Ship power system equipment fault diagnosis method based on knowledge graph and convolutional neural network

Abstract

The invention provides a ship power system equipment fault diagnosis method based on a knowledge graph and a convolutional neural network. The ship power system equipment fault diagnosis method comprises the following steps: (1) preprocessing a large amount of data acquired by sensors and monitoring equipment on ship power system equipment; (2) extracting knowledge graph entities and relationships from the processed data, training a relationship prediction model to obtain triple elements of a knowledge graph, and constructing the knowledge graph; (3) training the constructed knowledge graph, converting data features into entity vectors as input of a convolutional neural network for training, and obtaining a fault diagnosis and fault prediction model of the ship power system; and (4) giving a user side query instruction, performing real-time data prediction by utilizing the trained model, and adjusting model parameters to enable the model to be better suitable for an actual scene.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent fault diagnosis of ship power system equipment, and in particular to a ship power system equipment fault diagnosis method based on knowledge graph and convolutional neural network. Background Art

[0002] With the rapid development of the shipbuilding industry, the complexity and technical level of ship power systems are constantly improving, and the demand for fault diagnosis is also increasing. As the core component of ships, the stability and reliability of ship power systems are directly related to the safety and economy of ships. However, due to the wide variety of ship power system equipment, complex structure, and harsh operating environment, traditional fault diagnosis methods are often difficult to meet actual needs. Therefore, how to improve the accuracy and efficiency of ship power system fault diagnosis has become a hot topic and difficulty in current research. Traditional ship power system fault diagnosis methods mainly rely on expert experience and rule bases, but this method has certain limitations. First, the experience-based diagnosis method requires a lot of expert knowledge accumulation, and the diagnosis results are easily affected by subjective factors; second, the construction and maintenance of the rule base requires a lot of manpower and time, and it is difficult to cope with the challenges of system complexity and environmental variability. With the development of artificial intelligence and big data technology, data-driven fault diagnosis methods have gradually become the focus of research.

[0003] Knowledge graphs use nodes and lines to represent entities and relationships, and can achieve structured representation and reasoning of knowledge. Knowledge graphs can effectively integrate the relationships between various types of equipment and various fault types, and realize the visualization of text data. Deep learning has powerful data processing capabilities. For the large amount of multidimensional data generated by ship power system equipment, it can effectively extract and utilize data features, and achieve high-precision fault diagnosis and prediction through model training. Summary of the invention

[0004] The present invention mainly aims at the fault diagnosis problem of ship power system equipment. On the basis of feature extraction of historical data of ship power system equipment, the present invention integrates knowledge graph and convolutional neural network to analyze the useful information in multi-dimensional data of ship power system equipment and establish a fault diagnosis model.

[0005] A method for fault diagnosis of ship power system equipment based on knowledge graph and convolutional neural network, the main steps of which include:

[0006] S1: Preprocess a large amount of historical operating data obtained by sensors and monitoring equipment on ship power system equipment, including the elimination of erroneous data and the extraction of data features.

[0007] S2: Extract knowledge graph entities and relationships based on the processed data, train the relationship prediction model, obtain the triple elements of the knowledge graph, and construct the knowledge graph.

[0008] S3: Train the constructed knowledge graph, convert the data features into entity vectors as the input of the convolutional neural network for training, obtain the fault diagnosis and fault prediction model of the ship power system, and save the model training parameters and results.

[0009] S4: Give user-side query instructions, use the trained model to perform real-time data prediction, and adjust model parameters to make the model better applicable to actual scenarios.

[0010] Furthermore, in S1, the specific steps of preprocessing a large amount of historical data of the ship power system equipment are as follows:

[0011] S1.1: Collect maintenance manuals, pre-process the knowledge in the field of ship power system failures based on expert experience, and define the data set required for knowledge extraction.

[0012] S1.2: Divide the large amount of historical data of ship power system equipment into structured data and unstructured data.

[0013] S1.3: Perform data cleaning, including missing value processing, noise removal and outlier detection; perform data feature extraction to serve as training data for the model.

[0014] Furthermore, in S2, the specific steps of constructing the knowledge graph are as follows:

[0015] S2.1: For structured data, the content in the table can be directly regarded as entity attributes, entity names, and relationship names as part of the data required to build the knowledge graph.

[0016] S2.2: For unstructured data, it is necessary to train an entity recognition model to extract the entity attributes in the data as the constituent elements of the knowledge graph triples.

[0017] S2.3: For unstructured data, it is necessary to train a relationship extraction model. This step extracts entity relationships from the data to build a complete knowledge graph.

[0018] S2.4: Integrate all valid data extracted from structured data and unstructured data, divide the data into three categories: head entity, tail entity and relationship, and store them in the Neo4j database to realize the visualization of text data.

[0019] S2.5: Further perform knowledge reasoning on the knowledge graph constructed in S2, strengthen the correlation between data, and mine the potential information of the data.

[0020] Furthermore, when performing entity recognition and relationship extraction on unstructured data in S2.2 and S2.3, the text of the integrated unstructured data is firstly annotated.

[0021] Furthermore, the BERT-BiLSTM-MHA-CRF model is used for named entity recognition. The BERT-BiLSTM-MHA-CRF model is divided into three layers: BERT layer, BiLSTM layer and CRF layer. The BERT layer is used as a word embedding layer to build text vectors. By learning the input text data, it finds the regular features between named entities and data labels, preparing for entity extraction and relationship prediction; the BiLSTM layer is composed of a multi-layer neural network, which is mainly used to process contextual information and extract text entities; the CRF layer establishes the migration rules of named entities and corresponding relationship labels by learning the relationship between labels, and finally realizes the prediction of the extraction result labels.

[0022] Furthermore, when S2.5 performs knowledge reasoning on the constructed knowledge graph, the knowledge base data is imported into the Apache Jena Fuseki reasoning tool, and the rules to be inferred are manually set to complete the reasoning of the knowledge graph.

[0023] Furthermore, the knowledge graph and convolutional neural network are integrated in S3, and the neural network model is trained. The specific steps are as follows:

[0024] S3.1: Use the TransD method to embed the knowledge graph of ship power system equipment. The specific content is: for the triple (h, r, t) of the ship power system knowledge graph, through the projection matrix M r Map the head entity h and tail entity t in the triple to the space of relation r to obtain the low-dimensional embedding vector c of the knowledge graph i ∈R k×1 .

[0025] S3.2: Fault information of a certain equipment in the ship power system t=w 1n =[w1,w2…,w n ], its feature vector w 1n =[w1w2…w n ]∈R d×n Combined with the low-dimensional embedding vector, we get W = [w1w2…w n c1c2…], as the input of the CNN convolutional layer, filter h∈R d×l The activation function is:

[0026]

[0027] Then, the optimal vector data is obtained through the maximum pooling operation, and all the features are The final combination result is represented by e(t), and the specific formula is as follows:

[0028]

[0029] Among them, p is the number of convolution kernels.

[0030] Further, add the triple vector Provide more training data, the formula is as follows:

[0031]

[0032] Among them, context(c) is the set of adjacent nodes of node c, and then c i and Perform a nonlinear transformation:

[0033]

[0034] Among them, g is the transformation function of entity linking, and its formula is:

[0035] g(e)=tanh(Me+b)

[0036] Where M∈R d×k is the mapping matrix, b∈R k×l is the deviation. The three vectors are input into the CNN model through the tanh function. The specific process is:

[0037]

[0038] Repeat the above maximum pooling operation, input the final combined features into the Softmax function, and obtain the fault diagnosis results of the ship power system equipment:

[0039]

[0040] Among them, s j and b j Respectively represent the output y j The parameters and bias of , n represents the number of fault categories.

[0041] Advantages of the invention:

[0042] Knowledge graphs and convolutional neural networks are both important tools in the field of artificial intelligence. The present invention constructs a knowledge graph and extracts fault text information to visualize the fault data of ship power system equipment. The constructed knowledge graph is combined with the convolutional neural network to train a fault diagnosis model for the ship power system equipment, thereby improving the accuracy of fault prediction and diagnosis results.

[0043] The present invention combines knowledge graphs with convolutional neural networks to establish a ship power system equipment fault diagnosis model that can monitor in real time and respond quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart for specific implementation.

[0045] Figure 2 Establish a flow chart for the ship power system fault diagnosis model. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The present invention provides a method for diagnosing faults of ship power system equipment based on knowledge graph and convolutional neural network. The method extracts features from multi-dimensional data of ship power system equipment, constructs a knowledge graph and integrates a convolutional neural network training model to analyze the fault types of the ship power system.

[0048] The technology used in the present invention is: a method for diagnosing faults of ship power system equipment based on knowledge graph and convolutional neural network. Please refer to the attached figure for the specific steps as follows:

[0049] S1: Preprocess a large amount of historical operating data obtained by sensors and monitoring equipment on ship power system equipment, including the elimination of erroneous data and the extraction of data features.

[0050] S2: Extract knowledge graph entities and relationships based on the processed data, train the relationship prediction model, obtain the triple elements of the knowledge graph, and construct the knowledge graph.

[0051] S3: Train the constructed knowledge graph, convert the data features into entity vectors as the input of the convolutional neural network for training, obtain the fault diagnosis and fault prediction model of the ship power system, and save the model training parameters and results.

[0052] S4: Give user-side query instructions, use the trained model to perform real-time data prediction, and adjust model parameters to make the model better applicable to actual scenarios.

[0053] Furthermore, in S1, the specific steps of preprocessing a large amount of historical data of the ship power system equipment are as follows:

[0054] S1.1: Collect maintenance manuals, pre-process the knowledge in the field of ship power system failures based on expert experience, and define the data set required for knowledge extraction.

[0055] S1.2: Divide the large amount of historical data of ship power system equipment into structured data and unstructured data.

[0056] S1.3: Perform data cleaning, including missing value processing, noise removal and outlier detection; perform data feature extraction to serve as training data for the model.

[0057] Furthermore, in S2, the specific steps of constructing the knowledge graph are as follows:

[0058] S2.1: For structured data, the content in the table can be directly regarded as entity attributes, entity names, and relationship names as part of the data required to build the knowledge graph.

[0059] S2.2: For unstructured data, it is necessary to train an entity recognition model to extract the entity attributes in the data as the constituent elements of the knowledge graph triples.

[0060] S2.3: For unstructured data, it is necessary to train a relationship extraction model. This step extracts entity relationships from the data to build a complete knowledge graph.

[0061] S2.4: Integrate all valid data extracted from structured data and unstructured data, divide the data into three categories: head entity, tail entity and relationship, and store them in the Neo4j database to realize the visualization of the knowledge graph.

[0062] S2.5: Further perform knowledge reasoning on the knowledge graph constructed in S2, strengthen the correlation between data, and mine the potential information of the data.

[0063] Furthermore, when performing entity recognition and relationship extraction on unstructured data in S2.2 and S2.3, the text of the integrated unstructured data is first annotated. The specific contents of the annotation are as follows:

[0064] Use text annotation tools to annotate the entities and relations that need to be extracted in the text with specific labels as training data for the BERT-BiLSTM-MHA-CRF model.

[0065] The model is trained based on the labeled text data. The BERT layer converts the data into character encoding, and then uses the BERT model to train information outputs such as text vectors and word embedding representations. The bidirectional propagation of the BiLSTM layer is used to simultaneously process the text features extracted from the context information. Finally, the CRF layer label transfer rules are trained to achieve the extraction of entities and prediction of relationships in the final text.

[0066] Furthermore, in the process of realizing knowledge graph visualization in S2.4, the Python programming language is integrated, and the head entity, tail entity and relationship in the text data are imported into the Neo4j graph database using the data trained by the model to realize the visualization of the knowledge graph.

[0067] Furthermore, when S2.5 performs knowledge reasoning on the constructed knowledge graph, the knowledge base data is imported into the Apache Jena Fuseki reasoning tool, and the rules to be inferred are manually set to complete the reasoning of the knowledge graph.

[0068] Furthermore, the knowledge graph and convolutional neural network are integrated in S3, and the neural network model is trained. The specific steps are as follows:

[0069] S3.1: Use the TransD method to embed the knowledge graph of ship power system equipment. The specific content is: for the triple (h, r, t) of the ship power system knowledge graph, through the projection matrix M r Map the head entity h and tail entity t in the triple to the space of relation r to obtain the low-dimensional embedding vector c of the knowledge graph i ∈R k×1 .

[0070] S3.2: Fault information of a certain equipment in the ship power system t=w 1n =[w1,w2…,w n ], its feature vector w 1n =[w1w2…w n ]∈R d×n Combined with the low-dimensional embedding vector, we get W = [w1w2…w n c1c2…], as the input of the CNN convolutional layer, filter h∈R d×l The activation function is:

[0071]

[0072] Then, the optimal vector data is obtained through the maximum pooling operation, and all the features are The final combination result is represented by e(t), and the specific formula is as follows:

[0073]

[0074] Among them, p is the number of convolution kernels.

[0075] Further, add the triple vector Provide more training data, the formula is as follows:

[0076]

[0077] Among them, context(c) is the set of adjacent nodes of node c, and then c i and Perform a nonlinear transformation:

[0078]

[0079] Among them, g is the transformation function of entity linking, and its formula is:

[0080] g(e)=tanh(Me+b)

[0081] Where M∈R d×k is the mapping matrix, b∈R k×l is the deviation. The three vectors are input into the CNN model through the tanh function. The specific process is:

[0082]

[0083] Repeat the above maximum pooling operation, input the final combined features into the Softmax function, and obtain the fault diagnosis results of the ship power system equipment:

[0084]

[0085] Among them, s j and b j Respectively represent the output y j The parameters and bias of , n represents the number of fault categories.

[0086] The present invention discloses a method for diagnosing faults of ship power system equipment based on knowledge graph and convolutional neural network. The method effectively utilizes multi-dimensional data of ship power system by combining knowledge graph and convolutional neural network, and significantly improves the accuracy of model fault prediction and diagnosis results.

Claims

1. A method for diagnosing faults in ship power system equipment based on knowledge graph and convolutional neural network as described in this claim is characterized by: Collect a large amount of historical operating data of ship power system equipment collected by sensors and monitoring equipment, and pre-process the data. The knowledge graph entities and relationships are extracted from the processed data, and the relationship prediction model is trained to obtain the triple elements of the knowledge graph and construct the knowledge graph. The constructed knowledge graph is trained, and the data features are converted into entity vectors as the input of the convolutional neural network for training. The fault diagnosis and fault prediction model of the ship power system is obtained, and the model training parameters and results are saved.

2. The main steps of constructing the knowledge graph described in this claim are: S1: For structured data, the content in the table can be directly regarded as entity attributes, entity names, and relationship names as part of the data required to build the knowledge graph. S2: For unstructured data, it is necessary to train an entity recognition model to extract the entity attributes in the data as the constituent elements of the knowledge graph triples. S3: For unstructured data, it is necessary to train a relationship extraction model. This step extracts entity relationships from the data to build a complete knowledge graph. S4: Integrate all valid data extracted from structured data and unstructured data, divide the data into three categories: head entity, tail entity and relationship, and store them in the Neo4j database to realize the visualization of text data. S5: Further perform knowledge reasoning on the knowledge graph constructed in S2, strengthen the correlation between data, and mine the potential information of the data.

3. The knowledge graph combined with the convolutional neural network described in this claim has the main feature of using the TransD method to embed the knowledge graph of ship power system equipment, using the training results of the knowledge graph model as the input of the convolutional neural network, and further training through convolutional layers, pooling layers and Softmax functions to obtain a ship power system equipment fault diagnosis model.

Images