordinated manner to make the monitoring of individual devices adaptively match the risk status of related devices.

[0040] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A monitoring and maintenance system for a substation, characterized in that, The system includes: The quantitative classification module (100) collects electrical connection data and parameter correlation data, calculates the electrical connection degree and data correlation degree of the equipment, quantifies the fault impact range of the equipment, obtains a comprehensive score, compares the comprehensive score with the preset threshold, and classifies the equipment type into single equipment type and associated equipment type; The difference analysis module (200) uses the local threshold comparison method to monitor the status of a single device. It collects the functional parameters of the single device in real time and compares them with the rated parameters of the single device to determine the abnormal situation of the single device. The rated parameters are corrected based on the cumulative operating time of a single device. The load loss coefficient is calculated based on the percentage of cumulative full-load time to determine the final corrected parameters: Calculate the load loss coefficient. If the load-related thresholds such as current and power are to be corrected, the impact of equipment aging must be taken into account. In this case, the final correction parameter = K loss × aging coefficient. If only the load loss-related thresholds are to be corrected, K loss can be used directly as the final correction parameter. A global neural network (GNN) model is used to monitor the status of associated devices. The associated devices and their connections are modeled as a graph structure. The node feature matrix and adjacency matrix are used as inputs, and the graph convolutional layer is used to aggregate information and output the node health and fault propagation probability. The graded adjustment module (300) classifies associated equipment into high-risk, medium-risk, and low-risk anomalies based on node health and fault propagation probability. Based on the anomaly level, it takes adjustment measures with the goal of suppressing faults while reducing the impact on the normal operation of the power grid.

2. The substation monitoring and maintenance system according to claim 1, characterized in that: The quantitative classification module (100) includes: The data acquisition and calculation unit (110) extracts the static electrical connection relationship and dynamic switch status between equipment from the substation design drawings. It calculates the electrical connection degree by statistically analyzing the number of equipment that can be directly and indirectly affected by the equipment through electrical connection and the connection strength. It collects equipment operating parameters from sensors and smart meters, and filters out strongly correlated parameter pairs. It calculates the data correlation degree by statistically analyzing the number of strongly correlated parameter pairs. The quantitative comparison unit (120) quantifies the scope of the equipment's fault impact by weighted summation, calculates the equipment's comprehensive score, and compares the equipment's comprehensive score with a preset threshold. If the comprehensive score is lower than the preset threshold, it is a single equipment type; if the comprehensive score exceeds the preset threshold, it is an associated equipment type.

3. The substation monitoring and maintenance system according to claim 2, characterized in that: The quantitative comparison unit (120) collects historical operating data of substation equipment, calculates the fault impact range of each piece of equipment, and uses data statistical analysis to draw a distribution histogram of equipment fault impact, using the inflection point of the distribution curve as the initial preset threshold.

4. The substation monitoring and maintenance system according to claim 3, characterized in that: The quantitative comparison unit (120) collects the actual operating load of the equipment in the substation in real time, calculates the load rate of the equipment, classifies the operating conditions of the equipment into levels, determines the adjustment range of the initial threshold according to the magnitude of the impact of the operating condition level on the equipment failure, and establishes a rule mapping table between the operating conditions and the threshold.

5. The substation monitoring and maintenance system according to claim 1, characterized in that: The difference analysis module (200) includes: The single device monitoring unit (210) determines the correction parameters based on the cumulative running time of the single device, corrects the rated parameters, compares the real-time functional parameters of the single device with the corrected rated parameters, outputs the abnormal status of the single device, and issues an early warning signal. The associated device monitoring unit (220) takes each associated device as a node, the electrical connection relationship between the devices as an edge and assigns weights, and then converts the nodes and edges into an adjacency matrix. The adjacency matrix and the node feature matrix are input into the GNN global propagation model. Through the graph convolutional layer, each node integrates its own and neighboring device status information and outputs the node health and fault propagation probability of the associated devices.

6. The substation monitoring and maintenance system according to claim 5, characterized in that: The single device monitoring unit (210) acquires the cumulative running time of the single device, calculates the load loss coefficient based on the percentage of the cumulative full load time, and determines the final correction parameters.

7. The substation monitoring and maintenance system according to claim 5, characterized in that: The associated equipment monitoring unit (220) classifies the connections into hard connections, semi-hard connections and soft connections according to the functional attributes of the electrical connections, determines the basic weight of the edge between the devices according to the connection type, and corrects the basic weight through the key parameters of different connections.

8. The substation monitoring and maintenance system according to claim 1, characterized in that: The graded adjustment module (300) aims to quickly isolate faults and minimize the scope of power outages for high-risk anomalies, to prevent fault escalation and fine-tune to maintain stability for medium-risk anomalies, and to minimize intervention and maintain normal operation for low-risk anomalies.

9. The substation monitoring and maintenance system according to claim 8, characterized in that: After taking adjustment measures, the hierarchical adjustment module (300) determines the adjustment direction of the threshold in a single device according to the abnormality level of the associated devices, determines the threshold adjustment range according to the node health and fault propagation probability, and adjusts the threshold of the single device.

10. A method for monitoring and maintaining a substation, characterized in that, Includes the following steps: S1. Collect electrical connection data and parameter correlation data, quantify the scope of equipment failure impact, obtain a comprehensive score, compare the comprehensive score with a preset threshold, and classify the equipment type into single equipment type and associated equipment type; S2. Use the threshold comparison method to monitor the status of a single device and determine the abnormal situation of the single device. Use the GNN global propagation model to monitor the status of related devices. Take the node feature matrix and adjacency matrix as input, use the graph convolutional layer to aggregate information, and output the node health and fault propagation probability. S3. Based on the node health and fault propagation probability, identify high-risk, medium-risk, and low-risk anomalies of associated equipment. Based on the anomaly level, take adjustment measures with the goal of suppressing faults while reducing the impact on the normal operation of the power grid.

Images