A method, system, and storage medium for monitoring a high voltage gas relay

Abstract

The application relates to the technical field of relay monitoring and control, and discloses a method and system for monitoring a high-voltage gas relay and a storage medium. The method comprises the following steps: collecting closed-loop control data of SF6 gas parameters of the high-voltage gas relay to obtain control loop operation data; calculating control deviation and residual vectors based on the data to obtain a control performance index; inputting the performance index into an isolation forest algorithm to identify abnormalities and obtain a health state evaluation result; and performing adaptive parameter adjustment according to the evaluation result, optimizing PID parameters by using iterative learning control, and generating a control parameter set. The application solves the problem that the traditional fixed threshold monitoring method is prone to false positives and false negatives by collecting closed-loop control system data of SF6 gas parameters of the high-voltage gas relay and using the isolation forest algorithm to identify abnormalities.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of relay monitoring and control, and particularly relates to a method and system for monitoring a high-voltage gas relay and a storage medium. BACKGROUND

[0002] A high-voltage gas relay is an important protection device in a power system, usually uses sulfur hexafluoride (SF6) as an insulating medium, and is widely used in high-voltage power transmission and distribution systems. The traditional high-voltage gas relay monitoring method mainly relies on regular inspection and offline testing, and evaluates the device state by detecting parameters such as SF6 gas density, pressure, and purity. With the advancement of smart grid construction, automated monitoring technology has been widely applied, and existing technologies have begun to use real-time monitoring systems to collect online data of operating parameters of high-voltage gas relays, and to detect abnormalities through simple threshold methods. These monitoring systems usually use fixed thresholds to determine the device state, and use traditional PID controllers to maintain the stability of parameters such as gas pressure and density.

[0003] However, the existing high-voltage gas relay monitoring technology has obvious deficiencies. First, the abnormal detection method based on fixed thresholds is difficult to adapt to complex and changing operating environments and device aging processes, and is prone to false positives and false negatives. Second, the traditional PID controller uses fixed parameters and cannot adaptively adjust according to the device health state and changes in the operating environment, resulting in gradually deteriorating control performance as the device ages. In addition, the existing monitoring method lacks sufficient data preprocessing, fails to fully utilize the patterns and knowledge contained in historical operating data, lacks accurate assessment and prediction capabilities for device states, and is difficult to support predictive maintenance decisions.

[0004] At the same time, how to accurately collect and control SF6 gas parameters in a high-voltage gas relay, how to scientifically analyze the collected data to obtain control performance indicators, how to use advanced algorithms to identify abnormal states and degrees, and how to dynamically adjust control parameters based on evaluation results to establish a complete closed-loop intelligent monitoring and control system. In particular, in terms of control parameter adjustment, how to adopt differentiated control strategies based on different abnormal levels, and how to use iterative learning algorithms to optimize parameters, are key links that need to be broken through at present. SUMMARY

[0005] The present application provides a method and system for monitoring a high-voltage gas relay and a storage medium, which collects data of a closed-loop control system for SF6 gas parameters of a high-voltage gas relay and uses an isolation forest algorithm for abnormal identification, solving the problem of false positives and false negatives in traditional fixed threshold monitoring methods. At the same time, an iterative learning control algorithm is used to adaptively adjust PID parameters based on health state evaluation results, overcoming the defect that traditional fixed parameter controllers cannot cope with device aging and environmental changes.

[0006] In a first aspect, the application provides a method for monitoring a high-voltage gas relay, the method comprising: collecting data based on a closed-loop control system for SF6 gas pressure, density, and temperature parameters of the high-voltage gas relay to obtain relay control loop operation data; performing control deviation calculation and residual vector analysis on the relay control loop operation data to obtain a relay control performance index; inputting the relay control performance index into an isolation forest algorithm for abnormal pattern recognition, determining the degree of abnormality by calculating the sample isolation path length, and obtaining a control loop health state evaluation result; performing adaptive control parameter adjustment according to the control loop health state evaluation result, optimizing the PID parameters using an iterative learning control algorithm, and generating a relay control parameter set.

[0007] Optionally, the collecting data based on a closed-loop control system for SF6 gas pressure, density, and temperature parameters of the high-voltage gas relay to obtain relay control loop operation data comprises:

[0008] The physical parameters of the SF6 gas are collected by the pressure sensor, the density sensor, and the temperature sensor installed in the gas chamber of the high-voltage gas relay, and the collected data is time-stamped to obtain original parameter time series data;

[0009] The original parameter time series data is subjected to three standard deviation principle outlier detection, single-point outliers are replaced with average values at previous and subsequent time points, and continuous outliers are repaired using linear interpolation to obtain cleaned parameter data;

[0010] The cleaned parameter data is input into a Kalman filter, state estimation is performed based on a process noise covariance matrix and a measurement noise covariance matrix, and filtered parameter data is obtained;

[0011] The filtered parameter data is subjected to closed-loop control system standardization processing, different dimension data is converted into standard distribution data, and standardized control loop parameter characteristics are obtained;

[0012] Based on the standardized control loop parameter characteristics, statistical quantities in a ten-minute sliding window are calculated, including mean, variance, maximum, minimum, kurtosis, and skewness, and a multi-dimensional control loop feature vector is obtained;

[0013] Based on the multi-dimensional control loop feature vector and the action time data recorded by the relay action mechanism sensor, the deviation between the controller output signal and the actual system response is calculated, and the relay control loop operation data is generated.

[0014] Optionally, the performing control deviation calculation and residual vector analysis on the relay control loop operation data to obtain a relay control performance index comprises:

[0015] The relay control loop operation data is processed in segments, the data is divided into multiple pressure regulation periods according to the SF6 gas pressure control instruction change point, and gas pressure control segmented data is obtained;

[0016] The difference between the pressure set value and the actual pressure value is calculated for each regulation period in the gas pressure control segmented data, a gas pressure control deviation sequence is formed, and gas pressure control deviation data is obtained;

[0017] The integral absolute error and the integral time absolute error of the gas relay pressure control loop are calculated according to the gas pressure control deviation data, the error data is weighted and summed, and a gas pressure control precision evaluation value is obtained;

[0018] The gas pressure control precision evaluation value is compared with the historical reference value of the relay, the gas pressure control performance degradation rate is calculated, and the relay insulation performance attenuation index is obtained in combination with the gas insulation strength analysis;

[0019] A gas relay residual vector is constructed based on the gas pressure control deviation data, and a structured analysis is performed on the residual vector by calculating the difference between the theoretical gas pressure response and the actual gas pressure response, and a relay leakage characteristic distribution is obtained;

[0020] According to the relay leakage characteristic distribution and the relay insulation performance attenuation index, the state of the relay gas system is comprehensively evaluated by a weighted fusion algorithm, and the relay control performance index is generated.

[0021] Optionally, the relay control performance index is input into an isolation forest algorithm for abnormal pattern recognition, the abnormality degree is determined by calculating the sample isolation path length, and a control loop health state evaluation result is obtained, including:

[0022] The relay control performance index is processed by data standardization and feature space conversion to obtain a standardized control performance feature set;

[0023] An isolation forest model is constructed based on the standardized control performance feature set, multiple decision trees are constructed by randomly selecting a sample subset and randomly selecting a feature, a feature and a split threshold are randomly selected at each node for data splitting until the sample is completely isolated or the preset maximum depth is reached, and an abnormal sample detection structure is obtained;

[0024] The standardized control performance feature set is input into the abnormal sample detection structure, the path length of each feature sample point from the root node to the leaf node is recorded, the average value of the path length of each decision tree is taken as the isolation path length, and the abnormal separation data of the feature sample is obtained;

[0025] According to the abnormal separation data of the feature sample, normalization processing is performed, the average path length is converted into a standardized abnormality index between zero and one, and the closer the value is to one, the higher the abnormality possibility is, and a relay control abnormality probability value is obtained;

[0026] Based on the relay control abnormality probability value, a multi-level abnormality threshold is set, the abnormality is divided into four levels of slight abnormality, moderate abnormality, severe abnormality and critical abnormality, the threshold is calibrated according to the characteristics of the high-voltage gas relay, and a relay abnormality level result is obtained;

[0027] According to the relay abnormality level result, the contribution of each control performance index to the abnormality is analyzed, the key parameters affecting the relay performance are determined through feature importance calculation, and the control loop health state evaluation result is generated.

[0028] Optionally, the isolation forest model is constructed based on the standardized control performance feature set, a plurality of decision trees are constructed by randomly selecting sample subsets and randomly selecting features, a feature and a split threshold are randomly selected at each node for data splitting until the samples are completely isolated or a preset maximum depth is reached, and an abnormal sample detection structure is obtained, including:

[0029] The sample subsets are divided from the standardized control performance feature set, a plurality of sample subsets are extracted from the feature set by using a random sampling method, the number of samples in each subset is kept equal, and a training sample subset set is obtained;

[0030] The decision tree forest is constructed based on the training sample subset set, a complete random decision tree is constructed for each sample subset, the total number of decision trees is set to be the logarithm value of the total number of samples, and an initial forest structure is obtained;

[0031] The construction process of each decision tree in the initial forest structure is controlled, the maximum tree depth is set to be the logarithm value of the feature dimension to prevent overfitting, and a depth-restricted decision tree set is obtained;

[0032] Based on the depth-restricted decision tree set, a node splitting operation is performed, a feature dimension is randomly selected for each internal node, and a split threshold is randomly selected in the value range of the feature to divide the data into left and right subtrees, and a randomly split decision tree structure is obtained;

[0033] According to the randomly split decision tree structure, sample path calculation is performed, the average path length of each normal sample from the root node to the leaf node in the training set is recorded, the path length distribution of the normal operation state of the high-voltage gas relay is established, and a normal state reference model is obtained;

[0034] The normal state reference model is calibrated, the abnormality determination boundary is adjusted according to the gas relay characteristics, the conversion coefficient of the abnormality degree is set in combination with the expert knowledge, and the abnormal sample detection structure is formed.

[0035] Optionally, the adaptive control parameter adjustment is performed according to the control loop health state evaluation result, the PID parameter is optimized by using an iterative learning control algorithm, and a relay control parameter set is generated, including:

[0036] The control loop health state evaluation result is processed in a hierarchical manner, the control parameter adjustment strategy is determined according to the abnormality level and the abnormality type, different adjustment weights are determined for different types of abnormalities, and a control parameter adjustment scheme is obtained;

[0037] The control parameter adjustment scheme is subjected to iterative learning control, a learning gain matrix is designed for the control characteristics of the high-voltage gas relay, historical control data is analyzed and processed, and control trajectory optimization data is obtained;

[0038] The control trajectory optimization data is applied to the PID controller, the deviation trend of the current control period is predicted by analyzing the error trajectory of the previous control period, the control output is adjusted, and the compensated control signal is obtained;

[0039] PID parameter automatic adjustment calculation is performed according to the compensated control signal, the error convergence speed and the overshoot index are used, the proportional coefficient, the integral coefficient and the differential coefficient are dynamically optimized by the gradient descent method, and an optimized PID parameter group is obtained;

[0040] The optimized PID parameter group is subjected to robustness verification, simulation tests are performed for multiple working points in the SF6 gas pressure and temperature variation range, the control performance stability is evaluated, and a PID controller parameter with robustness is obtained;

[0041] A complete control strategy is formed according to the PID controller parameter with robustness and the gas relay characteristics, including a normal mode parameter group, a slight abnormality mode parameter group, a moderate abnormality mode parameter group and a safety mode parameter group, and the relay control parameter set is generated.

[0042] Optionally, the iterative learning control is performed on the control parameter adjustment scheme, the learning gain matrix is designed for the control characteristics of the high-voltage gas relay, the historical control data is analyzed and processed, and the control trajectory optimization data is obtained, including:

[0043] The historical operation data of the high-voltage gas relay is processed in batches, each batch contains control data under similar working conditions, and a control data learning set is obtained;

[0044] The control error trajectory features are extracted based on the control data learning set, the time domain error curve shape features and the frequency domain error distribution features of each control period are calculated, and an error mode feature library is obtained;

[0045] Similarity analysis is performed on the error mode feature library, the similarity between the current error trajectory and the historical error trajectory is calculated by using a dynamic time warping algorithm, the historical error mode with the highest similarity is identified, and an error mode matching result is obtained;

[0046] A learning gain matrix is constructed according to the error mode matching result, the learning gain value is dynamically adjusted according to the control loop health state, the higher the abnormality degree is, the smaller the learning gain is, and an adaptive learning gain matrix is obtained;

[0047] The historical control input is iteratively calculated based on the adaptive learning gain matrix, the control input of the next period is corrected by an error propagation correction formula, ten times of iterative operation are performed, and an optimized control input sequence is obtained;

[0048] The optimized control input sequence is fused with the current control strategy, the fusion weight is determined according to the control loop health state evaluation result, the control trajectory is smoothed, and the control trajectory optimization data is generated.

[0049] In a second aspect, the application provides a system for monitoring a high-voltage gas relay, the system comprising:

[0050] A collection module configured to collect SF6 gas pressure, density and temperature parameters of the high-voltage gas relay based on a closed-loop control system to obtain relay control loop operation data;

[0051] An analysis module configured to calculate control deviation and analyze residual vector based on the relay control loop operation data to obtain a relay control performance index;

[0052] An identification module configured to input the relay control performance index into an isolation forest algorithm to identify an abnormal mode, determine an abnormality degree by calculating a sample isolation path length, and obtain a control loop health state evaluation result;

[0053] An execution module configured to perform adaptive control parameter adjustment based on the control loop health state evaluation result, optimize PID parameters by using an iterative learning control algorithm, and generate a relay control parameter set.

[0054] In a third aspect, a device for monitoring a high-voltage gas relay is provided, comprising a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to enable the device for monitoring a high-voltage gas relay to perform the above-mentioned method for monitoring a high-voltage gas relay.

[0055] In a fourth aspect, a computer readable storage medium is provided, and the computer readable storage medium stores instructions which, when executed on a computer, cause the computer to perform the method for monitoring a high-voltage gas relay.

[0056] In the technical scheme provided in the present application, the SF6 gas pressure, density and temperature parameters are accurately collected through a closed-loop control system, the running state of the relay control loop is comprehensively perceived, the accuracy and integrity of data collection are greatly improved, the control deviation calculation and residual vector analysis are performed on the relay control loop running data, the control performance index is generated, the device performance evaluation is changed from qualitative to quantitative, and the problem of fuzzy evaluation standard in the traditional monitoring method is solved. The core innovation point of the present scheme is that the relay control performance index is input into the isolation forest algorithm for abnormal mode recognition. As an unsupervised learning algorithm specially designed for anomaly detection, the isolation forest algorithm determines the abnormality degree by calculating the sample isolation path length, discards the limitations of the traditional fixed threshold monitoring method, and significantly improves the accuracy of anomaly detection. The algorithm uses the distribution characteristics of the data itself to construct a random decision tree forest, has good processing capability for high-dimensional data, is especially suitable for complex state monitoring of high-voltage gas relays, and has high calculation efficiency and small required sample size, which meets the real-time monitoring demand of power equipment. The control loop health state evaluation result obtained through the isolation forest algorithm accurately reflects the actual condition of the equipment, and provides a reliable basis for subsequent control optimization. According to the control loop health state evaluation result, adaptive control parameter adjustment is performed, the PID parameter is optimized by using the iterative learning control algorithm, intelligent self-adaptation of the control strategy is realized, and the problem that the traditional fixed parameter PID controller cannot cope with equipment aging and running environment changes is solved. The iterative learning control algorithm fully utilizes the experience knowledge in the historical running data of the equipment, optimizes the control input by continuously learning the historical error trajectory, and has strong pertinence and adaptability. The application of the algorithm in the control of the high-voltage gas relay enables the control parameters to be dynamically adjusted according to the health state of the equipment, different control strategies are adopted for different abnormal types and levels, and the control precision and reliability are greatly improved. The finally generated relay control parameter set covers multiple working conditions such as normal mode, slight abnormal mode, moderate abnormal mode and safe mode, realizes seamless switching of the control strategy, and ensures the safe and stable operation of the equipment in various states.

[0057] The present application combines artificial intelligence algorithm with traditional control theory, forms an intelligent monitoring and control method according to the characteristics and application scenarios of the high-voltage gas relay, effectively solves the technical problems faced by the high-voltage gas relay monitoring and control in the power system, and provides strong support for improving the reliability of the power grid operation. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0059] Figure 1 An embodiment of the method for monitoring the high-voltage gas relay in the present application;

[0060] Figure 2 An embodiment of the system for monitoring the high-voltage gas relay in the present application;

[0061] Figure 3 An embodiment of the structure schematic diagram of the device for monitoring the high-voltage gas relay in the present application. DETAILED DESCRIPTION

[0062] The embodiments of the present application provide a method, a system and a storage medium for monitoring a high-voltage gas relay. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily mean a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variation thereof is intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0063] For the convenience of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 An embodiment of the method for monitoring the high-voltage gas relay in the present application includes:

[0064] Step S101, data acquisition based on a closed-loop control system is performed on SF6 gas pressure, density and temperature parameters of the high-voltage gas relay to obtain relay control loop operation data;

[0065] Step S102, control deviation calculation and residual vector analysis are performed on the relay control loop operation data to obtain relay control performance indicators;

[0066] In step S103, the relay control performance index is input into the isolation forest algorithm for abnormal mode identification, the abnormality degree is determined by calculating the sample isolation path length, and the control loop health state evaluation result is obtained.

[0067] In step S104, the adaptive control parameter adjustment is performed according to the control loop health state evaluation result, the PID parameter is optimized by using the iterative learning control algorithm, and the relay control parameter set is generated.

[0068] It can be understood that the execution subject of the present application can be a system for monitoring a high-voltage gas relay, and can also be a terminal or a server, which is not limited here. The server is taken as an example for description of the embodiments of the present application.

[0069] Specifically, the SF6 gas pressure, density and temperature parameters of the high-voltage gas relay are collected based on a closed-loop control system. Firstly, the physical parameters of the SF6 gas are collected by the pressure sensor, the density sensor and the temperature sensor installed in the gas chamber of the high-voltage gas relay, and the collected data is marked with a time stamp accurate to milliseconds to form original parameter time series data. Then, the three standard deviation principle outlier detection is performed on the original time series data, the data points deviating from the mean value by more than three standard deviations are marked as outliers, the average value of the previous and subsequent time points is used to replace the single-point abnormal value, and the linear interpolation method is used to repair the continuous abnormal values. The repaired data is input into the Kalman filter for noise elimination, and after standardization, the data of different dimensions is uniformly converted into standard distribution data. The statistical characteristics in a ten-minute sliding window are calculated, including mean, variance, maximum, minimum, kurtosis and skewness, and finally the deviation between the controller output signal and the actual system response is calculated based on the action time data recorded by the relay action mechanism, and the relay control loop operation data is generated. The control deviation calculation and residual vector analysis are performed on the relay control loop operation data. Firstly, the data is divided into multiple pressure regulation periods according to the SF6 gas pressure control instruction change point, the difference between the pressure set value and the actual pressure value in each regulation period is calculated to form a gas pressure control deviation sequence. The integral absolute error and the integral time absolute error are calculated based on the sequence, and the gas pressure control precision evaluation value is obtained by weighted summation. The evaluation value is compared with the historical reference value of the relay, the gas pressure control performance degradation rate is calculated, and the relay insulation performance attenuation index is obtained by combining the gas insulation strength analysis. The difference between the theoretical gas pressure response and the actual gas pressure response is calculated, the gas relay residual vector is constructed, and the relay leakage characteristic distribution is obtained after the structural analysis of the residual vector. Finally, the relay leakage characteristic distribution and the insulation performance attenuation index are integrated by a weighted fusion algorithm to evaluate the state of the relay gas system, and the relay control performance index is generated.

[0070] The relay control performance index is input into the isolation forest algorithm for abnormal mode identification. This step first performs data standardization processing and feature space conversion on the relay control performance index to obtain a standardized control performance feature set. Based on this feature set, an isolation forest model is constructed, multiple decision trees are constructed by randomly selecting sample subsets and features, and at each node, a feature and a split threshold are randomly selected for data splitting until the samples are completely isolated or the preset maximum depth is reached, and an abnormal sample detection structure is obtained. The standardized feature set is input into this structure, the path length of each feature sample from the root node to the leaf node is recorded, the average value is taken as the isolation path length, and abnormal separation data is obtained. The abnormal separation data is normalized to convert the path length to an abnormal index between 0 and 1, and the closer the value is to 1, the higher the abnormality probability. According to the abnormal probability value, a multi-level abnormal threshold is set, the abnormality is divided into four levels of slight, moderate, severe and critical, and the threshold is calibrated according to the characteristics of the high-pressure gas relay. Finally, the contribution of each index to the abnormality is determined to determine the key parameters affecting the performance of the relay, and a control loop health state evaluation result is generated.

[0071] According to the control loop health state evaluation result, adaptive control parameter adjustment is performed. This step first determines the control parameter adjustment strategy according to the abnormality level and type, determines different adjustment weights for different types of abnormalities, and obtains a control parameter adjustment scheme. This scheme is applied to the iterative learning control process, a learning gain matrix is designed for the control characteristics of the high-pressure gas relay, and control trajectory optimization data is obtained by analyzing historical control data. This data is applied to the PID controller to predict the deviation trend of the current period by analyzing the error trajectory of the previous control period, and the control signal is adjusted to obtain the compensated control signal. Based on this signal, PID parameter automatic adjustment calculation is performed to optimize the proportional, integral and derivative coefficients. After robustness verification of the optimized parameter set, a complete control strategy is formed, including four parameter sets for normal mode, slight abnormal mode, moderate abnormal mode and safety mode, and a relay control parameter set is generated.

[0072] In a specific embodiment, the process of step S101 can specifically include the following steps:

[0073] The physical parameters of SF6 gas are collected by pressure sensors, density sensors and temperature sensors installed in the gas chamber of the high-pressure gas relay, and the collected data is time-stamped to obtain original parameter time series data;

[0074] The original parameter time series data is subjected to three times standard deviation principle outlier detection, single point outliers are replaced by average values at previous and subsequent time points, and continuous outliers are repaired using linear interpolation method to obtain cleaned parameter data;

[0075] The cleaned parameter data is input into a Kalman filter, state estimation is performed based on a process noise covariance matrix and a measurement noise covariance matrix, and filtered parameter data is obtained;

[0076] The filtered parameter data is subjected to closed-loop control system standardization processing, different dimensional data is converted into standard distribution data, and standardized control loop parameter characteristics are obtained.

[0077] Based on the standardized control loop parameter characteristics, statistics in a ten-minute sliding window are calculated, including mean, variance, maximum value, minimum value, kurtosis and skewness, and a multi-dimensional control loop feature vector is obtained.

[0078] According to the multi-dimensional control loop feature vector and the action time data recorded by the relay action mechanism sensor, the deviation between the controller output signal and the actual system response is calculated, and relay control loop operation data is generated.

[0079] Specifically, pressure sensors, density sensors and temperature sensors are installed in the gas chamber of the high-voltage gas relay, and the pressure, density and temperature parameters of SF6 gas are collected respectively. The pressure sensor uses a piezoresistive sensor, the measurement range is 0-1 MPa, and the accuracy is 0.01 MPa; the density sensor uses a thermal conductivity sensor, the measurement range is 0-50 kg / m³, and the accuracy is 0.001 kg / m³; the temperature sensor uses a PT100 platinum resistance, the measurement range is -40℃ to +120℃, and the accuracy is 0.1℃. The collection frequency is set to 5 times per second for pressure, and 1 time per second for density and temperature, and a timestamp accurate to milliseconds is added to each data point to form raw parameter time series data. The raw parameter time series data is subjected to three-sigma principle outlier detection. This method first calculates the mean and standard deviation of each type of parameter data, and marks data points deviating from the mean by more than three times the standard deviation as outliers. For single-point outliers, i.e. isolated abnormal points with normal values before and after, the average value of the previous and next time points is used to replace them; for continuous outliers, i.e. a situation where multiple consecutive data points are abnormal, a linear interpolation method is used for repair. For example, if the previous normal value of the continuous abnormal point is a, the next normal value is b, and there are n abnormal points in between, then the repair value of the ith abnormal point is a+(b-a)×i / (n+1). After processing in this way, the cleaned parameter data is obtained.

[0080] Kalman filter is a recursive estimation algorithm, which can make optimal estimation of system state by establishing system state equation and observation equation, combining process noise covariance matrix and measurement noise covariance matrix. In this method, process noise covariance matrix represents the uncertainty in the system state transition process, and measurement noise covariance matrix represents the uncertainty in the sensor measurement process. These two matrices are obtained by offline calibration of historical data. Kalman filter effectively eliminates random noise in sensor data through the iterative process of prediction-correction, and obtains smoothed parameter data. The filtered parameter data is standardized for closed-loop control system. Because the units and dimensions of pressure, density and temperature are different, they need to be converted into dimensionless standard distribution data. Standardization processing adopts Z-score method, which subtracts the mean of each parameter data point from the standard deviation, so that different parameter data is converted into standard normal distribution data with mean of 0 and standard deviation of 1. The data processed in this way is convenient for comparison and fusion in subsequent analysis, and the standardized control loop parameter characteristics are obtained.

[0081] The statistical quantities in a ten-minute sliding window are calculated based on the standardized control loop parameter characteristics. The sliding window refers to a fixed length time period sliding on the time series, which is set to ten minutes in this method, and the window moves one minute each time. In each window, six statistical characteristics of each parameter are calculated: mean reflects the average level of the parameter, variance reflects the fluctuation degree of the parameter, maximum and minimum reflect the extreme value range of the parameter, kurtosis reflects the sharpness of the data distribution, and skewness reflects the symmetry of the data distribution. By calculating these statistical quantities, the time series data is converted into a feature vector, and a multi-dimensional control loop feature vector is obtained. According to the multi-dimensional control loop feature vector and the action time data recorded by the relay action mechanism sensor, the deviation between the controller output signal and the actual system response is calculated. The controller output signal refers to the control instruction issued by the system, and the actual system response refers to the action actually performed by the relay. By comparing the difference between the theoretical response time and the actual response time, the time delay is calculated; by comparing the difference between the theoretical response amplitude and the actual response amplitude, the amplitude error is calculated. The time delay and amplitude error are used as control deviation indicators, combined with the aforementioned multi-dimensional feature vector, to generate complete relay control loop operation data.

[0082] In a specific embodiment, the process of step S102 can specifically include the following steps:

[0083] The relay control loop operation data is processed in segments, and the data is divided into multiple pressure regulation periods according to the SF6 gas pressure control instruction change point, to obtain gas pressure control segmented data;

[0084] The difference between the pressure set value and the actual pressure value is calculated for each adjustment period in the segmented gas pressure control data to form a gas pressure control deviation sequence, obtaining gas pressure control deviation data;

[0085] The integral absolute error and the integral time absolute error of the gas relay pressure control loop are calculated according to the gas pressure control deviation data, and the error data is weighted and summed to obtain a gas pressure control precision evaluation value;

[0086] The gas pressure control precision evaluation value is compared with the historical reference value of the relay, the gas pressure control performance degradation rate is calculated, and the relay insulation performance attenuation index is obtained by combining the gas insulation strength analysis;

[0087] Based on the gas pressure control deviation data, a residual vector of the gas relay is constructed, and the residual vector is structurally analyzed by calculating the difference between the theoretical gas pressure response and the actual gas pressure response, to obtain a relay leakage characteristic distribution;

[0088] According to the relay leakage characteristic distribution and the relay insulation performance attenuation index, the state of the relay gas system is comprehensively evaluated by a weighted fusion algorithm to generate a relay control performance index.

[0089] Specifically, the relay control loop operation data is processed in segments, and the data is divided by identifying the change point of the SF6 gas pressure control instruction. The change point refers to the time when the control instruction is adjusted from one pressure set value to another set value, representing the beginning of a control adjustment period. The change point is identified by using the slope change detection method, and the first-order difference of the pressure set value is calculated. When the difference value exceeds the preset threshold (usually set to 0.05 MPa), it is determined as a change point. The data between the adjacent two change points is divided into an adjustment period to form the segmented gas pressure control data. Each segmented data contains time stamp, pressure set value and actual pressure value and other information. The segmented gas pressure control segmented data is processed, and the pressure control deviation in each adjustment period is calculated. The pressure control deviation is defined as the difference between the pressure set value and the actual pressure value, and is arranged in time sequence to form a gas pressure control deviation sequence. For the sampling time t, the deviation value is equal to the pressure set value at that time minus the actual pressure value. The deviation sequence reflects the tracking accuracy of the control system to the instruction, and the smaller the deviation, the better the control performance. The gas pressure control deviation data generated by these calculations is the basic data for evaluating the performance of the control loop.

[0090] Based on the pressure control deviation data, two important control performance indicators are calculated: integral absolute error and integral time absolute error. The integral absolute error refers to the cumulative sum of the absolute values of all deviations within a regulation period, reflecting the overall size of the deviation; the integral time absolute error is the cumulative sum of the absolute values of each time multiplied by the corresponding time weight, paying more attention to the deviation with long duration. The calculation of integral absolute error is to sum the absolute values of the deviations at each sampling time within the regulation period; the calculation of integral time absolute error is to multiply each absolute value by its corresponding time (relative to the time of the start of regulation) and then sum. The two error indicators are weighted and fused, generally using weights of 0.6 and 0.4 for integral absolute error and integral time absolute error respectively, and performing weighted summation to obtain the gas pressure control precision evaluation value. The smaller the value, the higher the control precision. Compare the calculated gas pressure control precision evaluation value with the historical reference value of the relay. The historical reference value refers to the typical control precision evaluation value of the relay under normal working conditions, usually taken from the test data at the initial stage of the device operation or the reference data provided by the manufacturer. By calculating the ratio of the current evaluation value to the reference value, the control performance degradation rate is obtained. At the same time, combined with the gas insulation strength analysis, the gas insulation strength is affected by the purity, density and temperature of SF6 gas. The content of decomposition products such as carbon tetrafluoride and sulfur tetrafluoride in SF6 gas is analyzed by professional gas analyzer. The increase of the content of these decomposition products indicates the decline of the gas insulation ability. The control performance degradation rate and the gas insulation analysis results are considered comprehensively to obtain the relay insulation performance decay indicator, which reflects the degree of deterioration of the insulation performance of the device over time.

[0091] Based on the pressure control deviation data, a gas relay residual vector is constructed. The residual vector refers to the difference sequence between the theoretical model predicted gas pressure response and the actual gas pressure response. The theoretical response is calculated based on the physical model of the relay and the control parameters, reflecting the expected performance of the device under ideal conditions. When constructing the residual vector, first establish the dynamic model of the relay, then calculate the theoretical response by substituting the actual control input into the model, and subtract the actual response point by point to obtain the residual. The residual vector is analyzed structurally, including frequency domain analysis, time series feature extraction, etc., to identify regular patterns in the residual. These patterns are usually related to specific types of device defects. Through residual analysis, the relay leakage characteristic distribution is obtained. The leakage characteristic in the residual is manifested as a slow downward trend in pressure. According to the relay leakage characteristic distribution and the relay insulation performance decay indicator, the state of the relay gas system is comprehensively evaluated by a weighted fusion algorithm. The weighted fusion algorithm assigns weights according to the influence of the two indicators on the performance of the device. The leakage characteristic is directly related to the pressure maintenance of the gas, and is given a higher weight (such as 0.7); the insulation performance decay is closely related to the safe operation of the device, and is given a proper weight (such as 0.3). The relay control performance indicator is calculated by weighted summation.

[0092] In a specific embodiment, the process of performing step S103 can specifically include the following steps:

[0093] Data standardization processing is performed on the relay control performance indicators, and feature space conversion is performed to obtain a standardized control performance feature set;

[0094] Based on the standardized control performance feature set, an isolation forest model is constructed, a plurality of decision trees are constructed by randomly selecting sample subsets and randomly selecting features, a feature and a split threshold are randomly selected at each node for data splitting until the samples are completely isolated or a preset maximum depth is reached, and an abnormal sample detection structure is obtained;

[0095] Specifically, the standardized control performance feature set is divided into sample subsets, a plurality of sample subsets are extracted from the feature set using a random sampling method, the number of samples in each subset is kept equal, a training sample subset set is obtained; based on the training sample subset set, a decision tree forest is constructed, a completely random decision tree is constructed for each sample subset, the total number of decision trees is set to be the logarithmic value of the total number of samples, and an initial forest structure is obtained; the construction process of each decision tree in the initial forest structure is controlled, the maximum tree depth is set to be the logarithmic value of the feature dimension to prevent overfitting, and a depth-restricted decision tree set is obtained; based on the depth-restricted decision tree set, a node splitting operation is performed, a feature dimension is randomly selected for each internal node, and a split threshold is randomly selected within the value range of the feature to divide the data into left and right subtrees, and a randomly split decision tree structure is obtained; sample path calculation is performed according to the randomly split decision tree structure, the average path length of each normal sample in the training set from the root node to the leaf node is recorded, the path length distribution of the normal operating state of the high-pressure gas relay is established, and a normal state reference model is obtained; the normal state reference model is calibrated, the abnormality judgment boundary is adjusted according to the characteristics of the gas relay, the conversion coefficient of the abnormality degree is set in combination with expert knowledge, and an abnormal sample detection structure is formed.

[0096] The standardized control performance feature set is input into the abnormal sample detection structure, the path length of each feature sample point from the root node to the leaf node is recorded, the average value of the path length of each decision tree is taken as the isolation path length, and abnormal separation data of the feature sample is obtained;

[0097] According to the abnormal separation data of the feature sample, normalization processing is performed, the average path length is converted into a standardized abnormality index between zero and one, the value closer to one indicates a higher abnormality possibility, and a relay control abnormality probability value is obtained;

[0098] The abnormality is divided into four levels of slight abnormality, moderate abnormality, serious abnormality and critical abnormality based on setting multi-level abnormal threshold values according to the abnormal probability values of the relay control, and the threshold values are calibrated according to the characteristics of the high-voltage gas relay to obtain the relay abnormality level results.

[0099] According to the relay abnormality level results, the contribution of each control performance index to the abnormality is analyzed, the key parameters affecting the relay performance are determined through feature importance calculation, and the control loop health state evaluation results are generated.

[0100] Specifically, the relay control performance indexes are subjected to data standardization processing to eliminate the influence of the dimension differences of different indexes. The standardization processing adopts the maximum and minimum value normalization method, each control performance index is subtracted by its minimum value and then divided by the range (the difference between the maximum value and the minimum value), and is converted into a value between 0 and 1. For example, for the integral absolute error index, if its range is 10 to 100 MPa·s, a data point of 40 MPa·s is subjected to standardization processing to obtain (40-10) / (100-10)=0.33. After the standardization processing of all control performance indexes, feature space conversion is performed, the multi-dimensional features are reduced through principal component analysis method to retain the main information, and the standardized control performance feature set is obtained. The isolation forest model is constructed based on the standardized control performance feature set, which is an algorithm specially designed for anomaly detection, and the basic principle is that abnormal data points are more easily "isolated". The specific implementation includes six detailed steps: first, the sample subset division is performed on the standardized control performance feature set, a random sampling method is used to extract multiple sample subsets from the feature set, each subset contains a part of the original data samples, and the sample quantities of the subsets are kept equal. For example, if there are 1000 high-voltage gas relay historical running data samples, they can be divided into 100 subsets, each subset contains 250 randomly selected samples, so that the training sample subset set is obtained.

[0101] The decision tree forest is constructed based on the training sample subset set, and a completely random decision tree is constructed for each sample subset. The completely random refers to randomly selecting features and splitting thresholds when splitting nodes, without searching for the optimal split. The total number of decision trees is usually set to the logarithmic value of the total number of samples, such as about 7 trees for 1000 samples, forming an initial forest structure. The third step controls the construction process of each decision tree in the initial forest structure, and sets the maximum tree depth to the logarithmic value of the feature dimension, to prevent overfitting caused by overgrowth of the decision tree. For example, if the feature dimension is 16, the maximum tree depth is set to log(16)≈4, obtaining a set of depth-limited decision trees. Based on the set of depth-limited decision trees, a node splitting operation is performed, a feature dimension such as pressure control accuracy, insulation performance index, etc. is randomly selected for each internal node, and a splitting threshold is randomly selected in the value range of the feature to divide the data into left and right sub-trees. This random splitting is different from the traditional decision tree method of searching for the optimal split point, but emphasizes randomness, which is beneficial to detecting abnormal values. Through this splitting method, a random splitting decision tree structure is formed. The fifth step calculates the sample path based on the random splitting decision tree structure, and records the average path length of each normal sample in the training set from the root node to the leaf node. Normal samples usually need more splitting to be isolated, and therefore have a longer average path length. By counting these path lengths, the path length distribution of the normal operating state of the high-voltage gas relay is established, and a normal state reference model is obtained.

[0102] The normal state reference model is calibrated, and the abnormality determination boundary is adjusted according to the characteristics of the gas relay. Different models and different operating environments of the gas relay may require different determination standards, and the conversion coefficient of the abnormality degree is set in combination with expert knowledge to form a final abnormal sample detection structure. After the construction of the isolation forest model is completed, the standardized control performance feature set is input into the abnormal sample detection structure, the path length of each feature sample point from the root node to the leaf node in all decision trees is recorded, the average path length of each decision tree is taken as the isolation path length, and the abnormal separation data of the feature sample is obtained.

[0103] According to the abnormal separation data of the characteristic sample, normalization processing is performed to convert the average path length into a standardized abnormality index between 0 and 1. The conversion process takes into account the influence of the size of the data set on the path length, and a correction factor is introduced when calculating the abnormality index. The closer the numerical value is to 1, the higher the abnormality possibility, and the relay control abnormality probability value is obtained after this processing. Based on the relay control abnormality probability value, multi-level abnormality threshold values are set to divide the abnormality into four levels: slight abnormality (0.6-0.75), moderate abnormality (0.75-0.85), severe abnormality (0.85-0.95), and critical abnormality (>0.95). The threshold values are calibrated according to the characteristics of the high-pressure gas relay, taking into account factors such as the importance of the equipment and the operating environment, to obtain the relay abnormality level results. According to the relay abnormality level results, the contribution of each control performance indicator to the abnormality is analyzed. The key parameters affecting the relay performance are determined by calculating the feature importance, which is achieved by analyzing the frequency of each feature being selected as a split feature and the change amplitude of the abnormality score caused by the split. The identified key parameters will help determine the fault type and cause, and form a complete control loop health status evaluation result.

[0104] In a specific embodiment, the process of performing step S104 can specifically include the following steps:

[0105] The control loop health status evaluation result is processed by grading, and the control parameter adjustment strategy is determined according to the abnormality level and the abnormality type. Different adjustment weights are determined for different types of abnormalities to obtain a control parameter adjustment scheme;

[0106] The control parameter adjustment scheme is subjected to iterative learning control, and a learning gain matrix is designed according to the control characteristics of the high-pressure gas relay. The historical control data is analyzed and processed to obtain control trajectory optimization data;

[0107] Specifically, the historical operation data of the high-voltage gas relay is batch-processed, each batch containing control data under similar working conditions to obtain a control data learning set; control error trajectory features are extracted based on the control data learning set, time-domain error curve shape features and frequency-domain error distribution features of each control period are calculated to obtain an error mode feature library; similarity analysis is performed on the error mode feature library, the similarity between the current error trajectory and the historical error trajectory is calculated using the dynamic time warping algorithm, and the highest similarity historical error mode is identified to obtain an error mode matching result; a learning gain matrix is constructed according to the error mode matching result, the learning gain value is dynamically adjusted according to the control loop health state, the higher the abnormality degree, the smaller the learning gain, and an adaptive learning gain matrix is obtained; the historical control input is iteratively calculated based on the adaptive learning gain matrix, the control input of the next period is corrected through an error propagation correction formula, ten iterations are performed, and an optimized control input sequence is obtained; the optimized control input sequence is fused with the current control strategy, the fusion weight is determined according to the control loop health state evaluation result, the control trajectory is smoothed, and control trajectory optimization data is generated.

[0108] The control trajectory optimization data is applied to the PID controller, the deviation trend of the current control period is predicted by analyzing the error trajectory of the previous control period, the control output is adjusted, and the compensated control signal is obtained;

[0109] PID parameter automatic adjustment calculation is performed according to the compensated control signal, the error convergence speed and the overshoot index are used to dynamically optimize the proportional coefficient, the integral coefficient and the differential coefficient by the gradient descent method, and an optimized PID parameter group is obtained;

[0110] The optimized PID parameter group is verified for robustness, simulation tests are performed for multiple working points within the SF6 gas pressure and temperature variation range, the control performance stability is evaluated, and the PID controller parameters with robustness are obtained;

[0111] According to the PID controller parameters with robustness and the gas relay characteristics, a complete control strategy is formed, including normal mode parameter group, slight abnormal mode parameter group, moderate abnormal mode parameter group and safety mode parameter group, and a relay control parameter set is generated.

[0112] Specifically, the control loop health state evaluation results are classified, and the abnormality level (mild, moderate, severe, and critical) and the abnormality type (pressure abnormality, density abnormality, temperature abnormality, etc.) obtained by the isolation forest algorithm are used to determine the control parameter adjustment strategy. For a slight abnormality, a small amplitude parameter fine tuning is adopted; for a moderate abnormality, a moderate amplitude parameter adjustment is performed; for a severe abnormality, a large amplitude parameter reorganization is implemented; and for a critical abnormality, a safe mode parameter group is switched to. Different adjustment weights are determined for different types of abnormality, for example, for a pressure abnormality, the pressure control loop related parameters are mainly adjusted, and for a temperature abnormality, the temperature control parameters are mainly adjusted. Through this classified strategy, a targeted control parameter adjustment scheme is obtained. Iterative learning control is a control strategy for repetitive processes, which optimizes the next control input by learning the error of the previous control cycle. The learning gain matrix is designed according to the control characteristics of the high-voltage gas relay, and the learning gain matrix determines the degree of the control system absorbing historical experience. Too large will lead to system instability, and too small will not have obvious learning effect. The historical control data are analyzed and processed to obtain control trajectory optimization data. This process includes six detailed sub-steps:

[0113] The historical operation data of the high-voltage gas relay is batch-processed, and the continuous control data is grouped according to the similarity of the working conditions, and each batch contains control data under similar working conditions. Similar working conditions refer to the running state in which the external environment temperature, load condition, control target and other factors are similar. For example, control data with an environment temperature in the range of 5-15°C and a load rate between 70% and 80% is classified into one batch, and the control data learning set is obtained after such classification. Based on the control data learning set, the control error trajectory features are extracted, and the control error trajectory refers to the curve of the difference between the control target value and the actual value over time. By calculating the time domain error curve shape features (such as rise time, peak error, steady state error, etc.) and the frequency domain error distribution features (obtained by Fourier transform to obtain the distribution of error at different frequency components) of each control period, a systematic error mode feature library is formed. Similarity analysis is performed on the error mode feature library, and the dynamic time warping algorithm is used to calculate the similarity between the current error trajectory and the historical error trajectory. The dynamic time warping algorithm is a method for measuring the similarity of two time series, which can handle sequences of different lengths and rhythms, and is particularly suitable for analyzing control error trajectories. By calculating the similarity between the error trajectory of the current control period and all error trajectories in the historical library, the error mode with the highest similarity is identified, and the error mode matching result is obtained. According to the error mode matching result, a learning gain matrix is constructed, and the learning gain value is dynamically adjusted according to the health state of the control loop. The higher the abnormality, the smaller the learning gain. This design is because when the system has a high degree of abnormality, the reference value of historical experience decreases, and excessive dependence may exacerbate the problem. Different learning gains are assigned according to the importance and sensitivity of different control variables to form an adaptive learning gain matrix. Based on the adaptive learning gain matrix, the historical control input is iteratively calculated, and the control input of the next period is corrected through the error propagation correction formula. The error propagation correction formula multiplies the control error of the current period by the learning gain and adds it to the control input of the next period to realize feedforward compensation. This process usually undergoes ten iterations, and each iteration is based on the results of the previous iteration to further optimize the control input sequence. The optimized control input sequence is fused with the current control strategy, and the fusion process determines the fusion weight according to the health state evaluation result of the control loop. For a healthy control loop, the current control strategy is preferred; for a control loop with poor health state, the optimized input sequence is adopted more. The control trajectory is smoothed to eliminate possible jump points, and a smooth control trajectory optimization data is generated.

[0114] The control trajectory optimization data is applied to a PID controller, which is the most commonly used controller type in industrial control systems and is composed of three parts: proportion, integral, and differential. By analyzing the error trajectory of the previous control period to predict the deviation trend of the current control period, the control output is adjusted to obtain the compensated control signal. According to the compensated control signal, automatic adjustment calculation of PID parameters is performed, and error convergence speed and overshoot indicators are used to evaluate control performance. Error convergence speed reflects the speed of system response, and overshoot reflects system stability. The proportion coefficient, integral coefficient, and differential coefficient are dynamically optimized by gradient descent method, which adjusts the parameters in the direction of the fastest performance index decline, iteratively solves the optimal value, and obtains the optimized PID parameter group. The robustness of the optimized PID parameter group is verified, and simulation tests are performed for multiple working points within the temperature and pressure range of SF6 gas. The working points usually cover extreme conditions of device operation, such as combinations of the highest temperature, the lowest temperature, the highest pressure, and the lowest pressure. By evaluating the control performance stability at different working points, such as step response and disturbance suppression ability, the PID controller parameters with good robustness are selected. According to the PID controller parameters with robustness and the characteristics of the gas relay, a complete control strategy is formed, including normal mode parameter group, slight abnormal mode parameter group, moderate abnormal mode parameter group, and safety mode parameter group, to generate a set of relay control parameters.

[0115] The method for monitoring the high-voltage gas relay in the embodiments of the present application is described above, and the system for monitoring the high-voltage gas relay in the embodiments of the present application is described below. Please refer to Figure 2 An embodiment of the system for monitoring the high-voltage gas relay in the embodiments of the present application includes:

[0116] The acquisition module 201 is configured to perform data acquisition based on a closed-loop control system for SF6 gas pressure, density, and temperature parameters of the high-voltage gas relay to obtain relay control loop operation data.

[0117] The analysis module 202 is configured to perform control deviation calculation and residual vector analysis on the relay control loop operation data to obtain relay control performance indicators.

[0118] The identification module 203 is configured to input the relay control performance indicators into an isolation forest algorithm for abnormal mode identification, determine the degree of abnormality by calculating the sample isolation path length, and obtain a control loop health state evaluation result.

[0119] The execution module 204 is configured to perform adaptive control parameter adjustment according to the control loop health state evaluation result, optimize PID parameters using an iterative learning control algorithm, and generate a set of relay control parameters.

[0120] The above Figure 2The system for monitoring high-voltage gas relay in the embodiment of the present application is described in detail from the perspective of modular functional entities, and the device for monitoring high-voltage gas relay in the embodiment of the present application is described in detail from the perspective of hardware processing.

[0121] Figure 3 Fig. 1 is a structural schematic diagram of a device for monitoring high-voltage gas relay provided by the embodiment of the present application. The device 300 for monitoring high-voltage gas relay can be different in configuration or performance and can include one or more central processing units (CPU) 310 (for example, one or more processors) and a memory 320, one or more storage media 330 (for example, one or more mass storage device ends) for storing application programs 333 or data 332. The memory 320 and the storage media 330 can be temporary storage or persistent storage. The programs stored in the storage media 330 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations for the device 300 for monitoring high-voltage gas relay. Furthermore, the processor 310 can be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the device 300 for monitoring high-voltage gas relay to implement the steps of the method for monitoring high-voltage gas relay described above.

[0122] The device 300 for monitoring high-voltage gas relay can also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that the device for monitoring high-voltage gas relay can include more or fewer components than those shown in the figure, or some components can be combined, or different components can be arranged. Figure 3 The device structure for monitoring high-voltage gas relay shown in the figure does not constitute a limitation on the device for monitoring high-voltage gas relay provided by the present application, and can include more or fewer components than those shown in the figure, or some components can be combined, or different components can be arranged.

[0123] The present application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the method for monitoring high-voltage gas relay.

[0124] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0125] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a device (which can be a personal computer, a server, or a network device, etc.) for monitoring high-voltage gas relays to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0126] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for monitoring a high-pressure gas relay, characterized in that, The method includes: S1. Data acquisition of SF6 gas pressure, density, and temperature parameters of the high-pressure gas relay based on a closed-loop control system is performed to obtain the relay control loop operation data; S2. Perform control deviation calculation and residual vector analysis on the relay control loop operating data to obtain relay control performance indicators, including: segmenting the relay control loop operating data; dividing the data into multiple pressure regulation cycles based on the SF6 gas pressure control command change points to obtain gas pressure control segmented data; calculating the difference between the pressure setpoint and the actual pressure value for each regulation cycle in the gas pressure control segmented data to form a gas pressure control deviation sequence to obtain gas pressure control deviation data; calculating the integral absolute error and integral time absolute error of the gas relay pressure control loop based on the gas pressure control deviation data, and performing error data analysis. Weighted summation yields the gas pressure control accuracy evaluation value; the gas pressure control accuracy evaluation value is compared with the relay's historical benchmark value to calculate the gas pressure control performance degradation rate, and combined with gas insulation strength analysis, the relay insulation performance degradation index is obtained; a gas relay residual vector is constructed based on the gas pressure control deviation data, and the residual vector is subjected to structured analysis by calculating the difference between the theoretical gas pressure response and the actual gas pressure response to obtain the relay leakage characteristic distribution; based on the relay leakage characteristic distribution and the relay insulation performance degradation index, the relay gas system state is comprehensively evaluated through a weighted fusion algorithm to generate the relay control performance index; S3. Input the relay control performance indicators into the isolated forest algorithm for abnormal pattern recognition. Determine the degree of abnormality by calculating the sample isolation path length to obtain the control loop health status assessment result. Specifically, this includes: standardizing the relay control performance indicators and performing feature space transformation to obtain a standardized control performance feature set; constructing an isolated forest model based on the standardized control performance feature set; building multiple decision trees by randomly selecting sample subsets and features; randomly selecting a feature and splitting threshold at each node for data splitting until the samples are completely isolated or the preset maximum depth is reached to obtain an abnormal sample detection structure; inputting the standardized control performance feature set into the abnormal sample detection structure and recording the path length from the root node to the leaf node for each feature sample point. The average path length of each decision tree is taken as the isolation path length to obtain anomaly separation data for feature samples. This anomaly separation data is then normalized to convert the average path length into a standardized anomaly index between zero and one. A value closer to one indicates a higher probability of an anomaly, resulting in a relay control anomaly probability value. Based on this probability value, multi-level anomaly thresholds are set, classifying anomalies into four levels: minor, moderate, severe, and critical. The thresholds are calibrated to suit the characteristics of the high-pressure gas relay, yielding relay anomaly level results. The contribution of each control performance index to the anomaly is analyzed based on these results. Key parameters affecting relay performance are determined through feature importance calculation, generating a health status assessment result for the control loop. Specifically, in S3, obtaining the abnormal sample detection structure includes: dividing the standardized control performance feature set into sample subsets; using a random sampling method to extract multiple sample subsets from the feature set, keeping the number of samples in each subset equal, to obtain a training sample subset set; constructing a decision tree forest based on the training sample subset set, building a completely random decision tree for each sample subset, setting the total number of decision trees to the logarithm of the total number of samples, to obtain an initial forest structure; controlling the construction process of each decision tree in the initial forest structure, setting the maximum tree depth to the logarithm of the feature dimension to prevent overfitting, to obtain a depth-limited decision tree set; based on the depth-limited... The decision tree set is subjected to node splitting operation. For each internal node, a feature dimension is randomly selected, and a splitting threshold is randomly selected within the value range of the feature to divide the data into left and right subtrees, resulting in a randomly split decision tree structure. Sample path calculation is performed based on the randomly split decision tree structure. The average path length from the root node to the leaf node for each normal sample in the training set is recorded to establish the path length distribution of the normal operating state of the high-pressure gas relay, thus obtaining a normal state reference model. The normal state reference model is then parameter-calibrated. The anomaly judgment boundary is adjusted according to the characteristics of the gas relay, and the conversion coefficient of the anomaly degree is set in combination with expert knowledge to form the anomaly sample detection structure. S4. Based on the health status assessment results of the control loop, perform adaptive control parameter adjustment, optimize PID parameters using an iterative learning control algorithm, and generate a relay control parameter set. This includes: classifying the health status assessment results of the control loop, determining control parameter adjustment strategies based on the anomaly level and type, assigning different adjustment weights to different types of anomalies, and obtaining a control parameter adjustment scheme; performing iterative learning control on the control parameter adjustment scheme, designing a learning gain matrix for the control characteristics of the high-pressure gas relay, analyzing and processing historical control data to obtain optimized control trajectory data; applying the optimized control trajectory data to the PID controller, predicting the deviation trend of the current control cycle by analyzing the error trajectory of the previous control cycle, and adjusting the control... The system outputs a compensated control signal. Based on this compensated control signal, it performs automatic PID parameter adjustment calculations, dynamically optimizing the proportional, integral, and derivative coefficients using the gradient descent method with the error convergence rate and overshoot index to obtain an optimized PID parameter set. The robustness of the optimized PID parameter set is verified by performing simulation tests at multiple operating points within the SF6 gas pressure and temperature variation range to evaluate control performance stability and obtain robust PID controller parameters. Based on the robust PID controller parameters and the characteristics of the gas relay, a complete control strategy is formed, including normal mode parameter sets, slightly abnormal mode parameter sets, moderately abnormal mode parameter sets, and safe mode parameter sets, generating the relay control parameter set.

2. The method for monitoring high-pressure gas relays according to claim 1, characterized in that, The data acquisition of SF6 gas pressure, density, and temperature parameters for the high-pressure gas relay based on a closed-loop control system yields relay control loop operation data, including: The physical parameters of SF6 gas are collected by pressure sensors, density sensors and temperature sensors installed in the gas chamber of the high-pressure gas relay, and the collected data is timestamped to obtain the raw parameter time series data. The original parameter time series data is subjected to outlier detection based on the three-standard-deviation principle. The average value of the preceding and following time points is used to replace single-point outliers, and linear interpolation is used to repair continuous outliers, resulting in cleaned parameter data. The cleaned parameter data is input into a Kalman filter, and state estimation is performed based on the process noise covariance matrix and the measurement noise covariance matrix to obtain the filtered parameter data. The filtered parameter data is subjected to closed-loop control system standardization processing to convert data with different dimensions into standard distribution data, thereby obtaining standardized control loop parameter characteristics. Based on the standardized control loop parameter characteristics, the statistics within a ten-minute sliding window are calculated, including mean, variance, maximum value, minimum value, kurtosis, and skewness, to obtain a multidimensional control loop feature vector; Based on the feature vector of the multidimensional control loop and the action time data recorded by the relay action mechanism sensor, the deviation between the controller output signal and the actual system response is calculated, and the relay control loop operation data is generated.

3. The method for monitoring high-pressure gas relays according to claim 1, characterized in that, The iterative learning control of the control parameter adjustment scheme involves designing a learning gain matrix based on the control characteristics of the high-pressure gas relay, analyzing and processing historical control data to obtain optimized control trajectory data, including: The historical operating data of the high-voltage gas relay is processed in batches, with each batch containing control data under similar operating conditions, to obtain a control data learning set. Based on the control data learning set, control error trajectory features are extracted, and the time-domain error curve shape features and frequency-domain error distribution features of each control cycle are calculated to obtain an error pattern feature library. A similarity analysis is performed on the error pattern feature library. The similarity between the current error trajectory and the historical error trajectory is calculated using the dynamic time warping algorithm. The historical error pattern with the highest similarity is identified, and the error pattern matching result is obtained. A learning gain matrix is ​​constructed based on the error pattern matching results. The learning gain value is dynamically adjusted according to the health status of the control loop. The higher the degree of abnormality, the smaller the learning gain, thus obtaining an adaptive learning gain matrix. Based on the adaptive learning gain matrix, the historical control input is iteratively calculated, and the control input for the next cycle is corrected by the error propagation correction formula. After ten iterations, the optimized control input sequence is obtained. The optimized control input sequence is fused with the current control strategy, the fusion weight is determined based on the control loop health status assessment result, the control trajectory is smoothed, and the optimized control trajectory data is generated.

4. A system for monitoring high-pressure gas relays, characterized in that, A method for implementing a high-pressure gas relay as described in any one of claims 1-3, wherein the system for monitoring the high-pressure gas relay comprises: The data acquisition module is used to acquire SF6 gas pressure, density, and temperature parameters of the high-pressure gas relay based on the closed-loop control system, and obtain the relay control loop operation data. The analysis module is used to calculate control deviations and perform residual vector analysis on the operating data of the relay control loop to obtain relay control performance indicators. The identification module is used to input the relay control performance indicators into the isolation forest algorithm for abnormal pattern identification, determine the degree of abnormality by calculating the sample isolation path length, and obtain the control loop health status assessment result. The execution module is used to perform adaptive control parameter adjustment based on the health status assessment result of the control loop, optimize PID parameters using an iterative learning control algorithm, and generate a relay control parameter set.

5. A device for monitoring high-pressure gas relays, characterized in that, The device includes a memory and a processor, the memory storing a computer program that can run on the processor, the processor executing the computer program to implement the method for monitoring a high-pressure gas relay as described in any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by a processor, it causes the processor to perform the method of monitoring a high-pressure gas relay as described in any one of claims 1 to 4.

Images