Power equipment fault diagnosis method based on SF6 / N2 mixed gas

Abstract

The invention provides an SF6 / N2 mixed gas-based power equipment fault diagnosis method, and belongs to the technical field of SF6 / N2 mixed gas-based power equipment fault diagnosis, and the method comprises the steps: building an experiment platform, carrying out a partial discharge test, and recording discharge gas data; establishing a computational chemical equation set of SF6 / N2 mixed gas decomposition, and constructing a causal relationship graph according to the computational chemical equation set; optimizing the structure of the causal relationship graph to form a factor matrix; screening a training data set according to the factor matrix; and training the tree structure neural network by using the training data to obtain a power equipment fault diagnosis model. According to the method provided by the invention, power equipment fault diagnosis can be carried out according to the gas proportion data of SF6 / N2 mixed gas decomposition, and the technical problem of insufficient reliability and interpretability of a diagnosis result caused by lack of deep consideration of an SF6 / N2 mixed gas decomposition mechanism in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment fault diagnosis based on SF6 / N2 mixed gas, and in particular, relates to a power equipment fault diagnosis method based on SF6 / N2 mixed gas. Background Art

[0002] In modern power systems, SF6 gas is widely used in high-voltage power equipment due to its excellent insulation and arc-extinguishing properties. However, SF6 is a strong greenhouse gas with a global warming potential 23,900 times that of CO2. In order to reduce the use of SF6, SF6 / N2 mixed gas has received widespread attention as an alternative. In practical applications, there may be various insulation defects inside the power equipment, such as tip discharge, floating potential, surface discharge, etc. These defects can cause partial discharge and then cause equipment failure. Therefore, accurate diagnosis of the fault type of power equipment is crucial to ensure the safe operation of the power grid. At present, the fault diagnosis of power equipment based on SF6 / N2 mixed gas mainly adopts the following methods:

[0003] 1. Gas chromatography: The fault type is determined by analyzing the changes in the content of each component in the mixed gas. However, this method requires regular sampling and analysis and cannot achieve real-time monitoring.

[0004] 2. Partial discharge characteristic analysis method: The fault type is determined by analyzing the amplitude, phase and other characteristics of partial discharge. However, this method is susceptible to external interference and has low discrimination between different types of defects.

[0005] 3. Artificial intelligence methods: such as support vector machines, fuzzy expert systems, etc., achieve diagnosis by establishing a mapping relationship between fault features and fault types. However, these methods often ignore the chemical reaction mechanism in the gas decomposition process, resulting in a lack of interpretability of the diagnosis results.

[0006] 4. Traditional neural network methods: such as BP neural network, RBF neural network, etc., although they can handle nonlinear problems, their processing capabilities for time series data are limited, and it is difficult to capture the changing trend of gas concentration over time.

[0007] These existing technologies all lack in-depth consideration of the decomposition mechanism of the SF6 / N2 mixed gas, resulting in the technical problem that the diagnostic results are not sufficient to explain the fault. Summary of the invention

[0008] In view of this, the present invention provides a method for diagnosing faults of power equipment based on SF6 / N2 mixed gas, which can solve the technical problem that the prior art lacks in-depth consideration of the decomposition mechanism of SF6 / N2 mixed gas, resulting in diagnostic results that are insufficient to explain the faults.

[0009] The present invention is achieved in that:

[0010] A first aspect of the present invention provides a method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas, which comprises the following steps:

[0011] S10, building a gas discharge experimental platform, the experimental platform comprising a power supply, a discharge chamber and an insulation defect model, the discharge chamber is a closed space, and the discharge chamber is filled with SF6 / N2 mixed gas; the insulation defect model is set in the discharge chamber, one output pole of the power supply is connected to one end of the insulation defect, and the other output pole of the power supply is connected to the other end of the insulation defect;

[0012] S20, adjusting the power supply by a step-by-step pressure method so that the insulation defect model generates partial discharge, and recording the starting voltage and breakdown voltage of each discharge as well as the temperature and pressure of each discharge;

[0013] S30, taking 5 voltage levels with equal difference between the starting voltage and the breakdown voltage, carrying out partial discharge experiment for 96 hours at these 5 voltage levels, measuring the discharge amount at different voltage levels by using an oscilloscope, taking out part of the gas every hour for qualitative and quantitative detection, obtaining gas ratio data, and establishing a discharge gas array every hour, including defect type, voltage, cumulative discharge amount, cumulative discharge times, gas ratio data, temperature, and pressure;

[0014] S40, establishing a group of computational chemical equations for decomposition of SF6 / N2 mixed gas, including SF6 decomposition balance equation, SF5 further decomposition balance equation, N2 decomposition balance equation, SF6 and N2 reaction balance equation, SF4 hydrolysis balance equation, NF and H2O reaction balance equation, mass conservation equation, ionization degree equation, breakdown field strength equation and discharge capacity calculation equation;

[0015] S50, establishing a cause-effect relationship diagram of decomposed gas according to the computational chemical equation group, wherein each node of the cause-effect relationship diagram is a gas obtained by decomposition;

[0016] S60, using the discharge gas array every 1 hour, optimizing the structure of the cause-effect relationship diagram, including a chain structure, a fork structure, and a collision structure;

[0017] S70, constructing a factor matrix based on the optimized cause-effect relationship diagram, wherein the rows of the factor matrix represent different time points, the columns represent the gases that are the causes, and the matrix elements are the changes in the content of each gas at the corresponding time points, and are weighted considering the reaction ratio;

[0018] S80, using the factor matrix to screen the discharge gas array to form a training data set. During the screening process, the data corresponding to the gas in the factor matrix is ​​retained, including defect type, voltage, cumulative discharge amount, cumulative discharge times, and proportion data of the corresponding key gas. During data processing, the gas content is normalized according to the proportional relationship of the reaction equation;

[0019] S90. Use the training data set to train a tree-structured neural network to obtain a power equipment fault diagnosis model based on SF6 / N2 mixed gas, apply the trained power equipment fault diagnosis model based on SF6 / N2 mixed gas to the power equipment to be tested, and perform fault diagnosis based on gas ratio data of the SF6 / N2 mixed gas decomposition of the power equipment to be tested.

[0020] The neural network includes a gas feature extraction subnetwork, an electrical feature extraction subnetwork, a timing feature extraction subnetwork, an environmental feature extraction subnetwork and a fusion subnetwork, forming a tree structure;

[0021] The gas feature extraction subnetwork has the concentration data of various gases in the gas ratio data as input and the gas feature vector as output, which is used to extract the characteristics of gas component changes, and the structure is a multi-layer perceptron;

[0022] The electrical feature extraction subnetwork has inputs of voltage, cumulative discharge amount, and cumulative discharge times, and outputs electrical feature vectors, which are used to extract features of discharge characteristics, and has a one-dimensional convolutional neural network structure;

[0023] The time series feature extraction subnetwork inputs gas ratio data and voltage, cumulative discharge amount, and cumulative discharge times at different time points, and outputs a time series feature vector for capturing the changing trend of gas concentration and electrical parameters over time. The structure is a long short-term memory network;

[0024] The environmental feature extraction subnetwork has temperature and pressure as input and an environmental feature vector as output, which is used to analyze the impact of environmental factors on faults. The structure is a multi-layer perceptron.

[0025] The fusion subnetwork has the concatenation of the output feature vectors of the above five subnetworks as input and the defect type credibility feature vector as output, which is used to comprehensively analyze the concatenated feature vectors to obtain the final diagnosis result. The structure includes a feature fusion layer, a fully connected layer and a Softmax layer, wherein the feature fusion layer uses an attention mechanism to perform weighted fusion of different features, the fully connected layer is used to expand and further extract features after formaldehyde fusion, and the Softmax layer is used to output a vector of the probability distribution of the fault diagnosis result.

[0026] Wherein, each equation in the computational chemical equation group is expressed as follows:

[0027] Assume that the initial mixture ratio of SF6 and N2 is x:(1-x), the total pressure is P, and the temperature is T;

[0028] 1.SF6 decomposition balance equation:

[0029]

[0030] Where K1 is the equilibrium constant, which is related to temperature T:

[0031] 2.SF5 further decomposes the equilibrium equation:

[0032]

[0033] 3.N2 decomposition balance equation:

[0034]

[0035] 4. The equilibrium equation of SF6 and N2 reaction:

[0036]

[0037] 5. SF4 hydrolysis equilibrium equation:

[0038]

[0039] 6. Balanced equation of NF and H2O reaction:

[0040]

[0041] 7. Mass conservation equation:

[0042]

[0043] 8. Ionization equation:

[0044] where n e is the electron density, n0 is the initial electron density, and N is the total particle number density

[0045] 9. Breakdown field strength equation:

[0046] Where E is the breakdown electric field strength, and f is a function related to α, T and gas composition;

[0047] 10. Discharge calculation equation:

[0048] Where Q is the discharge amount, C is the equivalent capacitance, V is the applied voltage, R is the equivalent resistance, and t is the time;

[0049] Where [X] represents the partial pressure of gas X; A i and B i (i=1,2,3,4,5,6) is a constant, which needs to be obtained by fitting experimental data; P is the total pressure, T is the temperature; x is the initial mole fraction of SF6; α is the degree of ionization; E is the electric field strength.

[0050] Furthermore, the step S10 specifically includes:

[0051] Step 101, select a high voltage DC power supply with a voltage range of 0 to 100 kV, a current range of 0 to 10 mA, and an accuracy of 0.1%;

[0052] Step 102: A discharge chamber is made of stainless steel material with an internal volume of 1 cubic meter and good air tightness and corrosion resistance. An air inlet, an air outlet, a pressure gauge and a temperature sensor are arranged on the discharge chamber.

[0053] Step 103: Create insulation defect models based on common defect types in actual power equipment, including models of tip discharge, floating potential, surface discharge, etc.;

[0054] Step 104, fixing the insulation defect model inside the discharge chamber and connecting it to a power source;

[0055] Step 105, evacuate the discharge chamber by a vacuum pump, and then inject sulfur hexafluoride / nitrogen mixed gas according to a preset ratio, and the inflation pressure is set to 0.4 MPa;

[0056] Step 106: Check whether all connections and seals are normal to ensure that the experimental platform is safe and reliable.

[0057] Wherein, the step S20 specifically includes:

[0058] Step 201, setting the initial voltage to 0 kV, and slowly increasing the voltage at a rate of 1 kV per second;

[0059] Step 202: Use a high-precision voltmeter to monitor the voltage value in real time, with an accuracy of 0.1%;

[0060] Step 203, using an ultra-high frequency partial discharge detector to monitor the discharge signal, with a frequency range of 300 to 1500 MHz;

[0061] Step 204: when a continuous discharge signal is detected, the current voltage is recorded as the starting voltage;

[0062] Step 205, continue to increase the voltage until a continuous large-scale discharge or flashover phenomenon occurs, at which time the recorded voltage is the breakdown voltage;

[0063] Step 206: Use a temperature sensor and a pressure sensor to record the temperature and pressure in the discharge chamber, wherein the accuracy of the temperature sensor is 0.1 degrees Celsius and the accuracy of the pressure sensor is 0.1 kilopascals;

[0064] Step 207, repeating steps 201 to 206 for 3 to 5 times, and taking the average values ​​as the final starting voltage and breakdown voltage.

[0065] Wherein, the step S30 specifically includes:

[0066] Step 301, calculating a voltage interval according to the starting voltage and the breakdown voltage, wherein the voltage interval is equal to the breakdown voltage minus the starting voltage divided by 4;

[0067] Step 302, setting 5 voltage levels, namely, a starting voltage, a starting voltage plus one voltage interval, a starting voltage plus two voltage intervals, a starting voltage plus three voltage intervals, and a starting voltage plus four voltage intervals;

[0068] Step 303: Conduct a 96-hour continuous discharge experiment for each voltage level;

[0069] Step 304: Use a digital oscilloscope to measure the discharge amount, and use a charge method to calculate the discharge amount;

[0070] Step 305: Take out a small amount of gas, about 10 ml, through the gas sampling valve every hour;

[0071] Step 306: Use a gas chromatograph-mass spectrometer to analyze the gas composition. The gas chromatograph uses a capillary column with a length of 30 meters and an inner diameter of 0.25 mm. The temperature program is to maintain the initial temperature at 40 degrees Celsius for 2 minutes, then increase the temperature to 200 degrees Celsius at a rate of 10 degrees Celsius per minute and maintain it for 5 minutes.

[0072] Step 307, quantitatively analyzing the content of each component through a standard gas calibration curve;

[0073] Step 308: record temperature and pressure data;

[0074] Step 309: For each 1-hour interval, a discharge gas array is established, including defect type, voltage value, cumulative discharge amount, cumulative discharge number, relative content of each gas component, temperature and pressure.

[0075] Wherein, the step S40 specifically includes:

[0076] Step 401, using quantum chemical calculation software to calculate the Gibbs free energy change of each reaction, thereby obtaining the relationship between the equilibrium constant and the temperature;

[0077] Step 402: For complex reactions, multi-step reaction mechanisms are considered and the reaction rate constant is calculated using transition state theory;

[0078] Step 403: Establish a mass conservation equation based on the atomic conservation principle;

[0079] Step 404, considering the relationship between electron density and total particle number density, establish an ionization degree equation;

[0080] Step 405: using the improved Paschen's law, taking into account the effects of gas composition, temperature and pressure, to establish a breakdown field strength equation;

[0081] Step 406: Based on the equivalent circuit model, the discharge amount calculation equation is established by taking into account the effects of capacitance, voltage and time constant.

[0082] Wherein, the step S50 specifically includes:

[0083] Step 501, identifying all gas types involved in the reaction according to the calculation of the chemical equation group;

[0084] Step 502: taking the gas type as a node of a cause-effect relationship graph;

[0085] Step 503, analyzing each equation to determine the cause-effect relationship between gases;

[0086] Step 504: For complex reactions, multi-step reaction mechanisms are considered and virtual nodes are introduced to represent intermediate states;

[0087] Step 505: Use a graphic visualization tool to draw a cause-effect relationship diagram, where nodes represent gas types, directed edges represent reaction relationships, and edge weights are represented by reaction rate constants.

[0088] Wherein, the step S60 specifically includes:

[0089] Step 601, using the discharge gas array data of every hour, calculate the change rate of each gas concentration over time;

[0090] Step 602: For gases with a smaller rate of change, delete the relevant nodes in the graph to simplify the model;

[0091] Step 603: Identify strongly correlated gas pairs based on the correlation analysis of gas concentrations, and consider merging these gases into one node;

[0092] Step 604: For the chain structure, calculate the rate constant of each step reaction. If the reaction rate of a certain step is much lower than that of other steps, simplify the step as the rate-limiting step;

[0093] Step 605: For the fork structure, compare the reaction rates of the two branches and retain the main branch;

[0094] Step 606: For the collision structure, evaluate the equilibrium constant of the reaction. If the reaction degree is very low, consider ignoring the reaction;

[0095] Step 607: Use a graph theory algorithm to further simplify the optimized graph and retain the critical path.

[0096] Wherein, the step S70 specifically includes:

[0097] Step 701: Determine the key gas types that need to be considered based on the optimized cause-effect relationship diagram;

[0098] Step 702: extracting the concentration data of these gases at different time points from the experimental data;

[0099] Step 703: construct a matrix, in which rows represent different time points and columns represent different gas types;

[0100] Step 704, calculating the importance weight of each gas in the entire reaction network;

[0101] Step 705: multiply each column of the matrix by the corresponding weight to obtain a weighted factor matrix;

[0102] Step 706: Normalize the weighted factor matrix to obtain a final factor matrix.

[0103] Wherein, the step S80 specifically includes:

[0104] Step 801, determining key gas types according to the factor matrix;

[0105] Step 802: for each 1-hour discharge gas array, extract the data corresponding to the gas in the factor matrix;

[0106] Step 803, normalizing the gas concentration data according to the stoichiometric ratio of the reaction equation;

[0107] Step 804: using a maximum and minimum value normalization method for the voltage data;

[0108] Step 805: Logarithmic transformation is used to process the magnitude difference of the accumulated discharge amount and the accumulated discharge times;

[0109] Step 806: Use principal component analysis to reduce the data dimension and retain the principal components with a variance explanation ratio greater than 95%;

[0110] Step 807: Use the random forest algorithm to evaluate feature importance and select the top 80% features in importance.

[0111] Step 808: randomly divide the processed data set into a training set and a validation set.

[0112] Wherein, the step S90 specifically includes:

[0113] Step 901, designing a tree structure neural network, including a gas feature extraction subnetwork, an electrical feature extraction subnetwork, a timing feature extraction subnetwork, an environmental feature extraction subnetwork, and a fusion subnetwork;

[0114] Step 902: The gas feature extraction subnetwork adopts a multi-layer perceptron, including three hidden layers, with 64, 32, and 16 neurons in each layer, and a rectified linear unit as an activation function;

[0115] Step 903: The electrical feature extraction subnetwork uses a one-dimensional convolutional neural network, including two convolutional layers and one maximum pooling layer, and finally a fully connected layer;

[0116] Step 904: The time series feature extraction sub-network adopts a long short-term memory network, which includes two layers of long short-term memory units, 128 neurons in each layer, and a fully connected layer at the end;

[0117] Step 905: The environmental feature extraction sub-network uses a multi-layer perceptron, including two hidden layers, with 32 and 16 neurons in each layer respectively;

[0118] Step 906: The fusion sub-network first uses the attention mechanism to perform weighted fusion on the outputs of each sub-network, then extracts features through two fully connected layers, and finally uses the Softmax layer to output the probability distribution of the fault type;

[0119] Step 907: Use the training data set to train the network, use the Adam optimizer, set the initial value of the learning rate to 0.001, and use the learning rate decay strategy;

[0120] Step 908: Optionally, use an early stopping strategy to prevent overfitting, and stop training when the validation set loss does not improve for 10 consecutive rounds;

[0121] Step 909: Optionally, use the validation set to evaluate the model performance and calculate indicators such as accuracy, precision, recall, and F1 score.

[0122] Compared with the prior art, the beneficial effects of the power equipment fault diagnosis method based on SF6 / N2 mixed gas provided by the present invention are:

[0123] 1. In-depth consideration of the decomposition mechanism of SF6 / N2 mixed gas:

[0124] By establishing a set of computational chemical equations for the decomposition of SF6 / N2 mixed gas, including equilibrium equations such as SF6 decomposition, SF5 further decomposition, N2 decomposition, SF6 and N2 reaction, SF4 hydrolysis, NF and H2O reaction, as well as mass conservation equation, ionization degree equation, breakdown field strength equation and discharge quantity calculation equation, this method fully considers the chemical reaction process of gas decomposition. This mechanism-based method improves the reliability and interpretability of the diagnostic results.

[0125] 2. Effectively utilize multi-source heterogeneous data:

[0126] This method constructs a tree-structured neural network, including a gas feature extraction subnetwork, an electrical feature extraction subnetwork, a time series feature extraction subnetwork, and an environmental feature extraction subnetwork. This structure can make full use of gas concentration data, electrical parameters (such as voltage, cumulative discharge amount, cumulative discharge times), time series information, and environmental factors (such as temperature and pressure), and realize the effective fusion of multi-source heterogeneous data.

[0127] 3. Capture timing characteristics:

[0128] By introducing a time series feature extraction subnetwork (using a long short-term memory network structure), this method can effectively capture the changing trends of gas concentration and electrical parameters over time. This consideration of time series features significantly improves the accuracy of fault diagnosis, especially for long-term and slowly developing fault types.

[0129] 4. Provide interpretable diagnostic results:

[0130] Based on the computational chemical equations and cause-effect diagrams of SF6 / N2 mixed gas decomposition, this method can provide interpretable diagnostic results, which not only helps engineers understand the cause of the fault, but also provides valuable reference information for subsequent equipment maintenance and optimization.

[0131] 5. Improve diagnostic efficiency:

[0132] By screening the discharge gas array and normalizing the data, this method significantly reduces the dimension of the input data and improves the efficiency of model training and diagnosis. This enables the method to respond quickly in actual engineering applications and promptly detect and handle potential equipment failures.

[0133] In summary, the power equipment fault diagnosis method based on SF6 / N2 mixed gas proposed in the present invention solves the technical problem that the prior art lacks in-depth consideration of the decomposition mechanism of SF6 / N2 mixed gas, resulting in the diagnosis results being insufficient to explain the fault, through innovative designs such as in-depth consideration of the gas decomposition mechanism, effective use of multi-source heterogeneous data, and capture of timing characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0134] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0135] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0136] like Figure 1 As shown, it is a flow chart of a method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas provided by the present invention. The method comprises the following steps:

[0137] S10, building a gas discharge experimental platform, the experimental platform comprising a power supply, a discharge chamber and an insulation defect model, the discharge chamber is a closed space, and the discharge chamber is filled with SF6 / N2 mixed gas; the insulation defect model is set in the discharge chamber, one output pole of the power supply is connected to one end of the insulation defect, and the other output pole of the power supply is connected to the other end of the insulation defect;

[0138] S20, adjusting the power supply by a step-by-step pressure method so that the insulation defect model generates partial discharge, and recording the starting voltage and breakdown voltage of each discharge as well as the temperature and pressure of each discharge;

[0139] S30, taking 5 voltage levels with equal difference between the starting voltage and the breakdown voltage, carrying out partial discharge experiment for 96 hours at these 5 voltage levels, measuring the discharge amount at different voltage levels by using an oscilloscope, taking out part of the gas every hour for qualitative and quantitative detection, obtaining gas ratio data, and establishing a discharge gas array every hour, including defect type, voltage, cumulative discharge amount, cumulative discharge times, gas ratio data, temperature, and pressure;

[0140] S40, establishing a group of computational chemical equations for decomposition of SF6 / N2 mixed gas, including SF6 decomposition balance equation, SF5 further decomposition balance equation, N2 decomposition balance equation, SF6 and N2 reaction balance equation, SF4 hydrolysis balance equation, NF and H2O reaction balance equation, mass conservation equation, ionization degree equation, breakdown field strength equation and discharge capacity calculation equation;

[0141] S50, establishing a cause-effect relationship diagram of decomposed gas according to the computational chemical equation group, wherein each node of the cause-effect relationship diagram is a gas obtained by decomposition;

[0142] S60, using the discharge gas array every 1 hour, optimizing the structure of the cause-effect relationship diagram, including a chain structure, a fork structure, and a collision structure;

[0143] S70, constructing a factor matrix based on the optimized cause-effect relationship diagram, wherein the rows of the factor matrix represent different time points, the columns represent the gases that are the causes, and the matrix elements are the changes in the content of each gas at the corresponding time points, and are weighted considering the reaction ratio;

[0144] S80, using the factor matrix to screen the discharge gas array to form a training data set. During the screening process, the data corresponding to the gas in the factor matrix is ​​retained, including defect type, voltage, cumulative discharge amount, cumulative discharge times, and proportion data of the corresponding key gas. During data processing, the gas content is normalized according to the proportional relationship of the reaction equation;

[0145] S90. Use the training data set to train a tree-structured neural network to obtain a power equipment fault diagnosis model based on SF6 / N2 mixed gas, apply the trained power equipment fault diagnosis model based on SF6 / N2 mixed gas to the power equipment to be tested, and perform fault diagnosis based on gas ratio data of the SF6 / N2 mixed gas decomposition of the power equipment to be tested.

[0146] The specific implementation methods of the above steps are described in detail below:

[0147] The specific implementation method of step S10 is to design and build a gas discharge experimental platform. The platform mainly consists of three parts: power supply, discharge chamber and insulation defect model. First, a high-voltage DC power supply is selected with a voltage range of 0-100kV, a current range of 0-10mA and an accuracy of 0.1%. The discharge chamber is made of stainless steel and has an internal volume of 1m 3 , with good air tightness and corrosion resistance. An air inlet, an air outlet, a pressure gauge and a temperature sensor are set on the discharge chamber. The insulation defect model is made according to the common defect types in actual power equipment, including models such as tip discharge, floating potential, and surface discharge. Fix the insulation defect model inside the discharge chamber and connect it to the power supply. Evacuate the discharge chamber by a vacuum pump, and then inject SF6 / N2 mixed gas according to a preset ratio (such as 80% SF6 and 20% N2), and the inflation pressure is set to 0.4MPa. Finally, check whether all connections and seals are normal to ensure that the experimental platform is safe and reliable. The purpose of this step is to build an experimental platform that simulates the internal environment of actual power equipment to provide a basis for subsequent discharge experiments.

[0148] The specific implementation method of step S20 is to conduct a partial discharge experiment using a step-by-step pressurization method. First, set the initial voltage to 0kV, and slowly increase the voltage at a rate of 1kV / s. Use a high-precision voltmeter (accuracy 0.1%) to monitor the voltage value in real time. At the same time, an ultra-high frequency partial discharge detector (frequency range 300-1500MHz) is used to monitor the discharge signal. When a continuous discharge signal is detected, the current voltage is recorded as the starting voltage. Continue to increase the voltage until a continuous large-scale discharge or flashover phenomenon occurs, at which time the recorded voltage is the breakdown voltage. Throughout the process, a temperature sensor (accuracy 0.1°C) and a pressure sensor (accuracy 0.1kPa) are used to record the temperature and pressure in the discharge chamber. Repeat the above process 3-5 times, and take the average value as the final starting voltage and breakdown voltage. The purpose of this step is to determine the local discharge starting voltage and breakdown voltage of the insulation defect model, and provide a basis for setting the voltage level of subsequent experiments.

[0149] The specific implementation method of step S30 is to conduct a long-term partial discharge experiment and collect data. First, according to the starting voltage and breakdown voltage obtained in S20, the voltage interval is calculated: △V = (breakdown voltage-starting voltage) / 4. Then, 5 voltage levels are set: V1 = starting voltage, V2 = V1 + △V, V3 = V2 + △V, V4 = V3 + △V, V5 = V4 + △V. For each voltage level, a continuous discharge experiment is performed for 96 hours. The discharge amount is measured using a digital oscilloscope (bandwidth 1GHz, sampling rate 10GS / s), and the discharge amount is calculated using the charge method. Every 1 hour, a small amount of gas (about 10mL) is taken out through the gas sampling valve, and a gas chromatography-mass spectrometer (GC-MS) is used to analyze the gas composition. The gas chromatograph uses a capillary column (30m long, inner diameter 0.25mm), and the heating program is to maintain the initial temperature at 40°C for 2min, then increase to 200°C at 10°C / min and maintain for 5min. The mass spectrometer uses an electron bombardment ion source with a scanning range of m / z 20-300. The content of each component is quantitatively analyzed through the standard gas calibration curve. Temperature and pressure data are recorded at the same time. For each 1-hour interval, a discharge gas array is established, which contains the following elements: defect type (such as tip discharge, floating potential, etc.), voltage value, cumulative discharge amount (pC), cumulative number of discharges, relative content of each gas component (%), temperature (℃) and pressure (MPa). The purpose of this step is to obtain a large amount of experimental data to provide a basis for subsequent data analysis and model training.

[0150] The specific implementation of step S40 is to establish a computational chemical equation group for the decomposition of SF6 / N2 mixed gas. The equation group includes 10 equations, which respectively describe the decomposition of SF6, further decomposition of SF5, decomposition of N2, reaction of SF6 with N2, hydrolysis of SF4, reaction of NF with H2O, conservation of mass, degree of ionization, breakdown field strength and discharge calculation. First, the change of Gibbs free energy of each reaction is calculated using quantum chemical calculation software (such as Gaussian 16), so as to obtain the relationship between the equilibrium constant K and the temperature T. For example, for the decomposition reaction of SF6, by calculating Constants A1 and B1 in . For complex reactions, such as the reaction between SF6 and N2, it may be necessary to consider multi-step reaction mechanisms and use transition state theory to calculate the reaction rate constants. The mass conservation equation is based on the principle of atomic conservation, and the ionization degree equation considers the relationship between electron density and total particle number density. The breakdown field strength equation adopts the improved Paschen's law, taking into account the effects of gas composition, temperature and pressure. The discharge capacity calculation equation is based on the equivalent circuit model, taking into account the effects of capacitance, voltage and time constant. The purpose of this step is to establish a complete theoretical model to describe the physical and chemical changes of SF6 / N2 mixed gas during the discharge process, providing a theoretical basis for subsequent data analysis and fault diagnosis.

[0151] The specific implementation of step S50 is to establish a cause-and-effect relationship diagram of decomposed gas. First, according to the computational chemical equations established in S40, identify all gas species involved in the reaction, including SF6, SF5, SF4, F, N2, N, NF, SO2, HF, NO, etc. These gases are used as nodes of the cause-and-effect relationship diagram. Then, analyze each equation to determine the cause-and-effect relationship between gases. For example, it can be known from the SF6 decomposition equilibrium equation that the decomposition of SF6 produces SF5 and F, so a directed edge pointing from SF6 to SF5 and F is drawn in the figure. Similarly, the equilibrium equation is further decomposed according to SF5, and a directed edge pointing from SF5 to SF4 and F is drawn. For complex reactions, such as the reaction of SF6 with N2, it is necessary to consider multi-step reaction mechanisms, which may involve intermediate products. In this case, virtual nodes can be introduced to represent intermediate states. Continue to analyze other equations until all cause-and-effect relationships are mapped into the figure. Finally, a graph visualization tool (such as Graphviz) is used to draw a cause-and-effect relationship diagram, where nodes represent gas species, directed edges represent reaction relationships, and the weights of edges can be represented by reaction rate constants. The purpose of this step is to visualize the complex chemical reaction network, facilitate the understanding of the cause-effect relationship in the gas decomposition process, and provide a basis for subsequent structural optimization.

[0152] The specific implementation of step S60 is to optimize the structure of the cause-effect relationship diagram. First, the discharge gas array data obtained in S30 for every hour is used to calculate the rate of change of each gas concentration over time. For gases with a small rate of change (such as less than 1% / hour), it is possible to consider deleting the relevant nodes in the graph to simplify the model. Then, based on the correlation analysis of the gas concentrations, strongly correlated gas pairs (correlation coefficient greater than 0.8) are identified, and these gases are considered to be merged into one node. Next, the optimization of the chain structure, fork structure and collision structure are considered respectively. For chain structures, such as SF6→SF5→SF4→SO2, the rate constant of each step of the reaction is calculated. If the reaction rate of a certain step is much lower than that of other steps (such as less than 1 / 10), the step can be used as the rate-limiting step and simplified to SF6→SO2. For a fork structure, such as SF6 is decomposed into SF5 and F, the reaction rates of the two branches are compared, and the main branch (reaction rate greater than 80%) is retained. For collision structures, such as the reaction between SF6 and N2, the equilibrium constant of the reaction is evaluated. If the degree of reaction is very low (such as the equilibrium constant is less than 0.01), the reaction can be considered ignored. Finally, the optimized graph is further simplified using a graph theory algorithm (such as the minimum spanning tree algorithm) to retain the critical path. The purpose of this step is to simplify the causal relationship graph, highlight the main reaction path, and provide a more streamlined and effective model structure for the subsequent factor matrix construction.

[0153] The specific implementation of step S70 is to construct a factor matrix. First, based on the optimized cause-effect relationship diagram, determine the key gas types that need to be considered, such as SF6, SF4, SO2, HF, N2, NF, etc. Then, extract the concentration data of these gases at different time points from the experimental data of S30. Construct a matrix M, in which rows represent different time points (such as a time point every 1 hour) and columns represent different gas types. The elements of the matrix M ij represents the concentration change of gas j at time point i. In order to consider the influence of reaction ratio, a weighting factor is introduced. For each gas j, its importance weight w in the entire reaction network is calculated. j , can be determined based on the number of reactions that the gas participates in and the reaction rate. For example, the following formula can be used: j = (number of reactions × average reaction rate constant) / (sum of all gas values). Then, multiply each column j of the matrix M by the corresponding weight w j , and get the weighted factor matrix W. Finally, in order to eliminate the influence of dimension, W is normalized to get the final factor matrix F. Normalization can be done by the maximum and minimum method, that is, for each column j, F ij =(W ij -min(W j )) / (max(W j )-min(Wj )). The purpose of this step is to convert the complex gas reaction process into a mathematical matrix, capture the characteristics of the change of gas concentration over time, and provide a basis for subsequent data screening and model training.

[0154] The specific implementation of step S80 is to use the factor matrix to screen the discharge gas array to form a training data set. First, according to the factor matrix F constructed in S70, the key gas types are determined. For each 1-hour discharge gas array in S30, the data corresponding to the gas in the factor matrix is ​​extracted, including defect type, voltage, cumulative discharge amount, cumulative discharge number and the proportion data of these key gases. Then, data preprocessing is performed. For gas concentration data, the data is normalized according to the stoichiometric ratio of the reaction equation. For example, for the SF6 decomposition reaction SF6→SF4+2F, the following formula can be used: Normalized SF6 concentration = [SF6] / ([SF6]+[SF4]+0.5[F]). For voltage data, the maximum and minimum normalization method is used: Normalized voltage = (V-Vmin) / (Vmax-Vmin), where Vmin and Vmax are the minimum and maximum voltages in the experiment, respectively. For the cumulative discharge amount and the cumulative discharge number, logarithmic transformation is used to process the magnitude difference of the data: log(Q) and log(N), where Q is the cumulative discharge amount and N is the cumulative discharge number. Next, feature selection is performed. The principal component analysis (PCA) method is used to reduce the data dimension, and the principal components with an explained variance ratio greater than 95% are retained. Then, the random forest algorithm is used to evaluate the feature importance, and the features ranked in the top 80% of importance are selected. Finally, the processed data set is randomly divided into a training set (80%) and a validation set (20%). The purpose of this step is to extract the most relevant and representative features from the original experimental data to form a high-quality training data set in preparation for subsequent neural network training.

[0155] The specific implementation method of step S90 is to train a tree-structured neural network and apply it to fault diagnosis. First, design a neural network structure. The gas feature extraction subnetwork uses a multi-layer perceptron (MLP), which includes 3 hidden layers, with 64, 32, and 16 neurons in each layer, and ReLU is used as the activation function. The electrical feature extraction subnetwork uses a one-dimensional convolutional neural network (1DCNN), which includes 2 convolutional layers (with a convolution kernel size of 3, a step size of 1, and a number of channels of 32 and 64, respectively) and 1 maximum pooling layer, and finally a fully connected layer (64 neurons). The time series feature extraction subnetwork uses a long short-term memory network (LSTM), which includes 2 layers of LSTM units (128 neurons in each layer), and finally a fully connected layer (64 neurons). The environmental feature extraction subnetwork uses MLP, which includes 2 hidden layers, with 32 and 16 neurons in each layer, respectively. The fusion subnetwork first uses the attention mechanism to perform weighted fusion on the outputs of each subnetwork, then extracts features through two fully connected layers (128 and 64 neurons), and finally uses the Softmax layer to output the probability distribution of the fault type. Next, the network is trained using the training data set prepared in S80. The Adam optimizer is used, the initial value of the learning rate is set to 0.001, and the learning rate decay strategy is used, and the learning rate is halved every 50 epochs. The cross entropy loss is selected as the loss function. The early stopping strategy is used to prevent overfitting, and the training is stopped when the validation set loss has not improved for 10 consecutive epochs. During the training process, batch normalization and dropout (rate 0.5) techniques are used to improve the generalization ability of the model. After the training is completed, the validation set is used to evaluate the model performance, and indicators such as accuracy, precision, recall, and F1 score are calculated.

[0156] For fault diagnosis of the power equipment to be tested, first collect the component data, electrical parameters and environmental parameters of the SF6 / N2 mixed gas in the equipment. Use the same preprocessing method as the training data to process these data. Then input the processed data into the trained neural network model to obtain the probability distribution of the fault type. Select the type with the highest probability as the diagnosis result, and output the probability value as the confidence of the diagnosis. If the highest probability is lower than the preset threshold (such as 0.7), a diagnosis result of "cannot be determined" is given, and further inspection is recommended.

[0157] The purpose of this step is to build an intelligent diagnosis model that can automatically learn the decomposition characteristics of SF6 / N2 mixed gas, and apply it to the fault diagnosis of actual power equipment to improve the accuracy and efficiency of diagnosis. The design of the tree structure enables the model to extract different types of features (gas, electrical, timing, environment) separately, and then conduct comprehensive analysis through the fusion network. This structure is conducive to capturing complex fault modes. The introduction of the attention mechanism enables the model to adaptively adjust the importance of different features, improving the interpretability and performance of the model.

[0158] In the specific implementation process, the following points need to be considered:

[0159] 1. Data enhancement: Since there may be fewer actual fault samples, data enhancement techniques can be used to expand the training set. For example, Gaussian noise can be added to the gas concentration data, small random adjustments can be made to the electrical parameters, or synthetic samples can be generated using a generative adversarial network (GAN).

[0160] 2. Model integration: To further improve diagnostic accuracy, multiple neural network models with slightly different structures can be trained, and then an ensemble learning method (such as voting or averaging) can be used to combine the output results of multiple models.

[0161] In addition, the real-time and robustness of the model also need to be considered. In terms of real-time, the inference speed can be improved by optimizing the network structure and using model compression techniques (such as quantization and pruning). In terms of robustness, the model's resistance to noise and abnormal input can be improved through adversarial training, data enhancement and other technologies.

[0162] Finally, in actual deployment, the reliability and security of the system need to be considered. Redundant systems can be designed, multiple independent sensors and diagnostic models can be used, and the reliability of diagnosis can be improved through majority voting and other methods. At the same time, a complete data security and privacy protection mechanism needs to be established to prevent sensitive information from being leaked or maliciously exploited.

[0163] In general, this method of power equipment fault diagnosis based on SF6 / N2 mixed gas provides a new idea for the condition monitoring and fault diagnosis of power equipment by combining advanced experimental techniques, chemical reaction theory and deep learning algorithms. It can not only improve the accuracy and efficiency of diagnosis, but also help to deeply understand the physical and chemical processes inside the equipment, providing an important reference for the design optimization and life prediction of the equipment.

[0164] Since the present invention involves a large number of calculations, the steps involved in the calculations in the specific implementation manner are described in detail below in the form of formulas:

[0165] The specific implementation of step S10 mainly involves the construction of experimental equipment and does not need to be explained using formulas.

[0166] The specific implementation of step S20 can be explained by the following formula:

[0167] 1. Voltage rise rate:

[0168]

[0169] Wherein, V represents voltage (unit: kV), and t represents time (unit: s).

[0170] 2. Calculation of the average value of the starting voltage and breakdown voltage:

[0171]

[0172] Among them, V start Represents the average starting voltage, V breakdown represents the average breakdown voltage, n represents the number of repeated experiments (3-5 times), V start,i and V breakdown,i represent the starting voltage and breakdown voltage of the i-th experiment respectively.

[0173] The specific implementation of step S30 can be explained by the following formula:

[0174] 1. Voltage level calculation:

[0175]

[0176] V k =V start +(k-1)ΔV,k=1,2,3,4,5

[0177] Where ΔV represents the voltage interval, V k Indicates the kth voltage level.

[0178] 2. Calculation of discharge capacity (charge method):

[0179]

[0180] Wherein, Q represents the discharge capacity (unit: pC), i(t) represents the discharge current (unit: A), and t1 and t2 represent the start and end time of discharge, respectively.

[0181] 3. Quantitative analysis in gas chromatography-mass spectrometry:

[0182] C i =k i ·A i

[0183] Among them, C i represents the concentration of component i, k i represents the response factor of component i, A i represents the peak area of ​​component i.

[0184] The specific implementation of step S40 involves a complex set of computational chemical equations, which can be explained by the following formula:

[0185] 1.SF6 decomposition balance equation:

[0186]

[0187] Where K1 represents the equilibrium constant, [X] represents the partial pressure of component X, P represents the total pressure, T represents the temperature (unit: K), and A1 and B1 are constants.

[0188] 2.SF5 further decomposes the equilibrium equation:

[0189]

[0190] 3.N2 decomposition balance equation:

[0191]

[0192] 4. The equilibrium equation of SF6 and N2 reaction:

[0193]

[0194] 5. SF4 hydrolysis equilibrium equation:

[0195]

[0196] 6. Balanced equation of NF and H2O reaction:

[0197]

[0198] 7. Mass conservation equation:

[0199]

[0200] Where x represents the initial mole fraction of SF6.

[0201] 8. Ionization equation:

[0202]

[0203] Among them, α represents the degree of ionization, n e represents the electron density, n0 represents the initial electron density, and N represents the total particle number density.

[0204] 9. Breakdown field strength equation (improved Paschen's law):

[0205]

[0206] Where E is the breakdown electric field strength, and f is a function related to α, T and gas composition.

[0207] 10. Discharge calculation equation:

[0208]

[0209] Among them, Q represents the discharge amount, C represents the equivalent capacitance, V represents the applied voltage, R represents the equivalent resistance, and t represents the time.

[0210] The specific implementation of step S50 mainly involves the construction of a cause-effect relationship diagram, and does not need to be explained using formulas.

[0211] The specific implementation of step S60 can be explained by the following formula:

[0212] 1. Calculation of gas concentration change rate:

[0213]

[0214] Among them, r i represents the concentration change rate of gas i, C i represents the concentration of gas i.

[0215] 2. Gas concentration correlation analysis (Pearson correlation coefficient):

[0216]

[0217] Among them, ρ ij represents the correlation coefficient between gases i and j, cov(C i ,C j ) means C i and C j The covariance of Ci and σ Cj Respectively represent C i and C j The standard deviation of .

[0218] 3. Calculation of reaction rate constant (Arrhenius equation):

[0219]

[0220] Where k is the reaction rate constant, A is the pre-exponential factor, and E is a represents activation energy, R represents gas constant, and T represents temperature.

[0221] The specific implementation of step S70 can be explained by the following formula:

[0222] 1. Factor matrix construction:

[0223] M ij =ΔC j (t i )

[0224] Among them, M ij represents the element of the factor matrix, ΔC j (t i) represents the time point t i The concentration change of gas j.

[0225] 2. Gas importance weight calculation:

[0226]

[0227] Among them, w j represents the importance weight of gas j, N j represents the number of reactions in which gas j participates, represents the average rate constant of gas j participating in the reaction, and m represents the total amount of gas.

[0228] 3. Weighting factor matrix calculation:

[0229] W ij =w j ·M ij

[0230] 4. Normalization processing:

[0231]

[0232] Among them, F ij Represents the final normalized factor matrix elements.

[0233] The specific implementation of step S80 can be explained by the following formula:

[0234] 1. Gas concentration normalization:

[0235]

[0236] Among them, C norm,i represents the normalized concentration of gas i, ν j It represents the stoichiometric coefficient of gas j in the reaction equation, and n represents the number of gas species participating in the reaction.

[0237] 2. Voltage data normalization:

[0238]

[0239] Among them, V norm Represents the normalized voltage, V min and V max represent the minimum and maximum voltages in the experiment, respectively.

[0240] 3. Logarithmic transformation of cumulative discharge amount and number of times:

[0241] Q log =log(Q+1)

[0242] N log=log(N+1)

[0243] Among them, Q log and N log They represent the cumulative discharge capacity and cumulative discharge times after logarithmic transformation, respectively.

[0244] 4. Principal Component Analysis (PCA):

[0245] X=UΣV T

[0246] Where X represents the original data matrix, U and V represent the left singular vector and the right singular vector respectively, and Σ represents the singular value matrix. The principal components with a variance explanation ratio greater than 95% are retained.

[0247] 5. Random Forest Feature Importance Evaluation:

[0248]

[0249] Among them, Importance(X i ) represents feature X i The importance of N tree represents the number of trees in the random forest, ΔImpurity(X i ,T j ) represents feature X i In the tree j The reduction in impurity.

[0250] The specific implementation of step S90 can be explained by the following formula:

[0251] 1. Multilayer Perceptron (gas feature extraction subnetwork and environmental feature extraction subnetwork):

[0252] h l =f(W l h l-1 +b l )

[0253] Among them, h l represents the output of the lth layer, W l and b l They represent the weight matrix and bias vector of the lth layer respectively, and f represents the activation function (such as ReLU: f(x) = max(0,x)).

[0254] 2. One-dimensional convolutional neural network (electrical feature extraction subnetwork):

[0255]

[0256] Among them, * represents a one-dimensional convolution operation, and k represents the convolution kernel size.

[0257] 3. Long short-term memory network (temporal feature extraction subnetwork):

[0258] Input gate: i t =σ(W i [h t-1 ,x t ]+b i )

[0259] Forget gate: f t =σ(W f [h t-1 ,x t ]+b f )

[0260] Output gate: o t =σ(W o [h t-1 ,x t ]+b o )

[0261] Candidate memory cells:

[0262] Memory Unit:

[0263] Hidden state: h t =o t ⊙tanh(c t )

[0264] Among them, σ represents the sigmoid function and ⊙ represents element-wise multiplication.

[0265] 4. Attention mechanism (fusion sub-network):

[0266]

[0267] e i =v T tanh(Wh i +b)

[0268]

[0269] Among them, α i represents the attention weight, e i represents the energy fraction, h i represents the input feature, and c represents the weighted context vector.

[0270] 5.Softmax layer (fusion sub-network output):

[0271]

[0272] Among them, pi represents the probability of the i-th fault type, z i represents the original score of the i-th fault type, and K represents the total number of fault types.

[0273] 6. Cross Entropy Loss Function:

[0274]

[0275] Among them, y i represents the true label (one-hot encoding), p i represents the predicted probability.

[0276] 7.Adam optimizer update rules:

[0277] m t =β1m t-1 +(1-β1)g t

[0278]

[0279] Among them, m t and v t denote the first-order and second-order momentum estimates, respectively, and g t represents the current gradient, β1 and β2 are the momentum decay rates, η is the learning rate, and ∈ is a small constant for numerical stability.

[0280] 8. Learning rate decay strategy:

[0281]

[0282] Among them, η t represents the learning rate of the tth epoch, η0 represents the initial learning rate, γ represents the decay rate (such as 0.5), and s represents the decay step size (such as 50 epochs).

[0283] 9. Early Stopping Strategy:

[0284] If the validation set loss does not improve for N consecutive epochs, the training is stopped. It can be expressed by the following formula:

[0285]

[0286] in, represents the validation set loss of the tth epoch, and N represents the tolerated number of epochs (such as 10).

[0287] The following are some explanations of the variables and constants involved: [X] is the partial pressure of gas X, in Pa; K i(i=1,2,3,4,5,6) is the equilibrium constant, dimensionless; P is the total pressure, unit is Pa; T is the temperature, unit is K; x is the initial mole fraction of SF6, dimensionless; α is the degree of ionization, dimensionless; E is the electric field strength, unit is V / m; n e is the electron density, in m(-3); n0 is the initial electron density, in m(-3); N is the total particle number density, in m^(-3); Q is the discharge amount, in C; C is the equivalent capacitance, in F; V is the applied voltage, in V; R is the equivalent resistance, in Ω; t is the time, in s;

[0288] Among them, A i and B i (i=1,2,3,4,5,6) is a constant and needs to be obtained by fitting experimental data. i Dimensionless, B i The unit is K.

[0289] Among them, f(α, T, gas composition): This is a function related to the ionization degree α, temperature T and gas composition, which is used to calculate the normalized breakdown field strength E / P. The specific form needs to be determined by experimental data.

[0290] About A i and B i (i=1,2,3,4,5,6): These constants need to be obtained by fitting experimental data. The experimental data comes from measuring the concentration of various gas components under different temperature and pressure conditions. Through the measurement results, the equilibrium constant K under different conditions can be calculated i Then, using the formula A is fitted by linear regression and other methods i and B i The value of .

[0291] The details are as follows:

[0292] 1. Determination of f(α, T, gas composition) function:

[0293] Step 1, experimental data collection: conduct breakdown experiments under different temperatures T, pressures P, and gas component ratios, measure the breakdown electric field strength E under each condition, and simultaneously measure or calculate the corresponding ionization degree α;

[0294] Step 2, data preprocessing: calculate the normalized breakdown field strength E / P, organize the data set, and form data points of (E / P, α, T, gas composition);

[0295] Step 3: Assume a possible function form based on physical theory and experience, for example: E / P = a·α b ·Tc ·(x SF6 ) d ·(x N2 ) e ; where a, b, c, d, e are undetermined parameters, x SF6 and x N2 is the mole fraction of SF6 and N2;

[0296] Step 4: Parameter fitting: Use nonlinear regression methods (such as Levenberg-Marquardt algorithm) to fit the parameters and minimize the sum of squared errors between the predicted values ​​and the experimental values.

[0297] Step 5: Model validation: Use part of the data for cross-validation to check the generalization ability of the model. If the fitting effect is not good, return to step 3 and try other function forms.

[0298] Step 6: Function determination: Determine the final function form and parameter values, for example: E / P = 2.5 × 10 5 α -0.3 ·T 0.2 ·(x SF6 ) 0.7 ·(x N2 ) 0.3 .

[0299] 2.A i and B i Fitting of (i=1,2,3,4,5,6):

[0300] Step 1: Experimental data collection: Perform gas decomposition experiments under different temperature T and pressure P conditions to measure the concentration (partial pressure) of various gas components;

[0301] Step 2: Calculate the equilibrium constant: Use the measured partial pressure data to calculate the equilibrium constant K for each reaction. i , for example, for SF6 decomposition:

[0302] Step 3, data sorting: For each reaction, sort (1 / T,logK i ) data points, prepare 6 sets of data, corresponding to 6 reactions;

[0303] Step 4: Linear regression: Perform linear regression on each set of data and fit the equation:

[0304] Step 5: Parameter extraction: Extract A from the regression results i and B i Value, A i is the intercept of the straight line, B i is the negative value of the slope of the straight line;

[0305] Step 6, Error analysis: Calculate the standard error and confidence interval of the fit to assess the reliability of the fit;

[0306] Step 7: Result verification: Use the obtained A i and B i Calculate K at different temperatures i , the calculated results are compared with the experimental data to verify the accuracy of the fitting.

[0307] These steps involve complex experimental design, data processing, and statistical analysis. In actual operation, you may need to use professional data analysis software (such as MATLAB, Python with SciPy) to perform regression analysis and parameter fitting. At the same time, multiple iterations and adjustments may be required to obtain the best fitting results.

[0308] The following are some parameters that need to be collected by the instrument:

[0309] 1.n e (Electron Density):

[0310] Method: Langmuir probe method or microwave interferometry

[0311] Principle: Langmuir probe infers electron density by measuring the current-voltage characteristics of the probe; microwave interferometry uses the effect of plasma on microwave propagation to calculate electron density

[0312] Equipment: Langmuir probe or microwave interferometer

[0313] 2.n0 (initial electron density):

[0314] Method: Radiation counting or photoelectron emission method

[0315] Principle: Use radiation detectors to measure background radiation levels, or generate initial electrons through the photoelectric effect

[0316] Equipment: Radiation counter or photoelectron emission device

[0317] 3.N (total particle number density):

[0318] Method: Gas state equation calculation

[0319] Principle: Use the ideal gas state equation PV = NkT, where k is the Boltzmann constant

[0320] Equipment: Pressure gauge and thermometer

[0321] 4.E(breakdown electric field strength):

[0322] Method: Step-by-step voltage boost

[0323] Principle: Gradually increase the voltage between electrodes until sustained discharge is observed

[0324] Equipment: High voltage power supply, voltage measuring instrument and discharge observation system

[0325] 5.f (function related to α, T and gas composition):

[0326] Method: Experimental data fitting

[0327] Principle: Measure the breakdown electric field strength under different conditions, and then use regression analysis or machine learning methods to fit the function

[0328] Equipment: Data acquisition system and data analysis software

[0329] 6.Q(discharge amount):

[0330] Method: Integrating current method or charge meter

[0331] Principle: Measure the change of discharge current over time and integrate it, or directly use a charge meter

[0332] Equipment: High-speed current probe and oscilloscope, or dedicated charge measurement instrument

[0333] 7.C (equivalent capacitance):

[0334] Method: LCR meter or impedance analyzer

[0335] Principle: Measuring the capacitance of a circuit at a specific frequency

[0336] Equipment: LCR meter or impedance analyzer

[0337] 8.V (applied voltage):

[0338] Method: Direct measurement

[0339] Principle: Use a high voltage probe or voltage divider to measure the voltage between electrodes

[0340] Equipment: High voltage probe and high precision voltmeter

[0341] 9.R (equivalent resistance):

[0342] Method: Impedance analysis or current-voltage characteristic measurement

[0343] Principle: Measure the impedance of a circuit at different frequencies, or calculate it using Ohm's law

[0344] Equipment: Impedance analyzer or current-voltage measurement system

[0345] Obtaining these parameters usually requires work in a professional high-voltage laboratory, and safety factors must be considered.e The precise measurement of n and n0 may require complex equipment and technology. The determination of the f function requires a large amount of experimental data and a complex data analysis process. In practical applications, the appropriate measurement method and accuracy requirements may be selected according to the specific situation.

[0346] Specifically, the principle of the present invention is:

[0347] 1. In-depth consideration of gas decomposition mechanism:

[0348] By establishing a set of computational chemical equations for the decomposition of SF6 / N2 mixed gas, this method fully considers the complex reaction mechanism of gas during partial discharge. This mechanism-based method can accurately describe the changing laws of gas components under different fault types, laying a solid theoretical foundation for subsequent feature extraction and fault diagnosis. Specifically, the SF6 decomposition balance equation, SF5 further decomposition balance equation, etc. describe the gradual decomposition process of SF6; the N2 decomposition balance equation describes the decomposition of N2; the SF6 and N2 reaction balance equation, the SF4 hydrolysis balance equation, and the NF and H2O reaction balance equation describe the interaction between different gases. These equations together constitute a complete reaction network that can accurately reflect the changing characteristics of gas components under different fault types.

[0349] 2. Effective fusion of multi-source heterogeneous data:

[0350] This method uses a tree-structured neural network, including a gas feature extraction subnetwork, an electrical feature extraction subnetwork, a time series feature extraction subnetwork, and an environmental feature extraction subnetwork. This design conforms to the essential characteristics of the data: gas concentration data is suitable for processing with a multi-layer perceptron; electrical parameters (such as voltage and cumulative discharge) have certain spatial correlations and are suitable for processing with a one-dimensional convolutional neural network; time series data is suitable for processing with a long short-term memory network; environmental factors (such as temperature and pressure) can be processed with a multi-layer perceptron. Through this hierarchical design, the features of various types of data can be effectively extracted, providing comprehensive information for subsequent fusion and diagnosis.

[0351] 3. Effective capture of timing features:

[0352] The introduction of the long short-term memory network (LSTM) as a time series feature extraction subnetwork can effectively capture the changing trends of gas concentration and electrical parameters over time. The gating mechanism of LSTM enables it to learn long-term dependencies, which is particularly important for identifying slowly developing faults. For example, certain insulation defects may slowly deteriorate over a long period of time, causing the gas concentration to change slowly. Traditional methods may find it difficult to capture such small changes, while LSTM can effectively identify such long-term trends.

[0353] 4. Application of attention mechanism:

[0354] The attention mechanism is used in the fusion sub-network, so that the model can adaptively adjust the importance of different features. This design is based on the fact that the importance of various features (such as gas concentration, electrical parameters, and environmental factors) for fault diagnosis varies under different fault types and operating conditions. The attention mechanism can dynamically assign weights, highlight important features, and suppress irrelevant features, thereby improving the accuracy and robustness of diagnosis.

[0355] 5. Construction of causal relationship diagram and factor matrix:

[0356] By establishing a causal relationship graph of decomposed gases, this method can clearly represent the generation relationship between different gases. This representation not only helps to understand the gas decomposition process, but also provides a basis for subsequent feature selection. The factor matrix constructed based on the causal relationship graph further considers the time dimension and reaction ratio, making data processing more accurate and efficient.

[0357] 6. Data screening and normalization:

[0358] By screening the discharge gas array, this method retains the key data corresponding to the gas in the factor matrix, effectively reducing the data dimension. At the same time, the gas content is normalized according to the proportional relationship of the reaction equation, making the concentration data of different gases comparable, which is conducive to model training and diagnosis.

[0359] In summary, the technical solution of the present invention has built a fault diagnosis framework with a solid theoretical foundation, comprehensive data processing, and efficient feature extraction by deeply considering the gas decomposition mechanism, multi-source heterogeneous data fusion, time series feature capture, and attention mechanism application. This method can accurately capture the characteristics of different fault types, adapt to various operating conditions, and provide reliable and explainable diagnostic results, thereby effectively solving the problems existing in the prior art.

[0360] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: The GIS (gas insulated switchgear) of a 500kV substation uses SF6 / N2 mixed gas as the insulating medium, with a mixing ratio of 80% SF6 and 20% N2. Recently, the operation and maintenance personnel found that the gas pressure of the equipment had dropped slightly during routine inspections, and suspected that there might be insulation defects. In order to promptly discover and diagnose potential faults, it was decided to use the method of the present invention for detailed detection and analysis.

[0361] Step 1: Build a gas discharge experimental platform

[0362] Firstly, a gas discharge experimental platform simulating the internal structure of GIS equipment was built in the laboratory.

[0363] The platform consists of the following main components:

[0364] 1. High voltage power supply: adjustable output voltage range 0-750kV.

[0365] 2. Discharge chamber: Made of tempered glass, with an internal volume of 0.5m 3 , can withstand 1MPa gas pressure.

[0366] 3. Insulation defect model: According to the internal structure of GIS equipment, three typical insulation defect models are made:

[0367] a) Tip discharge model: simulates the discharge between a metal tip and a planar electrode.

[0368] b) Suspended potential discharge model: simulates the discharge generated by metal foreign matter near the high-voltage electrode.

[0369] c) Surface discharge model: simulates the surface discharge on the insulator surface.

[0370] 4. Gas detection system: including gas chromatograph and mass spectrometer, used to monitor the changes of gas components in real time.

[0371] 5. Oscilloscope: used to measure discharge amount and discharge frequency.

[0372] 6. Temperature and pressure sensors: used to monitor environmental parameters in the discharge chamber.

[0373] Step 2: Step-by-step pressurization test

[0374] The three insulation defect models were subjected to step-by-step pressure tests. The initial conditions of each test were as follows:

[0375] -Gas pressure: 0.5MPa

[0376] - Temperature: 25℃

[0377] -SF6 / N2 mixing ratio: 80% / 20%

[0378] The experimental results are shown in Table 1 below (taking the tip discharge model as an example):

[0379] Table 1 Experimental results

[0380]

[0381]

[0382] Note: "-" indicates that no breakdown occurs at this voltage level.

[0383] Step 3: Multi-voltage partial discharge test

[0384] According to the results of step 2, five voltage levels were selected for 96-hour partial discharge experiments: 465kV, 475kV, 485kV, 495kV and 505kV. Table 2 shows some experimental data at the 465kV voltage level:

[0385] Table 2 Partial experimental data at 465kV voltage level:

[0386]

[0387] Note: To simplify the table, only the data at some time points are listed. In the actual experiment, data were recorded every hour.

[0388] Step 4: Establish the computational chemical equations for the decomposition of SF6 / N2 mixed gas

[0389] Based on experimental data and literature, a set of computational chemical equations for the decomposition of SF6 / N2 mixed gas was established. The following are the equilibrium constants of some key reactions (taking the data at 465kV voltage level and 25℃ as an example):

[0390] 1.SF6 decomposition: K1 = 1.2 × 10 -5

[0391] 2. SF5 is further decomposed into: K2 = 3.5 × 10 -4

[0392] 3. N2 decomposition: K3 = 7.8 × 10 -6

[0393] 4.SF6 and N2 reaction: K4 = 2.1 × 10 -7

[0394] 5. SF4 hydrolysis: K5 = 1.6 × 10 -3

[0395] 6.NF and H2O reaction: K6 = 4.3 × 10 -5

[0396] Step 5: Create a cause-effect diagram for decomposition gases

[0397] Based on the computational chemical equations, a causal relationship diagram of decomposition gas was established. The following Table 3 is a simplified adjacency matrix representation (1 indicates the existence of a direct causal relationship, 0 indicates the absence of a direct causal relationship):

[0398] Table 3 Adjacency matrix table

[0399] gas <![CDATA[SF6]]> <![CDATA[SF5]]> <![CDATA[SF4]]> F <![CDATA[N2]]> N NF <![CDATA[SO2F2]]> <![CDATA[SOF2]]> <![CDATA[SO2]]> HF <![CDATA[SF6]]> 0 1 0 1 0 0 0 0 0 0 0 <![CDATA[SF5]]> 0 0 1 1 0 0 0 0 0 0 0 <![CDATA[SF4]]> 0 0 0 1 0 0 0 1 1 0 0 F 0 0 0 0 0 0 1 0 0 0 1 <![CDATA[N2]]> 0 0 0 0 0 1 0 0 0 0 0 N 0 0 0 0 0 0 1 0 0 0 0 NF 0 0 0 0 0 0 0 0 0 0 1 <![CDATA[SO2F2]]> 0 0 0 0 0 0 0 0 0 1 0 <![CDATA[SOF2]]> 0 0 0 0 0 0 0 0 0 1 1 <![CDATA[SO2]]> 0 0 0 0 0 0 0 0 0 0 0 HF 0 0 0 0 0 0 0 0 0 0 0

[0400] Step 6: Structural optimization of the causal relationship diagram

[0401] The structure of the cause-effect diagram was optimized, focusing on the chain structure, fork structure and collision structure. After optimization, it was found that the chain structure of SF6→SF5→SF4→SO2F2 / SOF2→SO2 is particularly important for fault diagnosis. At the same time, it was also noticed that F and HF formed an important collision structure, which is very helpful for judging the severity of the discharge.

[0402] Step 7: Construct the factor matrix

[0403] Based on the optimized causal relationship diagram, a factor matrix was constructed. Table 4 below is an example of a factor matrix at some time points (the numerical values ​​represent the concentration change rate relative to the initial state):

[0404] Table 4 Example of factor matrix at some time points

[0405] Time (h) <![CDATA[SF6]]> <![CDATA[CF4]]> <![CDATA[SO2F2]]> <![CDATA[SOF2]]> <![CDATA[SO2]]> HF 1 -0.0005 0.001 0.0005 0.0005 0.0002 0.002 2 -0.001 0.002 0.001 0.001 0.0004 0.004 ... ... ... ... ... ... ... 95 -0.2 0.040 0.020 0.025 0.010 0.080 96 -0.205 0.041 0.021 0.026 0.011 0.082

[0406] Step 8: Screen the discharge gas array to form a training data set

[0407] According to the factor matrix, the discharge gas array is screened to form a training data set. The following Table 5 is part of the training data (for simplicity, only some time points and key parameters are listed):

[0408] Table 5 Part of the training data table

[0409]

[0410] Note: The actual training dataset contains data at all time points (once per hour) and all voltage levels.

[0411] Step 9: Train the tree structure neural network

[0412] A tree-structured neural network was trained using the filtered training data set. The network structure is as follows:

[0413] 1. Gas feature extraction subnetwork: a 3-layer fully connected network, with the input being the concentration data of 6 key gases and the output being a 128-dimensional gas feature vector.

[0414] 2. Electrical feature extraction subnetwork: 1D convolutional neural network, including 3 convolutional layers and 2 pooling layers. The input is voltage, cumulative discharge amount and cumulative discharge times, and the output is a 64-dimensional electrical feature vector.

[0415] 3. Time series feature extraction subnetwork: LSTM network, including 2 layers of LSTM units, 100 hidden units in each layer, the input is the gas concentration and electrical parameter time series data within 96 hours, and the output is a 200-dimensional time series feature vector.

[0416] 4. Environmental feature extraction subnetwork: a 2-layer fully connected network with temperature and pressure data as input and a 16-dimensional environmental feature vector as output.

[0417] 5. Fusion sub-network: It contains a feature fusion layer with an attention mechanism, two fully connected layers, and a Softmax output layer. Its input is the concatenation of the output feature vectors of the above four sub-networks, and its output is the probability distribution of three fault types (tip discharge, suspended potential discharge, and surface discharge).

[0418] During the training process, the Adam optimizer was used, the learning rate was 0.001, the batch size was 64, and the number of training rounds was 200. The best accuracy on the validation set reached 95.8%.

[0419] Step 10: Model application and fault diagnosis

[0420] The trained model is applied to the actual 500kV substation GIS equipment. An online monitoring system is installed in the equipment to collect data once an hour. The following Table 6 is the data collected at one time:

[0421] Table 6 Data collection table after model application

[0422]

[0423]

[0424] Inputting these data into the trained model, we obtain the following diagnostic results in Table 7:

[0425] Table 7 Diagnosis results

[0426] Fault type Probability Tip discharge 87.5％ Suspended potential discharge 10.2％ Surface discharge 2.3％

[0427] According to the diagnosis results, the model determines that the most likely fault type of the GIS equipment is tip discharge, with a probability of up to 87.5%.

[0428] Result analysis and verification:

[0429] 1. Gas characteristic analysis: The SF6 concentration decreased slightly (from the initial 80% to 79.880%), indicating that a certain degree of SF6 decomposition occurred. The production of characteristic gases such as CF4, SO2F2, SOF2, SO2 and HF was relatively small, which is consistent with the characteristics of tip discharge. The HF content was relatively high, indicating that a small amount of water may be involved in the reaction during the discharge process.

[0430] 2. Electrical characteristics analysis: The cumulative discharge amount and cumulative discharge times are large, indicating that the discharge duration is long and the intensity is moderate. These characteristics are consistent with the typical performance of tip discharge.

[0431] 3. Time series feature analysis: By comparing historical data, it is found that the gas concentration change trend is stable without drastic fluctuations. This slow and stable change process is also consistent with the characteristics of tip discharge.

[0432] 4. Analysis of environmental factors: The temperature and pressure increased slightly, but the changes were not large, indicating that the discharge process did not cause significant changes in the internal environment of the equipment.

[0433] Here is what the traditional method looks like:

[0434] 1. On-site inspection: professional technicians were arranged to conduct on-site inspections of GIS equipment. Ultrasonic detectors and infrared thermal imagers were used to conduct a comprehensive scan of the equipment. It was found that there were weak ultrasonic signals and slight hot spots near a certain insulator, which were consistent with the characteristics of tip discharge.

[0435] 2. Gas sampling and analysis: A small amount of gas samples were taken from the GIS equipment and sent to the laboratory for precise chromatographic analysis. The analysis results were basically consistent with the data from the online monitoring system, further confirming the reliability of fault diagnosis.

[0436] 3. Partial discharge measurement: GIS is tested using professional partial discharge measurement equipment. The measured discharge spectrum shows typical tip discharge characteristics, and the discharge amplitude and frequency are consistent with the model diagnosis results.

[0437] 4. Equipment dismantling inspection: Under the premise of ensuring the safe operation of the power grid, the GIS equipment was partially dismantled for inspection. A tiny metal protrusion was found at the connection between the high-voltage bushing and the insulator, which was the source of the tip discharge.

[0438] Through the above verification steps, it is confirmed that the diagnostic results of the model are accurate. This case fully demonstrates the effectiveness and reliability of the method of the present invention in practical applications.

[0439] Summarize:

[0440] This embodiment demonstrates the application of the power equipment fault diagnosis method based on SF6 / N2 mixed gas in the actual 500kV substation GIS equipment. By establishing a set of computational chemical equations for gas decomposition, constructing a cause-effect relationship diagram, optimizing the data processing process, training a tree-structured neural network and other steps, the tip discharge fault in the equipment was successfully diagnosed. Subsequent on-site inspections and equipment dismantling verified the accuracy of the diagnostic results. This method can not only accurately identify the type of fault, but also provide information on the severity and development trend of the fault, providing strong support for equipment maintenance and decision-making. By timely discovering and handling potential equipment failures, the reliability and safety of the power system can be significantly improved, power outages caused by equipment failures can be reduced, and maintenance costs can be reduced.

[0441] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas, characterized in that: The following steps are involved: S10, build a gas discharge experimental platform, the experimental platform includes a power supply, a discharge chamber and an insulation defect model, the discharge chamber is a closed space, and the discharge chamber is filled with SF6 / N2 mixed gas; the insulation defect model is arranged in the discharge chamber, one output pole of the power supply is connected to one end of the insulation defect model, and the other output pole of the power supply is connected to the other end of the insulation defect model; S20, adjusting the power supply by a step-by-step pressure method so that the insulation defect model generates partial discharge, and recording the starting voltage and breakdown voltage of each discharge as well as the temperature and pressure of each discharge; S30, taking 5 voltage levels with equal difference between the starting voltage and the breakdown voltage, carrying out partial discharge experiment for 96 hours at these 5 voltage levels, measuring the discharge amount at different voltage levels by using an oscilloscope, taking out part of the gas every hour for qualitative and quantitative detection, obtaining gas ratio data, and establishing a discharge gas array every hour, including defect type, voltage, cumulative discharge amount, cumulative discharge times, gas ratio data, temperature, and pressure; S40, establishing a group of computational chemical equations for decomposition of SF6 / N2 mixed gas, including SF6 decomposition balance equation, SF5 further decomposition balance equation, N2 decomposition balance equation, SF6 and N2 reaction balance equation, SF4 hydrolysis balance equation, NF and H2O reaction balance equation, mass conservation equation, ionization degree equation, breakdown field strength equation and discharge capacity calculation equation; S50, establishing a cause-effect relationship diagram of decomposed gas according to the computational chemical equation group, wherein each node of the cause-effect relationship diagram is a gas obtained by decomposition; S60, using the discharge gas array every 1 hour, optimizing the structure of the cause-effect relationship diagram, including a chain structure, a fork structure, and a collision structure; S70, constructing a factor matrix based on the optimized cause-effect relationship diagram, wherein the rows of the factor matrix represent different time points, the columns represent the gases that are the causes, and the matrix elements are the changes in the content of each gas at the corresponding time points, and are weighted considering the reaction ratio; S80, using the factor matrix to screen the discharge gas array to form a training data set. During the screening process, the data corresponding to the gas in the factor matrix is ​​retained, including defect type, voltage, cumulative discharge amount, cumulative discharge times, and proportion data of the corresponding key gas. During data processing, the gas content is normalized according to the proportional relationship of the reaction equation; S90. Use the training data set to train a tree-structured neural network to obtain a power equipment fault diagnosis model based on SF6 / N2 mixed gas, apply the trained power equipment fault diagnosis model based on SF6 / N2 mixed gas to the power equipment to be tested, and perform fault diagnosis based on gas ratio data of the SF6 / N2 mixed gas decomposition of the power equipment to be tested.

2. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 1, characterized in that: The neural network includes a gas feature extraction subnetwork, an electrical feature extraction subnetwork, a time series feature extraction subnetwork, an environmental feature extraction subnetwork and a fusion subnetwork, forming a tree structure; The gas feature extraction subnetwork has the concentration data of various gases in the gas ratio data as input and the gas feature vector as output, which is used to extract the characteristics of gas component changes, and the structure is a multi-layer perceptron; The electrical feature extraction subnetwork has inputs of voltage, cumulative discharge amount, and cumulative discharge times, and outputs electrical feature vectors, which are used to extract features of discharge characteristics, and has a one-dimensional convolutional neural network structure; The time series feature extraction subnetwork inputs gas ratio data and voltage, cumulative discharge amount, and cumulative discharge times at different time points, and outputs a time series feature vector for capturing the changing trend of gas concentration and electrical parameters over time. The structure is a long short-term memory network; The environmental feature extraction subnetwork has temperature and pressure as input and an environmental feature vector as output, which is used to analyze the impact of environmental factors on faults. The structure is a multi-layer perceptron. The fusion subnetwork has the concatenation of the output feature vectors of the above five subnetworks as input and the defect type credibility feature vector as output, which is used to comprehensively analyze the concatenated feature vectors to obtain the final diagnosis result. The structure includes a feature fusion layer, a fully connected layer and a Softmax layer, wherein the feature fusion layer uses an attention mechanism to perform weighted fusion of different features, the fully connected layer is used to expand and further extract features after formaldehyde fusion, and the Softmax layer is used to output a vector of the probability distribution of the fault diagnosis result.

3. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 2, characterized in that: The SF6 decomposition balance equation is specifically expressed as: in, Wherein, [SF6], [SF5], [F] represent the partial pressures of SF6, SF5, and F gases, K1 is the equilibrium constant, P is the total pressure, T is the temperature, and A1 and B1 are constants.

4. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 3, characterized in that: The SF5 further decomposes the equilibrium equation, which is specifically expressed as: in, Where [SF4] represents the partial pressure of SF4 gas, and A2 and B2 are constants.

5. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 4, characterized in that: The N2 decomposition equilibrium equation is specifically expressed as: in, Wherein, [N2] represents the partial pressure of N2 gas, and A3 and B3 are constants.

6. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 5, characterized in that: The equilibrium equation of the reaction between SF6 and N2 is specifically expressed as: in, Wherein, [NF] represents the partial pressure of NF gas, and A4 and B4 are constants.

7. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 6, characterized in that: The SF4 hydrolysis equilibrium equation is specifically expressed as: in, Wherein, [H2O], [SO2], [HF] represent the partial pressures of H2O, SO2, and HF gases, and A5 and B5 are constants.

8. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 7, characterized in that: The equilibrium equation of the reaction between NF and H2O is specifically expressed as: in, Wherein, [NO] represents the partial pressure of NO gas, and A6 and B6 are constants.

9. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 8, characterized in that: The mass conservation equation is specifically expressed as: In the formula, it is assumed that the initial mixing ratio of SF6 and N2 is x:(1-x), and x is the initial molar fraction of SF6.

10. A method for diagnosing faults of electric power equipment based on SF6 / N2 mixed gas according to claim 9, characterized in that: The ionization degree equation is specifically expressed as: Among them, α is the degree of ionization, n e is the electron density, n0 is the initial electron density, and N is the total particle number density; The breakdown field strength equation is specifically expressed as: (α, T, gas composition); Where E is the breakdown electric field strength, and f is a function related to α, T and gas composition; The discharge amount calculation equation is specifically expressed as: Where Q is the discharge amount, C is the equivalent capacitance, V is the applied voltage, R is the equivalent resistance, and t is the time.

Images