New energy power generation complete set power equipment fault diagnosis method based on semi-supervised SVM model

Abstract

The invention relates to a new energy power generation complete power equipment fault diagnosis method based on a semi-supervised SVM model. Feature extraction and analysis are performed on state monitoring data of power equipment by using a principal component analysis method; after the new energy power generation complete set of power equipment breaks down, the generated outage state is analyzed; and diagnosing the fault of the new energy power generation complete set of power equipment according to the extracted features and the outage state. According to the method, feature extraction and selection are performed by using unmarked data and a small amount of marked data, and then a model is trained and optimized by using a semi-supervised learning algorithm. Meanwhile, through experimental verification and performance evaluation, the fault state of the power equipment can be effectively identified and classified, and the accuracy and reliability of diagnosis are improved. And the fault state of the power equipment is classified and predicted.

Description

Technical Field

[0001] The invention belongs to the technical field of electric power diagnosis, in particular to a fault diagnosis method for a complete set of electric power equipment for new energy power generation based on a semi-supervised SVM model. Background Art

[0002] With the adjustment of global energy structure and the enhancement of environmental awareness, new energy power generation technology is becoming increasingly important in the energy structure due to its clean and renewable characteristics. New energy power generation equipment is widely used in power systems, but its complexity and environmental uncertainty lead to frequent failures, challenging system stability. Traditional diagnostic methods rely on manual and experience and are difficult to meet modern needs. Machine learning technology, especially support vector machines, has achieved remarkable results, but is limited by supervised learning. As an extension of SVM, semi-supervised support vector machines can use unlabeled data to improve performance, which has research value and practical significance for fault diagnosis of new energy power generation equipment.

[0003] In the work of power equipment fault diagnosis, there is a lack of special diagnostic technology for new energy power generation complete sets of power equipment, and traditional diagnostic methods have certain defects:

[0004] 1. The fault diagnosis method of PCA combined with SVM uses principal component analysis and support vector machine to perform fault diagnosis, but the parameter setting has a great influence on the accuracy of the diagnosis results and the misjudgment rate is high.

[0005] 2. Equipment fault diagnosis is performed by improving the wavelet transform method. The fruit fly algorithm is added to the wavelet transform algorithm to decompose the fault signal of the power equipment. However, the filtering characteristics of the wavelet transform determine that the fault diagnosis accuracy is not high.

[0006] 3. The information entropy power equipment fault data acquisition system uses information entropy theory and combines the operating characteristics of power equipment in the distribution network to collect, process and analyze equipment fault data in real time. However, the information entropy calculation is complex, the efficiency of processing equipment fault signals is low, and the response speed is not high.

[0007] 4. Combine big data technology and Internet of Things technology to diagnose power equipment faults. However, the data integration level in the power industry is not high and data sharing is not smooth.

[0008] In summary, in order to solve the problems existing in the fault diagnosis of power equipment, the semi-supervised SVM model is applied to the fault diagnosis of complete sets of power equipment for renewable energy power generation. By making full use of the information in the unlabeled data, the accuracy and efficiency of fault diagnosis are improved; at the same time, the parameters of the model are optimized to further improve its performance. Summary of the invention

[0009] The purpose of the present invention is to overcome the shortcomings of the prior art and propose a fault diagnosis method for a complete set of power equipment for renewable energy power generation based on a semi-supervised SVM model, which can effectively identify and classify the fault status of power equipment, improve the accuracy and reliability of diagnosis, and classify and predict the fault status of power equipment.

[0010] The present invention solves the technical problem by adopting the following technical solutions:

[0011] The fault diagnosis method of a complete set of power equipment for renewable energy power generation based on a semi-supervised SVM model comprises the following steps:

[0012] Step 1: Use the principal component analysis method to extract and analyze the features of the state monitoring data of the power equipment;

[0013] Step 2: Analyze the outage status generated after the failure of the new energy power generation complete set of power equipment;

[0014] Step 3: Diagnose the fault of the new energy power generation complete set of power equipment based on the features extracted in step 1 and the outage status in step 2.

[0015] Moreover, the specific implementation method of step 1 is: collect n data samples, each of which has p attribute variables, and its sample data vector is:

[0016]

[0017] Among them, x np represents sample data; the dimensions of the selected sample data vectors are different, so before the principal component analysis calculation, the original power equipment data is standardized to eliminate the impact of the dimension:

[0018]

[0019] in, is the average value of the sample data;

[0020] When selecting eigenvalues, the factor with the largest eigenvalue is selected as the principal component, and the load of the principal component is:

[0021]

[0022] In the formula: α1, α2, ..., α m represents the characteristic value, ε1,ε2,...ε m Represents the eigenvector; calculate the weight of each data indicator and analyze the selected m principal components; find the weight z of each state indicator parameter in the principal component, and then sort them according to the size of each state indicator parameter; the weight of each indicator is:

[0023]

[0024] The weight z of each state index parameter obtained is standardized to [0,1]. Among them, the state index parameter with a larger weight indicates a higher degree of correlation. When extracting features of the power equipment state index, the key parameters of the equipment state are extracted based on the actual requirements and the weights of each state index parameter, thereby obtaining the key indicator parameters of the equipment state.

[0025] Moreover, the specific implementation method of step 2 is: the reliability function is the probability of normal operation of the power equipment in time t, the continuous random variable T (T ≥ 0) is the continuous time period from the normal operation of the power equipment to the shutdown state, and its use reliability function is:

[0026] λ(t)=P(T≥t)

[0027] Among them, P() represents probability; the fault distribution function F(t) represents the probability that the power equipment is out of service before time t. The relationship between the fault distribution function and the reliability function is:

[0028] F(t)=P(T≤t)=1-λ(t)

[0029] The outage rate reliability index is obtained from the fault distribution function:

[0030]

[0031] The reliability level of power equipment operation is calculated based on the reliability index. At the same time, by strengthening the operation monitoring and maintenance of power equipment, problems with the equipment can be discovered and dealt with in a timely manner to ensure the normal operation of the equipment.

[0032] Moreover, the specific implementation method of step 3 is as follows: after extracting the characteristic value in step 1, step 2 analyzes the fault situation, uses the semi-supervised SVM model to compare the characteristic value with the fault analysis situation of the fault, establishes a database, and records the fault situation; after the fault occurs again, extracts the characteristic value and compares it with the database to accurately diagnose the fault, and uses the semi-supervised learning algorithm to train and optimize the model to improve the accuracy;

[0033] The association rule algorithm is used to process data, and the data sequence of the new energy power generation equipment is:

[0034] K={k1,k2,...,k m}

[0035] Where m represents the set of different data items of power equipment, each element in the set is called an item, and the set of items is called an item set; the association rule between the data of the complete set of new energy power generation equipment and fault judgment is a transaction V, then the transaction database is:

[0036] DK={V1,V2,...,V m}

[0037] In D, the total number of transactions is |D|; each subset of the itemset K represents a transaction V; if a transaction V contains both μ and Then the association rule of fault diagnosis is established:

[0038]

[0039] The mining process of power equipment fault diagnosis rules involves a data processing phase, which includes cleaning and preprocessing the collected fault data; in terms of new energy power generation complete sets of power equipment, mining and analysis are carried out based on fault sample data, elements that meet the standards are selected and appropriate rules are established; after the model is established, the new rules are verified with the original rule base to ensure that there are no duplications or incompatibilities in the rules, and compared with the effective diagnostic rules in the original rule base;

[0040] At the same time, in the power system, the positive sequence of zero-sequence information has a higher priority than the negative sequence; in fault diagnosis, rare faults are regarded as matching failures, and appropriate rules are added to prevent such rare faults from occurring; after each fault data collection, the number of faults in the rules in the database is updated to realize digital power equipment fault diagnosis based on big data.

[0041] The advantages and positive effects of the present invention are:

[0042] The present invention uses the principal component analysis method to extract and analyze the features of the state monitoring data of the power equipment; analyzes the outage status generated after the failure of the new energy power generation complete set of power equipment; and diagnoses the failure of the new energy power generation complete set of power equipment based on the extracted features and outage status. The present invention uses unlabeled data and a small amount of labeled data to extract and select features, and then uses a semi-supervised learning algorithm to train and optimize the model. At the same time, the present invention can effectively identify and classify the fault status of the power equipment through experimental verification and performance evaluation, thereby improving the accuracy and reliability of the diagnosis. The fault status of the power equipment is classified and predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of a test of an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below with reference to the accompanying drawings.

[0045] Support vector machine is an important theoretical model, which originated from the linearly separable case. Its main goal is to separate more than two types of samples. In support vector machine, it is necessary not only to correctly classify samples, but also to maximize the classification interval, which is one of its important goals. In supervised problems, support vector machine separates two types of samples by solving the maximum margin hyperplane. In semi-supervised problems, support vector machine looks for a hyperplane that can separate two types of labeled samples and pass through the low-density sample area.

[0046] The fault diagnosis method of a complete set of power equipment for renewable energy power generation based on a semi-supervised SVM model comprises the following steps:

[0047] Step 1: Use the principal component analysis method to extract and analyze the features of the status monitoring data of power equipment.

[0048] The semi-supervised SVM model collects data from power equipment and divides the data into structured and unstructured data according to its structure. Structured data refers to the status monitoring data that is commonly found in power grids, of which digital monitoring data is the most typical. Compared with existing structured data, unstructured data is an emerging and rapidly developing type of data.

[0049] n data samples are collected, each of which has p attribute variables, and its sample data vector is:

[0050]

[0051] Among them, x np represents sample data; the dimensions of the selected sample data vectors are different. Therefore, before performing principal component analysis, the influence of the dimensions needs to be eliminated. The common method is to standardize the original power equipment data.

[0052]

[0053] in, is the average value of the sample data;

[0054] According to the principle of covariance vector, any variable is standardized and transformed, and the resulting variable covariance vector is its correlation coefficient vector. The standardized covariance correlation coefficient is equivalent. The standard for selecting the number of principal components should be determined based on the eigenvalue of the principal component vector. If the eigenvalue of the principal component vector is less than a certain value, then it cannot explain any indicator. Therefore, when selecting the eigenvalue, the factor with a larger eigenvalue should be selected as the principal component. The load of its principal component is:

[0055]

[0056] In the formula: α1, α2, ..., α m represents the characteristic value, ε1,ε2,...ε m Represents the eigenvector; calculate the weight of each data indicator and analyze the selected m principal components; find the weight z of each state indicator parameter in the principal component, and then sort them according to the size of each state indicator parameter; the weight of each indicator is:

[0057]

[0058] The weights z of each state index parameter obtained after normalization belong to [0,1]. Among them, the state index parameter with a larger weight indicates a higher degree of correlation, which means that this state index parameter is more representative among many indicators. When extracting features of power equipment state indicators, the key parameters of the equipment state should be extracted in combination with the actual requirements and the weights of each state index parameter, so as to obtain the key indicator parameters of the equipment state. Taking the transformer as an example, the feature types of its operating state are shown in Table 1 below.

[0059] Table 1 Transformer operating status characteristics

[0060]

[0061] Step 2: Analyze the outage status generated after a failure of a complete set of new energy power generation equipment.

[0062] After a failure of a complete set of power equipment for renewable energy power generation, it is necessary to analyze the resulting outage state, in which the calculation of various reliability indicators of the equipment outage probability is crucial. First, the reliability function refers to the probability of normal operation of the power equipment in time t. The continuous random variable T (T ≥ 0) is set as the continuous time period from the normal operation of the power equipment to the outage state. Its reliability function can be expressed as:

[0063] λ(t)=P(T≥t)

[0064] Among them, P() represents probability; the fault distribution function F(t) represents the probability that the power equipment is out of service before time t. The relationship between the fault distribution function and the reliability function is:

[0065] F(t)=P(T≤t)=1-λ(t)

[0066] The outage rate reliability index is obtained from the fault distribution function:

[0067]

[0068] The reliability level of power equipment operation is calculated based on the reliability index. At the same time, by strengthening the operation monitoring and maintenance of power equipment, problems with the equipment can be discovered and dealt with in a timely manner to ensure the normal operation of the equipment.

[0069] Step 3: Diagnose the fault of the new energy power generation complete set of power equipment based on the features extracted in step 1 and the outage status in step 2.

[0070] After extracting the characteristic value in step 1, step 2 analyzes the fault situation, uses the semi-supervised SVM model to compare the characteristic value with the fault analysis situation of the fault, establishes a database, and records the fault situation; after the fault occurs again, extracts the characteristic value and compares it with the database to accurately diagnose the fault, and uses the semi-supervised learning algorithm to train and optimize the model to improve the accuracy;

[0071] Fault diagnosis is the information reflected by the characteristic signals of power equipment collected by condition monitoring and the results of data calculation and processing, combined with the understanding and experience of relevant knowledge of power equipment, and the selection of appropriate maintenance and processing strategies by power experts. It can also be understood as using the characteristic information of power equipment obtained by condition monitoring, evaluating the characteristic information through fault diagnosis, comparing it with previous accidents and similar devices, and determining the maintenance and processing strategy to be adopted to complete the condition maintenance of the equipment. Fault diagnosis of complete sets of power equipment for renewable energy power generation is usually carried out in the dispatching center and the centralized control center. By monitoring the current and voltage of power equipment, the global voltage, current and location of operating faults in the equipment can be determined and analyzed. The data is processed using the association rule algorithm, and the data sequence of the complete set of power equipment for renewable energy power generation is:

[0072] K={k1,k2,...,k m}

[0073] Where m represents the set of different data items of power equipment, each element in the set is called an item, and the set of items is called an item set; the association rule between the data of the complete set of new energy power generation equipment and fault judgment is a transaction V, then the transaction database is:

[0074] DK={V1,V2,...,V m}

[0075] In D, the total number of transactions is |D|; each subset of the itemset K represents a transaction V; if a transaction V contains both μ and Then the association rule of fault diagnosis is established:

[0076]

[0077] The mining process of power equipment fault diagnosis rules involves a data processing phase, which includes cleaning and preprocessing the collected fault data; in terms of new energy power generation complete sets of power equipment, mining and analysis are carried out based on fault sample data, elements that meet the standards are selected and appropriate rules are established; after the model is established, the new rules are verified with the original rule base to ensure that there are no duplications or incompatibilities in the rules, and compared with the effective diagnostic rules in the original rule base;

[0078] At the same time, in the power system, the positive sequence of zero-sequence information has a higher priority than the negative sequence; in fault diagnosis, rare faults are regarded as matching failures, and appropriate rules are added to prevent such rare faults from occurring; after each fault data collection, the number of faults in the rules in the database is updated to realize digital power equipment fault diagnosis based on big data.

[0079] According to the above-mentioned method for fault diagnosis of complete sets of power equipment for renewable energy power generation based on the semi-supervised SVM model, the effect of the present invention has been illustrated by conducting tests.

[0080] This embodiment takes a 110kV substation in City A as an example. First, it is necessary to collect the operating data of the new energy power generation complete set of power equipment in the substation. These data include electrical parameters such as voltage, current, and temperature of the equipment. These data can be collected by sensors and measuring instruments installed on the equipment. The collected detailed data of partial discharge detection of the substation is shown in Table 2 below.

[0081] Table 2 Substation data

[0082]

[0083] In order to verify that the proposed method for fault diagnosis of complete sets of power equipment for renewable energy power generation based on semi-supervised SVM model has certain feasibility, the 10kV#3 capacitor group of the 110kV substation is taken as an example, and four fault states, including capacitor breakdown, capacitor casing burst, fuse blowout, and shell oil leakage, are selected and tested. In the fault diagnosis of complete sets of power equipment for renewable energy power generation, it is very important to identify and classify faults. Therefore, the present invention is compared with the traditional method to verify the accuracy of fault classification of the two methods in the experiment.

[0084] Based on the above experimental preparation, a sample data set of the above four fault states is randomly selected and its detailed data is shown in Table 3 below.

[0085] Table 3 Sample dataset

[0086]

[0087] The comparative test is as follows Figure 1As shown. Based on the above experimental results, it can be seen that the present invention can accurately diagnose and classify the fault status of power equipment with an accuracy of more than 90%, while the accuracy of traditional methods is lower, with a maximum of only 80%. Since the semi-supervised SVM model does not need to mark all data during the training process, it can save a lot of time to improve the classification accuracy of fault diagnosis.

[0088] Therefore, the effectiveness of the fault diagnosis method for complete sets of power equipment for renewable energy power generation based on the semi-supervised SVM model can be proved. At the same time, the fault diagnosis method for complete sets of power equipment for renewable energy power generation based on the semi-supervised SVM model shows good performance in terms of accuracy and real-time performance. However, in practical applications, the impact of factors such as data quality and feature selection on model performance still needs to be considered.

[0089] It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation manner. Any other implementation manners derived by those skilled in the art based on the technical solution of the present invention also fall within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for a complete set of power equipment for renewable energy power generation based on a semi-supervised SVM model, characterized by: The following steps are involved: Step 1: Use the principal component analysis method to extract and analyze the features of the state monitoring data of the power equipment; Step 2: Analyze the outage status generated after the failure of the new energy power generation complete set of power equipment; Step 3: Diagnose the fault of the new energy power generation complete set of power equipment based on the features extracted in step 1 and the outage status in step 2.

2. The method for fault diagnosis of a complete set of power equipment for renewable energy power generation based on a semi-supervised SVM model according to claim 1 is characterized in that: The specific implementation method of step 1 is: collect n data samples, each of which has p attribute variables, and its sample data vector is: Among them, x np represents sample data; the dimensions of the selected sample data vectors are different, so before the principal component analysis calculation, the original power equipment data is standardized to eliminate the impact of the dimension: in, is the average value of the sample data; When selecting eigenvalues, the factor with the largest eigenvalue is selected as the principal component, and the load of the principal component is: In the formula: α1, α2, ..., α m represents the characteristic value, ε1,ε2,...ε m Represents the eigenvector; calculate the weight of each data indicator and analyze the selected m principal components; find the weight z of each state indicator parameter in the principal component, and then sort them according to the size of each state indicator parameter; the weight of each indicator is: The weight z of each state index parameter obtained is standardized to [0,1]. Among them, the state index parameter with a larger weight indicates a higher degree of correlation. When extracting features of the power equipment state index, the key parameters of the equipment state are extracted based on the actual requirements and the weights of each state index parameter, thereby obtaining the key indicator parameters of the equipment state.

3. The method for fault diagnosis of a complete set of power equipment for renewable energy power generation based on a semi-supervised SVM model according to claim 1 is characterized in that: The specific implementation method of step 2 is: the reliability function is the probability of normal operation of the power equipment in time t, the continuous random variable T (T ≥ 0) is the continuous time period from the normal operation of the power equipment to the shutdown state, and its use reliability function is: λ(t)=P(T≥t) Among them, P() represents probability; the fault distribution function F(t) represents the probability that the power equipment is out of service before time t. The relationship between the fault distribution function and the reliability function is: F(t)=P(T≤t)=1-λ(t) The outage rate reliability index is obtained from the fault distribution function: The reliability level of power equipment operation is calculated based on the reliability index. At the same time, by strengthening the operation monitoring and maintenance of power equipment, problems with the equipment can be discovered and dealt with in a timely manner to ensure the normal operation of the equipment.

4. The method for fault diagnosis of a complete set of power equipment for renewable energy power generation based on a semi-supervised SVM model according to claim 1 is characterized in that: The specific implementation method of step 3 is as follows: after extracting the characteristic value in step 1, in step 2, the fault condition is analyzed, the characteristic value and the fault analysis condition of the fault are compared by using the semi-supervised SVM model, a database is established, and the fault condition is recorded; after the fault occurs again, the characteristic value is extracted and compared with the database to accurately diagnose the fault, and the model is trained and optimized by using the semi-supervised learning algorithm to improve the accuracy; Among them, the association rule algorithm is used to process the data, and the data sequence of the new energy power generation complete set of power equipment is: K={k1,k2,...,k m } Where m represents the set of different data items of power equipment, each element in the set is called an item, and the set of items is called an item set; the association rule between the data of the complete set of new energy power generation equipment and fault judgment is a transaction V, then the transaction database is: <h2 style=";text-align:left;direction:ltr">DK = {V1,V2,...,V<h2 style=";text-align:left;direction:ltr"> m <h2 style=";text-align:left;direction:ltr">} In D, the total number of transactions is |D|; each subset of the itemset K represents a transaction V; if a transaction V contains both μ and Then the association rule of fault diagnosis is established: The mining process of power equipment fault diagnosis rules involves a data processing phase, which includes cleaning and preprocessing the collected fault data; in terms of new energy power generation complete sets of power equipment, mining and analysis are carried out based on fault sample data, elements that meet the standards are selected and appropriate rules are established; after the model is established, the new rules are verified with the original rule base to ensure that there are no duplications or incompatibilities in the rules, and compared with the effective diagnostic rules in the original rule base; At the same time, in the power system, the positive sequence of zero-sequence information has a higher priority than the negative sequence; in fault diagnosis, rare faults are regarded as matching failures, and appropriate rules are added to prevent such rare faults from occurring; after each fault data collection, the number of faults in the rules in the database is updated to realize digital power equipment fault diagnosis based on big data.

Images