CN118781469A - A method and system for identifying transmission line loss types - Google Patents

Abstract

The invention relates to a method and system for identifying loss types of power transmission lines, belonging to the technical field of loss type identification, and comprising: obtaining historical loss data of power transmission lines, and dividing the data into a training set, a verification set and a test set; constructing an improved CNN-LSTM neural network, adding a second convolutional layer to the improved CNN-LSTM neural network while retaining the original network structure, training the improved CNN-LSTM neural network with the training set, optimizing the parameters of the improved CNN-LSTM neural network with the verification set, and testing the improved CNN-LSTM neural network with the test set to obtain a final improved CNN-LSTM neural network as a loss identification model; obtaining loss data of the current power transmission line, inputting the data into the loss identification model, outputting an initial identification result, and processing the data using a weighted average method to obtain a final loss type identification result. The invention improves the problems of overfitting and slow calculation speed, improves the generalization ability of the neural network, eliminates noise in the result, and reduces the error.

Description

A method and system for identifying transmission line loss types

Technical Field

The present invention relates to the technical field of loss type identification, and in particular to a method and system for identifying the loss type of a power transmission line.

Background Art

The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

The current power system is becoming more and more complex. Line losses will occur in the power generation, distribution and transmission process of the power system, especially transmission line losses. In the process of generating electric energy and transmitting it to the user end, the power plant may lose electric energy in other forms such as heat energy due to various reasons. The electric energy in other surrounding media becomes transmission line loss. Generally, the causes of loss are divided into force majeure factors and human factors. Force majeure factors may include the loss caused by line resistance when electric energy passes through the transmission line. Human factors may include the theft of electricity by individuals or enterprises for their own interests. The loss of transmission lines will reduce the quality of power supply and bring economic and energy losses. Therefore, it is of great significance in this field to timely identify the type of transmission line loss and achieve early identification and early intervention.

With the continuous development of neural network technology, the technology of loss type recognition based on neural network is developing faster and faster. However, traditional neural networks, such as CNN, have defects such as gradient vanishing, overfitting, large amount of calculation and weak model generalization ability. How to extract deep features in the model and enhance the generalization ability of the model is one of the directions that technicians in this field are constantly exploring. In the study, it was also found that in the existing technical solutions, the final output results of the model are mixed with noise, which causes a large error between the recognition results and the actual situation. Therefore, how to extract deeper features, enhance the generalization ability of the model and solve the problem of noise in the output results is a technical problem that technicians in this field urgently need to solve.

Summary of the invention

In order to solve the above problems, the present invention proposes a method and system for identifying the type of transmission line loss. By improving the traditional neural network and adding a convolution layer on the basis of retaining the original convolution layer, the deep features can be effectively extracted, the generalization ability of the neural network is enhanced, and the output initial recognition result is processed to eliminate the interference of noise and improve the recognition accuracy.

In order to achieve the above object, the present invention adopts the following technical solution:

In a first aspect, the present invention provides a method for identifying a type of transmission line loss, comprising:

Acquire historical loss data of the transmission line, preprocess the acquired historical loss data of the transmission line, construct a historical loss data set, and divide the historical loss data set into a training set, a validation set, and a test set;

Constructing an improved CNN-LSTM neural network, wherein the improved CNN-LSTM neural network sequentially includes an input layer, a feature extraction layer, a second convolutional layer, an LSTM layer, a fully connected layer, and an output layer, wherein the feature extraction layer sequentially includes a first convolutional layer and a pooling layer, training the improved CNN-LSTM neural network using a training set, optimizing the parameters of the improved CNN-LSTM neural network using a validation set, and then testing and verifying the improved CNN-LSTM neural network using a test set, obtaining a final improved CNN-LSTM neural network, and using it as a loss recognition model;

The loss data of the current transmission line is obtained, input into the loss identification model, the initial identification result is output, and then the initial identification result is processed using the weighted average method to obtain the final loss type identification result.

According to a further technical solution, the historical loss data of the transmission line includes resistance data, inductance data, current data, voltage data, load data and power data, and the loss type identification result includes conductor resistance loss, inductance loss, insulation loss, load loss, no-load loss and power theft loss.

A further technical solution is to train the improved CNN-LSTM neural network using the training set, and the specific process is as follows:

The input layer processes the input training set into a first feature, and inputs it into the feature extraction layer for feature extraction, wherein the first convolution layer in the feature extraction layer extracts the first feature to obtain a first feature map, and then obtains a second feature map through an activation function; the pooling layer in the feature extraction layer performs secondary processing on the second feature map to obtain a third feature map and outputs it to the second convolution layer; the second convolution layer further extracts features from the third feature map to obtain a fourth feature map, and then obtains a fifth feature map through an activation function and inputs it to the LSTM layer; the LSTM layer automatically learns the feature information in the fifth feature map, and then completes the classification task through a fully connected layer, and finally outputs it by the output layer; the fully connected layer is also used to calculate the loss function, and uses the minimum loss function as the learning target.

In a further technical solution, the secondary processing of the second feature map includes data division and compression, and obtaining A third feature map, where is the number of the first feature, and y is the size of the pooling layer.

A further technical solution is described in which the parameters of the CNN-LSTM neural network are optimized and improved by using a validation set: the parameters of the CNN-LSTM neural network are optimized and improved by using a stochastic gradient descent method and a back propagation method, wherein the back propagation method performs several iterative calculations and solves the minimum loss function value according to the direction of the stochastic gradient descent method.

A further technical solution is to use a test set to test and verify the improved CNN-LSTM neural network, specifically including evaluating performance indicators based on the results of the test verification, and the performance indicators include precision, accuracy, F1 value and recall rate.

A further technical solution is that the method of processing the initial recognition result using the weighted average method is:

= ;

in, is the final loss type identification result obtained after weighted average processing. is the initial recognition result at time j, i is the time length, and k is the weight coefficient.

In a second aspect, the present invention provides a transmission line loss type identification system, comprising:

The data acquisition module is configured to: acquire historical loss data of the transmission line, preprocess the acquired historical loss data of the transmission line, construct a historical loss data set, and divide the historical loss data set into a training set, a validation set, and a test set;

The network construction module is configured to: construct an improved CNN-LSTM neural network, wherein the improved CNN-LSTM neural network includes an input layer, a feature extraction layer, a second convolutional layer, an LSTM layer, a fully connected layer and an output layer in sequence, wherein the feature extraction layer includes a first convolutional layer and a pooling layer in sequence, train the improved CNN-LSTM neural network using a training set, optimize the parameters of the improved CNN-LSTM neural network using a validation set, and test and verify the improved CNN-LSTM neural network using a test set to obtain a final improved CNN-LSTM neural network as a loss recognition model;

The identification module is configured to: obtain the loss data of the current transmission line, input it into the loss identification model, output the initial identification result, and then use the weighted average method to process the initial identification result to obtain the final loss type identification result.

Compared with the prior art, the present invention has the following advantages and beneficial effects:

The present invention improves the traditional CNN-LSTM neural network. On the basis of retaining the original network structure, a second convolution layer is added between the feature extraction layer and the LSTM layer. The second convolution layer can extract deep features of the model and effectively enhance the generalization ability of the model. In addition, the present application performs data division and compression in the pooling layer, which reduces the risk of overfitting to a certain extent and speeds up the calculation speed. The loss identification model obtained in this way improves the problems of overfitting and slow calculation speed in the process of transmission line loss identification, and also improves the generalization ability of the neural network.

The present invention processes the initial recognition results output by the loss recognition model using the weighted average method and uses the processed results as the final loss type recognition results. This can effectively eliminate the noise in the initial recognition results, reduce the error between the final loss type recognition results and the actual situation, and improve the recognition accuracy.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their description are used to explain the present invention but do not constitute a limitation of the present invention.

FIG1 is a flow chart of a method according to a first embodiment of the present invention.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

Embodiment 1

As shown in FIG1 , this embodiment proposes a method for identifying a type of power transmission line loss, which specifically includes the following steps:

S1: Acquire historical loss data of the transmission line, preprocess the acquired historical loss data of the transmission line, construct a historical loss data set, and divide the historical loss data set into a training set, a validation set, and a test set;

S2: construct an improved CNN-LSTM neural network, wherein the improved CNN-LSTM neural network includes an input layer, a feature extraction layer, a second convolutional layer, an LSTM layer, a fully connected layer and an output layer in sequence, wherein the feature extraction layer includes a first convolutional layer and a pooling layer in sequence, train the improved CNN-LSTM neural network using a training set, optimize the parameters of the improved CNN-LSTM neural network using a validation set, and test and verify the improved CNN-LSTM neural network using a test set to obtain a final improved CNN-LSTM neural network as a loss recognition model;

S3: Obtain the loss data of the current transmission line, input it into the loss identification model, output the initial identification result, and then use the weighted average method to process the initial identification result to obtain the final loss type identification result.

Wherein, in step S1, the acquired historical loss data of the transmission line includes resistance data, inductance data, current data, voltage data, load data and power data, wherein the power data also includes active power and reactive power. The method of preprocessing the acquired historical loss data of the transmission line includes but is not limited to data segmentation and data cleaning, the purpose of which is to obtain enough samples for neural network learning, because the neural network requires the input data length to be consistent, so the historical loss data at different times must be segmented. The above historical loss data set is divided into a training set, a validation set and a test set in a ratio of 6:2;2.

In step S2, this embodiment improves the CNN-LSTM neural network by adding a new convolutional layer to the original network structure. The improved CNN-LSTM neural network in this embodiment includes an input layer, a feature extraction layer, a second convolutional layer, an LSTM layer, a fully connected layer and an output layer in sequence, wherein the feature extraction layer includes a first convolutional layer and a pooling layer in sequence.

In step S2, the improved CNN-LSTM neural network is trained using the training set. The specific process is: first, the historical loss data of the transmission line in the training set is input into the input layer. The input layer is the first layer of the improved CNN-LSTM neural network. The input layer can process the input historical loss data of the transmission line into a first feature, and input the first feature into the feature extraction layer for feature extraction, wherein the first convolution layer in the feature extraction layer extracts the first feature to obtain a first feature map. Specifically: the first convolution layer obtains a first feature, a is a freely definable parameter value, and the calculation process of obtaining the first feature map is as follows:

= × + ;in, is the first feature map, is the i-th first feature obtained by the first convolutional layer, is the convolution kernel on the i-th first feature, is the bias matrix.

Secondly, after obtaining the first feature map, normalization is performed and the normalized first feature map is The second feature map is obtained through the activation function. The activation function here selects the ReLU activation function. The calculation process of obtaining the second feature map is as follows:

=f（ ) = max(0, ),in, is the i-th second feature map, The first feature map after normalization.

In this embodiment, the pooling layer in the feature extraction layer selects the maximum pooling layer, and the pooling layer performs secondary processing on the acquired second feature map to obtain the third feature map, wherein the secondary processing includes data partitioning and compression, and the data partitioning can select rectangular segmentation, and the third feature map is obtained after data partitioning and compression. The third feature map is is the number of the first feature, and y is the size of the pooling layer. The calculation process for obtaining the third feature map is as follows:

=max{ (r)}, where For the The third feature map, ∈[1, ], (r) is the value of the rth neuron in the i-th second feature map, where r is y.

After obtaining the third feature map, it is output to the second convolutional layer. The second convolutional layer further extracts features from the third feature map to obtain the fourth feature map. It should be noted that the second convolutional layer is to extract deep features. The calculation process of the second convolutional layer is the same as the calculation process of the first convolutional layer in the feature extraction layer, which will not be repeated here. After obtaining the fourth feature map, the fifth feature map is obtained through the ReLU activation function and input into the LSTM layer. The LSTM layer automatically learns the feature information in the fifth feature map, and then completes the classification task through the fully connected layer, and is finally output by the output layer. The newly added second convolutional layer can extract deep features of the model and effectively enhance the generalization ability of the model. In addition, the present application performs data division and compression in the pooling layer, which reduces the risk of overfitting to a certain extent and speeds up the calculation speed. The loss recognition model obtained in this way improves the problems of overfitting and slow calculation speed in the process of transmission line loss identification, and also improves the generalization ability of the neural network.

In this embodiment, the fully connected layer is designed after the LSTM layer. After the LSTM layer learns the feature information in the fifth feature map, it is sent to the fully connected layer, and the possibility of the loss type is calculated using the softmax function. Specifically, the probability of input return is calculated using the softmax function, and the input is divided into any mutually exclusive class to obtain the probability string { , , …, ; , , …, ; , , …, }, where A is the number of input data, and A=p+1, B is the number of loss types, and B=q+1, p and q are mutually exclusive, It is the output of the b-th type of input data a, that is, the output of softmax, which can represent the possibility that the input data a of the neural network in this embodiment is type b, 1≤a≤A, 1≤b≤B. The fully connected layer is also used to calculate the loss function, and the minimum loss function is used as the learning goal.

At the same time, in step S2, the parameters of the CNN-LSTM neural network are optimized and improved using the validation set. The specific method is: the parameters of the CNN-LSTM neural network are continuously optimized and improved by the stochastic gradient descent method and the back propagation method, so that the improved CNN-LSTM neural network can better adapt to the characteristics of the historical loss data of the power transmission line. According to the direction of the stochastic gradient descent method, the back propagation method will perform several iterative calculations and solve the minimum loss function value. The back propagation method enables the loss function at the end of the improved CNN-LSTM neural network to obtain the optimal solution. The curve of the stochastic gradient descent method matches the loss function, and the direction of the gradient is the same as the direction of the loss function reduction. The gradient value is calculated, and the braking gradient value is repeatedly reset to zero. The gradient value of 0 corresponds to the minimum value point of the loss function, and the minimum loss function value is solved.

In step S2, the improved CNN-LSTM neural network is tested and verified using a test set. The specific method is: use an independent test set to test and verify the trained improved CNN-LSTM neural network, evaluate performance indicators such as precision, accuracy, F1 value and recall rate, so as to determine the loss identification model. The loss types include wire resistance loss, inductance loss, insulation loss, load loss, no-load loss and power theft loss. This embodiment compares and evaluates various performance indicators and finally determines the improved CNN-LSTM neural network as the loss identification model. The precision, accuracy, F1 value and recall rate of the loss identification model for various types of transmission line loss types are shown in Table 1:

Table 1

In step S3, the initial identification result is processed using the weighted average method, specifically: the loss data of the current transmission line is obtained, input into the above-mentioned loss identification model, and the initial identification result is output. It is found in the study that directly inputting the loss data of the current transmission line into the loss identification model for type identification will cause noise in the output result, resulting in a large error between the identification result and the actual situation. In this regard, this embodiment introduces a weighted average method to process the initial recognition result to eliminate the noise in the initial recognition result. The essence of weighted average is to replace the current result with the weighted average of the recognition result in the previous time as the current recognition result. According to the essence of weighted average, in this embodiment, the length of the time window is set to 5, that is, the weighted average of the current result and the recognition results of the previous 4 adjacent moments is calculated. If the current moment and the moment adjacent to it meet 5 time points, the weight coefficient is set to [1, 2, 3, 4, 5]. If the current moment and the moment adjacent to it do not meet 5 time points, the weight coefficient is set to [1, 2, ..., n], n < 5. The method of processing the initial recognition result using the weighted average method is:

= ;

in, is the final loss type identification result obtained after weighted average processing. is the initial recognition result at time j, i is the time length, and k is the weight coefficient.

Embodiment 2

This embodiment provides a transmission line loss type identification system, which specifically includes the following modules:

The data acquisition module is configured to: acquire historical loss data of the transmission line, preprocess the acquired historical loss data of the transmission line, construct a historical loss data set, and divide the historical loss data set into a training set, a validation set, and a test set;

The network construction module is configured to: construct an improved CNN-LSTM neural network, wherein the improved CNN-LSTM neural network includes an input layer, a feature extraction layer, a second convolutional layer, an LSTM layer, a fully connected layer and an output layer in sequence, wherein the feature extraction layer includes a first convolutional layer and a pooling layer in sequence, train the improved CNN-LSTM neural network using a training set, optimize the parameters of the improved CNN-LSTM neural network using a validation set, and test and verify the improved CNN-LSTM neural network using a test set to obtain a final improved CNN-LSTM neural network as a loss recognition model;

The identification module is configured to: obtain the loss data of the current transmission line, input it into the loss identification model, output the initial identification result, and then use the weighted average method to process the initial identification result to obtain the final loss type identification result.

The implementation of the specific modules in this embodiment refers to the steps of a method for identifying the type of transmission line loss described in Embodiment 1, and will not be described in detail here.

The above-described embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.

Claims (10)

1. The method for identifying the loss type of the power transmission line is characterized by comprising the following steps of:

Acquiring historical loss data of a power transmission line, preprocessing the acquired historical loss data of the power transmission line, constructing a historical loss data set, and dividing the historical loss data set into a training set, a verification set and a test set;

Constructing an improved CNN-LSTM neural network, wherein the improved CNN-LSTM neural network sequentially comprises an input layer, a feature extraction layer, a second convolution layer, an LSTM layer, a full connection layer and an output layer, the feature extraction layer sequentially comprises a first convolution layer and a pooling layer, the improved CNN-LSTM neural network is trained by using a training set, parameters of the improved CNN-LSTM neural network are optimized by using a verification set, and the improved CNN-LSTM neural network is tested and verified by using a test set to obtain a final improved CNN-LSTM neural network and serve as a loss identification model;

and acquiring loss data of the current power transmission line, inputting the loss data into a loss identification model, outputting an initial identification result, and processing the initial identification result by using a weighted average method to obtain a final loss type identification result.

2. The transmission line loss type identification method according to claim 1, wherein the transmission line historical loss data includes current data, voltage data, load data and power data, and the loss type identification result includes load loss, no-load loss and power theft loss.

3. The method for identifying the loss type of the power transmission line according to claim 1, wherein the training of the improved CNN-LSTM neural network by using the training set comprises the following specific steps:

The input layer processes the input training set into first features, and inputs the first features into the feature extraction layer for feature extraction, wherein a first convolution layer in the feature extraction layer extracts the first features to obtain a first feature map, and then an activation function is used to obtain a second feature map; the pooling layer in the feature extraction layer carries out secondary treatment on the second feature map to obtain a third feature map, and outputs the third feature map to the second convolution layer; the second convolution layer performs further feature extraction on the third feature map to obtain a fourth feature map, and then obtains a fifth feature map through an activation function and inputs the fifth feature map into the LSTM layer; the LSTM layer automatically learns the feature information in the fifth feature mapping, and then completes the classification task through the full-connection layer, and finally is output by the output layer; the fully connected layer is also used for calculating a loss function and taking the minimum loss function as a learning target.

4. A transmission line loss type identification method according to claim 3, wherein said secondary processing of the second feature map includes data division and compression, and obtainingA third feature map in whichFor the first number of features, y is the pooling layer size.

5. The transmission line loss type identification method according to claim 1, wherein the parameters of the CNN-LSTM neural network are improved by verification set optimization: parameters of the CNN-LSTM neural network are optimized and improved through a random gradient descent method and a back propagation method, wherein the back propagation method can perform iterative computation for a plurality of times according to the direction of the random gradient descent method, and the minimum loss function value is solved.

6. The method for identifying the loss type of the power transmission line according to claim 1, wherein the test set is used for testing and verifying the improved CNN-LSTM neural network, and specifically comprises evaluating performance indexes according to the test and verification result, wherein the performance indexes comprise accuracy rate, F1 value and recall rate.

7. The method for identifying the loss type of the power transmission line according to claim 1, wherein the method for processing the initial identification result by using a weighted average method comprises the following steps:

=；

Wherein, For the final loss type recognition result obtained after the weighted average method processing,And i is the time length, and k is the weight coefficient, which is the initial identification result of the moment j.

8. A transmission line loss type identification system, comprising:

A data acquisition module configured to: acquiring historical loss data of a power transmission line, preprocessing the acquired historical loss data of the power transmission line, constructing a historical loss data set, and dividing the historical loss data set into a training set, a verification set and a test set;

A network construction module configured to: constructing an improved CNN-LSTM neural network, wherein the improved CNN-LSTM neural network sequentially comprises an input layer, a feature extraction layer, a second convolution layer, an LSTM layer, a full connection layer and an output layer, the feature extraction layer sequentially comprises a first convolution layer and a pooling layer, the improved CNN-LSTM neural network is trained by using a training set, parameters of the improved CNN-LSTM neural network are optimized by using a verification set, and the improved CNN-LSTM neural network is tested and verified by using a test set to obtain a final improved CNN-LSTM neural network and serve as a loss identification model;

An identification module configured to: and acquiring loss data of the current power transmission line, inputting the loss data into a loss identification model, outputting an initial identification result, and processing the initial identification result by using a weighted average method to obtain a final loss type identification result.

9. A computer readable storage medium having stored thereon a program, which when executed by a processor, implements the steps of a transmission line loss type identification method according to any of claims 1-7.

10. An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor performs the steps of a transmission line loss type identification method according to any of claims 1-7 when the program is executed.