Line fault positioning analysis method based on artificial intelligence

Abstract

The invention discloses a line fault positioning analysis method based on artificial intelligence, and relates to the technical field of power system fault detection. Comprising the steps of constructing a tree-shaped undirected network graph based on a power transmission line network; a data acquisition device is installed in the power transmission line to acquire and process data; determining feature data with high sensitivity to the fault type and the fault position from the historical operation and maintenance data; constructing a multi-task joint autonomous learning model; and using model analysis, outputting a fault type and a fault position, and assisting in forming an operation and maintenance scheme. According to the method, the accuracy of line fault positioning and the accuracy of fault type judgment are improved through methods of constructing a tree-shaped undirected graph, installing various sensors to collect data, determining sensitive features from historical data, constructing a multi-task joint autonomous learning model, actively learning the model and the like; and meanwhile, the adaptive capacity to a complex and changeable environment is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault detection, and specifically to a method for analyzing line fault location based on artificial intelligence. Background Art

[0002] In a power system, line fault location is a key link to ensure the stable operation of the power grid and quickly restore power supply. With the continuous expansion and complexity of the power system, traditional line fault location methods have gradually revealed many deficiencies.

[0003] Traditional line fault location methods, such as the bridge method and the traveling wave method, can achieve fault location to a certain extent, but have limitations. The bridge method is mainly applicable to low-resistance faults and has poor effects on high-resistance faults and flashover faults; while the traveling wave method has a fast positioning speed and high accuracy, but the equipment cost is high, and the technical requirements for operators are also high. In addition, these methods often only rely on single current or voltage data and ignore other factors that may affect fault location, such as temperature and environmental conditions. Therefore, we propose a method for analyzing line fault location based on artificial intelligence. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for analyzing line fault location based on artificial intelligence. By constructing a tree-shaped undirected graph, installing a variety of sensors to collect data, determining sensitive features from historical data, constructing a multi-task joint autonomous learning model, and actively learning the model, etc., the accuracy of line fault location and the accuracy of fault type judgment are improved, and at the same time, the adaptability to complex and changeable environments is enhanced.

[0005] The technical problems solved by the present invention are: the problems of insufficient positioning accuracy, inaccurate fault type judgment, and inability to adapt to complex and changeable environments existing in the existing line fault location methods.

[0006] The present invention can be realized by the following technical solutions: A method for analyzing line fault location based on artificial intelligence, including the following steps:

[0007] Step 1: Based on the transmission line network, construct a tree-shaped undirected network graph, taking the substation and the transmission towers as the root node and child nodes respectively, and taking the modulus of the feature vector constructed by the line length, wire type, and impedance between each node as the length of the edge;

[0008] Step 2: Install data acquisition equipment in the transmission line to collect and process data, and collect the temperature of the line equipment, the temperature and humidity of the surrounding environment, the wind force, and the fault traveling wave occurring during the fault.

[0009] Step 3: Determine the characteristic data highly sensitive to the fault type and fault location from the historical operation and maintenance data, including traveling wave characteristic data and environment - related data, and use the ReliefF algorithm to screen the sensitive features;

[0010] Step 4: Build a multi - task joint self - learning model. The input includes traveling wave characteristic data and environment - related data, and the output is fault type classification and fault location. The model uses the PyTorch framework, and a shared layer and two task - branch layers are designed, which are used for fault type classification and fault location regression respectively;

[0011] Step 5: Analyze the real - time collected data using the multi - task joint self - learning model, output the fault type and fault location, and perform visual marking in the tree - shaped undirected network graph to assist the operation and maintenance department in quickly forming an operation and maintenance plan.

[0012] A further technical improvement of the present invention is that: the line length between each node is the actual length L of the transmission line under the current environment, and the calculation method is:

[0013] where l represents the horizontal distance between two towers, h represents the height difference between the connection points of the line between the two towers, g is the specific load of the overhead line; δ0 is the stress value at the lowest point of the overhead line, and its calculation method is:

[0014]

[0015] In the formula, g m is the specific load of the overhead line under the meteorological conditions of the maximum stress, T m is the temperature of the overhead line at the maximum stress, [δ] is the allowable stress of the overhead line when the specific load is g m and the temperature is T m . E is the elastic modulus, and α is the linear expansion coefficient.

[0016] A further technical improvement of the present invention is that: the installation density of the environmental sensor array for detecting environmental temperature, humidity and wind force in Step 2 is set according to the terrain and topography and the installation detection radius:

[0017] As long as the terrain and topography change, install the environmental sensor array at the corresponding node;

[0018] In the same terrain and topography, install only one environmental sensor array among the nodes within the set installation detection radius.

[0019] A further technical improvement of the present invention is that: in Step 3, the analysis and processing of the traveling wave characteristic data include:

[0020] The wavelet packet decomposition algorithm is used to decompose the fault traveling wave signal to obtain the energy distribution in the high-frequency band, and the energy distribution is related to the fault type;

[0021] Calculate the wavefront distortion coefficient, which is related to the fault location along with the waveform state;

[0022] According to the node range of the received traveling wave signals within a certain time neighborhood after the earliest time of receiving the traveling wave, calculate the traveling wave time difference between any two different nodes, and then obtain the fault location points to be arranged to determine the fault line to be arranged.

[0023] A further technical improvement of the present invention lies in: analyzing and processing the environment-related data in step three, including:

[0024] Calculate the resistance temperature drift rate of the line equipment: wherein, R0 is the resistance value at temperature T0, α is the coefficient of change of the resistance of the line equipment with temperature, T is the current temperature of the line equipment, and ΔR is the drift resistance;

[0025] Calculate the insulation performance index of the line equipment:

[0026] wherein, V peak represents the peak voltage, tanδ(T,H) is a correction function with respect to temperature and humidity, representing the loss characteristics of the material under different temperature and humidity conditions, T and H respectively represent temperature and humidity, and μ represents the humidity influence coefficient.

[0027] A further technical improvement of the present invention lies in: when constructing the multi-task joint autonomous learning model, feature weighting is performed on the features with high sensitivity according to the sensitivity evaluation results of the fault type or fault location distance in different task branch layers, and the influence proportion of this feature on the output result is enlarged.

[0028] A further technical improvement of the present invention lies in: the multi-task joint autonomous learning model conducts active learning once a month, collects the data samples with the fault type not matching the actual situation and the fault location accuracy exceeding the set value in the output results, and performs incremental update.

[0029] A further technical improvement of the present invention lies in: the fault location distance in the output result of the multi-task joint autonomous learning model is calibrated according to the actual length of the transmission line in step one.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. The present invention constructs a tree-shaped undirected graph, considers characteristic parameters such as the actual length of the line, wire type, and impedance, and uses a multi-task joint self-learning model to weight sensitive features, thereby improving the accuracy of line fault location; through intelligent data collection and analysis, it reduces the frequency and cost of manual inspections. At the same time, through accurate fault location, it reduces unnecessary line repairs and maintenance work.

[0032] 2. The present invention can simultaneously achieve fault type classification and fault location through a multi-task joint self-learning model, reduce the model complexity, and improve the analysis efficiency; at the same time, through incremental learning and dynamic weight adjustment, the model can adapt to equipment aging and environmental changes, avoid the problem of "catastrophic forgetting", and maintain high performance in the long term.

[0033] 3. The present invention can significantly shorten the fault handling time and improve the reliability and stability of the power grid by visualizing and marking the fault location and quickly forming an operation and maintenance plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings.

[0035] Figure 1 It is a schematic diagram of the execution flow of the method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0036] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, will detail the specific implementation manners, structures, features, and effects according to the present invention.

[0037] Please refer to Figure 1 As shown, a method for analyzing line fault location based on artificial intelligence specifically includes the following steps:

[0038] Step 1: Construct a tree-shaped undirected graph based on the transmission line network

[0039] Construct an undirected graph G = [V, E], where V represents the set of points and E represents the set of edges; G is a binary tuple composed of the set of points V = {v i} and the set of edges E = {e j}, the element v i in V is called a vertex, the element e j in E is called an edge, and the number of edges associated with a certain vertex v i is set to s(v i );

[0040] Taking the substations in the transmission line network as the root nodes, and the transmission towers and terminal loads as the child nodes to form a point set, constructing a feature vector with the characteristic parameters such as the line length, conductor type, and impedance between the nodes of the transmission line network, and taking the modulus of this feature vector as the length of the edge;

[0041] Among them, the conductor types include copper conductors, aluminum conductors, and aluminum conductor steel-reinforced conductors, and three corresponding characteristic values are set and allocated for the unit conductor values of each different type;

[0042] The length of the transmission line is not simply equivalent to the distance between the towers. Since the conductor is suspended between the towers in a catenary shape, the actual length L of the transmission line is:

[0043]

[0044] where l represents the horizontal distance between two towers, h represents the height difference between the attachment points of the line on the two towers, and g is the specific weight of the overhead line;

[0045] δ0 is the stress value at the lowest point of the overhead line, and its calculation method is:

[0046]

[0047] In the formula, g m is the specific weight of the line under the maximum stress meteorological conditions of the overhead line, T m is the temperature of the overhead line at the maximum stress, [δ] is the allowable stress of the overhead line when the specific weight is g m and the temperature is T m , E is the elastic modulus, and α is the coefficient of linear expansion.

[0048] Step 2: Install a variety of sensors in the transmission line and perform data collection and preprocessing

[0049] Install infrared thermal imagers and environmental sensor arrays at the positions of the root nodes and child nodes in the tree-shaped network diagram; among them, the infrared thermal imagers are used to detect the temperature of the line equipment, and the environmental sensor array includes temperature, humidity sensors, and wind speed sensors;

[0050] The installation density of the environmental sensor array is lower than that of other sensors. The environmental sensor array is selected for installation according to the changes in the terrain and landform. In the same terrain and landform, an installation detection radius is set. After exceeding the range of this radius, the environmental sensor array can be selected to be installed again to cover the detection range; in different terrains and landforms, this installation detection radius does not need to be considered. As long as the terrain and landform change, the environmental sensor array is installed at the corresponding nodes; due to installation cost issues, traveling wave recorders are generally installed near the root node position;

[0051] Use an infrared thermal imager to collect the working temperatures of transmission line equipment (such as transformers, circuit breakers, insulator strings) in real time. Additionally, since environmental data generally changes gradually, the environmental sensor array collects data at regular time intervals; the traveling wave recorder will record the arrival time of the wavefront of the fault current traveling wave in the line when a line fault occurs;

[0052] Upload the collected data to the cloud database based on 5G communication technology. Before storage, denoise the data, delete duplicate and redundant data, and use interpolation to fill in missing data. At the same time, perform intelligent and normalization processing on the data to eliminate the difference in data dimensions.

[0053] Step 3: Determine the characteristic data that is highly sensitive to the fault type and location from the historical operation and maintenance data, including:

[0054] Traveling wave characteristic data:

[0055] According to the transient traveling wave signal based on time series received by the traveling wave recorder when a specific fault occurs in the historical data, use the wavelet packet decomposition algorithm to decompose the traveling wave signal to obtain the energy distribution in the decomposed high-frequency band:

[0056] If the energy distribution in the high-frequency band is symmetric and concentrated, it indicates single-phase grounding;

[0057] If the energy distribution in the high-frequency band is concentrated but asymmetric, it indicates interphase short circuit;

[0058] If there are continuous harmonic components in the energy distribution in the high-frequency band, it indicates an arc fault.

[0059] At the same time, calculate the wavefront distortion coefficient. Generally, the Zernike polynomial is used to simulate and describe the wavefront distortion. When the wavefront distortion coefficient is greater than 0.25 and the wavefront steepness increases, the fault point occurs at the tower; when the wavefront even coefficient is less than 0.15 and the waveform is relatively smooth, the fault point occurs at the line position between two towers.

[0060] Furthermore, extract the moments when the transient current traveling wave signal at the time of fault occurs is received by each root node in the transmission line network from the historical data. Select the root node that receives the traveling wave earliest at time t0, and summarize the set of root nodes that receive this fault traveling wave and whose reception times are within the time neighborhood t0 + Δt. Assume there are k root nodes in the root node set. Arbitrarily select two different root nodes from the root nodes to form a group of double-ended nodes, and thus groups of double-ended nodes;

[0061] Obtain the time difference of the received fault traveling waves at each group of double-ended nodes, so as to obtain the possible fault points on the line between the two nodes, and mark them as the fault points to be arranged. If there are fault points to be arranged on the transmission line between the corresponding two nodes (not necessarily the root nodes) in the tree-shaped network diagram, then mark this section of the line and number it. If all the fault points to be arranged are located in the same section of the line, then the fault point can be directly determined to be located in this section of the line.

[0062] Environment-related data:

[0063] The temperature change of the line equipment will affect the resistance change of the line equipment. The calculation method of the resistance temperature drift rate is as follows: Among them, R0 is the resistance value at temperature T0, α is the temperature change coefficient of the resistance of the line equipment, T is the current temperature of the line equipment, and ΔR is the drift resistance;

[0064] The changes in temperature and humidity will also affect the insulation performance of the line equipment. The calculation formula of its insulation performance index PI is:

[0065] Among them, V peak represents the peak voltage, tanδ(T,H) is a correction function regarding temperature and humidity, represents the tangent value of the loss angle, represents the loss characteristics of the material under different temperature and humidity conditions, T and H respectively represent temperature and humidity, and μ represents the humidity influence coefficient;

[0066] Perform sensitivity quantization evaluation on the energy distribution, wavefront distortion coefficient, traveling wave arrival time difference and the set of lines to be arranged in the above traveling wave characteristic data, the temperature data, humidity data, drift resistance and insulation performance index in the environment-related data, and the wind force data (including wind direction and wind speed). Use the ReliefF algorithm to screen the sensitive features and obtain the set of sensitive features with the highest sensitivity weight for a certain fault type or a certain fault location distance.

[0067] Step Four: Construct a multi-task joint autonomous learning model

[0068] The model input includes traveling wave characteristic data and environment-related data; the model output is fault type classification (including single-phase grounding, phase-to-phase short circuit, arc fault, and open-circuit fault, etc.) and fault location (fault example). Extract the relevant data situation when a fault occurs in the historical operation and maintenance database to construct a training set and a test set in a ratio of 7:3;

[0069] The model adopts the PyTorch deep learning framework and designs a shared layer and two task branch layers. The shared layer extracts common features in the input based on a convolutional neural network. The two task branch layers are respectively a fault type classification task (using a fully connected layer and a softmax activation function) and a fault location regression task (using a fully connected layer and a linear activation function). In different task branch layers, according to the sensitivity evaluation results for the fault type or fault location distance in step three, feature weighting is performed on highly sensitive features to enhance the expressive ability of the corresponding features, thereby expanding their influence proportion on the output results.

[0070] For the fault type classification task, the cross-entropy loss function is used; for the fault location regression task, the mean squared error loss function is used; the loss functions of the two tasks are weighted and summed to obtain the total loss function and convergence analysis is carried out. The selection of the weight can be dynamically adjusted according to the difficulty of the task.

[0071] The model is trained until the classification accuracy and location accuracy reach the set values and then stopped to obtain a multi-task joint autonomous learning model based on the deep learning framework.

[0072] The model conducts active learning once a month, collects data samples with mismatched fault types and actual situations and fault location accuracies exceeding the set values in the output results, and performs incremental updates. The role of incremental updates can prevent the "catastrophic forgetting problem" caused by changes in important parameters of the previous model due to equipment aging and environmental changes.

[0073] Step Five: Use the multi-task joint autonomous learning model to analyze and simulate data

[0074] When a fault occurs, after the collected electrical signals are denoised, the model input data is sorted out. The multi-task joint autonomous learning model outputs the fault type and the fault occurrence location, and visual markings are made in the tree-shaped network diagram. The operation and maintenance department quickly forms an operation and maintenance plan based on the fault type and the fault occurrence location, and coordinates operation and maintenance personnel and operation and maintenance tools for quick operation and maintenance processing; it should be noted that the fault location distance is calibrated according to the actual length of the transmission line in step one.

[0075] The above is only a preferred embodiment of the present invention, and does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to be equivalent change equivalent embodiments within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A line fault location and analysis method based on artificial intelligence, characterized by: The steps include: Step 1: Construct a tree-like undirected network graph based on the transmission line network, with substations and transmission towers as root nodes and child nodes respectively, and the modulus of the characteristic vector constructed by the line length, conductor type, and impedance between nodes as the edge length; Step 2: Install data acquisition equipment in the transmission line to collect and process data, including the temperature of the line equipment, the temperature and humidity of the environment, wind speed, and the fault traveling wave caused by the fault; Step 3: Determine characteristic data that is highly sensitive to the fault type and location from historical operation and maintenance data, including traveling wave characteristic data and environmental correlation data, and use the Reli efF algorithm to screen sensitive features; Step 4: Build a multi-task joint autonomous learning model. The input includes traveling wave feature data and environmental correlation data, and the output is fault type classification and fault location. The model uses the PyTorch framework and designs a shared layer and two task branch layers for fault type classification and fault location regression respectively. Step 5: Use the multi-task joint autonomous learning model to analyze the real-time collected data, output the fault type and location, and visually mark them in the tree-like undirected network diagram to assist the operation and maintenance department in quickly formulating an operation and maintenance plan.

2. The method for analyzing and locating line faults based on artificial intelligence according to claim 1, characterized in that, The line length between each node is the actual length L of the transmission line in the current environment, which is calculated as follows: Where l represents the horizontal distance between the two towers, h represents the height difference between the tie points of the line on the two towers, g is the specific load of the overhead line, and δ0 is the stress value at the lowest point of the overhead line, which is calculated as follows: Where, g m is the specific weight of the overhead line under the meteorological conditions of the maximum stress, T m is the temperature of the overhead line at the maximum stress, [δ] is the allowable stress of the overhead line when the specific weight is g m and the temperature is T m , and E is the elastic modulus and α is the coefficient of linear expansion.

3. A method for analyzing line fault location based on artificial intelligence according to claim 1, characterized in that, In step 2, the installation density of the environmental sensor array used to detect ambient temperature, humidity, and wind speed is set according to the terrain and the installation detection radius: Whenever the terrain changes, an environmental sensor array is installed at the corresponding node; In the same terrain, only one environmental sensor array is installed in each node within the set installation detection radius.

4. An analysis method for line fault location based on artificial intelligence according to claim 1, characterized in that In step three, the traveling wave characteristic data is analyzed and processed, including: The wavelet packet decomposition algorithm is used to decompose the fault traveling wave signal to obtain the energy distribution in the high frequency band. The energy distribution is related to the fault type. Calculate the wavefront distortion coefficient, which is related to the waveform state and the fault location; According to the range of nodes that receive the traveling wave signal within a certain time neighborhood after the earliest time of receiving the traveling wave, the traveling wave time difference between any two different nodes is calculated, and then the fault point to be repaired is obtained, thereby determining the fault line to be repaired.

5. A method for analyzing line fault location based on artificial intelligence according to claim 4, characterized in that, In step 3, the environmental related data is analyzed and processed, including: Calculate the resistance temperature drift rate of the line equipment: where R0 is the resistance value at temperature T0, α is the temperature change coefficient of the resistance of the line equipment, T is the current temperature of the line equipment, and ΔR is the drift resistance; Calculate the insulation performance index of circuit equipment: Among them, V peak represents the peak voltage, tanδ(T,H) is the correction function for temperature and humidity, which represents the loss characteristics of the material under different temperature and humidity conditions. T,H represent temperature and humidity respectively, and μ represents the humidity influence coefficient.

6. The line fault location and analysis method based on artificial intelligence according to claim 1, characterized in that: When constructing a multi-task joint autonomous learning model, the sensitivity evaluation results of the fault type or fault location distance in different task branch layers are used to weight the highly sensitive features to increase the influence of the feature on the output result.

7. The line fault location and analysis method based on artificial intelligence according to claim 2, characterized in that: The multi-task joint autonomous learning model performs active learning once a month, collects data samples in which the fault type in the output results does not match the actual situation and the fault location accuracy exceeds the set value, and performs incremental updates.

8. An analysis method for line fault location based on artificial intelligence according to claim 7, characterized in that, The fault location distance in the output result of the multi-task joint autonomous learning model is calibrated according to the actual length of the transmission line in step 1.

Images