Simulation model-based switch cabinet static contact temperature prediction method and system

Abstract

The invention discloses a simulation model-based switch cabinet static contact temperature prediction method and system. The method comprises the following steps of: establishing a high-voltage switch cabinet electromagnetic-thermal-fluid multi-physics field coupling simulation model and calibrating through a temperature rise test; simulating different working conditions to obtain a sample data set containing environment temperature, current-carrying capacity, tulip contact temperature and moving and static contact temperature; normalizing the sample data and dividing the sample data into a training set and a test set; constructing and training an LSTM neural network model and an SVM (Support Vector Machine) model; determining an optimal weight coefficient of the two models by adopting a least square method, and establishing an LSTM-SVM combined prediction model; and inputting the environment temperature, the current-carrying capacity and the tulip contact temperature which are acquired in real time into the combined model to obtain a moving and static contact temperature predicted value. The method does not need to damage the original structure of the switch cabinet, gives consideration to time sequence characteristics and nonlinear mapping capability, is high in prediction precision and high in real-time performance, can timely discover overheating hidden dangers, and guarantees the reliable operation of equipment.

Description

Technical Field

[0001] This invention relates to the field of condition monitoring technology for high-voltage electrical equipment, and in particular to a method and system for predicting the temperature of stationary contacts in switchgear based on a simulation model. Background Technology

[0002] High-voltage switchgear is a crucial piece of equipment in power systems. The moving and stationary contacts within the vacuum circuit breaker's arc-extinguishing chamber are prone to overheating during long-term operation due to contact resistance and arcing effects. Because the moving and stationary contacts are in a sealed vacuum environment, it is impossible to directly install temperature sensors for real-time monitoring, posing a significant challenge to equipment condition monitoring and fault early warning.

[0003] In high-voltage switchgear, the moving and stationary contacts in a vacuum circuit breaker are primarily used for circuit breaking. Under normal operating conditions, the stationary contact transmits current through the circuit, while the moving contact opens and closes with each switch operation. When the moving and stationary contacts separate, the vacuum environment effectively suppresses arcing, ensuring rapid current interruption. Because vacuum has a low degree of ionization, it significantly reduces the duration and temperature of the arc, thereby extending the switch's lifespan and improving equipment safety. Therefore, the moving and stationary contacts play a crucial role in vacuum switchgear, determining its breaking capacity and reliability.

[0004] Existing temperature monitoring methods mainly include infrared thermometry and fiber Bragg grating thermometry, but these methods all have certain limitations. Infrared thermometry cannot measure the temperature inside a sealed structure, and fiber Bragg grating thermometry requires the placement of optical fibers inside the equipment, which is difficult and costly to implement. Some methods estimate contact temperature by establishing thermal path models or finite element models, but their accuracy is limited and cannot meet the needs of practical engineering.

[0005] It should be noted that the method for predicting the static contact temperature of switchgear based on simulation model described in this invention is only applicable to the prediction of the dynamic and static contact temperature of the arc-extinguishing chamber of high-voltage switchgear vacuum circuit breaker under stable conditions. However, this method is no longer applicable under transient conditions such as circuit breaker opening and closing operations and circuit short-circuit faults.

[0006] Therefore, it is necessary to develop a method that can accurately predict the temperature of moving and stationary contacts in the vacuum interrupter without damaging the original structure of the switchgear. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for predicting the temperature of stationary contacts in switchgear based on a simulation model, which aims to solve the technical problem that the temperature of moving and stationary contacts in the sealed vacuum interrupter chamber cannot be directly measured in the prior art.

[0008] To address the aforementioned problems, according to one aspect of this application, an embodiment of the present invention provides a method for predicting the temperature of stationary contacts in switchgear based on a simulation model, comprising the following steps:

[0009] An electromagnetic-thermal-fluid multiphysics coupling simulation model of a high-voltage switchgear was established, and the simulation model was calibrated using temperature rise test data.

[0010] By simulating different operating conditions using a calibrated simulation model, a sample dataset containing ambient temperature, current carrying capacity, plum blossom contact temperature, and moving and stationary contact temperature was obtained.

[0011] The sample data is preprocessed by normalization and then divided into training and test sets according to a certain ratio;

[0012] LSTM neural network model and SVM support vector machine model are constructed respectively, and the training set is used to train the model;

[0013] The optimal weight coefficients of the LSTM and SVM models are determined by the least squares method, and an LSTM-SVM combined prediction model is established.

[0014] By collecting real-time ambient temperature, current carrying capacity, and plum blossom contact temperature data, and inputting them into a combined prediction model, the predicted values ​​of the moving and stationary contact temperatures are obtained.

[0015] In this embodiment, the establishment of the multiphysics coupling simulation model includes:

[0016] A simplified physical model was established based on the actual structure of the high-voltage switchgear.

[0017] Set material property parameters, including thermal conductivity, specific heat capacity, electrical conductivity, and magnetic permeability;

[0018] Perform electromagnetic field simulation calculations and temperature-fluid field simulation calculations;

[0019] Solve multiphysics coupling equations using the finite element method or the finite volume method.

[0020] In some implementations, the specific process of calibrating the simulation model is as follows:

[0021] Multiple temperature monitoring points are installed on the surface of the switch cabinet;

[0022] Conduct actual temperature rise tests and record temperature data at each monitoring point;

[0023] Simulation calculations were performed under the same operating conditions to obtain simulated temperature data for the corresponding monitoring points.

[0024] Adjust the boundary condition parameters of the simulation model to make the error between the simulation data and the experimental data less than 5%.

[0025] In some implementations, the different operating conditions include:

[0026] The ambient temperature range is from -25℃ to 50℃;

[0027] The current carrying capacity varies from 0 to 1.2 times the rated current;

[0028] The contact resistance of the plum blossom contact varies from 5 to 50 μΩ.

[0029] In some implementations, the sample data processing employs a max-min normalization method:

[0030] Xnorm = (X - Xmin) / (Xmax - Xmin)

[0031] Where Xnorm is the normalized data value after processing by the max-min normalization method; X is the original data, and Xmin and Xmax are the minimum and maximum values ​​of the feature parameter, respectively. The feature parameter refers to the input variable that is currently being normalized, and the input variables include ambient temperature, current carrying capacity, upper or lower clover contact temperature.

[0032] In some implementations, the construction parameters of the LSTM neural network model include:

[0033] Number of input layer nodes: 4, corresponding to ambient temperature, current carrying capacity, upper plum blossom contact temperature, and lower plum blossom contact temperature respectively.

[0034] There are two hidden layers, with 32 and 16 neurons in the two hidden layers, respectively;

[0035] The output layer has one node, which is used to directly output the predicted temperature values ​​of the moving and stationary contacts;

[0036] The activation function is the ReLU function;

[0037] The training iterations are 500-1000 times;

[0038] Alternatively, the model structure can be adjusted as follows: the number of input layer nodes can be expanded to 15, the number of hidden layers can be increased to 10, and the number of neurons can be adjusted accordingly to enhance the ability to extract temporal features.

[0039] In some implementations, the construction parameters of the SVM support vector machine model include:

[0040] Kernel function type: Radial basis function (RBF);

[0041] Penalty parameter C: 2 -2 Up to 2 9 The optimal value is determined within the range using cross-validation.

[0042] Kernel function parameter γ: 2 -10 Up to 2 9 The optimal value is determined through cross-validation within the specified range.

[0043] In some implementations, the optimal weighting coefficients are determined as follows:

[0044] Calculate the sum of squared prediction errors SSE_lstm of the LSTM model on the test set;

[0045] Calculate the sum of squared prediction errors of the SVM model on the test set, SSE_svm;

[0046] The optimal weighting coefficients are determined by minimizing the sum of squared errors of the two models using the least squares method.

[0047] The optimal weight coefficients satisfy the condition w_lstm+w_svm=1.

[0048] Where w_lstm represents the weights of the LSTM model, and w_svm represents the weights of the SVM model.

[0049] In some implementations, the real-time acquisition of ambient temperature, current carrying capacity, and plum blossom contact temperature data is achieved through the following sensors:

[0050] Ambient temperature sensor: Installed outside the switch cabinet to measure the ambient temperature;

[0051] Current sensor: Installed on the bus line to measure real-time current carrying capacity;

[0052] Temperature sensor: Installed on the surface of the plum blossom contact to measure the contact temperature.

[0053] This invention also provides a simulation model-based system for predicting the temperature of stationary contacts in switchgear, used to implement the method described above, including:

[0054] Multiphysics simulation module: used to build and calibrate the temperature field simulation model of the switchgear;

[0055] Data acquisition module: includes ambient temperature sensor, current sensor and temperature sensor;

[0056] Data processing module: used for sample data processing and feature extraction;

[0057] Model training module: used for training and optimizing LSTM and SVM models;

[0058] Prediction execution module: used for real-time temperature prediction and result display;

[0059] Early warning module: Issues an alarm when the predicted temperature exceeds the set threshold.

[0060] Compared with the prior art, the simulation model-based method for predicting the stationary contact temperature of switchgear of the present invention has at least the following beneficial effects:

[0061] It can accurately predict the temperature of moving and stationary contacts in the vacuum interrupter without damaging the original structure of the switchgear.

[0062] The accuracy of the model was improved by combining multiphysics coupling simulation with experimental calibration;

[0063] The LSTM-SVM combined prediction model takes into account both time series characteristics and nonlinear mapping capabilities, resulting in high prediction accuracy.

[0064] It has strong real-time capabilities, enabling timely detection of potential overheating risks and preventing equipment damage.

[0065] It should be noted that the method described in this invention is only applicable to temperature prediction of the moving and stationary contacts in the arc-extinguishing chamber of a vacuum circuit breaker in a high-voltage switchgear under stable operating conditions. This method is no longer applicable during transient processes such as circuit breaker opening and closing operations or short-circuit faults.

[0066] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0067] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is a flowchart of the method for predicting the temperature of stationary contacts in switchgear based on a simulation model, as described in this invention.

[0069] Figure 2 This is a flowchart illustrating the construction and calibration process of the multiphysics coupling simulation model of this invention.

[0070] Figure 3 This is a flowchart illustrating the construction process of the LSTM-SVM combined prediction model of this invention.

[0071] Figure 4 This is a block diagram of the switchgear stationary contact temperature prediction system based on a simulation model according to the present invention.

[0072] Figure 5 This is a simplified physical model diagram of the KYN28A-12 high-voltage switchgear in an embodiment of the present invention;

[0073] Figure 6 This is a comparison curve between the predicted values ​​of the LSTM-SVM combined model and the actual values ​​of the moving and stationary contact temperatures in an embodiment of the present invention. Detailed Implementation

[0074] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments, structures, features, and effects according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "an embodiment" or "an embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0075] In the description of this invention, it should be clearly stated that the terms "first," "second," etc., in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence; the terms "vertical," "lateral," "longitudinal," "front," "rear," "left," "right," "up," "down," "horizontal," etc., indicate orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, and are merely for the convenience of describing this invention, and do not mean that the device or element referred to must have a specific orientation or position, and therefore should not be construed as a limitation of this invention.

[0076] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0077] like Figures 1-6 As shown, this embodiment of the invention provides a method for predicting the temperature of stationary contacts in switchgear based on a simulation model, including the following steps:

[0078] An electromagnetic-thermal-fluid multiphysics coupling simulation model of a high-voltage switchgear was established, and the simulation model was calibrated using temperature rise test data.

[0079] By simulating different operating conditions using a calibrated simulation model, a sample dataset containing ambient temperature, current carrying capacity, plum blossom contact temperature, and moving and stationary contact temperature was obtained.

[0080] The sample data is preprocessed by normalization and then divided into training and test sets according to a certain ratio;

[0081] LSTM neural network model and SVM support vector machine model are constructed respectively, and the training set is used to train the model;

[0082] The optimal weight coefficients of the LSTM and SVM models are determined by the least squares method, and an LSTM-SVM combined prediction model is established.

[0083] By collecting real-time ambient temperature, current carrying capacity, and plum blossom contact temperature data, and inputting them into a combined prediction model, the predicted values ​​of the moving and stationary contact temperatures are obtained.

[0084] In this embodiment, firstly, based on the actual structure of the high-voltage switchgear, bolts are removed, screw holes are filled, unnecessary rounded corners are removed, and irregular components are transformed into regular shapes. Components such as surge arresters and grounding switches, which have no substantial impact on the temperature of critical parts, are omitted, while retaining key thermal characteristics to establish a simplified physical model. Then, referring to material property parameters (such as the thermal conductivity of copper 386W / (m×K) and the specific heat capacity of epoxy resin 1200J / (kg×K)), the thermal conductivity, specific heat capacity, electrical conductivity, and magnetic permeability of each component are set. Electromagnetic field and temperature-fluid field simulations are performed using the finite element method or finite volume method to construct an electromagnetic-thermal-fluid multiphysics coupled simulation model. Next, temperature monitoring points are set at locations such as the busbar connection points on the switchgear surface, the circuit breaker housing surface, and cable joints. Temperature rise tests are conducted to record the temperature at each monitoring point. The simulation model is run under the same operating conditions (such as ambient temperature 25℃ and current carrying capacity 1250A), and the boundary conditions are adjusted to ensure that the error between the simulation and experimental data is less than 5%, thus completing the simulation model calibration. Then, a calibration model was used to simulate operating conditions with ambient temperatures ranging from -25℃ to 50℃, current carrying capacity from 0 to 1.2 times the rated current, and contact resistance of the stud contact from 5 to 50 μΩ. 125 sets of sample data were obtained, including ambient temperature, current carrying capacity, stud contact temperature, and moving / stationary contact temperature. The samples were preprocessed using the formula Xnorm=(X-Xmin) / (Xmax-Xmin) to normalize to the range [0,1]. They were then divided into a training set (100 sets) and a test set (Table 1 shows a comparison of the actual temperature values, LSTM predicted values, SVM predicted values, and LSTM-SVM predicted values ​​for the 25 test set samples) at an 8:2 ratio. An LSTM model (4 nodes in the input layer, 32 and 16 neurons in each of the 2 hidden layers, 1 node in the output layer, ReLU activation function, 500-1000 iterations) and an SVM model (RBF kernel function, determined through cross-validation, C=4.0, γ=0.8) were constructed and trained using the training set. The sum of squared errors of the two models on the test set (SSE_lstm=0.5999, SSE_svm=0.6636) was calculated. With the goal of minimizing the error, and considering the constraint w_lstm+w_svm=1, the optimal weights w_lstm=0.3652 and w_svm=0.6348 were obtained using the least squares method, and an LSTM-SVM combined model was established. Finally, data was collected in real time from external ambient temperature sensors, bus current sensors, and plum blossom contact surface temperature sensors, and input into the combined model to obtain predicted values ​​for the moving and stationary contact temperatures. This process does not require damaging the switchgear structure. Multiphysics simulation combined with experimental calibration improves model accuracy. The combined model balances temporal and nonlinear capabilities, achieving prediction accuracy of MAE=0.349 and RMSE=0.4456, and can promptly detect potential overheating risks.

[0085] The LSTM-SVM combined prediction model achieves prediction accuracy of MAE ≤ 0.35℃ and RMSE ≤ 0.45℃ on the test set.

[0086] By quantifying the prediction accuracy requirements of the combined model, the technical effects of the model performance were clarified, enhancing the practicality and verifiability of the invention.

[0087] Table 1. Comparison of actual temperature values, LSTM predicted values, SVM predicted values, and LSTM-SVM predicted values ​​for 25 test set samples.

[0088]

[0089] In some implementations, the establishment of the multiphysics coupling simulation model includes:

[0090] A simplified physical model was established based on the actual structure of the high-voltage switchgear.

[0091] Set material property parameters, including thermal conductivity, specific heat capacity, electrical conductivity, and magnetic permeability;

[0092] Perform electromagnetic field simulation calculations and temperature-fluid field simulation calculations;

[0093] Solve multiphysics coupling equations using the finite element method or the finite volume method.

[0094] In this embodiment, referring to the actual structural dimensions of the high-voltage switchgear, and following the principle of "retaining key thermal characteristics and simplifying unnecessary structures," bolts were removed and bolt holes were filled. Irregular components were transformed into regular shapes, and components such as surge arresters and grounding switches, which have no substantial impact on the temperature of key parts, were omitted. A simplified physical model was established to reduce the amount of subsequent simulation calculations while ensuring that the core temperature field characteristics were not lost. Next, based on the materials of each component of the switchgear, corresponding material property parameters were set: for example, the thermal conductivity of copper for conductive components is 386 W / (m×K), and the electrical conductivity is 5.8E+7S / m; the thermal conductivity of epoxy resin for insulating components is 0.9 W / (m×K), and the specific heat capacity is 1200 J / (kg×K); the magnetic permeability of galvanized steel sheet for the outer shell is 2500, and the density is 8030 kg / m³, ensuring that the simulation data closely matches the actual material properties. Subsequently, electromagnetic field simulation calculations were conducted to analyze the Joule heating and electromagnetic induction effects generated when current flows through the busbars and contacts. Simultaneously, temperature and fluid field simulation calculations were performed to simulate heat conduction, air convection, and radiative heat dissipation between components, fully presenting the temperature generation and transfer mechanism within the cabinet. Finally, the finite element method or finite volume method was used to solve the electromagnetic-thermal-fluid multiphysics coupling equations. Numerical calculations quantified the interactions between various physical fields (such as the influence of electromagnetic heat on temperature field distribution, and temperature changes altering material conductivity and thus affecting the electromagnetic field), ultimately completing the construction of the multiphysics coupling simulation model. This model can accurately reproduce the temperature field distribution of the switchgear, providing reliable basic simulation data for subsequent temperature prediction.

[0095] The simplified physical model omits surge arresters and grounding switches, fills in screw holes and removes fillets to improve computational efficiency while retaining key thermal properties.

[0096] By simplifying the specific operations of the model, the invention highlights the technical means to improve simulation efficiency while ensuring accuracy, demonstrating the practicality and innovation of the invention.

[0097] In some implementations, the specific process of calibrating the simulation model is as follows:

[0098] Multiple temperature monitoring points are installed on the surface of the switch cabinet;

[0099] Conduct actual temperature rise tests and record temperature data at each monitoring point;

[0100] Simulation calculations were performed under the same operating conditions to obtain simulated temperature data for the corresponding monitoring points.

[0101] Adjust the boundary condition parameters of the simulation model to make the error between the simulation data and the experimental data less than 5%.

[0102] In this embodiment, key monitoring locations are first selected on the switchgear surface: busbar connections (where current is concentrated and easily generates heat), the circuit breaker housing surface (close to the arc-extinguishing chamber and indirectly reflecting internal temperature), and cable joints (load current transmission nodes), etc., are chosen to set up multiple temperature monitoring points to ensure that the monitoring data can comprehensively reflect the temperature change trend inside the cabinet. Then, a temperature rise test platform is built, and the switchgear is placed in a standard test environment with a rated current (e.g., 1250A) applied. Temperature data at each monitoring point is recorded from equipment startup to temperature stabilization, collected every 5 minutes, forming a complete test temperature curve. Next, under the exact same operating conditions as the test (ambient temperature, current carrying capacity, and initial state are identical), the constructed multiphysics coupling simulation model is run to calculate the simulated temperature data at each monitoring point and generate a simulated temperature curve. Then, the test and simulated temperature curves for the same monitoring point are compared, the deviation values ​​at each time point are calculated, and the causes of the deviation are analyzed. Finally, based on the deviation analysis results, the boundary condition parameters of the simulation model were adjusted (such as optimizing the air convection heat dissipation coefficient inside the cabinet and correcting the contact resistance value of the stud contactor). The simulation and comparison operations were iterated repeatedly until the error between the simulation and experimental data at all monitoring points was less than 5%, thus completing the simulation model calibration. The calibrated model can more realistically reflect the actual temperature field of the switchgear, providing a high-precision foundation for subsequent sample data generation.

[0103] The temperature monitoring points include at least the busbar connection points, the surface of the circuit breaker housing, and the cable joints.

[0104] The specific locations of the monitoring points were clearly defined, ensuring the representativeness and comprehensiveness of the calibration data and improving the reliability and practicality of the model.

[0105] In some implementations, the different operating conditions include:

[0106] The ambient temperature range is from -25℃ to 50℃;

[0107] The current carrying capacity varies from 0 to 1.2 times the rated current;

[0108] The contact resistance of the plum blossom contact varies from 5 to 50 μΩ.

[0109] In this embodiment, based on the calibrated multiphysics coupling simulation model, the ambient temperature range is first set to -25℃ to 50℃. This range covers the low temperatures of winter in northern my country, the high temperatures of summer in southern China, and the ambient temperature indoors, simulating the impact of different regional and seasonal climates on the switchgear temperature. Secondly, the current carrying capacity range is set to 0 to 1.2 times the rated current. 0 current carrying capacity corresponds to the equipment's no-load standby state, while 1.2 times the rated current corresponds to the equipment's short-term overload operation scenario, consistent with the current changes caused by actual power grid load fluctuations. Furthermore, the contact resistance of the stud contact is set to 5 to 50 μΩ. Considering that oxidation and wear of the contacts after long-term use will lead to increased contact resistance, this range can simulate the impact of different aging degrees of the contacts on heat generation. Subsequently, an orthogonal experimental design method is used to combine the above three types of parameters (such as ambient temperature -10℃ + current carrying capacity 1000A + contact resistance 10μΩ, ambient temperature 35℃ + current carrying capacity 1500A + contact resistance 30μΩ, etc.) to ensure that the operating condition combinations cover the main operating scenarios. For each operating condition combination, the simulation model was run until the temperature stabilized. Data on ambient temperature, current carrying capacity, upper and lower pendant contact temperatures, and moving and stationary contact temperatures were collected, ultimately forming 125 representative sample datasets. This rich dataset comprehensively covers the actual operating conditions of the equipment, providing ample data support for subsequent predictive model training and improving the model's adaptability to different operating conditions.

[0110] The operating condition combinations employ orthogonal experimental design to cover the main operating scenarios.

[0111] By introducing orthogonal experimental design, a variety of working conditions can be scientifically covered, which improves the representativeness of the sample data and the generalization ability of the model.

[0112] In some implementations, the sample data processing employs a max-min normalization method:

[0113] Xnorm = (X - Xmin) / (Xmax - Xmin);

[0114] Where Xnorm is the normalized data value after processing by the max-min normalization method; X is the original data, and Xmin and Xmax are the minimum and maximum values ​​of the feature parameter, respectively. The feature parameter refers to the input variable that is currently being normalized, and the input variables include ambient temperature, current carrying capacity, upper or lower clover contact temperature.

[0115] In this embodiment, 125 sets of sample data generated by simulation are first collected. The data includes characteristic parameters such as ambient temperature (-25℃ to 50℃), current carrying capacity (0 to 1500A), contact temperature (10℃ to 110℃), and moving and stationary contact temperature (15℃ to 120℃). The dimensions and numerical ranges of each parameter vary greatly, which can easily affect the model training efficiency and accuracy. Then, for each type of characteristic parameter, all sample data are traversed to determine the minimum value Xmin and the maximum value Xmax of the parameter. Next, the maximum-min normalization formula Xnorm=(X-Xmin) / (Xmax-Xmin) is used to perform a linear transformation on each original data X. For example, for an ambient temperature of X=25℃, substituting it into the formula yields Xnorm=(25-(-25)) / (50-(-25))≈0.67; for a current carrying capacity of X=750A, Xnorm=(750-0) / (1500-0)=0.5. This transformation maps all feature parameters to the range [0,1]. Finally, the normalized dataset is randomly shuffled and divided into a training set (100 sets, used for model parameter learning) and a test set (25 sets, used for model performance validation) in an 8:2 ratio. Normalization eliminates the interference caused by differences in parameter dimensions, enabling the model to learn the influence of each feature on the temperature of the moving and stationary contacts in a balanced way. At the same time, it narrows the data range, accelerates the model training convergence speed, and improves training stability.

[0116] After normalization, the data is randomly shuffled and divided into training and test sets in an 8:2 ratio.

[0117] The data partitioning method was clarified, which avoids training bias caused by data order and improves the scientific nature and repeatability of model training.

[0118] In some implementations, the construction parameters of the LSTM neural network model include:

[0119] The input layer has four nodes, corresponding to ambient temperature, current carrying capacity, upper plum blossom contact temperature, and lower plum blossom contact temperature, respectively.

[0120] There are two hidden layers, with 32 and 16 neurons in the two hidden layers, respectively;

[0121] The output layer has one node, which is used to directly output the predicted temperature values ​​of the moving and stationary contacts;

[0122] The activation function is the ReLU function;

[0123] The training iterations are 500-1000 times;

[0124] Alternatively, the model structure can be adjusted as follows: the number of input layer nodes can be expanded to 15, the number of hidden layers can be increased to 10, and the number of neurons can be adjusted accordingly to enhance the ability to extract temporal features.

[0125] In this embodiment, the LSTM model is first configured with four input layer nodes, corresponding to four key input features: ambient temperature, current carrying capacity, upper cloverleaf contact temperature, and lower cloverleaf contact temperature. This ensures the model can receive the core variables affecting the temperatures of the moving and stationary contacts. Next, a hidden layer structure is set: two hidden layers are included. The first hidden layer has 32 neurons and is responsible for initially extracting temporal features from the data (such as the temperature trend over time). The second hidden layer has 16 neurons and further filters and integrates the initially extracted features, reducing redundant information and improving feature representation. The output layer has one node, directly outputting the predicted temperatures of the moving and stationary contacts, making the model's output target clear. The ReLU activation function is chosen, as it effectively solves the gradient vanishing problem that occurs with the traditional Sigmoid function in deep networks, ensuring stable parameter updates during 500-1000 training iterations and avoiding training stagnation. The training process employs batch gradient descent, updating parameters with 20 sets of training samples each time, and iterating 800 times (balancing training accuracy and efficiency; the model error stabilizes after 800 iterations). Furthermore, to prevent overfitting, a Dropout layer is added to the hidden layers with a dropout probability of 0.2, randomly disabling some neuron connections to improve the model's generalization ability. These reasonable parameter settings enable the LSTM model to accurately capture the temporal correlations of temperature data, providing reliable temporal prediction capabilities for the combined prediction model.

[0126] A Dropout layer was added to the hidden layer with a dropout probability of 0.2 to prevent overfitting.

[0127] By introducing the Dropout mechanism, the generalization ability of the model can be improved, overfitting can be avoided, and the stability of the model in practical applications can be enhanced.

[0128] In another embodiment, the number of input layer nodes of the LSTM model can be expanded to 15, the number of hidden layers can be increased to 10, and the number of neurons can be adjusted accordingly to further enhance the model's ability to extract complex temporal features and adapt to a wider range of working conditions.

[0129] In some implementations, the construction parameters of the SVM support vector machine model include:

[0130] Kernel function type: Radial basis function (RBF);

[0131] Penalty parameter C: 2 -2 Up to 2 9 The optimal value is determined within the range using cross-validation.

[0132] Kernel function parameter γ: 2 -10 Up to 2 9 The optimal value is determined through cross-validation within the specified range.

[0133] In this embodiment, the kernel function type of the SVM model is first determined to be a radial basis function (RBF). The RBF kernel function has good nonlinear mapping capabilities, enabling it to map sample data from a low-dimensional input space (4-dimensional features such as ambient temperature and current carrying capacity) to a higher-dimensional feature space, thereby effectively handling the complex nonlinear relationships between various input features and the temperatures of moving and stationary contacts (such as the nonlinear correlation between increased current carrying capacity and increased temperature). Secondly, the parameter search range is set: the penalty parameter C has a value range of 2. -2 Up to 2 9 C is used to balance the model's fit to the training set and its generalization ability. A C value that is too large will lead to overfitting the training data (large prediction errors on new samples), while a C value that is too small will lead to underfitting (inability to fully learn data patterns). The kernel function parameter γ ranges from 2. -10 Up to 2 9 γ determines the radial range of the RBF kernel function, affecting the model's ability to capture local features of samples. Next, a 5-fold cross-validation method is used to select the optimal parameters: the 100 training sets are divided into 5 subsets, with 4 subsets used to train the model and 1 subset used to validate its performance. All combinations of C and γ are iterated, and the mean absolute error (MAE) of the model under each combination is calculated. Finally, the parameter combination with the smallest MAE is selected as the optimal parameters. Validation shows that when C=4.0 and γ=0.8, the model has the smallest MAE on the validation set, thus completing the SVM model parameter settings. The optimal parameters achieve a balance between fitting accuracy and generalization ability in the SVM model, enabling accurate fitting of nonlinear features and providing reliable nonlinear prediction capabilities for the combined prediction model.

[0134] The cross-validation method used is 5-fold cross-validation, with mean absolute error (MAE) as the evaluation metric.

[0135] The specific methods and evaluation metrics for cross-validation were clarified, which improved the scientific rigor and reliability of model parameter optimization.

[0136] In some implementations, the optimal weighting coefficients are determined as follows:

[0137] Calculate the sum of squared prediction errors SSE_lstm of the LSTM model on the test set;

[0138] Calculate the sum of squared prediction errors of the SVM model on the test set, SSE_svm;

[0139] The optimal weighting coefficients are determined by minimizing the sum of squared errors of the two models using the least squares method.

[0140] The optimal weight coefficients satisfy the condition w_lstm+w_svm=1.

[0141] Where w_lstm represents the weights of the LSTM model, and w_svm represents the weights of the SVM model.

[0142] In this embodiment, 25 sets of test data are first input into the trained LSTM and SVM models respectively to obtain two sets of predicted temperatures for the moving and stationary contacts: the LSTM model outputs prediction set 1, and the SVM model outputs prediction set 2. Then, the sum of squared prediction errors (SSE) of the two models is calculated: the predicted value of each sample in prediction set 1 is subtracted from the corresponding actual moving and stationary contact temperature value to obtain the error vector E1, and the sum of squares of each element in E1 is calculated as SSE_lstm (SSE_lstm = 0.5999); similarly, the error vector E2 is obtained by subtracting the actual value from prediction set 2, and SSE_svm = 0.6636 is calculated. Next, a weight optimization objective function is established: Let the weights of the LSTM model be w_lstm and the weights of the SVM model be w_svm. The objective is to minimize the total sum of squared errors of the combined model (w_lstm²×SSE_lstm+w_svm²×SSE_svm+2×w_lstm×w_svm×Cov(E1,E2), where Cov(E1,E2) is the covariance of the two error vectors). Simultaneously, the constraint w_lstm+w_svm=1 is introduced to ensure reasonable weight allocation. Finally, the least squares method is used to solve this constrained optimization problem: substituting the constraints into the objective function transforms it into a single-variable optimization problem. By calculating the extreme values ​​through differentiation, the weight coefficients w_lstm=0.3652 and w_svm=0.6348 that minimize the total error are finally obtained. This method focuses on minimizing error and determines the weights through rigorous mathematical derivation, enabling the combined model to fully integrate the temporal advantages of LSTM and the nonlinear advantages of SVM, thereby significantly reducing prediction errors.

[0143] Furthermore, the covariance term of the errors of the two models is considered in the objective function to more accurately reflect the correlation between the models.

[0144] Introducing a covariance term more scientifically reflects the correlation between model errors, improving the accuracy of weight allocation and the prediction accuracy of the combined model.

[0145] In some implementations, the real-time acquisition of ambient temperature, current carrying capacity, and plum blossom contact temperature data is achieved through the following sensors:

[0146] Ambient temperature sensor: Installed outside the switch cabinet to measure the ambient temperature;

[0147] Current sensor: Installed on the bus line to measure real-time current carrying capacity;

[0148] Temperature sensor: Installed on the surface of the plum blossom contact to measure the contact temperature.

[0149] In this embodiment, an ambient temperature sensor with an accuracy of ±0.5℃ and an operating temperature range of -40℃ to 85℃ is first selected. This sensor is mounted on the external side wall of the high-voltage switchgear using an insulating bracket, away from the cooling fan on the top of the cabinet and heat-generating components inside the cabinet (such as busbars and circuit breakers). This avoids direct heat radiation or airflow affecting the sensor, ensuring that the collected ambient temperature data accurately reflects the external ambient temperature of the switchgear. Secondly, a through-hole current sensor with a range of 0 to 2000A and an accuracy of 0.2 is selected. A wire-free installation method is used, where the sensor's ring probe is fitted onto the switchgear busbar without cutting the busbar or damaging the original circuit structure. The sensor collects the busbar current in real time through electromagnetic induction, and after signal conversion, obtains the real-time current-carrying capacity data of the switchgear, with a transmission error of less than 0.5%. Furthermore, a surface-mount temperature sensor with a diameter of less than 5mm and a response time of less than 1s was selected. It was adhered to the outer surface of the L-shaped contact using thermally conductive adhesive resistant to high temperatures (above 200℃). The sensor probe was tightly fitted to the contact surface, and the sensor leads were wrapped with an insulating sleeve to prevent short circuits with other conductive components. This sensor can acquire the surface temperature of the L-shaped contact in real time with a measurement accuracy of ±0.3℃. Finally, each sensor transmits the acquired ambient temperature, current carrying capacity, and L-shaped contact temperature data to the system's data processing module in real time via a shielded RS485 communication cable. The communication rate was set to 9600bps, and the transmission cycle was 10s. The shielding layer was grounded to reduce electromagnetic interference and ensure stable and real-time data transmission. The sensor selection is suitable for high-voltage environments, the installation location is reasonable, and the data acquisition is accurate and real-time, providing high-quality real-time input data for the combined model.

[0150] The temperature sensor is a patch type, with a diameter of less than 5mm, a response time of less than 1s, and is attached to the contact surface with thermally conductive adhesive.

[0151] The selection and installation methods of sensors were clarified to ensure the accuracy, real-time performance, and engineering feasibility of data acquisition, thereby improving the practicality of the system.

[0152] This invention also provides a simulation model-based system for predicting the temperature of stationary contacts in switchgear, used to implement the method described above, including:

[0153] Multiphysics simulation module: used to build and calibrate the temperature field simulation model of the switchgear;

[0154] Data acquisition module: includes ambient temperature sensor, current sensor and temperature sensor;

[0155] Data processing module: used for sample data processing and feature extraction;

[0156] Model training module: used for training and optimizing LSTM and SVM models;

[0157] Prediction execution module: used for real-time temperature prediction and result display;

[0158] Early warning module: Issues an alarm when the predicted temperature exceeds the set threshold.

[0159] In this embodiment, the multiphysics simulation module first establishes a simplified model based on the actual structure of the switchgear, sets material properties, and performs electromagnetic-thermal-fluid coupling simulation. It then adjusts boundary conditions using temperature rise tests, completes simulation model calibration, and outputs a high-precision simulation model for sample generation. The data acquisition module consists of an ambient temperature sensor (installed outside the cabinet), a current sensor (busbar assembly), and a perforated contact temperature sensor (contact adhesive). It collects three types of data in real time and transmits them to the data processing module via a shielded communication line. After receiving the data, the data processing module uses a maximum-minimum normalization method to map the data to the [0,1] range, simultaneously removing outliers (such as abnormal data caused by sudden current surges), and organizes it into model input features according to a preset format. The model training module loads the preprocessed training set and trains an LSTM model (with parameters set as input layer 4, double hidden layers 32 / 16, and ReLU activation function) and an SVM model (with parameters set as RBF kernel, C=4.0, γ=0.8). Then, it calculates the weights using the least squares method, constructs an LSTM-SVM combined model, and stores it in the model library. The prediction execution module calls combined models from the model library in real time, inputting real-time features output by the data processing module into the model to calculate the predicted temperature values ​​of the moving and stationary contacts. The predicted temperature and its change curve are displayed in real time on the interface. The early warning module presets a safety threshold for the moving and stationary contact temperatures (e.g., 120℃), comparing the predicted temperature with the threshold in real time. When the predicted temperature exceeds the threshold or the temperature rise rate is greater than 5℃ / min, an audible and visual alarm is immediately triggered, and maintenance personnel are notified via SMS. The alarm time and temperature data are also recorded for subsequent analysis. All modules of the system work collaboratively to achieve full automation from data acquisition to prediction and early warning, ensuring the safe operation of the switchgear.

[0160] It should be noted that the method described in this invention is only applicable to temperature prediction of the moving and stationary contacts in the arc-extinguishing chamber of a vacuum circuit breaker in a high-voltage switchgear under stable operating conditions. This method is no longer applicable during transient processes such as circuit breaker opening and closing operations or short-circuit faults.

[0161] The system also includes a model library for storing trained LSTM-SVM combined models for use by the prediction execution module.

[0162] A new model library module has been added, which clarifies that the system has the function of storing and retrieving models, improving the system's integration and practical efficiency, and meeting the needs of industrial applications.

[0163] The present invention also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to realize the functions of the data processing unit, the model building unit, and the prediction unit.

[0164] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed, implements the method for predicting the stationary contact temperature of a switchgear based on a simulation model as described above.

[0165] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0166] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the temperature of stationary contacts in a switchgear based on a simulation model, characterized in that, Includes the following steps: An electromagnetic-thermal-fluid multiphysics coupling simulation model of a high-voltage switchgear was established, and the simulation model was calibrated using temperature rise test data. By simulating different operating conditions using a calibrated simulation model, a sample dataset containing ambient temperature, current carrying capacity, plum blossom contact temperature, and moving and stationary contact temperature was obtained. The sample data is preprocessed by normalization and then divided into training and test sets according to a certain ratio; LSTM neural network model and SVM support vector machine model are constructed respectively, and the training set is used to train the model; The optimal weight coefficients of the LSTM and SVM models are determined by the least squares method, and an LSTM-SVM combined prediction model is established. By collecting real-time ambient temperature, current carrying capacity, and plum blossom contact temperature data, and inputting them into a combined prediction model, the predicted values ​​of the moving and stationary contact temperatures are obtained.

2. The method according to claim 1, characterized in that, The establishment of the multiphysics coupling simulation model includes: A simplified physical model was established based on the actual structure of the high-voltage switchgear. Set material property parameters, including thermal conductivity, specific heat capacity, electrical conductivity, and magnetic permeability; Perform electromagnetic field simulation calculations and temperature-fluid field simulation calculations; Solve multiphysics coupling equations using the finite element method or the finite volume method.

3. The method according to claim 1, characterized in that, The specific process for calibrating the simulation model is as follows: Multiple temperature monitoring points are installed on the surface of the switch cabinet; Conduct actual temperature rise tests and record temperature data at each monitoring point; Simulation calculations were performed under the same operating conditions to obtain simulated temperature data for the corresponding monitoring points. Adjust the boundary condition parameters of the simulation model to make the error between the simulation data and the experimental data less than 5%.

4. The method according to claim 1, characterized in that, The different operating conditions include: The ambient temperature range is from -25℃ to 50℃; The current carrying capacity varies from 0 to 1.2 times the rated current; The contact resistance of the plum blossom contact varies from 5 to 50 μΩ.

5. The method according to claim 1, characterized in that, The sample data processing employs the max-min normalization method: Xnorm = (X - Xmin) / (Xmax - Xmin); Where Xnorm is the normalized data value after processing by the max-min normalization method; X is the original data, and Xmin and Xmax are the minimum and maximum values ​​of the feature parameter, respectively. The feature parameter refers to the input variable that is currently being normalized, and the input variables include ambient temperature, current carrying capacity, upper or lower clover contact temperature.

6. The method according to claim 1, characterized in that, The construction parameters of the LSTM neural network model include: The input layer has four nodes, corresponding to ambient temperature, current carrying capacity, upper plum blossom contact temperature, and lower plum blossom contact temperature, respectively. There are two hidden layers, with 32 and 16 neurons in the two hidden layers, respectively; The output layer has one node, which is used to directly output the predicted temperature values ​​of the moving and stationary contacts; The activation function is the ReLU function; The training iterations are 500-1000 times; Alternatively, the model structure can be adjusted as follows: the number of input layer nodes can be expanded to 15, the number of hidden layers can be increased to 10, and the number of neurons can be adjusted accordingly to enhance the ability to extract temporal features.

7. The method according to claim 1, characterized in that, The construction parameters of the SVM support vector machine model include: Kernel function type: Radial basis function (RBF); Penalty parameter C: 2 -2 Up to 2 9 The optimal value is determined within the range using cross-validation. Kernel function parameter γ: 2 -10 Up to 2 9 The optimal value is determined through cross-validation within the specified range.

8. The method according to claim 1, characterized in that, The method for determining the optimal weight coefficient is as follows: Calculate the sum of squared prediction errors SSE_lstm of the LSTM model on the test set; Calculate the sum of squared prediction errors of the SVM model on the test set, SSE_svm; The optimal weighting coefficients are determined by minimizing the sum of squared errors of the two models using the least squares method. The optimal weight coefficients satisfy the condition w_lstm+w_svm=1. Where w_lstm represents the weights of the LSTM model, and w_svm represents the weights of the SVM model.

9. The method according to claim 1, characterized in that, The real-time acquisition of ambient temperature, current carrying capacity, and plum blossom contact temperature data is achieved through the following sensors: Ambient temperature sensor: Installed outside the switch cabinet to measure the ambient temperature; Current sensor: Installed on the bus line to measure real-time current carrying capacity; Temperature sensor: Installed on the surface of the plum blossom contact to measure the contact temperature.

10. A simulation model-based system for predicting the temperature of stationary contacts in switchgear, used to implement the method described in any one of claims 1-9, characterized in that, include: Multiphysics simulation module: used to build and calibrate the temperature field simulation model of the switchgear; Data acquisition module: includes ambient temperature sensor, current sensor and temperature sensor; Data processing module: used for sample data processing and feature extraction; Model training module: used for training and optimizing LSTM and SVM models; Prediction execution module: used for real-time temperature prediction and result display; Early warning module: Issues an alarm when the predicted temperature exceeds the set threshold.

Images