Ceramic insulator temperature self-adaptive control method and system based on twin simulation

Abstract

The application discloses a ceramic insulator temperature self-adaptive control method and system based on twin simulation, and relates to the technical field of process control. The method calculates the equivalent conductivity of the pollution layer by coupling the leakage current amplitude change rate and the environmental humidity data to construct a first pollution layer state matrix; adopts a long short-term memory network to extract the time sequence evolution law of the surface resistance distribution, generates a temperature gradient distribution map in combination with the local discharge pulse frequency to obtain a first temperature evolution trend; generates and corrects a target voltage value when the first temperature evolution trend exceeds a dynamic threshold value based on the electric field distortion coefficient; then, the humidity compensation correction is performed to obtain a second pollution layer state matrix, the time integral features of the heat accumulation density are extracted to obtain a second temperature evolution trend; and finally, the target voltage value is dynamically updated according to a flashover risk assessment index. The application realizes the advanced prediction of the pollution layer temperature evolution trend and the adaptive active adjustment of the voltage parameter, effectively suppresses the local overheating, reduces the flashover risk and prolongs the service life of the equipment.

Description

Technical Field

[0001] This application belongs to the field of process control technology, specifically a method and system for adaptive temperature control of ceramic insulators based on twin simulation. Background Technology

[0002] The safe and stable operation of the power system is the core foundation of national energy security. Ceramic insulators, as key components in transmission lines that provide electrical insulation and mechanical support, directly determine the reliability of the power grid. During long-term operation, industrial dust, salt spray, and other contaminants accumulate on the surface of insulators, forming a contaminant layer. When the ambient humidity changes, this contaminant layer absorbs moisture and becomes conductive, generating heat under the influence of leakage current. The increased temperature accelerates the electrochemical reaction and surface discharge of the contaminant layer, potentially leading to flashover accidents and widespread power outages.

[0003] Current methods for addressing insulator contamination primarily rely on regular cleaning or replacement, but this reactive approach is costly and poses significant safety hazards. While online monitoring-based early warning technologies have emerged in recent years, these methods can only monitor the static state parameters of the contaminant layer and cannot predict its thermodynamic evolution under different environmental conditions, making effective intervention difficult before the contaminant layer becomes uncontrollable. More importantly, the heating process of the contaminant layer is influenced by multiple factors, including humidity, contaminant composition, and leakage current, resulting in highly nonlinear and dynamic temperature changes.

[0004] This dynamic coupling characteristic presents a dual challenge: predicting the temperature evolution of the contaminant layer and controlling the process. On the one hand, the conductivity and heat generation of the contaminant layer can change drastically under different humidity conditions. Traditional fixed threshold control strategies cannot adapt to this dynamic change, often leading to control lag or over-intervention. On the other hand, although adjusting the voltage across the insulator can change the leakage current and thus affect the heating of the contaminant layer, there is a complex time-varying relationship between voltage regulation and temperature response. A lack of accurate understanding of this dynamic response process will make it difficult to achieve the expected control effect.

[0005] Therefore, how to predict future temperature change trends based on real-time environmental conditions during the dynamic evolution of contaminant heating, and adaptively adjust voltage parameters accordingly to achieve active process control of contaminant temperature, has become a key issue in ensuring the safe operation of insulators. Summary of the Invention

[0006] To address the above problems, this application provides a method and system for adaptive temperature control of ceramic insulators based on twin simulation. To achieve the above objectives, the technical solution adopted in this application is as follows: According to a first aspect of this application, a method for adaptive temperature control of ceramic insulators based on twin simulation is provided, comprising: S1. Construct a twin simulation model corresponding to the ceramic insulator on site, obtain the environmental humidity data and leakage current data of the insulator surface, and input the environmental humidity data and leakage current data into the twin simulation model; calculate the leakage current amplitude change rate, and calculate the equivalent conductivity of the pollution layer based on the coupling relationship between the leakage current amplitude change rate and humidity, and construct the first pollution layer state matrix. S2. For the surface resistance distribution characteristics of the first contamination layer state matrix, a long short-term memory network is used to extract the temporal evolution law. Combined with the partial discharge pulse frequency to generate a temperature gradient distribution map, the temperature rise rate is extracted to obtain the first temperature evolution trend. S3. When the first temperature evolution trend exceeds the dynamic threshold based on the electric field distortion coefficient, the temperature response under different voltage regulation strategies is deduced in the twin simulation model. An initial voltage value is generated based on the hot spot location and the thickness of the contaminant layer in the temperature gradient distribution map. The initial voltage value is then corrected according to the first temperature evolution trend to obtain the target voltage value. S4. Collect the first temperature response data after the target voltage value is applied, extract the slope of the temperature response curve, and when the slope of the temperature response curve exceeds the safe range, stitch together the ambient humidity data and correct it with humidity compensation to obtain the second dirt layer state matrix. S5. For the heat accumulation density distribution of the second contamination layer state matrix, the time integral features are extracted using a long short-term memory network, and the temperature change trajectory of the local temperature rise peak position is determined by combining the surface water film evaporation rate, so as to obtain the second temperature evolution trend. S6. Calculate the flashover risk assessment index based on the spatial correspondence between the second temperature evolution trend and the location of the local temperature rise peak, generate active control commands, update the target voltage value based on the voltage adjustment increment of the active control commands, and control the spatial uniformity of the heat accumulation density distribution on the insulator surface.

[0007] Preferably, the construction of a twin simulation model corresponding to the field ceramic insulator, obtaining environmental humidity data and leakage current data of the insulator surface, and inputting the environmental humidity data and leakage current data into the twin simulation model; calculating the leakage current amplitude change rate, and estimating the equivalent conductivity of the pollution layer based on the coupling relationship between the leakage current amplitude change rate and humidity, and constructing the first pollution layer state matrix includes: The first data segment is obtained by extracting the ambient humidity data and the leakage current data according to the preset time window, and the second data segment is obtained by smoothing the first data segment using the moving average filtering method. Calculate the leakage current amplitude change rate corresponding to the second data segment. If the leakage current amplitude change rate is greater than the preset change rate threshold, then the second data segment is determined to be the target sudden change data segment. For the target abrupt change data segment, the random forest algorithm is used to extract the coupling degree between the leakage current amplitude change rate and the environmental humidity data to obtain a first coupling degree sequence. The equivalent conductivity of the fouling layer is calculated based on the first coupling degree sequence. The surface resistance of the insulator surface is calculated based on the equivalent conductivity of the contamination layer. The equivalent conductivity and surface resistance of the contamination layer of each grid are normalized and then weighted and fused according to preset weights to obtain the matrix elements of the first state matrix. The first state matrix is ​​then dimensionality-reduced using the principal component analysis algorithm to obtain the first contamination layer state matrix.

[0008] Preferably, the step of extracting coupling degree using the random forest algorithm includes: analyzing the coupling relationship between the leakage current amplitude change rate and the environmental humidity data using a Pearson correlation coefficient model to obtain a coupling coefficient; inputting the coupling coefficient as a feature into the random forest algorithm to obtain the first coupling degree sequence; calculating the equivalent conductivity of the fouling layer based on the first coupling degree sequence; and if the equivalent conductivity of the fouling layer is less than a preset conductivity threshold, then performing linear compensation on the equivalent conductivity of the fouling layer to obtain a target conductivity.

[0009] Preferably, the step of extracting the temporal evolution law of the surface resistance distribution characteristics of the first contamination state matrix using a long short-term memory network, generating a temperature gradient distribution map by combining the partial discharge pulse frequency, and extracting the temperature rise rate to obtain the first temperature evolution trend includes: extracting surface resistance values ​​from the first contamination state matrix to generate a first resistance sequence, and processing the first resistance sequence using a long short-term memory network to obtain a resistance time evolution sequence. The number of partial discharge pulses is obtained based on the resistance time evolution sequence. If the number of partial discharge pulses is greater than a preset pulse threshold, the pulse frequency change characteristics are tensor-concatenated with the resistance time evolution sequence and mapped to generate the temperature gradient distribution map. The temperature rise rate of the core heating region corresponding to the estimated thickness of the contaminant layer is extracted from the temperature gradient distribution map. Using the temperature rise rate as the initial slope parameter, the first temperature evolution trend is obtained by iterative calculation using an exponential smoothing prediction model.

[0010] Preferably, when the first temperature evolution trend exceeds the dynamic threshold based on the electric field distortion coefficient, the temperature response under different voltage regulation strategies is deduced in the twin simulation model, an initial voltage value is generated based on the hot spot location and contaminant thickness in the temperature gradient distribution map, and the initial voltage value is corrected according to the first temperature evolution trend to obtain the target voltage value, including: processing the electric field distortion data using a support vector machine model to obtain the electric field distortion coefficient, and substituting the electric field distortion coefficient into the exponential decay model to calculate the dynamic threshold; When the first temperature evolution trend is greater than the dynamic threshold, the temperature gradient distribution map is processed by the image edge detection operator to obtain the hot spot location. The hot spot location and the dirt layer thickness are spatially superimposed to obtain the high-risk area and an initial voltage value is assigned. The first temperature evolution trend is converted into an equivalent voltage drop using a dimensionless conversion factor. The voltage regulation increment is obtained by multiplying the equivalent voltage drop by a safety margin factor. The target voltage value is obtained by subtracting the voltage regulation increment from the initial voltage value and then smoothing and filtering it.

[0011] Preferably, the process of collecting the first temperature response data after the target voltage value is applied, extracting the slope of the temperature response curve, and when the slope of the temperature response curve exceeds the safe range, splicing the environmental humidity data and correcting it with humidity compensation to obtain the second contamination state matrix includes: processing the first temperature response data with a linear regression algorithm to obtain the slope of the temperature response curve, obtaining the safe range pre-established based on the contamination drying rate parameter, and determining whether the slope of the temperature response curve is within the safe range. If the slope of the temperature response curve is not within the safe range, then extract the environmental humidity data corresponding to the first temperature response data, and concatenate the first temperature response data and the environmental humidity data in a column direction to obtain an initial concatenated dataset. A humidity compensation correction coefficient is introduced to correct the initial spliced ​​dataset, and a support vector machine algorithm is used to classify the corrected data to obtain the second contamination state matrix.

[0012] Preferably, the heat accumulation density distribution of the second contamination state matrix is ​​obtained by extracting time integral features using a long short-term memory network and combining the surface water film evaporation rate to determine the temperature change trajectory of the local temperature rise peak position, thereby obtaining the second temperature evolution trend. This includes: extracting the heat accumulation density distribution of each spatial node from the second contamination state matrix to obtain a first density distribution sequence, and extracting the time integral features of the first density distribution sequence using a long short-term memory network to obtain a first integral feature vector. The first integral feature vector and the surface water film evaporation rate are used to determine the location of the local temperature rise peak, and the temperature change trajectory corresponding to the local temperature rise peak location is extracted. Obtain the dynamic decay curve of conductivity. When the slope of the temperature change trajectory is greater than a preset slope threshold, extract the period of continuous temperature rise from the dynamic decay curve of conductivity, and fit the temperature change trajectory within the period to obtain the second temperature evolution trend.

[0013] Preferably, the step of calculating the flashover risk assessment index based on the spatial correspondence between the second temperature evolution trend and the local temperature rise peak position, generating an active control command, updating the target voltage value according to the voltage adjustment increment of the active control command, and controlling the spatial uniformity of the heat accumulation density distribution on the insulator surface includes: using a support vector machine algorithm to classify the second temperature evolution trend and the local temperature rise peak position to obtain a spatial correspondence degree; when the spatial correspondence degree is greater than a preset correspondence degree threshold, calculating the flashover risk assessment index corresponding to the spatial correspondence degree. The active control command, which includes voltage amplitude and application timing, is generated by encoding the command based on the flashover risk assessment index. Extract the voltage adjustment increment corresponding to the active control command, and use the product of the voltage adjustment increment and the dirt drying rate parameter as the voltage correction value to update the target voltage value.

[0014] Preferably, the method further includes: obtaining the product of the voltage regulation increment and the contaminant drying rate parameter; when the absolute value of the product is less than a preset product threshold, updating the target voltage value according to the absolute value of the product to obtain an updated target voltage value; using a random forest algorithm to distribute the updated target voltage value, determining the spatial uniformity of the heat accumulation density distribution on the insulator surface, and reducing the leakage current intensity on the insulator surface through the updated target voltage value.

[0015] According to a second aspect of this application, a temperature adaptive control system for ceramic insulators based on twin simulation is provided, comprising: The state perception module is configured to acquire ambient humidity data and leakage current data of the insulator surface, calculate the rate of change of leakage current amplitude and estimate the equivalent conductivity of the pollution layer, and construct the first pollution layer state matrix. The trend prediction module is configured to use a long short-term memory network to extract the temporal evolution law of surface resistance distribution, and combine it with the frequency of partial discharge pulses to generate a temperature gradient distribution map to obtain the first temperature evolution trend. The voltage correction module is configured to generate an initial voltage value based on the hot spot location and dirt layer thickness in the temperature gradient distribution map when the first temperature evolution trend exceeds the dynamic threshold based on the electric field distortion coefficient, and then correct the initial voltage value according to the first temperature evolution trend to obtain a target voltage value. The state correction module is configured to collect the first temperature response data after the target voltage value is applied, and when the slope of the temperature response curve exceeds the safe range, the second dirt layer state matrix is ​​obtained after humidity compensation correction. The trend update module is configured to use a long short-term memory network to extract the time integral features of the heat accumulation density and combine them with the surface water film evaporation rate to obtain the second temperature evolution trend. The uniformity control module is configured to generate active control commands based on the flashover risk assessment index and update the target voltage value according to the voltage adjustment increment to control the spatial uniformity of the surface heat accumulation density distribution.

[0016] The beneficial effects of this application include: By coupling the rate of change of leakage current amplitude with environmental humidity data to calculate the equivalent conductivity of the fouling layer, and by using a long short-term memory network to extract the temporal evolution law of surface resistance distribution, the advanced prediction of the temperature evolution trend of the fouling layer is realized, overcoming the limitation of existing technologies that can only monitor static parameters. By replacing the fixed threshold with a dynamic threshold based on the electric field distortion coefficient, the control strategy can adaptively follow the changes in the state of the fouling layer, avoiding control lag or over-intervention. By collecting temperature response data in a closed loop and introducing humidity compensation correction, a complete process control closed loop is formed, from state perception and trend prediction to voltage correction, which effectively suppresses local overheating, reduces the risk of flashover, and extends the service life of equipment. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for adaptive temperature control of ceramic insulators based on twin simulation, according to an embodiment of this application. Figure 2 This is a structural block diagram of a temperature adaptive control system for ceramic insulators based on twin simulation, according to an embodiment of this application. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solution, the present application will be described in detail below with reference to the embodiments. The description in this section is only exemplary and explanatory, and should not be used to limit the scope of protection of the present application in any way.

[0019] The twin simulation environment upon which this application is based refers to the establishment of a digital mirror model on the server side that corresponds one-to-one with the field insulator. This model is driven by data such as ambient humidity, leakage current, partial discharge and surface temperature uploaded in real time by field sensors, and synchronously reproduces the electro-thermal-humidity multi-physics field coupling evolution process of the pollution layer in virtual space.

[0020] The purpose of the twin simulation is to: predict the temperature response under different voltage regulation strategies on the virtual model in advance, thereby finding the target voltage value without interfering with the field equipment; and compare the residual between the measured values ​​on site and the predicted values ​​of the twin model. When the residual exceeds the allowable range, the model parameters are automatically corrected online to ensure the prediction accuracy.

[0021] Intermediate quantities such as the first contamination state matrix, temperature gradient distribution map, and temperature evolution trend in each step of this application are not only a representation of the field conditions but also state variables of the twin model. The two are kept consistent through data reinjection.

[0022] See Figure 1 This embodiment provides a method for adaptive temperature control of ceramic insulators based on twin simulation, including steps S1 to S6. The steps are described in detail below: Step S1: Construct a twin simulation model corresponding to the ceramic insulator on site, obtain the environmental humidity data and leakage current data of the insulator surface, and input the environmental humidity data and leakage current data into the twin simulation model; calculate the leakage current amplitude change rate, and calculate the equivalent conductivity of the pollution layer based on the coupling relationship between the leakage current amplitude change rate and humidity, and construct the first pollution layer state matrix; The system uses miniature weather sensors and high-frequency through-core current transformers deployed on transmission line towers to collect real-time environmental humidity and leakage current data on the insulator surface at a sampling frequency of once per second. For example, in the early morning, the humidity was collected to be 85.5%, and the synchronous leakage current amplitude was 12.3 mA. The sampling frequency is set to once per second, which is based on the fact that the time constant of the hygroscopic conductivity of the contaminant layer is usually on the order of seconds to minutes. Sampling once per second can fully capture the abrupt change in leakage current while avoiding the storage and computational burden caused by excessively high sampling rates.

[0023] The system sets a sliding time series window of 30 minutes, automatically identifying and extracting continuous periods where humidity shows a monotonically increasing trend to obtain the first data segment. For example, it extracts 30 minutes of data showing a continuous increase in humidity from 70.0% to 90.0%. The time window length is set to 30 minutes, based on the fact that the typical process of an insulator's contamination layer absorbing moisture and reaching conductivity saturation takes approximately 20 to 40 minutes. Taking the median of 30 minutes can cover a complete moisture evolution cycle without introducing interference from adjacent weather processes. Subsequently, a moving average filtering method is used to smooth the first data segment to obtain the second data segment. The moving average window uses 5 sampling points, which is based on achieving a balance between suppressing high-frequency noise from the sensor and preserving the true trend of current change.

[0024] The system uses a first-order difference algorithm to calculate the rate of change of leakage current amplitude in the leakage current sequence of the second data segment. The calculation method is the ratio of the difference in current values ​​between adjacent moments to the corresponding time interval. The formula for calculating the increment of current amplitude per unit time is as follows:

[0025] in, and They are time points and The leakage current amplitude. Taking the current increasing from 10.1 mA to 15.6 mA during this period as an example, the average rate of change is calculated to be 0.18 mA per minute. If the rate of change of the leakage current amplitude is greater than the preset rate of change threshold, the second data segment is determined to be the target abrupt change data segment; the rate of change threshold is set to 0.1 mA per minute, and its value is based on the statistics of historical normal operation data. Under normal operating conditions, three times the standard deviation of the rate of change is approximately 0.1 mA per minute. Exceeding this value indicates that the sludge layer has entered a rapid moisture absorption stage.

[0026] For the target abrupt change data segment, the system introduces a Pearson correlation coefficient model to analyze the coupling relationship between the leakage current amplitude change rate and the humidity rise sequence, and obtains the coupling coefficient. The calculation method is as follows:

[0027] in, This is a sample of the rate of change of leakage current amplitude. For the corresponding humidity sample, , The mean of each sample is denoted as . The calculated coupling coefficient is 0.85, indicating a high positive correlation between the two. The system uses the coupling coefficient as a feature input to the random forest algorithm to obtain the first coupling degree sequence. The random forest is tuned to 100 decision trees with a maximum depth of 10. The selection criteria are: 100 trees ensure convergence of the out-of-bag error without overfitting, and a maximum depth of 10 matches the dimension of the input features, avoiding excessive splitting of a single tree.

[0028] The system substitutes the first coupling degree sequence into a preset exponential conductivity estimation model to calculate the current equivalent conductivity of the contaminant layer to be 0.045 Siemens per meter. The coefficients of the exponential estimation model are obtained by least-squares fitting tuning of multiple sets of contaminant layer samples with known conductivity. If the equivalent conductivity of the contaminant layer is less than the preset conductivity threshold of 0.02 Siemens per meter, linear compensation is performed to obtain the target conductivity. The value of this conductivity threshold is based on the upper limit of the equivalent conductivity of a dry, clean ceramic surface measured in the laboratory. If it is lower than this value, it indicates that the contaminant layer has not yet effectively conducted electricity, and linear compensation according to the drying correction curve is required to eliminate measurement deviation.

[0029] The system calculates the surface resistance of the insulator surface based on the target conductivity, and fills a two-dimensional grid with the equivalent conductivity value and the estimated surface resistance values ​​(e.g., 500 kΩ) of four different shed regions of the insulator according to the geometric topology, constructing a 4x4 first state matrix. Subsequently, principal component analysis is used to reduce the dimensionality of the first state matrix, retaining the principal components with a cumulative variance contribution rate of 95%, resulting in the first contamination layer state matrix, whose diagonal elements are the core state values, such as 0.045 Siemens per meter. The cumulative variance contribution rate threshold is tuned to 95%, which is determined to retain the main characteristic information of the spatial distribution of the contamination layer while compressing redundant noise dimensions.

[0030] Specifically, the two-dimensional grid of the first state matrix is ​​defined according to the geometric topology of the insulator: the row direction is divided into four shed sections along the insulator axis from the high-voltage end to the grounding end, and the column direction is divided into four sectors along the insulator circumference at 0°, 90°, 180°, and 270°. Therefore, the matrix dimension is 4×4, with a total of 16 spatial grids. Matrix element x ij This represents the overall state quantity of the contaminant layer within the grid where the i-th umbrella skirt segment intersects with the j-th circumferential sector. It is determined by the equivalent conductivity σ of the contaminant layer in that grid. ij and surface resistance R ij x is obtained by normalized weighted fusion. ij The larger the value, the stronger the conductivity of the contaminant layer and the higher the risk of Joule heating and flashover. After construction, the 16 elements are flattened into a 16-dimensional vector in row-major order. This vector is then input into principal component analysis. The top k principal components with a cumulative variance contribution rate ≥ 95% are retained to obtain the k-dimensional first contaminant state matrix. Its elements are principal component scores, which are used to characterize the comprehensive features of the spatial distribution of the contaminant layer.

[0031] Step S2: For the surface resistance distribution characteristics of the first contamination state matrix, a long short-term memory network is used to extract the temporal evolution law, and a temperature gradient distribution map is generated by combining the partial discharge pulse frequency to extract the temperature rise rate and obtain the first temperature evolution trend. The system extracts surface resistance distribution features from the first contamination layer state matrix to generate a first resistance sequence. The resistance matrix sequence collected every minute over 120 consecutive minutes is then input into a Long Short-Term Memory (LSTM) network model. The LTM network is tuned to have three hidden layers, each with 128 neurons. The selection of these neurons is based on the following criteria: the evolution of contamination layer resistance exhibits a moderately complex temporal dependency; the three-layer, 128-neuron structure strikes a balance between fitting ability and training stability; and cross-validation shows that it has the lowest mean squared error in prediction. The network updates its weights using a backpropagation algorithm, extracting a dynamic vector representing the nonlinear decrease in resistance value over time—the resistance time evolution sequence.

[0032] The system synchronously receives partial discharge signals collected by a high-frequency current sensor and uses a fast Fourier transform algorithm to statistically analyze the frequency changes of pulse counts. If the number of partial discharge pulses exceeds a preset pulse threshold, the frequency change characteristics are tensor-concatenated with the aforementioned dynamic law vector and input into a multilayer perceptron for nonlinear mapping, generating a 64x64 temperature gradient distribution map. Each pixel in the map represents the expected heat generation power of the corresponding physical region. The pulse threshold is set to 100 times per second. The rationale for this setting is that during normal operation, the partial discharge frequency is generally below 50 times per second. When the frequency exceeds 100 times per second, it indicates that significant local arcing activity has occurred on the surface of the contaminant layer, and only then does the heat generation have analytical value. In the example, the discharge frequency was monitored to surge from 50 times per second to 300 times per second, meeting the triggering condition.

[0033] In one optional implementation, the multilayer perceptron adopts a three-layer fully connected structure. The input layer dimension is determined by tensor concatenation of the resistance time evolution sequence features and the partial discharge pulse frequency change features. For example, the 128-dimensional resistance time sequence features and 16-dimensional pulse frequency features output by the long short-term memory network are concatenated to form a 144-dimensional input vector. The number of neurons in the hidden layer is set to 256 and 512, respectively, both using the ReLU activation function. The number of neurons in the output layer is 4096. After normalizing each element to the 0 to 1 range using the Sigmoid activation function, it is reshaped into 64×64 pixels and mapped to the expected heat power of each pixel according to the preset heat power range. The training dataset is constructed in a twin simulation environment: different levels of pollution, humidity, voltage, and partial discharge intensity are set for ceramic insulators, and the corresponding resistance timing and pulse frequency are collected or simulated as input samples. The 64×64 heat generation power distribution obtained from finite element electro-thermal coupling simulation is used as the label. The mean square error is used as the training loss function, Adam is used as the optimizer, the learning rate is set to 0.001, the batch size is 32, and the iteration is until the validation set loss converges, so that the generation process of the temperature gradient distribution map can be reproduced.

[0034] The system uses an ultrasonic reflection time difference algorithm to calculate an estimated current contaminant thickness of 2 mm. It then automatically locates the core heating region matching this thickness in a temperature gradient distribution map, extracting the average temperature rise rate of this region as 3 degrees Celsius per minute. The system uses this temperature rise rate as the initial slope parameter and substitutes it into an exponential smoothing prediction model for iterative calculation. The calculation method involves weighting and recursively extrapolating future temperatures using a smoothing coefficient. The first temperature evolution trend is obtained by multiplying the smoothing coefficient by the current observation, plus the product of its complement and the previous predicted value, and outputting a continuous predicted curve of temperature change over time for the next 30 minutes. The smoothing coefficient of the exponential smoothing is set to 0.3. The reason for this value is that the larger the smoothing coefficient, the more sensitive it is to recent data. Setting it to 0.3 can achieve a trade-off between following the temperature rise trend and suppressing random fluctuations, avoiding excessive jitter in the predicted curve.

[0035] Step S3: When the first temperature evolution trend exceeds the dynamic threshold based on the electric field distortion coefficient, the temperature response under different voltage regulation strategies is deduced in the twin simulation model. An initial voltage value is generated based on the hot spot location and the thickness of the contaminant layer in the temperature gradient distribution map. The initial voltage value is then corrected according to the first temperature evolution trend to obtain the target voltage value. The system collects electric field data from the insulator surface and processes it using a support vector machine (SVM) model to obtain an electric field distortion coefficient of 2.5. The SVM employs a radial basis function kernel, with a penalty coefficient C tuned to 1.0 and a kernel parameter gamma tuned to 0.1. These values ​​are determined based on a set of parameters that minimizes classification error on the distortion coefficient label set through grid search combined with five-fold cross-validation. The system substitutes the electric field distortion coefficient into an exponential decay model to calculate a dynamic threshold of 3.2 degrees Celsius per minute. This exponential decay model characterizes the physical law that a larger distortion coefficient corresponds to a lower allowable safe temperature rise rate, and its decay constant is determined by inversion tuning from insulator flashover test data.

[0036] When the initial temperature evolution trend (e.g., an increase of 4.5 degrees Celsius per minute) exceeds the dynamic threshold of 3.2 degrees Celsius, an over-limit judgment is triggered. The system uses an image edge detection operator to process the temperature gradient distribution map to obtain hotspot locations, identifying a hotspot distribution location with an area of ​​15 square centimeters at the edge of the high-voltage end umbrella skirt. Simultaneously, the estimated value of the contaminant layer thickness in this area is calculated to be 0.8 millimeters using the microwave reflection method. The system inputs the hotspot area and contaminant layer thickness into a multiple linear regression model for spatial superposition, calculating an initial voltage value of 110 kV to maintain the insulation performance of this high-risk area. The regression coefficients of the multiple linear regression model are fitted and tuned using historical high-risk area voltage-insulation margin data.

[0037] To quantify the weakening effect of a sharp increase in temperature on insulation strength, a dimension conversion factor is introduced into the system. The first temperature evolution trend is converted into an equivalent voltage drop, and the calculation method is as follows:

[0038] in, The first temperature evolution trend value, dimension conversion coefficient. The value is set to 0.6, based on an empirically calculated ratio of the decrease in insulation strength per degree Celsius temperature rise under standard pollution levels. Substituting the example value of 4.5, the equivalent voltage drop is 2.7 kV. This equivalent voltage drop is then multiplied by the safety margin factor. The voltage regulation increment is obtained:

[0039] Among them, the safety margin coefficient The value is set to 1.2, based on a 20% adjustment margin reserved on the equivalent voltage reduction to accommodate model errors and operating condition disturbances. Multiplying 2.7 by 1.2 yields a voltage regulation increment of 3.24 kV. Finally, the target voltage value is obtained by subtracting the voltage regulation increment from the initial voltage value and applying a smoothing filter.

[0040] in, This is the initial voltage value. This is the voltage regulation increment. Subtracting 3.24 kV from 110 kV, the final output target voltage value is 106.76 kV, which is then sent to the intelligent voltage regulation terminal to perform a step-down operation to suppress the risk of thermal breakdown.

[0041] Step S4: Collect the first temperature response data after the target voltage value is applied, extract the slope of the temperature response curve, and when the slope of the temperature response curve exceeds the safe range, stitch together the ambient humidity data and correct it with humidity compensation to obtain the second dirt layer state matrix. After the system applies the target voltage to the insulator, it continuously collects the surface temperature for ten minutes at a frequency of once per minute using an infrared sensor array, generating initial temperature response data including 15.2, 16.5, 18.1, 20.3, 23.0, 26.1, 29.5, 33.2, 37.1, and 41.3 degrees Celsius. Subsequently, a linear regression algorithm (least squares method) is used to fit this time series, extracting the slope of the temperature response curve. The calculation method is as follows:

[0042] in, Sampling time, For the corresponding temperature, Represents all sampling times arithmetic mean Represents all temperature observations The arithmetic mean of the values ​​was used to calculate the slope of the temperature response curve, which was found to be 2.85 degrees Celsius per minute.

[0043] The system reads a safe range of 0.5 to 2.0 degrees Celsius per minute from the database, set based on historical contaminant drying rate parameters, and determines whether the slope of the temperature response curve falls within this range. The safe range is set by statistically analyzing the temperature response slopes of multiple groups of insulators with different pollution levels after safe voltage reduction, and taking the lower quartile and upper quartile of their normal fluctuation range as the lower limit (0.5) and upper limit (2.0) of the range. The criteria for these values ​​are: below 0.5 indicates excessive voltage reduction, and above 2.0 indicates excessively rapid heating of the contaminant layer, posing a risk of localized arcing. The current slope of 2.85 exceeds the upper limit, confirming that it is not within the safe range.

[0044] After the exception handling mechanism is triggered, the system synchronously retrieves 85% of the ambient humidity data collected within the same time period. Using a timestamp alignment algorithm, it concatenates the first temperature response data vector and the humidity data vector column-wise to obtain an initial concatenated dataset of ten rows and two columns. The system further introduces a humidity compensation correction coefficient. The method for correcting the spliced ​​dataset is as follows:

[0045] in, The value represents the ambient humidity, with a constant of 0.60 as the baseline humidity and 0.05 as the humidity sensitivity coefficient. The baseline humidity of 0.60 is determined based on the critical relative humidity at which the contaminant begins to show significant hygroscopic conductivity. The sensitivity coefficient of 0.05 is determined by linear fitting of measured data on the conductivity increment of the contaminant under different humidity levels. Substituting 85% humidity yields a correction coefficient of 1.0125. Matrix multiplication is used to uniformly multiply the temperature column elements in the initial spliced ​​dataset by 1.0125 for nonlinear correction. Then, a support vector machine algorithm is used to classify the corrected data, thereby generating a second contaminant state matrix containing corrected temperature and humidity characteristics.

[0046] Step S5: For the heat accumulation density distribution of the second contamination layer state matrix, the time integral features are extracted using a long short-term memory network, and the temperature change trajectory at the local temperature rise peak position is determined by combining the surface water film evaporation rate, thus obtaining the second temperature evolution trend. The system constructs a second contamination state matrix with a dimension of 50x50, extracts the heat accumulation density distribution of each spatial node to obtain the first density distribution sequence, and identifies that the initial density of the central region reaches 4500 joules per square meter.

[0047] In this step, the 50×50 dimension of the second contamination state matrix is ​​not a simple replacement of the first state matrix, but rather a newly subdivided grid oriented towards a finer distribution of heat accumulation density. The 4×4 dimension of the first state matrix is ​​a macroscopic contamination conductivity-resistance state grid, divided into four axial skirt sections and four circumferential sectors of the insulator, used for contamination state perception and temperature trend prediction from S1 to S3. The 50×50 dimension of the second contamination state matrix is ​​a fine temperature-heat grid, divided into 50 equal parts along the insulator axis and 50 equal parts circumferentially, with a total of 2500 spatial nodes, used for locating local temperature rise peaks and controlling spatial uniformity from S4 to S6. Specifically, the 4×4 grid of the first contamination state matrix is ​​used as the initial field and boundary conditions input into the twin simulation model. A high-resolution temperature field is obtained through electro-thermal-humidity coupling solution, and then discretized according to the aforementioned 50×50 rule. Each matrix element represents the heat accumulation density within the corresponding grid, and after humidity compensation correction, the second contamination state matrix is ​​formed. The two are kept consistent through spatial mapping and data backfeedback in the twin simulation model.

[0048] The system inputs matrix data from 120 consecutive sampling periods into a Long Short-Term Memory (LSTM) network. High-frequency environmental noise is filtered through a forget gate and an input gate. The system then extracts the integral feature of the accumulated heat density over time to obtain the first integral feature vector. This vector is calculated by integrating the density sequence over the time dimension.

[0049] in, Let be the cumulative heat density at time t. The integration duration was calculated to be 27,000 joules per square meter. The integration duration consisted of 120 sampling periods, and its value was determined based on the complete heat accumulation process of the covering contaminant layer from heating to the formation of a dry zone.

[0050] The system retrieves ambient humidity data and uses thermodynamic equations to calculate the current surface water film evaporation rate as 0.02 kg / s². It then performs a difference operation between the first integral eigenvector and the surface water film evaporation rate to deduce the local temperature rise peak location, locking it at matrix coordinates (25, 25), and generates a temperature trajectory at this point that non-linearly climbs from 25°C to 85°C. Simultaneously, the system analyzes the dynamic conductivity decay curve of this area, using a first-order derivative algorithm to calculate the curve slope in real time. When the slope is detected to be below -0.05 Siemens per meter for four consecutive cycles, the system automatically extracts the period from the 2nd to the 5th second as the core period of continuous temperature rise. The slope threshold of -0.05 Siemens per meter is set as the criterion for the expansion of the dry contamination zone, and its value is based on the inflection point of the conductivity decrease rate during the measured drying process.

[0051] It should be noted that the dynamic decay of conductivity represents the process of evaporation of moisture from the contaminant layer and expansion of the dry zone, while the continuous rise in temperature represents the accumulation of Joule heating and partial discharge heat. These two processes can occur simultaneously: localized drying causes a decrease in conductivity and an increase in resistance in the dry zone, leading to current transfer to adjacent wetted areas and inducing electric field distortion and arcing partial discharge at the edge of the dry zone, further generating additional heat. Simultaneously, the insulator ceramic body and contaminant layer possess thermal inertia, with the temperature response lagging behind the change in conductivity. Therefore, the local temperature can continue to rise even as conductivity decreases. When the conductivity decay slope is below -0.05 Siemens per meter for four consecutive cycles, it indicates that the dry zone has stabilized and expanded. The period during which the temperature continues to rise is then selected as the core period for heat accumulation, used for subsequent fitting of the second temperature evolution trend.

[0052] The system inputs the temperature change trajectory data within the core period into the least squares fitting model, outputting a parabolic equation with a quadratic coefficient of 0.15, thus obtaining the second temperature evolution trend characterizing the expansion process of the dry sludge zone. The fitting model is as follows:

[0053] in, , , These are the fitting coefficients, specifically the coefficients of the quadratic term in the example. For example, a value of 0.15 indicates an accelerating temperature increase. The coefficients in the quadratic fitting model... , , Instead of being pre-set fixed values, these values ​​are obtained by least-squares fitting of the temperature change trajectory within the captured core time period. Here, c is determined by the initial temperature value at the start of the core time period, b is determined by the temperature rise rate at the start of the core time period, and a is determined by the acceleration characteristics of the temperature trajectory. A positive value for a indicates an accelerating temperature rise. a=0.15 is merely illustrative and does not constitute a limitation on the scope of protection of this application. Under different operating conditions, a, b, and c vary with factors such as dirt layer thickness, humidity, leakage current, partial discharge intensity, and voltage regulation strategy, and are all updated in real time through least-squares fitting.

[0054] Step S6: Calculate the flashover risk assessment index based on the spatial correspondence between the second temperature evolution trend and the local temperature rise peak position, generate active control commands, update the target voltage value based on the voltage adjustment increment of the active control commands, and control the spatial uniformity of the heat accumulation density distribution on the insulator surface. The system uses an infrared thermal imaging matrix to obtain a second temperature evolution trend with a temperature rise slope of 5 degrees Celsius per minute at the third skirt edge of the high-voltage end of the insulator. The Pearson correlation coefficient algorithm is used to calculate the spatial correspondence between this trend and the location of the local temperature rise peak, yielding a spatial correspondence of 85%. When the spatial correspondence exceeds a preset correspondence threshold, the aggregation degree is substituted into an exponential smoothing model to calculate a flashover risk assessment index of 85. The correspondence threshold is set to 80%, based on the fact that when the heating trend and the peak location highly overlap (exceeding 80%), the probability of heat concentration causing flashover increases significantly.

[0055] In one optional implementation, the flashover risk assessment index is calculated as follows: Let the spatial correspondence of the support vector machine output be C(t), expressed as a percentage, ranging from 0 to 100%; a preset correspondence threshold Ct... th It is 80%. When C(t) is greater than C th At that time, C(t) is used as the current risk observation and substituted into the first-order exponential smoothing model: R(t) = α·C(t) + (1-α)·R(t-1) Where R(t) is the flashover risk assessment index for the current control period, R(t-1) is the flashover risk assessment index for the previous control period, and α is the smoothing coefficient, set to 0.7. Its value is determined by balancing the response speed to the current aggregation level with suppressing fluctuations in single measurements. Initially, R(0) = C(0). The risk index R(t) ranges from 0 to 100, and the preset risk threshold R0... th The value is 80, when R(t) is greater than R th Active control commands are triggered at certain times.

[0056] In the example, the spatial correspondence C(t) is 85%, and the risk index R(t-1) of the previous cycle is 85. Therefore, R(t) = 0.7 × 85 + 0.3 × 85 = 85, which exceeds the risk threshold of 80, triggering subsequent voltage regulation. This formula is applicable to different operating conditions; it can be calculated simply by inputting the current spatial correspondence and the risk index of the previous cycle.

[0057] The system determines that the flashover risk assessment index of 85 exceeds the risk threshold of 80, triggering the fuzzy logic controller to generate an active control command containing the voltage amplitude and application timing. This command sets the voltage amplitude to 90% of the rated value, and the application timing is to complete a step adjustment within the next 200 milliseconds. The risk threshold of 80 is determined based on the steep inflection point of the insulator flashover cumulative probability curve; exceeding this value requires immediate intervention. The control center analyzes the command, extracts the difference between the current operating voltage of 100 kV and the required amplitude of 90 kV, and obtains a voltage adjustment increment of -10 kV.

[0058] The system synchronously accesses environmental sensor data and calculates the dirt drying rate parameter under the current humidity using thermodynamic equations. The voltage adjustment increment is set at 60%. This parameter is multiplied to obtain the voltage correction value, which is then added to the reference voltage to obtain the updated target voltage value. The calculation method is as follows:

[0059] in, The current reference voltage, For voltage regulation increment, This is the desiccant drying rate parameter. The desiccant drying rate parameter is calculated in real-time from the current humidity using thermodynamic equations, and its physical meaning is the rate at which conductivity decreases due to moisture evaporation from the desiccant per unit time. Multiplying the increment of -10 kV by 60% yields a correction value of -6 kV, which is then added to the base voltage of 100 kV to update the target voltage value to 94 kV.

[0060] The system sends the updated target voltage value of 94 kV to the thyristor voltage regulation module. A random forest algorithm is used to distribute the updated target voltage value, determining the spatial uniformity of the heat accumulation density distribution on the insulator surface. This reduces the leakage current intensity on the insulator surface, narrowing the difference in Joule heating rates between the high-voltage and grounding ends to within 2 watts per square centimeter, thus achieving spatial uniformity control of the heat accumulation density distribution on the insulator surface. The above steps S1 to S6 constitute a complete closed-loop process control from state perception and trend prediction to voltage correction.

[0061] In an optional implementation, step S6 further includes: obtaining the product of the voltage adjustment increment and the contaminant drying rate parameter; updating the target voltage value only when the absolute value of the product is less than a preset product threshold, to avoid frequent ineffective actions caused by excessively small adjustment amplitudes. The product threshold is set to 8 kV, and its value is based on the minimum voltage correction amount corresponding to the perceptible change in insulator surface temperature caused by a single adjustment; adjustments below this value have no substantial impact on temperature distribution. The system uses a random forest algorithm to distribute the updated target voltage value among the various skirt regions of the insulator. The random forest uses the heat accumulation density, surface resistance, and humidity of each region as input features and outputs the voltage distribution weight of each region, thereby making the heat suppression effect caused by voltage reduction spatially more balanced, ultimately reducing the overall leakage current intensity.

[0062] In step S1, during the extraction of coupling degree using the random forest algorithm, as a refined implementation method, the system first uses the Pearson correlation coefficient model to analyze the coupling relationship between the leakage current amplitude change rate and the environmental humidity data to obtain the coupling coefficient. Then, the coupling coefficient, along with the filtered current and humidity features, is used as a feature vector and input into the random forest algorithm to obtain the first coupling degree sequence. Compared to training directly with the original sequence, calculating the Pearson coupling coefficient in advance provides the random forest with a normalized correlation prior, thereby improving the stability of coupling degree estimation when the sample is limited. If the equivalent conductivity of the fouling layer calculated accordingly is less than the conductivity threshold, linear compensation is performed according to the aforementioned drying correction curve to obtain the target conductivity.

[0063] See Figure 2 This application also provides a temperature adaptive control system for ceramic insulators based on twin simulation, including a state sensing module, a trend prediction module, a voltage correction module, a state correction module, a trend update module, and a uniformity control module. Each module is configured to execute steps S1 to S6 of the above method, wherein: The state perception module is configured to acquire ambient humidity data and leakage current data of the insulator surface, calculate the rate of change of leakage current amplitude and estimate the equivalent conductivity of the pollution layer, and construct the first pollution layer state matrix. The trend prediction module is configured to use a long short-term memory network to extract the temporal evolution law of surface resistance distribution, and combine it with the frequency of partial discharge pulses to generate a temperature gradient distribution map to obtain the first temperature evolution trend. The voltage correction module is configured to generate an initial voltage value based on the hot spot location and dirt layer thickness in the temperature gradient distribution map when the first temperature evolution trend exceeds the dynamic threshold based on the electric field distortion coefficient, and then correct the initial voltage value according to the first temperature evolution trend to obtain a target voltage value. The state correction module is configured to collect the first temperature response data after the target voltage value is applied, and when the slope of the temperature response curve exceeds the safe range, the second dirt layer state matrix is ​​obtained after humidity compensation correction. The trend update module is configured to use a long short-term memory network to extract the time integral features of the heat accumulation density and combine them with the surface water film evaporation rate to obtain the second temperature evolution trend. The uniformity control module is configured to generate active control commands based on the flashover risk assessment index and update the target voltage value according to the voltage adjustment increment to control the spatial uniformity of the surface heat accumulation density distribution.

[0064] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for adaptive temperature control of ceramic insulators based on twin simulation, characterized in that, include: A twin simulation model corresponding to the ceramic insulator in the field is constructed. The environmental humidity data and leakage current data of the insulator surface are obtained and input into the twin simulation model. The leakage current amplitude change rate is calculated. Based on the coupling relationship between the leakage current amplitude change rate and humidity, the equivalent conductivity of the pollution layer is calculated, and the first pollution layer state matrix is ​​constructed. The surface resistance distribution characteristics of the first contamination state matrix are analyzed by using a long short-term memory network to extract the temporal evolution law. Combined with the partial discharge pulse frequency to generate a temperature gradient distribution map, the temperature rise rate is extracted to obtain the first temperature evolution trend. When the first temperature evolution trend exceeds the dynamic threshold based on the electric field distortion coefficient, the temperature response under different voltage regulation strategies is deduced in the twin simulation model. An initial voltage value is generated based on the hot spot location and the thickness of the fouling layer in the temperature gradient distribution map. The initial voltage value is then corrected according to the first temperature evolution trend to obtain the target voltage value. The first temperature response data after the target voltage value is applied is collected, and the slope of the temperature response curve is extracted. When the slope of the temperature response curve exceeds the safe range, the ambient humidity data is spliced ​​and corrected by humidity compensation to obtain the second dirt layer state matrix. For the heat accumulation density distribution of the second contamination layer state matrix, the time integral features are extracted using a long short-term memory network, and the temperature change trajectory at the local temperature rise peak position is determined by combining the surface water film evaporation rate, thus obtaining the second temperature evolution trend. The flashover risk assessment index is calculated based on the spatial correspondence between the second temperature evolution trend and the location of the local temperature rise peak. Active control commands are generated, and the target voltage value is updated based on the voltage adjustment increment of the active control commands to control the spatial uniformity of the heat accumulation density distribution on the insulator surface.

2. The adaptive temperature control method for ceramic insulators based on twin simulation according to claim 1, characterized in that, The process involves constructing a twin simulation model corresponding to the ceramic insulator in the field, obtaining environmental humidity data and leakage current data on the surface of the insulator, and inputting the environmental humidity data and leakage current data into the twin simulation model. The leakage current amplitude change rate is calculated, and the equivalent conductivity of the fouling layer is estimated based on the coupling relationship between the leakage current amplitude change rate and humidity. The first fouling layer state matrix is ​​constructed as follows: The first data segment is obtained by extracting the ambient humidity data and the leakage current data according to the preset time window, and the second data segment is obtained by smoothing the first data segment using the moving average filtering method. Calculate the leakage current amplitude change rate corresponding to the second data segment. If the leakage current amplitude change rate is greater than the preset change rate threshold, then the second data segment is determined to be the target sudden change data segment. For the target abrupt change data segment, the random forest algorithm is used to extract the coupling degree between the leakage current amplitude change rate and the environmental humidity data to obtain a first coupling degree sequence. The equivalent conductivity of the fouling layer is calculated based on the first coupling degree sequence. The surface resistance of the insulator surface is calculated based on the equivalent conductivity of the contamination layer. The equivalent conductivity and surface resistance of the contamination layer of each grid are normalized and then weighted and fused according to preset weights to obtain the matrix elements of the first state matrix. The first state matrix is ​​then dimensionality-reduced using the principal component analysis algorithm to obtain the first contamination layer state matrix.

3. The adaptive temperature control method for ceramic insulators based on twin simulation according to claim 2, characterized in that, The extraction of coupling degree using the random forest algorithm includes: analyzing the coupling relationship between the leakage current amplitude change rate and the environmental humidity data using the Pearson correlation coefficient model to obtain the coupling coefficient; inputting the coupling coefficient as a feature into the random forest algorithm to obtain the first coupling degree sequence; calculating the equivalent conductivity of the fouling layer based on the first coupling degree sequence; if the equivalent conductivity of the fouling layer is less than a preset conductivity threshold, then linearly compensating the equivalent conductivity of the fouling layer to obtain the target conductivity.

4. The adaptive temperature control method for ceramic insulators based on twin simulation according to claim 1, characterized in that, The surface resistance distribution characteristics of the first contamination state matrix are extracted using a long short-term memory network to extract the temporal evolution law, and a temperature gradient distribution map is generated by combining the partial discharge pulse frequency. The temperature rise rate is then extracted to obtain the first temperature evolution trend, including: extracting surface resistance values ​​from the first contamination state matrix to generate a first resistance sequence, and processing the first resistance sequence using a long short-term memory network to obtain a resistance time evolution sequence. The number of partial discharge pulses is obtained based on the resistance time evolution sequence. If the number of partial discharge pulses is greater than a preset pulse threshold, the pulse frequency change characteristics are tensor-concatenated with the resistance time evolution sequence and mapped to generate the temperature gradient distribution map. The temperature rise rate of the core heating region corresponding to the estimated thickness of the contaminant layer is extracted from the temperature gradient distribution map. Using the temperature rise rate as the initial slope parameter, the first temperature evolution trend is obtained by iterative calculation using an exponential smoothing prediction model.

5. The adaptive temperature control method for ceramic insulators based on twin simulation according to claim 1, characterized in that, When the first temperature evolution trend exceeds the dynamic threshold based on the electric field distortion coefficient, the temperature response under different voltage regulation strategies is simulated in the twin simulation model. An initial voltage value is generated based on the hot spot location and the thickness of the contaminant layer in the temperature gradient distribution map. The initial voltage value is then corrected according to the first temperature evolution trend to obtain the target voltage value. This includes: processing the electric field distortion data using a support vector machine model to obtain the electric field distortion coefficient, and substituting the electric field distortion coefficient into an exponential decay model to calculate the dynamic threshold. When the first temperature evolution trend is greater than the dynamic threshold, the temperature gradient distribution map is processed by the image edge detection operator to obtain the hot spot location. The hot spot location and the dirt layer thickness are spatially superimposed to obtain the high-risk area and an initial voltage value is assigned. The first temperature evolution trend is converted into an equivalent voltage drop using a dimensionless conversion factor. The voltage regulation increment is obtained by multiplying the equivalent voltage drop by a safety margin factor. The target voltage value is obtained by subtracting the voltage regulation increment from the initial voltage value and then smoothing and filtering it.

6. The adaptive temperature control method for ceramic insulators based on twin simulation according to claim 1, characterized in that, The method involves collecting the first temperature response data after the target voltage value is applied, extracting the slope of the temperature response curve, and when the slope of the temperature response curve exceeds the safe range, splicing the ambient humidity data and correcting it with humidity compensation to obtain the second contamination state matrix. This includes: processing the first temperature response data using a linear regression algorithm to obtain the slope of the temperature response curve, obtaining the safe range pre-established based on the contamination drying rate parameter, and determining whether the slope of the temperature response curve is within the safe range. If the slope of the temperature response curve is not within the safe range, then extract the environmental humidity data corresponding to the first temperature response data, and concatenate the first temperature response data and the environmental humidity data in a column direction to obtain an initial concatenated dataset. A humidity compensation correction coefficient is introduced to correct the initial spliced ​​dataset, and a support vector machine algorithm is used to classify the corrected data to obtain the second contamination state matrix.

7. The adaptive temperature control method for ceramic insulators based on twin simulation according to claim 1, characterized in that, The heat accumulation density distribution of the second contamination state matrix is ​​obtained by extracting time integral features using a long short-term memory network and combining it with the surface water film evaporation rate to determine the temperature change trajectory of the local temperature rise peak position, thus obtaining the second temperature evolution trend. This includes: extracting the heat accumulation density distribution of each spatial node from the second contamination state matrix to obtain a first density distribution sequence, and extracting the time integral features of the first density distribution sequence using a long short-term memory network to obtain a first integral feature vector. The first integral feature vector and the surface water film evaporation rate are used to determine the location of the local temperature rise peak, and the temperature change trajectory corresponding to the local temperature rise peak location is extracted. Obtain the dynamic decay curve of conductivity. When the slope of the temperature change trajectory is greater than a preset slope threshold, extract the period of continuous temperature rise from the dynamic decay curve of conductivity, and fit the temperature change trajectory within the period to obtain the second temperature evolution trend.

8. The adaptive temperature control method for ceramic insulators based on twin simulation according to claim 1, characterized in that, The step of calculating the flashover risk assessment index based on the spatial correspondence between the second temperature evolution trend and the local temperature rise peak position, generating an active control command, updating the target voltage value according to the voltage adjustment increment of the active control command, and controlling the spatial uniformity of the heat accumulation density distribution on the insulator surface includes: using a support vector machine algorithm to classify the second temperature evolution trend and the local temperature rise peak position to obtain a spatial correspondence degree; when the spatial correspondence degree is greater than a preset correspondence degree threshold, calculating the flashover risk assessment index corresponding to the spatial correspondence degree. The active control command, which includes voltage amplitude and application timing, is generated by encoding the command based on the flashover risk assessment index. Extract the voltage adjustment increment corresponding to the active control command, and use the product of the voltage adjustment increment and the dirt drying rate parameter as the voltage correction value to update the target voltage value.

9. The adaptive temperature control method for ceramic insulators based on twin simulation according to claim 8, characterized in that, Also includes: The product of the voltage regulation increment and the contaminant drying rate parameter is obtained. When the absolute value of the product is less than a preset product threshold, the target voltage value is updated according to the absolute value of the product to obtain the updated target voltage value. The updated target voltage value is distributed using a random forest algorithm to determine the spatial uniformity of the heat accumulation density distribution on the insulator surface. The leakage current intensity on the insulator surface is reduced by using the updated target voltage value.

10. A temperature adaptive control system for ceramic insulators based on twin simulation, characterized in that, include: The state perception module is configured to acquire ambient humidity data and leakage current data of the insulator surface, calculate the rate of change of leakage current amplitude and estimate the equivalent conductivity of the pollution layer, and construct the first pollution layer state matrix. The trend prediction module is configured to use a long short-term memory network to extract the temporal evolution law of surface resistance distribution, and combine it with the frequency of partial discharge pulses to generate a temperature gradient distribution map to obtain the first temperature evolution trend. The voltage correction module is configured to generate an initial voltage value based on the hot spot location and dirt layer thickness in the temperature gradient distribution map when the first temperature evolution trend exceeds the dynamic threshold based on the electric field distortion coefficient, and then correct the initial voltage value according to the first temperature evolution trend to obtain a target voltage value. The state correction module is configured to collect the first temperature response data after the target voltage value is applied, and when the slope of the temperature response curve exceeds the safe range, the second dirt layer state matrix is ​​obtained after humidity compensation correction. The trend update module is configured to use a long short-term memory network to extract the time integral features of the heat accumulation density and combine them with the surface water film evaporation rate to obtain the second temperature evolution trend. The uniformity control module is configured to generate active control commands based on the flashover risk assessment index and update the target voltage value according to the voltage adjustment increment to control the spatial uniformity of the surface heat accumulation density distribution.

Images