A method for realizing data link fault classification and early warning of a data middle platform based on a knowledge graph

Abstract

The application relates to a method for realizing data link fault classification early warning of a data middle platform based on a knowledge graph. Firstly, the fault classification in the data link is taken as a target, a fault classification model is obtained through Kmeans-SVM model training, then a knowledge graph of the fault type is constructed through a Markov model and the like, the fault classification is based on the fault field knowledge graph and the Kmeans-SVM method, the correlation between faults is analyzed through fault reason similarity, and then the possible associated faults in the data link are predicted. Through mutual cooperation of the machine learning fault classification module and the knowledge graph fault early warning module, intelligent analysis of the data link is realized, a series of problems existing in the expert system and the machine learning are solved, the maintenance efficiency of the data link is effectively improved, and the fault classification accuracy is greatly improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of power internet of things, and relates to data middle platform technology, in particular to a method for realizing data link fault classification and early warning of data middle platform based on a knowledge graph. BACKGROUND

[0002] With the rapid development of power internet of things and the promotion of data middle platform, the amount of power resource data is continuously expanding, and more and more data manufacturers and data users appear. At the same time, there are problems such as scattered data distribution, huge data scale, complex data interaction, low data transmission efficiency, and difficult diagnosis of data link faults.

[0003] At present, the method for analyzing data link faults mainly focuses on fault analysis of entity hardware, and the analysis at the software level is relatively less. The fault classification methods for entity hardware mainly include expert systems and machine learning, among which the most widely used is the expert system, which is divided into expert knowledge based on shallow knowledge field and model knowledge based on deep knowledge analysis object. Machine learning does not need manual knowledge sorting and summarizing, but only needs to use related data sets for training to obtain a fault classification model, and good results have been achieved in the field of fault diagnosis.

[0004] Although the expert system can effectively simulate the fault diagnosis process of the fault diagnosis expert, there are still problems in practical application, such as difficulty in obtaining a complete knowledge base, slow diagnosis speed, high operation and maintenance difficulty, lack of learning ability, and poor fault tolerance. Therefore, machine learning models are gradually widely used in fault classification problems, but using a single machine learning method for fault prediction not only requires a large number of labeled data sets, but also greatly depends on the setting of training time, training parameters, and the fault diagnosis effect may not be good. SUMMARY

[0005] The purpose of the present application is to overcome the shortcomings of the prior art, and to provide a method for realizing data link fault classification and early warning of data middle platform based on a knowledge graph, which mainly designs and realizes a data link fault classification method based on deep learning, classifies the existing faults in the data link, and warns about possible future faults, thereby effectively improving the maintenance efficiency of the data link.

[0006] The technical problem of the present application is solved by adopting the following technical scheme:

[0007] A method for realizing data link fault classification and early warning of data middle platform based on a knowledge graph, and the specific method steps are as follows:

[0008] (1) Establishing a Kmeans-SVM fault classification model

[0009] First, the data is reduced using the PCA method, and the reduced data is clustered using the K-means method, which is an unsupervised learning method. The cluster number k is set to 2, and the Euclidean distance is used as the distance standard in the clustering process. The error sum of squares is used as a representative of the similarity of samples within the cluster. The smaller the error sum of squares of a certain cluster G u , the greater the similarity of samples within the cluster, and vice versa. The greater the error sum of squares of G u , the smaller the similarity of samples within the cluster. The calculation formula of the error sum of squares is as follows:

[0010]

[0011] These input unlabeled data become labeled data after K-means method, and the input data is finally divided into two categories, represented by numbers 1 and 2. Next, use these labeled data to train the SVM model to get the maximum interval hyperplane, and divide the 1 and 2 data on both sides of the hyperplane. After the SVM training is completed, use the maximum interval hyperplane to test the accuracy of the classification result obtained by K-means. The data predicted correctly is used to retrain the SVM hyperplane. Update the SVM hyperplane iteratively until the error rate of the data predicted by SVM no longer changes. The final maximum interval hyperplane is obtained. The specific algorithm steps are as follows:

[0012] Step 1: Preprocess the unlabeled data, and use the K-means algorithm to divide the data into two clusters, labeled as 1 and 2, respectively. Convert the unlabeled data into labeled data.

[0013] Step 2: Use the labeled data obtained in step 1 to train the SVM classifier model to get the maximum interval hyperplane.

[0014] Step 3: Test the two sample data obtained by K-means clustering using the trained SVM classification model. The data predicted accurately is used as the input data of step 2 to get a new maximum interval hyperplane. Repeat this process until the error rate of the data predicted by SVM no longer changes.

[0015] (2) Constructing the knowledge graph of the fault domain

[0016] The construction of the knowledge graph of the link fault domain mainly includes four parts: entity and attribute extraction, coreference resolution, knowledge processing, and data integration.

[0017] (3) Data link fault early warning

[0018] According to the logical relationship between the fault node information and the fault reason in the link fault knowledge graph, knowledge reasoning is performed.

[0019] Moreover, the extraction of entities and attributes includes: firstly, performing a word segmentation operation on the corpus of the fault domain, using a Markov model to extract entities and attributes, and using the extracted entities and attributes as the final entities and attributes applied to the knowledge graph, then performing part-of-speech tagging on all words, and dividing them into the following categories: fault noun entities, fault phenomenon verbs, fault degree adverbs, fault degree quantifiers, and original words of unextracted words.

[0020] Moreover, the main purpose of the co-reference resolution step is to find out synonymous words representing entities and attributes, and to classify words with high similarity into the same category, using a thesaurus form to represent synonyms.

[0021] Moreover, the main purpose of the knowledge processing step is to identify the corresponding relationship between entities and attributes, and to delete redundant inclusion relationships by taking whether there is an inclusion or non-inclusion relationship between each word as a standard.

[0022] Moreover, the data integration step is to merge the triples of fault entities, attributes and relationships, and to update the nodes through the concept layer and the entity attribute layer to realize the final graph construction and finally form the knowledge graph of the link fault domain.

[0023] Moreover, the logical relationship between the fault node information and the fault reason in the link fault knowledge graph is as follows: firstly, the known knowledge stored is extracted, and after rule extraction, rule matching is performed, if the matching is successful, it is added to the rule execution area, if the rule conflict is established, it is conflict resolution, if the rule conflict is not established, the reasoning result of not established is obtained, and similarly, if the rule matching is not successful, the reasoning result of not reasoning out new knowledge is obtained.

[0024] The advantages and positive effects of the present application are:

[0025] 1. The Kmeans-SVM fault classification model is trained, the model is automatically verified and evaluated and the parameters are adjusted through an optimization algorithm, the model is tested using a test set, and finally the current fault condition of the data link is classified, and the diagnosis result is transmitted to the knowledge graph fault early warning module to provide the current information basis for fault early warning. The unlabeled data is converted into labeled data using the K-means method, and the final maximum interval hyperplane is trained through iterative use of the SVM algorithm. The advantage is that the K-means method can save the cost of manual labeling, and the iterative solution of the maximum interval hyperplane greatly improves the fault classification accuracy.

[0026] 2、The application can find associated faults and effectively warn by constructing a fault domain knowledge graph and inputting the classification result data of the machine learning fault classification module into the knowledge graph fault warning module. The knowledge graph can link massive different types of information together and form a relationship network, so that the user can analyze the problem from the perspective of relationship. The application can not only classify the current fault type, but also warn future faults through the constructed fault domain knowledge graph, greatly improving the data link maintenance efficiency.

[0027] 3、The application uses the Kmeans-SVM combined machine learning method to construct a classification model, and then uses the constructed fault domain knowledge graph to warn associated faults. Through the mutual cooperation of the machine learning fault classification module and the knowledge graph fault warning module, the intelligent analysis of the data link is realized, a series of problems existing in the expert system and machine learning are solved, and the maintenance efficiency of the data link is effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 A Kmeans-SVM fault classification model flowchart is established for the method of the application.

[0029] Figure 2 A logic relationship reasoning graph between the fault node information and the fault reason in the link fault knowledge graph in the application. DETAILED DESCRIPTION

[0030] The application will be further described in detail through specific embodiments. The following embodiments are only descriptive and not limiting, and cannot limit the protection scope of the application.

[0031] The application first takes the fault classification in the data link as the target, and obtains the fault classification model through the Kmeans-SVM model training. Then, the knowledge graph of the fault type is constructed through the Markov model and other methods. Based on the fault domain knowledge graph and the Kmeans-SVM method, the fault classification is obtained, and then the associated relationship between faults is analyzed through the fault reason similarity, and the associated faults that may occur in the data link are predicted.

[0032] The application provides a method for realizing data link fault classification and warning based on a knowledge graph, and the specific method steps are as follows:

[0033] (1) Establishing a Kmeans-SVM fault classification model

[0034] The specific modeling method flowchart is as follows: Figure 1As shown, first, the data is dimensionally reduced using the PCA method. The data after dimension reduction is clustered using the K-means method, taking advantage of its unsupervised learning characteristics, and the untagged substation data is clustered, where the cluster number k is set to 2, and the distance standard used in the clustering process is the Euclidean distance. The error sum of squares is used as a representative of the similarity of the samples in the cluster. The smaller the error sum of squares of a certain cluster G u , the greater the similarity of the samples in the cluster; on the contrary, the greater the error sum of squares of G u , the smaller the similarity of the samples in the cluster. The calculation formula of the error sum of squares is as follows:

[0035]

[0036] These input unlabeled data become labeled data after the K-means method. Finally, the input data is divided into two categories, represented by numbers 1 and 2. Next, use these labeled data to train the SVM model to get the maximum interval hyperplane, and divide the 1 and 2 data on both sides of the hyperplane. After the SVM training is completed, the maximum interval hyperplane is used to test the accuracy of the classification result obtained by K-means. The data predicted correctly is used to retrain the SVM hyperplane, and the SVM hyperplane is updated iteratively in this way until the error rate of the data predicted by the SVM no longer changes, and the final maximum interval hyperplane is obtained. The algorithm steps are as follows:

[0037] Step 1: Preprocess the unlabeled data, and divide the data into two clusters using the K-means algorithm, labeled as 1 and 2 respectively, and convert the unlabeled data into labeled data.

[0038] Step 2: Train the SVM classifier model using the labeled data obtained in step 1 to obtain the maximum interval hyperplane.

[0039] Step 3: Test the two sample data obtained by K-means clustering using the trained SVM classification model, and use the data predicted accurately as the input data of step 2 to obtain a new maximum interval hyperplane until the error rate of the data predicted by the SVM no longer changes.

[0040] (2) Constructing the fault domain knowledge graph

[0041] When constructing the knowledge graph of the link fault domain, it mainly includes four parts: entity and attribute extraction, coreference resolution, knowledge processing and data integration.

[0042] Entity and attribute extraction: first, the corpus of the fault domain is segmented, and the Markov model is used for entity and attribute extraction. The extracted entity and attribute are used as the final entity and attribute applied to the knowledge graph. Then, all the words are tagged with parts of speech, which are divided into the following categories: fault noun entity, fault phenomenon verb, fault degree adverb, fault degree quantifier and original word dictionary of words not extracted.

[0043] Coreference resolution: the main purpose of this step is to find out the synonyms representing entities and attributes, and to classify words with high similarity into the same category. The synonym table is used to represent synonyms.

[0044] Knowledge processing: the main purpose of this step is to identify the corresponding relationship between entities and attributes, and to delete the redundant containing relationship by taking whether there is a containing or not containing relationship between each part of speech as a standard.

[0045] Data integration: this step combines the triples of fault entities, attributes and relationships, and updates the final graph construction through the concept layer and entity attribute layer nodes to form the final link fault domain knowledge graph.

[0046] (3) Data link fault early warning

[0047] According to the logical relationship between the fault node information and the fault reason in the link fault knowledge graph, knowledge reasoning is carried out, and the reasoning rule is as shown in Figure 2 The process of knowledge reasoning between the fault node information and the fault reason in the link fault knowledge graph is as follows: first, the known knowledge stored is extracted, and after rule extraction, rule matching is carried out. If the matching is successful, it is added to the rule execution area. If the rule conflict is established, the conflict is resolved. If the rule conflict is not established, the reasoning result of not established is obtained. Similarly, if the rule matching is not successful, the reasoning result of not reasoning out new knowledge is obtained.

[0048] Although the embodiments of the present application are disclosed for the purpose of illustration, those skilled in the art can understand that various alternatives, changes and modifications are possible without departing from the spirit and scope of the present application and the appended claims, therefore, the scope of the present application is not limited to the disclosed content of the embodiments.

Claims

1.A method for realizing data link fault classification and early warning of a data middle platform based on a knowledge graph, characterized in that: The specific method steps are as follows: (1) Establishing Kmeans-SVM fault classification model Firstly, the data is reduced dimension by PCA method, and the data after dimension reduction is clustered by K-means method, which is unsupervised learning, and the unlabeled substation data is clustered, wherein the cluster number k is set to 2, the distance standard used in the clustering process is Euclidean distance, and the error sum of squares is used as the representative of the similarity of the samples in the cluster, the error sum of squares of the divided cluster G u is smaller, indicating that the similarity of the samples in the cluster is larger, and on the contrary, the error sum of squares of G u is larger, indicating that the similarity of the samples in the cluster is smaller, and the calculation formula of the error sum of squares is as follows: These input incoming unlabeled data after K-means method becomes labeled data, finally the input data is divided into two categories, respectively represented by the number 1 and 2, next use these labeled data to train SVM model, get the maximum interval hyperplane, 1 and 2 two kinds of data are divided on both sides of the hyperplane, after SVM training using maximum interval hyperplane to test the accuracy of K-means classification results, the data predicted correctly is used to retrain the SVM hyperplane, in this way the SVM hyperplane is updated iteratively until the error rate of using SVM to predict data no longer changes, get the final maximum interval hyperplane, the specific algorithm steps are as follows: Step one: the preprocessed data is used to K-means algorithm and divided into two clusters, respectively marked as 1 and 2, the unlabeled data is converted into labeled data; Step two: using the labeled data obtained in step one to train SVM classifier model, get the maximum interval hyperplane; Step three: using the trained SVM classification model to test the two kinds of sample data obtained by K-means clustering, the data predicted accurately is used as the input data of step two again, get the new maximum interval hyperplane, until the error rate of using SVM to predict data no longer changes; (2) Constructing fault domain knowledge graph When constructing the knowledge graph of link fault domain, it mainly includes four parts: entity and attribute extraction, coreference resolution, knowledge processing and data integration; (3) Data link fault early warning According to the logical relationship between fault node information and fault reason in the link fault knowledge graph, knowledge reasoning is carried out, The extraction of entities and attributes includes: firstly, performing word segmentation on the corpus of the fault domain, using the Markov model to extract entities and attributes, and using the extracted entities and attributes as the final entities and attributes applied to the knowledge graph; then, performing part-of-speech tagging on all words, which are classified into the following categories: fault noun entities, fault phenomenon verbs, fault degree adverbs, fault degree quantifiers, and original words of unextracted words; the main purpose of the steps of coreference resolution is to find out the synonymous words representing entities and attributes, and to classify the words with high similarity into the same category; the synonymous word table is used to represent the synonymous words; the main purpose of the processing steps is to identify the corresponding relationship between entities and attributes, and to delete the redundant containing relationship according to whether there is a containing or not containing relationship between the words; the data integration step is to merge the triplets of fault entities, attributes and relationships, and to update the nodes through the concept layer and the entity attribute layer to realize the final graph construction, and finally to form the knowledge graph of the link fault domain; the logical relationship between the fault node information and the fault reason in the link fault knowledge graph is as follows: firstly, extracting rules from the stored known knowledge, then performing rule matching, if the matching is successful, adding to the rule execution area, if the rule conflict is established, conflict resolution, if the rule conflict is not established, obtaining the reasoning result of not reasoning out new knowledge, and similarly, if the rule matching is not successful, obtaining the reasoning result of not reasoning out new knowledge.

Images