10kV overhead line state sensing and fault inspection decision-making method and system

Abstract

The invention discloses a 10kV overhead line state perception and fault inspection decision method and system, and relates to the technical field of high-voltage circuit inspection, and the method comprises the following steps: S1, real-time data collection, S2, real-time data transmission, S3, data preprocessing, S4, data feature extraction, S5, data feature fusion, and S6, fault prediction and decision making. 4, features extracted in the step 4 are fused with features selected manually, the learning ability of the model can be improved, decision optimization can be effectively carried out, the decision accuracy is improved, and in fault prediction and decision making, a test set is input into the trained random forest model, so that the fault prediction efficiency is improved. The current, the voltage, the temperature, the humidity, the weight vectors of the extracted characteristic values and the manually selected weight vectors are used as indexes, the weight vectors of the extracted characteristic values are continuously increased, the manually selected weight vectors are continuously reduced, the learning ability of the model is further improved, and the accuracy of decision making is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of high-voltage circuit inspection, and in particular to a 10kV overhead line state perception and fault inspection decision-making method and system. Background Art

[0002] With the continuous development of modern science and technology, my country's high-voltage transmission network has also been developing continuously. As people's dependence on the power grid is getting higher and higher, all the fault requirements for the transmission network are also higher.

[0003] According to the patent number: CN118504803B-A high-voltage transmission line inspection method and system, which records "fitting discrete data space into continuous data space, designing differentiable algorithms, and reducing the amount of calculation in the model construction process; different from the common model self-optimization algorithm that can only adapt to a specific network architecture, the present invention designs a directed graph composed of an ordered sequence of N nodes, which can flexibly form various structures, and can update parameters in different heterogeneous architecture models to achieve autonomous iterative optimization of the model during use." From this, those skilled in the art can know that the prior art uses statistical models for processing. Although it can process large-scale data and is suitable for scenarios with high real-time requirements, if the data obeys a certain distribution, it may not be able to accurately describe the complexity and nonlinear relationship of high-voltage power grid failures, and it is highly sensitive to data, and noise data may affect the prediction results.

[0004] In summary, a 10kV overhead line state perception and fault inspection decision-making method and system were designed. Summary of the invention

[0005] In order to overcome the above-mentioned shortcomings, the present invention provides a 10kV overhead line state perception and fault inspection decision-making method and system.

[0006] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0007] A 10kV overhead line state perception and fault inspection decision-making method comprises the following steps:

[0008] S1, real-time data collection, responsible for measuring and collecting data on current, voltage, temperature and humidity in 10kV overhead lines through sensor networks and non-contact detection networks;

[0009] S2, real-time data transmission, is responsible for data stream transmission through the communication network and transmits data to the cloud platform;

[0010] S3, data preprocessing, preprocessing the data through data cleaning and data normalization;

[0011] S4, data feature extraction, feature extraction of input data through convolutional neural network to capture local features in the data;

[0012] S5, data feature fusion, fusing the features extracted in step S4 with the manually selected features;

[0013] S6. Fault prediction and decision-making: predict the fused features, output the probability of the fault mode and make a decision.

[0014] Preferably, in the step S3, the data cleaning step adopts the 3σ principle to clean the data, the data is normally distributed, the mean μ and standard deviation σ of the data are calculated, and then the points with |Z|>3 are defined as abnormal points, where

[0015] Preferably, in the step S3, the data normalization step is performed by means of minimum and maximum standardization, and the original data is linearly transformed to the interval [0,1], and the formula is: Where x is the original data, min is the minimum value in the data, max is the maximum value in the data, and x 1 It is normalized data, which is simple and easy to use. However, when new data is added, the maximum and minimum values ​​may change and need to be redefined.

[0016] Preferably, in step S4, the operation steps of data feature extraction are as follows:

[0017] S41. Design the convolution layer, use a 2×2 convolution kernel, and use current, voltage, temperature, and humidity as input data;

[0018] S42, add ReLU activation function, add ReLU activation function after the convolution layer to introduce nonlinear factors and enhance the expressiveness of the model;

[0019] S43, setting a pooling layer, using a maximum pooling layer to downsample the feature map, reducing the amount of calculation while retaining important feature information;

[0020] S44, add a fully connected layer. After the convolution layer and the pooling layer, add one or more fully connected layers to map the extracted features to the final output category;

[0021] S45, output layer, after the fully connected layer, add an output layer to output the probability of a fault occurring, and the output probability can be between 0 and 1.

[0022] Preferably, in step S5, the extracted feature value is x i , the weight vector of the extracted eigenvalues ​​is ω i , the artificially selected feature value is y j, the manually selected weight vector is p j , the weighted summation result is

[0023] Preferably, in the step S6, a random forest algorithm step is used to label the fused feature vector in the step S5, i.e., the actual probability of a fault occurring or a binary label of whether a fault occurs, 0 indicates that no fault occurs, and 1 indicates that a fault occurs. The labeled data is put into a data set, which includes a training set and a test set, and a random forest is set at the same time. The random forest parameters are manual selection, intelligent selection, and failure rate. Randomly selected samples are put back from the training set to form multiple sub-training sets, each of which constitutes a decision tree. The test set is input into the trained random forest model, and current, voltage, temperature, humidity, and the weight vector of the extracted eigenvalues ​​and the weight vector of the manually selected are used as indicators, wherein the weight vector of the extracted eigenvalues ​​is continuously increased, and the weight vector of the manually selected is continuously decreased, so as to make the model more intelligent.

[0024] A 10kV overhead line state perception and fault inspection decision system as described above includes a terminal layer, a transmission layer and an application layer. The terminal layer includes a sensor network and a non-contact detection network, the transmission layer includes a communication network, and the application layer includes a cloud platform. The sensor network and the non-contact detection network collect data and transmit the data to the cloud platform through a wireless communication network. The cloud platform performs overhead line state perception and fault inspection decision-making.

[0025] Preferably, the communication network adopts LoRa communication, GPRS communication or wireless radio frequency.

[0026] The beneficial effects of the present invention are as follows: in the 10kV overhead line state perception and fault inspection decision method and system:

[0027] 1. In data feature fusion, the features extracted in step S4 are fused with the manually selected features to increase the learning ability of the model, effectively optimize the decision, and improve the accuracy of the decision;

[0028] 2. In fault prediction and decision-making, the test set is input into the trained random forest model, and the current, voltage, temperature and humidity, as well as the weight vectors of the extracted eigenvalues ​​and the manually selected weight vectors are used as indicators. The weight vectors of the extracted eigenvalues ​​continue to increase, while the weight vectors of the manually selected weight vectors continue to decrease, which further increases the learning ability of the model and improves the accuracy of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0030] Figure 1 It is a diagram of the method steps of the present invention;

[0031] Figure 2 It is a step diagram of data feature extraction of the present invention;

[0032] Figure 3 It is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0033] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0034] like Figure 1 and Figure 2 As shown, a 10kV overhead line state perception and fault inspection decision method includes the following steps:

[0035] S1, real-time data collection, responsible for measuring and collecting data on current, voltage, temperature and humidity in 10kV overhead lines through sensor networks and non-contact detection networks;

[0036] S2, real-time data transmission, is responsible for data stream transmission through the communication network and transmits data to the cloud platform;

[0037] S3, data preprocessing, preprocessing the data through data cleaning and data normalization;

[0038] The 3σ principle is used in the data cleaning step. The data is normally distributed. The mean μ and standard deviation σ of the data are calculated. Then, the points with |Z|>3 are defined as abnormal points.

[0039] In the data normalization step, the data is sorted by minimum and maximum standardization, and the original data is linearly transformed to the [0,1] interval. The formula is Where x is the original data, min is the minimum value in the data, max is the maximum value in the data, and x 1 It is normalized data, which is simple and easy to use. However, when new data is added, the maximum and minimum values ​​may change and need to be redefined.

[0040] S4, data feature extraction, feature extraction of input data through convolutional neural network to capture local features in the data;

[0041] In step S4, the operation steps of data feature extraction are as follows:

[0042] S41. Design the convolution layer, use a 2×2 convolution kernel, and use current, voltage, temperature, and humidity as input data;

[0043] S42, add ReLU activation function, add ReLU activation function after the convolution layer to introduce nonlinear factors and enhance the expressiveness of the model;

[0044] S43, setting a pooling layer, using a maximum pooling layer to downsample the feature map, reducing the amount of calculation while retaining important feature information;

[0045] S44, add a fully connected layer. After the convolution layer and the pooling layer, add one or more fully connected layers to map the extracted features to the final output category;

[0046] S45, output layer, after the fully connected layer, add an output layer to output the probability of a fault occurring, and the output probability can be between 0 and 1;

[0047] S5, data feature fusion, the features extracted in step S4 are fused with the manually selected features, and the extracted feature value is x i , the weight vector of the extracted eigenvalues ​​is ω i , the artificially selected feature value is y j , the manually selected weight vector is p j , the weighted summation result is

[0048] S6, fault prediction and decision-making, predict the fused features, output the probability of the fault mode and make a decision, use the random forest algorithm step, label the fused feature vector in step S5, that is, the actual probability of the fault or the binary label of whether the fault occurs, 0 means no fault occurs, 1 means fault occurs. Put the labeled data into the data set, which includes the training set and the test set, and set the random forest at the same time. The random forest parameters are manual selection, intelligent selection and failure rate. Put back the random samples from the training set to form multiple sub-training sets. Each training set constitutes a decision tree. Input the test set into the trained random forest model, use the current, voltage, temperature and humidity, as well as the weight vector of the extracted eigenvalues ​​and the weight vector of the manual selection as indicators, where the weight vector of the extracted eigenvalues ​​increases continuously and the weight vector of the manual selection decreases continuously to make the model more intelligent.

[0049] The working principle of the invention is:

[0050] S1, real-time data collection, responsible for measuring and collecting data on current, voltage, temperature and humidity in 10kV overhead lines through sensor networks and non-contact detection networks;

[0051] S2, real-time data transmission, is responsible for data stream transmission through the communication network and transmits data to the cloud platform;

[0052] S3, data preprocessing, cleaning the parameters with large errors in the collected data through data cleaning, and organizing the data through data normalization method;

[0053] S4, data feature extraction;

[0054] S41. Design the convolution layer, use a 2×2 convolution kernel, and use current, voltage, temperature and humidity as input data to form x i The value is between -1 and 1, I is the standard current value, Ii is the i-th current value, U is the standard voltage value, Ui is the i-th voltage value, T is the standard temperature value, Ti is the i-th temperature value, P is the standard humidity value, Pi is the i-th humidity value, x i is the i-th eigenvalue, ω 1 ,ω 2 ,ω 3 ,ω 4 Respective weight coefficients;

[0055] S42, add ReLU activation function, add ReLU activation function after the convolution layer, the ReLU function will input value x i Mapped to max(0,x i ), that is, when x i Output x between -0.5 and 0.5 i itself, when x i Output 1 when it is greater than 0.5 or less than -0.5;

[0056] S43, setting a pooling layer to downsample the feature map, reducing the amount of calculation;

[0057] S44, add a fully connected layer to map the feature values ​​in step S43 to the output, so as to facilitate the output in S45;

[0058] S45, output layer, outputs the probability of failure occurrence, and the output probability can be between 0 and 1;

[0059] S5, data feature fusion, the features extracted in step S4 are fused with the manually selected features. Here, according to the work experience of the staff, a set of their own feature values ​​is designed. The weight vector selected manually is p j , which enables the entire system model to be manually educated, speeding up the model training progress and accuracy;

[0060] S6. Fault prediction and decision making.

[0061] like Figure 3As shown, a 10kV overhead line state perception and fault inspection decision-making system includes a terminal layer, a transmission layer and an application layer. The terminal layer includes a sensor network and a non-contact detection network, the transmission layer includes a communication network, and the application layer includes a cloud platform. The sensor network and the non-contact detection network collect data and transmit the data to the cloud platform through a wireless communication network. The cloud platform performs overhead line state perception and fault inspection decision-making. The communication network adopts LoRa communication, GPRS communication or wireless radio frequency.

[0062] The above is based on the present invention as an inspiration. Through the above description, relevant staff can make various changes and modifications without departing from the technical idea of ​​this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A 10kV overhead line state perception and fault inspection decision method, characterized by: The following steps are involved: S1, real-time data collection, responsible for measuring and collecting data on current, voltage, temperature and humidity in 10kV overhead lines through sensor networks and non-contact detection networks; S2, real-time data transmission, is responsible for data stream transmission through the communication network and transmits data to the cloud platform; S3, data preprocessing, preprocessing the data through data cleaning and data normalization; S4, data feature extraction, feature extraction of input data through convolutional neural network to capture local features in the data; S5, data feature fusion, fusing the features extracted in step S4 with the manually selected features; S6. Fault prediction and decision-making: predict the fused features, output the probability of the fault mode and make a decision.

2. A 10kV overhead line state perception and fault inspection decision method according to claim 1, characterized in that: In the step S3, the 3σ principle is used in the data cleaning step to clean the data. The data is normally distributed, the mean μ and standard deviation σ of the data are calculated, and then the points with |Z|>3 are defined as abnormal points, where 3. A 10kV overhead line state perception and fault inspection decision method according to claim 1, characterized in that: In the S3 step, the data normalization step is performed by minimum and maximum standardization to linearly transform the original data into the interval [0,1], and the formula is: Where x is the original data, min is the minimum value in the data, max is the maximum value in the data, and x 1 is the normalized data.

4. A 10kV overhead line state perception and fault inspection decision method according to claim 1, characterized in that: In step S4, the operation steps of data feature extraction are as follows: S41. Design the convolution layer, use a 2×2 convolution kernel, and use current, voltage, temperature, and humidity as input data; S42, add ReLU activation function; S43, setting a pooling layer; S44, add a fully connected layer; S45, output layer.

5. A 10kV overhead line state perception and fault inspection decision method according to claim 1, characterized in that: In the step S5, the extracted feature value is x i , the weight vector of the extracted eigenvalues ​​is ω i , the artificially selected feature value is y j , the manually selected weight vector is p j , the weighted summation result is 6. A 10kV overhead line state perception and fault inspection decision method according to claim 1, characterized in that: In the step S6, a random forest algorithm is used to label the feature vectors fused in the step S5.

7. A 10kV overhead line state perception and fault inspection decision system according to any one of claims 1 to 6, characterized in that: It includes terminal layer, transmission layer and application layer. The terminal layer includes sensor network and contactless detection network, the transmission layer includes communication network, and the application layer includes cloud platform. The sensor network and contactless detection network collect data and transmit the data to the cloud platform through wireless communication network. The cloud platform perceives the status of overhead lines and makes fault inspection decisions.

8. A 10kV overhead line state perception and fault inspection decision system according to claim 7, characterized in that: The communication network adopts LoRa communication, GPRS communication or wireless radio frequency.

Images