CN117031202A - K-SMOTE and depth forest based power transmission line fault multi-source diagnosis method and system - Google Patents

Abstract

The invention discloses a power transmission line fault multi-source diagnosis method and system based on K-SMOTE and deep forests, relates to the field of power transmission line fault diagnosis, and aims at solving the problems of unbalance classification and low diagnosis accuracy of the existing fault data set. Firstly, taking historical transmission line fault information of a national network as a database, dividing the historical transmission line fault information into multiple ranges of data sets according to fault types, wherein each data set is composed of time-frequency domain characteristics of transmission line fault voltage current waveform signals; secondly, clustering a few fault data sets by adopting a K-means method, expanding the data sets by using an SMOTE oversampling method, and balancing various fault sub-databases; finally, using the balanced fault sub-data set, and carrying out model training based on a depth forest algorithm to obtain a trained fault diagnosis model; and diagnosing the faults by using the trained fault diagnosis model. The method has good accuracy and effectiveness, and provides diagnosis with high accuracy and high coverage rate for the power transmission line faults.

Description

Multi-source diagnosis method for transmission line faults based on K-SMOTE and deep forest and system

Technical field

The present invention relates to the field of transmission and distribution line fault detection in power grid systems, and in particular to fault types and fault occurrence diagnosis. Specifically, it is a multi-source diagnosis method and system for transmission line faults based on K-SMOTE and deep forest.

Background technique

As China's power transmission and distribution system becomes increasingly large-scale, complex and intelligent, the operating characteristics of the power grid have undergone profound changes, and power grid dispatching is also facing new and major challenges. On the one hand, the power transmission and distribution system will be affected by factors such as regional climate variability and complex operating environments, and there is an urgent need to control the power grid's operating situation synchronously; on the other hand, the sending and receiving ends of the power grid are closely coupled with AC and DC, which can easily lead to overall security risks.

Therefore, power grid fault diagnosis is extremely important to assist the normal operation and dispatching of the power grid. When a power grid failure occurs, the massive fault alarm data collected by the monitoring system (covering correct alarm information, incorrect alarm information, repeated alarm information and irrelevant information) is sent from the local automatic device to the dispatch center. Existing fault diagnosis methods have unbalanced classification of fault data sets and have low fault diagnosis accuracy.

Contents of the invention

The present invention provides a multi-source diagnosis method and system for transmission line faults based on K-SMOTE and deep forest to solve the technical problems of unbalanced classification and low diagnosis accuracy of distribution line fault data sets in the prior art.

According to one aspect of the present invention, a multi-source diagnosis method for transmission line faults based on K-SMOTE and deep forest is provided. The method includes:

Obtain transmission line fault data;

Input the transmission line fault data into the trained fault diagnosis model to obtain fault diagnosis results;

The training of the fault diagnosis model includes:

Extract the time-frequency characteristics of voltage and current waveforms in historical fault data, classify faults based on the extracted time-frequency characteristics, and form fault waveform data sets of different fault types;

Use K-means clustering algorithm to cluster the unbalanced data set in the fault data set, and use SMOTE oversampling to expand the clustered unbalanced data set;

Form fault sub-data sets based on the balanced data set and the expanded unbalanced data set;

The fault sub-data set is used and model training is performed based on the deep forest algorithm to obtain a trained fault diagnosis model.

Preferably, the extracted time-frequency features include time domain, frequency domain and transient waveform features in the frequency domain.

Preferably, K-means clustering algorithm is used to cluster the imbalanced data set in the fault data set, further including:

Step 1: For the data set, extract k initial clustering center points as i ₁ , i ₂ ,..., i _k ;

Step 2: For other data in the data set except the cluster center, calculate the Euclidean of each data and i _j (j=1,2,...,k) through the formula c _i =argmin[|x _i -i _j |] ² distance, and classify the data closest to i, j into one category to obtain k category data;

Step 3: Calculate the average value of the data in each category, set the obtained average value as the central value of the category, and then use the formula in step 2 to calculate the sum of the Euclidean distances from each data point to the central value, calculated as S ;

Step 4: Repeat steps 2 and 3 until the result S does not change and output the K-means clustering subset.

Preferably, SMOTE oversampling is used to augment the clustered imbalanced data set, including:

Step 1: For each sample x of the minority class in the unbalanced data set, use the Euclidean distance as the standard, calculate its distance to all samples in the minority class sample set, and obtain its k nearest neighbors;

Step 2: Determine the sampling rate N. For each minority class sample x, randomly select several samples from its k nearest neighbors;

Step 3: Randomly select a nearest neighbor Xnew, which is compared with the original sample through the formula Construct a new sample; rand(0, N) is a random number generated between 0 and N,/> is the cluster center, and x is a sample of this type.

Preferably, the deep forest algorithm adopts a deep forest multi-granularity cascade forest model.

According to another aspect of the specification of the present invention, a transmission line fault multi-source diagnosis system based on K-SMOTE and deep forest is provided, which system includes:

Acquisition unit: obtain fault data;

Diagnosis unit: input the transmission line fault data into the trained fault diagnosis model to obtain fault diagnosis results;

Training unit: Extract the time-frequency characteristics of voltage and current waveforms in historical fault data, classify faults based on the extracted time-frequency characteristics and form fault data sets of different fault types;

Use K-means clustering algorithm to cluster the unbalanced data set in the fault data set, and use SMOTE oversampling to expand the clustered unbalanced data set;

Form fault sub-data sets based on the balanced data set and the expanded unbalanced data set;

The fault sub-data set is used and model training is performed based on the deep forest algorithm to obtain a trained fault diagnosis model.

Based on another aspect of the present invention, an electronic device is provided. The electronic device includes a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein the computer program is When the processor is executed, the steps of the multi-source diagnosis method for transmission line faults based on K-SMOTE and deep forest are implemented.

Based on another aspect of the present invention, a computer-readable storage medium is provided. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the K-SMOTE-based method is implemented. and steps of a multi-source diagnosis method for transmission line faults in Deep Forest.

Compared with the prior art, the beneficial effects of the present invention are:

The invention provides a multi-source diagnosis method and system for transmission line faults based on K-SMOTE and deep forest. By performing feature extraction and classification on transmission line faults and performing K-means cluster analysis to correct the unbalanced classification subset, it avoids It solves the problem of complex fault data classification and unbalanced data depth causing the SMOTE algorithm to strengthen noise data points when processing small samples, thereby improving the accuracy and coverage of subsequent deep forest-based training.

Description of the drawings

Figure 1 is a flow chart of a transmission line fault multi-source diagnosis method based on K-SMOTE and deep forest according to an embodiment of the present invention;

Figure 2 is a schematic diagram of a transmission line fault multi-source diagnosis system based on K-SMOTE and deep forest according to an embodiment of the present invention;

Figure 3 is a flow chart of the training process of the fault diagnosis model according to the embodiment of the present invention.

Detailed ways

The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

As shown in Figures 1 and 3, this embodiment provides a multi-source transmission line fault diagnosis method based on K-SMOTE and deep forest, including:

Obtain transmission line fault data;

Input the transmission line fault data into the trained fault diagnosis model to obtain fault diagnosis results;

The training of the fault diagnosis model includes:

Extract the time-frequency characteristics of voltage and current waveforms in historical fault data, classify faults based on the extracted time-frequency characteristics, and form fault waveform data sets of different fault types;

Use K-means clustering algorithm to cluster the unbalanced data set in the fault waveform data set, and use SMOTE oversampling to expand the clustered unbalanced data set;

Form fault sub-data sets based on the balanced data set and the expanded unbalanced data set;

The fault sub-data set is used and model training is performed based on the deep forest algorithm to obtain a trained fault diagnosis model.

Specifically, the waveform features extracted from the historical fault waveform database are transient waveform features in three aspects: time domain, frequency domain, and time-frequency domain. Among them, the extraction of transient waveform time domain characteristics includes waveform mean, mean square error, root square amplitude, peak value, skewness, crest factor and waveform factor time domain characteristic parameters, and the extraction of transient frequency domain includes center of gravity frequency, average Frequency, frequency standard deviation and root mean square frequency, frequency domain spectral power, the extraction of transient frequency domain includes wavelet entropy;

At the same time, the formula is used for normalization to eliminate dimensional inconsistencies. The formula is: In the formula, y _i is the normalized data; x _max and x _min are the maximum and minimum values of the extracted feature parameters;

And the above time-frequency characteristics are used for fault classification. The fault classification includes: lightning strike and non-lightning strike fault two classifications, lightning strike multi-classification, non-lightning strike multi-classification or overall multi-classification, thereby constructing different types of fault waveform data sets;

Specifically, the classified data set has a different number of data for each class. The class with the largest number of data is used as a balanced data set, and the data sets of other remaining classes are all unbalanced data sets. The remaining data sets are clustered and Oversampling processing;

Based on the historical fault information of the State Grid, a fault waveform data set has been initially established, with a total of 3,415 pieces of data, including a total of 994 pieces of lightning strike data and a total of 2,421 pieces of non-lightning strike data. After excluding data without specific counterattack types, 582 pieces of lightning strike data remain. There are 485 of them, including 485 for bypass and 97 for counterattack. After excluding the non-lightning data without detailed category labels, there are 898 remaining non-lightning data, including 120 for bird damage, 162 for ice damage, 186 for wind deflection, and 365 for external force damage. Articles, 65 others.

Specifically, the imbalanced data set is clustered through the K-means clustering algorithm, including,

Step 1: For the data set, extract k initial clustering center points as i ₁ , i ₂ ,..., i _k ;

Step 2: For other data in the data set except the cluster center, calculate the Euclidean of each data and i _j (j=1,2,...,k) through the formula c _i =argmin[|x _i -i _j |] ² distance, and classify the data closest to i, j into one category, and implement the data into k categories;

Step 3: Calculate the average value of the data in each category, set the obtained average value as the central value of the category, and then use the formula in step 2 to calculate the sum of the Euclidean distances from each data point to the central value, calculated as S ;

Step 4: Repeat steps 2 and 3 until the result S does not change and output the K-means clustering subset.

11. Specifically, use SMOTE oversampling to expand the clustered imbalanced data set, further including:

Step 1: For each sample x of the minority class in the imbalanced data set, use the Euclidean distance as the standard, calculate its distance to all samples in the minority class sample set, and obtain its k nearest neighbors;

Step 2: Determine the sampling rate N. For each minority class sample x, randomly select several samples from its k nearest neighbors;

Step 3: Randomly select a nearest neighbor Xnew, which is compared with the original sample through the formula Construct a new sample; rand(0, N) is a random number generated between 0 and N,/> is the cluster center, and x is a sample of this type.

The requirement of SMOTE oversampling is to increase the number of data in other categories except for the category with the largest number of data. Interpolation is performed on the connection between the cluster center and other sample points, the clustering samples are corrected, the data set is expanded, and the fault sub-data set is constructed;

Specifically, the deep forest algorithm adopts the deep forest multi-granularity cascade forest model, including,

The multi-granularity cascade forest model includes multi-granularity scanning and cascade forest; multi-granularity scanning partially filters features to generate feature quantities that are more closely related to classification. First, a sliding window of length L is used to scan the original data with dimension K, and each sliding step is S. After sliding sampling is completed, N L-dimensional feature subvectors are obtained, where N = (K-L)/S+1. Assume that the original data is distinguished by 10 fault feature classes, the input sample is a 300-dimensional vector, and the granularity of 100 is used for scanning. Each sliding step is 1. A total of 201 100-dimensional vectors can be generated. Each vector is processed through a random forest. Generate a 10-dimensional class vector, generating a total of 2010-dimensional input vectors;

The multi-granularity cascade forest model uses random forests as the base learner for integrated learning. Each forest of the cascade forest species is composed of multiple decision trees. The decision trees use the classification and regression tree (CAET) algorithm;

In the formula, Pm represents the proportion of samples of the mth category in the current sample set D, |y| is a non-zero integer, and Gini(D) is the Gini index. The smaller the value, the higher the purity of the data set D. When constructing CART node, calculate the Gini coefficient of the existing features of the current training sample set under all possible values, and select the feature corresponding to the minimum Gini coefficient and the corresponding split point as the splitting condition of the current node;

The function of the cascade forest is to process sample features layer by layer to enhance the accuracy of pattern recognition of the algorithm. Each layer has a random forest and a complete random forest. The input data of the first layer of the cascade forest is a 4020-dimensional vector spliced by the 2010-dimensional class vector of the random forest and the 2010-dimensional vector of the complete random forest. After the classification process, two two-dimensional category vectors are obtained; then these two two-dimensional category vectors are spliced with the 4020-dimensional initial feature vector to form a 4024-dimensional new feature vector as the input of the second layer; follow this method and analogy. Finally, the category vectors output by the Nth layer are averaged, and the category corresponding to the maximum value is selected as the final classification result of this type of fault.

In this embodiment, after processing different types of fault waveform data sets, the number of data in the minority class data set is increased, and the number of data in the training set is increased. Deep forest is used as a classification method for classification, which increases the accuracy of the classification results. . The processed data was tested for classification accuracy using Weka. Under the condition of 9:1 between training set and test set, the non-lightning multi-classification accuracy was 71.27%, and the highest accuracy after SMOTE processing of 39 features was 77.90%.

In this embodiment, the ratio of the training set and the test set of the deep forest algorithm is 7:3; K-means-SMOTE is combined with the deep forest algorithm, and its accuracy evaluation index is accuracy Precision=TP/(TP/TP+FP), that is Correct predictions are positive as a proportion of all positive predictions; the Weka simulation results show that after K-means-SMOTE is combined with the deep forest algorithm, the lightning strike/non-lightning strike classification accuracy can reach 90%; the lightning strike/counterattack classification accuracy can reach 90%. 86%; the non-lightning multi-classification accuracy can reach 73%; the overall multi-classification accuracy can reach 79%, and the accuracy is more than 10% higher than traditional random forest and other methods.

As shown in Figures 2 and 3, this embodiment also provides a transmission line fault multi-source diagnosis system based on K-SMOTE and deep forest, including:

Acquisition unit: obtain fault data;

Diagnosis unit: input the transmission line fault data into the trained fault diagnosis model to obtain fault diagnosis results;

The training of the fault diagnosis model includes:

Extract the time-frequency characteristics of voltage and current waveforms in historical fault data, classify faults based on the extracted time-frequency characteristics, and form fault data sets of different fault types;

Use K-means clustering algorithm to cluster the unbalanced data set in the fault data set, and use SMOTE oversampling to expand the clustered unbalanced data set;

Form fault sub-data sets based on the balanced data set and the expanded unbalanced data set;

The new fault data set is used and model training is performed based on the deep forest algorithm to obtain a trained fault diagnosis model.

This embodiment also provides a computer-readable storage medium. A computer program is stored on the storage medium. When the computer program is executed by a processor, the transmission line fault multiplication based on K-SMOTE and deep forest is realized. Source diagnostic method steps.

The computer-readable storage medium may be an internal storage unit of the computer device described in the previous embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMediaCard), a secure digital (Secure Digital) card, a flash memory card ( Flash Card), etc.

This embodiment also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein the computer program is executed by the processor. When, the steps of the multi-source diagnosis method for transmission line faults based on K-SMOTE and deep forest are implemented.

Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art will understand that various changes, modifications, and substitutions can be made to these embodiments without departing from the principles and spirit of the invention. and modifications, the scope of the invention is defined by the appended claims and their equivalents.

Claims (10)

1. The power transmission line fault multi-source diagnosis method based on the K-SMOTE and the depth forest is characterized by comprising the following steps of:

acquiring fault data of a power transmission line;

inputting the power transmission line fault data into a trained fault diagnosis model to obtain a fault diagnosis result;

the training of the fault diagnosis model comprises the following steps:

extracting time-frequency characteristics of voltage and current waveforms in historical fault data, and performing fault classification according to the extracted time-frequency characteristics to form fault waveform data sets of different fault types;

clustering the unbalanced data set in the fault waveform data set by using a K-means clustering algorithm, and performing data expansion on the clustered unbalanced data set by using SMOTE oversampling;

forming a failure sub-data set according to the balanced data set and the expanded unbalanced data set;

and performing model training by using the fault sub-data set and based on a depth forest algorithm to obtain a trained fault diagnosis model.

2. The K-SMOTE and depth forest based transmission line fault multisource diagnosis method according to claim 1, wherein the extracted time-frequency features include time domain, frequency domain and time-frequency domain transient waveform features.

3. The K-SMOTE and depth forest based transmission line fault multisource diagnosis method according to claim 1, wherein the fault classification mode comprises: lightning stroke non-lightning stroke fault two-classification, lightning stroke multi-classification, non-lightning stroke multi-classification or overall multi-classification.

4. The K-SMOTE and deep forest based transmission line fault multisource diagnosis method according to claim 1, wherein the fault waveform data set comprises a balanced data set and an unbalanced data set, the balanced data set is the data set with the largest number of fault data, and the rest data sets are all unbalanced data sets.

5. The K-SMOTE and depth forest based transmission line fault multisource diagnosis method according to claim 1, wherein the K-means clustering algorithm is used to cluster unbalanced data sets in the fault data set, further comprising:

step one: for the dataset, extracting k initial cluster center points as i ₁ ，i ₂ ，…，i _k；

Step two: for other data except the clustering center in the data set, the data is obtained through a formula c _i ＝argmin[|x _i -i _j |] ² Calculate each data and i _j (j=1, 2, …, k), and classifying the data nearest to i, j into one type to obtain k types of data;

step three: calculating the average value of data in various types, setting the calculated average value as the central value of the type, and calculating the sum of Euclidean distances from each data point to the central value by using the formula in the second step, and recording as S;

step four: and repeating the second and third steps until the result S is unchanged, and outputting a K-means cluster subset.

6. The K-SMOTE and depth forest based transmission line fault multisource diagnostic method of claim 1, wherein the data expansion of the clustered unbalanced data set using SMOTE oversampling further comprises:

step one: for each sample x of a minority class in the unbalanced data set, calculating the distance from the sample x to all samples in the minority class sample set by taking Euclidean distance as a standard to obtain k nearest neighbor;

step two: determining a sampling multiplying power N, and randomly selecting a plurality of samples from k neighbors of each minority sample x;

step (a)Thirdly,: randomly selecting a neighbor Xnew, wherein the neighbor Xnew and the original sample pass through the formulaConstructing a new sample; rand (0, N) is a random number generated between 0 and N, +.>Is the cluster center, x is the sample of this type.

7. The K-SMOTE and depth forest based transmission line fault multisource diagnosis method according to claim 1, wherein the depth forest algorithm adopts a depth forest multiscale cascade forest model.

8. The power transmission line fault multi-source diagnosis system based on K-SMOTE and deep forest is characterized in that the system comprises,

an acquisition unit: acquiring fault data;

diagnosis unit: inputting the power transmission line fault data into a trained fault diagnosis model to obtain a fault diagnosis result;

the training of the fault diagnosis model comprises the following steps:

extracting time-frequency characteristics of voltage and current waveforms in historical fault data, and carrying out fault classification according to the extracted time-frequency characteristics to form fault data sets of different fault types;

clustering the unbalanced data set in the fault data set by using a K-means clustering algorithm, and performing data expansion on the clustered unbalanced data set by using SMOTE oversampling;

forming a failure sub-data set according to the balanced data set and the expanded unbalanced data set;

and performing model training by using the fault sub-data set and based on a depth forest algorithm to obtain a trained fault diagnosis model.

9. An electronic device comprising a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein the computer program when executed by the processor implements the steps of the K-SMOTE and depth forest based transmission line fault multisource diagnostic method according to any one of claims 1 to 7.

10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program, wherein the computer program, when executed by a processor, implements the steps of the K-SMOTE and depth forest based transmission line fault multisource diagnostic method according to any one of claims 1 to 7.