SF6 density relay leakage determination method

Abstract

The invention discloses an SF6 density relay leakage determination method. The method comprises the following steps: establishing an SF6 gas chamber leakage multi-physics field coupling model and generating a multi-state sample set; a residual multi-layer perceptron is adopted as a classification model, and hyper-parameter optimization is carried out in combination with grid search and ten-fold cross validation; constructing a leakage state intelligent identification model, and deploying the leakage state intelligent identification model on an edge device or a cloud platform; and finally, real-time and accurate classified output of the SF6 leakage state is realized. According to the method, 'non-leakage ', 'slight leakage' and'severe leakage 'can be efficiently and accurately distinguished only by depending on original air chamber structure parameters and temperature sensing information without additionally arranging an expensive detection device, and the reliability and economical efficiency of leakage judgment are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power equipment state monitoring and fault diagnosis, and particularly relates to an SF6 density relay leakage determination method based on a residual multilayer perception machine, which is suitable for real-time, accurate and automatic identification of the leakage state of SF6 gas in high-voltage electrical equipment. BACKGROUND

[0002] SF6 gas is widely used in high-voltage power equipment such as gas-insulated switchgear and circuit breakers due to its excellent insulation and arc-extinguishing performance. However, SF6 gas leakage may occur due to aging of the seal, vibration or assembly defects during long-term operation of the equipment, which may lead to a decrease in the insulation strength of the equipment and even cause a major safety accident. Therefore, timely and accurate detection and determination of SF6 leakage are of great significance for ensuring the stable operation of the power system, the safety of the equipment and the protection of personnel.

[0003] Currently, SF6 leakage detection methods mainly include optical methods and non-optical methods. Optical methods such as photoacoustic spectroscopy and tunable diode laser absorption spectroscopy have high detection accuracy, but the equipment is expensive, has poor environmental adaptability and is difficult to deploy on a large scale. Non-optical methods such as the bandage method, the bubble method and the pressure comparison algorithm have lower costs, but generally have low sensitivity, are easily affected by temperature, cannot identify micro-leakage or can only make relative judgments. In addition, existing detection methods have poor adaptability to different sizes of gas chambers, have obvious "size effect" and are difficult to achieve low-cost and high-precision online leakage monitoring without modifying the structure of the original equipment. Therefore, an intelligent SF6 leakage determination method is needed that can adapt to different gas chamber structures, rely only on basic sensing information, does not require additional expensive detection devices and can run in real time on an embedded or edge side. SUMMARY

[0004] The technical problem to be solved by the application is to provide an SF6 leakage state determination method with high detection accuracy, fast response speed, strong environmental adaptability and easy deployment, to overcome the dependence on high-cost sensors, the obvious size effect and the difficulty in balancing accuracy and real-time performance in existing detection technologies.

[0005] To solve the technical problem, the solution of the application is to provide an SF6 density relay leakage determination method based on ResMLP.

[0006] The simulation of the leakage state of the SF6 gas chamber in a variable temperature environment by finite element simulation includes the following steps:

[0007] (1) Based on the multi-physical field coupling finite element simulation method, a flow-solid coupling three-dimensional model of SF6 gas chamber under various leakage states is established; the model comprehensively considers the structure parameters of the gas chamber height, radius, wall thickness, initial SF6 gas pressure, external environment temperature change, and the boundary conditions with different leakage rates set at the top of the gas chamber, and simulates three leakage states of no leakage, slight leakage and severe leakage;

[0008] (2) By changing the values of the above parameters within a reasonable engineering range, a large number of virtual sample data covering various working conditions are generated; the sample data not only includes the structure parameters of the gas chamber, the initial pressure, the environment temperature, but also includes the current time and the last time of the gas chamber external environment temperature and the gas chamber internal average temperature, which together constitute the multi-dimensional input features;

[0009] (3) The pre-processing and standardization of all sample data generated in the above step are carried out: first, the original data is normalized by linear transformation, which maps the data of each feature dimension to the interval [-1, 1] to eliminate the dimensional difference and the order of magnitude, accelerate the convergence of the subsequent training process of the model and improve the numerical stability;

[0010] (4) After normalization, the data is explicitly divided into input feature set and output label set, and the output label is encoded as an integer corresponding to "no leakage", "slight leakage" and "severe leakage".

[0011] The application further provides a ResMLP classification model constructed with residual connection as the core, including the following steps:

[0012] (1) The pre-processed sample set is trained on the model; the input matrix where h is the height of the gas chamber, r is the radius of the gas chamber, P is the initial pressure of the gas chamber, is the wall thickness of the gas chamber, T0 is the temperature at the last time, T1 is the temperature at this time, T t0 is the average temperature of the gas chamber at the last time, T t1 is the average temperature of the gas chamber at this time, and the output matrix Y={0, 1, 2} represents different leakage degrees;

[0013] (2) Before training, the grid search strategy is adopted and combined with the ten-fold cross-validation method to systematically optimize and automatically optimize the key hyperparameters of network depth, number of neurons in each layer, activation function type, learning rate, batch size and dropout rate, so as to determine the best model configuration;

[0014] (3) After the model training is completed, the accuracy, area under the curve and single calculation time are used to comprehensively evaluate and verify the performance of the model on the independent test set;

[0015] (4) After verification, the trained ResMLP classification model is deployed on an edge computing device or a cloud server; in actual application, the system receives the SF6 gas chamber structure parameters, internal pressure, environmental temperature and historical temperature data measured by the on-site sensor in real time, inputs them into the deployed model, and outputs the classification determination result of the current gas chamber leakage state online, automatically and quickly, thereby realizing low-cost and high-precision SF6 leakage intelligent monitoring.

[0016] The beneficial effects of the present application are: the high-precision finite element simulation generates samples covering multiple working conditions, overcoming the problems of high cost and few samples of real leakage experiments; the ResMLP residual structure is used to enhance the training stability and feature extraction capability of the deep network, significantly improving the classification accuracy under small sample and unbalanced data; only relying on the original gas chamber structure parameters and temperature sensor, without adding expensive special sensors, the system transformation cost and complexity are greatly reduced; the model has high lightweight degree, supports real-time operation on the edge side, and meets the dual requirements of response speed and accuracy in engineering applications. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The SF6 density relay leakage determination method implemented by the present application is a general flowchart;

[0018] Figure 2 The structure diagram of the SF6 gas chamber model is shown in the figure;

[0019] Figure 3 The ResMLP training flowchart is shown in the figure. DETAILED DESCRIPTION

[0020] The practical basis of the present application is to simulate the thermal-flow field characteristics of the SF6 gas chamber under different leakage states based on a multi-physical field coupling simulation model, which specifically includes the following steps:

[0021] (1) A three-dimensional model of the SF6 gas chamber is established in a multi-physical field coupling simulation software, and the gas chamber structure parameters including height, radius and wall thickness are defined; the SF6 gas and the gas chamber shell are given corresponding material properties, mainly including thermal conductivity, specific heat capacity and density, and the leakage boundary conditions are set; the k-ε turbulence model is used for leakage flow, the solid heat transfer and fluid heat transfer interfaces are coupled, and the initial SF6 gas pressure and the external environmental temperature change conditions are set; the structured grid division technology is used to encrypt the grid in the gas chamber area, especially in the near-leakage and wall contact area, and high-quality calculation grid is generated;

[0022] (2) Based on the model constructed in the previous step, the dynamic response data of the temperature and pressure field inside the gas chamber under different leakage states is obtained through finite element calculation; the structure size of the gas chamber, the leakage rate, the initial pressure, the external temperature and other parameters are adjusted in turn, batch simulation is performed, and a multi-physical field data set of the gas chamber containing "no leakage", "slight leakage" and "severe leakage" is generated.

[0023] (3) The pre-processing and standardization of all sample data generated in the previous step are performed: linear transformation normalization method is adopted to map the data in each feature dimension to the interval of [-1, 1];

[0024] (4) After normalization, the data is explicitly divided into input feature set and output label set, and the output label adopts integer coding to correspond to "no leakage", "slight leakage" and "severe leakage" respectively.

[0025] After the simulation model is constructed, the ResMLP classification model with residual connection as the core is constructed, including the following steps:

[0026] (1) The pre-processed sample set obtained is used to train the model; the input matrix X is defined as where h is the height of the gas chamber, r is the radius of the gas chamber, P is the initial pressure of the gas chamber, is the wall thickness of the gas chamber, T0 is the external temperature at the previous time, T1 is the external temperature at this time, T t0 is the average temperature inside the gas chamber at the previous time, T t1 is the average temperature inside the gas chamber at this time, and the output matrix Y = {0, 1, 2} represents different leakage degrees;

[0027] (2) Before training, the grid search strategy is adopted, combined with the ten-fold cross-validation method, to optimize the number of hidden layer neurons, learning rate, initial weight and bias in the classification neural network; the search range and step of each hyperparameter are set, and the average accuracy of cross-validation is used as the model performance evaluation index; by traversing all possible combinations of hyperparameters and evaluating their generalization performance, the best hyperparameter configuration is selected and applied to the construction and training of the network model;

[0028] (3) The trained neural network is verified on the test set to evaluate its classification accuracy, AUC and calculation time under different leakage states, to ensure that the model accuracy and response speed meet the requirements of engineering practice;

[0029] (4) After verification, the model is embedded into the SF6 density relay monitoring system, and the leakage state is determined online by inputting the gas chamber structure parameters, temperature and pressure sequence, to realize early leakage warning and state classification.

[0030] The application will be further described in conjunction with the accompanying drawings.

[0031] Figure 1 The flow chart of the SF6 density relay temperature compensation method is shown, and the specific process is as follows: 1) input SF6 gas chamber structure parameters; 2) establish SF6 gas chamber leakage model; 3) finite element calculation; 4) grid search and ten-fold cross-validation for hyperparameter optimization; 5) construct residual multilayer perception SF6 gas chamber leakage model; 6) output SF6 gas chamber leakage state.

[0032] Figure 2 The SF6 gas chamber model result schematic diagram is shown, and the structure parameters include gas chamber height, radius, wall thickness and initial pressure inside the gas chamber.

[0033] Figure 3 The GA-IPSO-RBF training flow chart is shown, and the specific process is as follows: 1) data preprocessing; 2) construct ResMLP network structure containing residual connection; 3) initialize hyperparameter search range; 4) grid search and ten-fold cross-validation; 5) model training; 6) evaluate validation set performance; 7) judge whether the accuracy requirement is reached; 8) if the accuracy requirement is not reached, then re-perform process 4; 9) if the accuracy requirement is reached, then end.

Claims

1. A method for determining leakage in an SF6 density relay, characterized in that, The finite element method simulation of the leakage state of SF6 gas chamber under variable temperature environment includes the following steps: (1) Based on the multi-physics field coupled finite element simulation method, a fluid-structure coupling three-dimensional model of SF6 gas chamber under various leakage states is established. The model comprehensively considers the structural parameters of gas chamber height, radius, wall thickness, initial SF6 gas pressure, external ambient temperature change, and boundary conditions with different leakage velocities set at the top of the gas chamber. The system simulates three leakage states: no leakage, slight leakage, and heavy leakage. (2) By changing the values ​​of the above parameters within a reasonable engineering range, virtual sample data covering multiple working conditions is generated in batches; the sample data includes not only the air chamber structural parameters, initial pressure, and ambient temperature, but also the ambient temperature outside the air chamber and the average temperature inside the air chamber at the current and previous times, which together constitute multi-dimensional input features. (3) Preprocess and standardize all sample data generated in the previous step: use the linear transformation normalization method to map the data of each feature dimension to the interval [-1, 1] to eliminate the difference in dimensions and the influence of order of magnitude, accelerate the convergence of the subsequent training process of the model and improve the numerical stability. (4) After normalization, the data is clearly divided into input feature set and output label set, where the output label uses integer encoding to correspond to the three states of "no leakage", "slight leakage" and "severe leakage".

2. A method for determining leakage in an SF6 density relay based on claim 1, characterized in that, Constructing a ResMLP classification model with residual connections as its core involves the following steps: (1) Train the model using the preprocessed sample set obtained; define the input matrix. Where h is the height of the air chamber, r is the radius of the air chamber, and P is the initial pressure of the air chamber. Where T is the wall thickness of the air chamber, T0 is the ambient temperature at the previous moment, T1 is the ambient temperature at this moment, and T... t0 T represents the average internal temperature of the chamber at the previous moment. t1 The average temperature inside the chamber at this moment is given by the output matrix Y = {0, 1, 2}, which represents different degrees of leakage. (2) Before training, a grid search strategy and ten-fold cross-validation method were used to systematically optimize and automatically find the key hyperparameters such as network depth, number of neurons per layer, activation function type, learning rate, batch size, and dropout rate in order to determine the best model configuration. (3) After the model training is completed, the model performance is comprehensively evaluated and verified on an independent test set using multiple indicators such as accuracy, area under the curve and single computation time. (4) After verification, the trained ResMLP classification model is deployed on edge computing devices or cloud servers. In practical applications, the system receives SF6 gas chamber structural parameters, internal pressure, ambient temperature and historical temperature data measured by field sensors in real time and inputs them into the deployed model. The system can then output the classification judgment results of the current gas chamber leakage status online, automatically and quickly, thereby achieving low-cost and high-precision intelligent monitoring of SF6 leakage.

Images