CN-121456770-B - Power grid new energy operation fault discrimination method and system based on knowledge enhancement

CN121456770BCN 121456770 BCN121456770 BCN 121456770BCN-121456770-B

Abstract

The invention discloses a power grid new energy operation fault judging method and system based on knowledge enhancement, and relates to the technical field of wind farm fault diagnosis. The method comprises the steps of obtaining time sequence data of operation of new energy equipment of a power grid, constructing a graph structure according to the time sequence data of the operation of the new energy equipment of the power grid to obtain graph data representation, encoding expert knowledge into a hierarchical rule set according to the characteristics of the operation data of the new energy equipment of the power grid, constructing a knowledge rule base according to the hierarchical rule set, constructing a fault diagnosis model by using a graph neural network with a knowledge enhancement mechanism, training the fault diagnosis model by using the hierarchical rule set in the knowledge rule base as a teacher and the graph neural network as a student, and detecting abnormality of the graph data representation by using the trained fault diagnosis model to obtain a wind power plant fault judgment result. The invention can realize the unsupervised and accurate identification of complex fault scenes such as gear box faults, abnormal pitch system and the like.

Inventors

Lv tianguang
ZHANG XIXIAN
ZHANG DUXI
LI YIXIAO
LI JING
AI QIAN
YANG MING
LI PENG

Assignees

山东大学

Dates

Publication Date: 20260508
Application Date: 20260105

Claims (10)

1. The power grid new energy operation fault judging method based on knowledge enhancement is characterized by comprising the following steps of: Acquiring time sequence data of the operation of the new energy equipment of the power grid, and constructing a graph structure according to the time sequence data of the operation of the new energy equipment of the power grid to obtain graph data representation; Expert knowledge is encoded into a layering rule set according to the characteristic of the operation data of the new energy equipment of the power grid, and a knowledge rule base is constructed according to the layering rule set; Constructing a fault diagnosis model by using a graph neural network with a knowledge enhancement mechanism, taking a layering rule set in a knowledge rule base as a teacher, taking the graph neural network as a student, and training the fault diagnosis model; an improved teacher-student training scheme is adopted, knowledge rules are used as teachers for evaluating the reasonableness of the state of the GNN output node, and the training process is guided by the improved teacher-student training scheme; Generating knowledge signals by converting knowledge rules into discriminant functions applicable to node state vectors, wherein each rule R corresponds to a function Its output value represents the node state The degree to which the rule is met; Design of loss function, total loss The model is used for initial training stage of model, and the loss function consists of two parts, namely data reconstruction loss ensures that the model can learn the normal operation mode in SCADA data, knowledge consistency loss embeds expert summarized physical rules and fault priori knowledge into the model in the form of soft constraint, and the weight parameters of the graphic neural network are directly optimized, and are the basis for building model capacity, and the total loss function is specific The method consists of data reconstruction loss and knowledge consistency loss: wherein λ is a superparameter for balancing the weights of the two losses; is a reconstruction error based on self-supervised learning of the input data; the confidence coefficient and expected value difference obtained after all knowledge rules are applied to the GNN output node embedded H are defined, and binary cross entropy loss is adopted: ; and carrying out anomaly detection on the graph data representation by using the trained fault diagnosis model to obtain a wind power plant fault discrimination result.
2. The knowledge-based enhanced power grid new energy operation fault discrimination method according to claim 1, wherein the specific steps of constructing a graph structure according to time sequence data of power grid new energy equipment operation are as follows: Node characteristics are designed according to the physical characteristics of the operation of the new energy equipment of the power grid; Constructing a directed graph by adopting a sliding window method based on node characteristics to obtain an adjacency matrix, specifically abstracting time sequence data of the operation of new energy equipment of a power grid into a graph structure, representing the state of the equipment at key time by the nodes, and representing causal dependency on a time sequence by the edges; and fusing the node characteristics based on an attention mechanism to obtain a dimension reduction characteristic matrix.
3. The method for discriminating new energy operation faults of power grid based on knowledge enhancement according to claim 1 wherein the specific step of encoding expert knowledge into a hierarchical rule set according to the characteristics of new energy equipment operation data of power grid is as follows: designing a knowledge rule level according to fault scenes with different complexity; Based on a knowledge rule hierarchy, expert knowledge related to wind farm anomaly detection is converted into continuous vector representation according to the characteristics of the operation data of the new energy equipment of the power grid, wherein the knowledge rule hierarchy comprises a unique alarm layer, a fixed combination layer and a complex combination layer, the unique alarm layer is suitable for a simple fault scene, the fixed combination layer is suitable for medium-complexity faults, and the complex combination layer is suitable for complex faults.
4. The knowledge-based power grid new energy operation fault discrimination method according to claim 1, wherein the specific steps of using the trained fault diagnosis model to perform abnormality detection on the graph data representation are: Processing the graph data representation by using the trained fault diagnosis model to generate a node low-dimensional embedded representation; calculating an anomaly score represented by node low-dimensional embedding through an energy function by adopting a limited Boltzmann machine; And obtaining a wind power plant fault discrimination result according to the abnormality score.
5. The method for discriminating new energy operation faults of a power grid based on knowledge enhancement as claimed in claim 4, wherein the specific steps of obtaining the wind power plant fault discrimination result according to the abnormality score are as follows: Setting a self-adaptive dynamic abnormality judgment threshold according to the historical detection result; and comparing the abnormality score with a dynamic threshold value to obtain a wind power plant fault discrimination result.
6. The utility model provides a new energy operation fault discrimination system of electric wire netting based on knowledge reinforcing which characterized in that includes: The data acquisition module is configured to acquire time sequence data of the operation of the new energy equipment of the power grid, and construct a graph structure according to the time sequence data of the operation of the new energy equipment of the power grid to obtain graph data representation; the rule base building module is configured to encode expert knowledge into a layering rule set according to the characteristics of the operation data of the new energy equipment of the power grid, and build a knowledge rule base according to the layering rule set; the model construction module is configured to construct a fault diagnosis model by using a graph neural network with a knowledge enhancement mechanism, takes a layering rule set in a knowledge rule base as a teacher, takes the graph neural network as a student, and trains the fault diagnosis model; an improved teacher-student training scheme is adopted, knowledge rules are used as teachers for evaluating the reasonableness of the state of the GNN output node, and the training process is guided by the improved teacher-student training scheme; Generating knowledge signals by converting knowledge rules into discriminant functions applicable to node state vectors, wherein each rule R corresponds to a function Its output value represents the node state The degree to which the rule is met; Design of loss function, total loss The model is used for initial training stage of model, and the loss function consists of two parts, namely data reconstruction loss ensures that the model can learn the normal operation mode in SCADA data, knowledge consistency loss embeds expert summarized physical rules and fault priori knowledge into the model in the form of soft constraint, and the weight parameters of the graphic neural network are directly optimized, and are the basis for building model capacity, and the total loss function is specific The method consists of data reconstruction loss and knowledge consistency loss: wherein λ is a superparameter for balancing the weights of the two losses; is a reconstruction error based on self-supervised learning of the input data; the confidence coefficient and expected value difference obtained after all knowledge rules are applied to the GNN output node embedded H are defined, and binary cross entropy loss is adopted: ; the abnormality detection module is configured to perform abnormality detection on the graph data representation by using the trained fault diagnosis model to obtain a wind farm fault discrimination result.
7. The knowledge-based enhanced grid new energy operation fault discrimination system of claim 6, wherein the data acquisition module is further configured to: Node characteristics are designed according to the physical characteristics of the operation of the new energy equipment of the power grid; Constructing a directed graph by adopting a sliding window method based on node characteristics to obtain an adjacency matrix, specifically abstracting time sequence data of the operation of new energy equipment of a power grid into a graph structure, representing the state of the equipment at key time by the nodes, and representing causal dependency on a time sequence by the edges; and fusing the node characteristics based on an attention mechanism to obtain a dimension reduction characteristic matrix.
8. The knowledge-based enhanced grid new energy operation fault discrimination system of claim 6, wherein the rule base creation module is further configured to: designing a knowledge rule level according to fault scenes with different complexity; Based on a knowledge rule hierarchy, expert knowledge related to wind farm anomaly detection is converted into continuous vector representation according to the characteristics of the operation data of the new energy equipment of the power grid, wherein the knowledge rule hierarchy comprises a unique alarm layer, a fixed combination layer and a complex combination layer, the unique alarm layer is suitable for a simple fault scene, the fixed combination layer is suitable for medium-complexity faults, and the complex combination layer is suitable for complex faults.
9. The knowledge-based enhanced grid new energy operational fault discrimination system of claim 6, wherein the model building module is further configured to: Processing the graph data representation by using the trained fault diagnosis model to generate a node low-dimensional embedded representation; calculating an anomaly score represented by node low-dimensional embedding through an energy function by adopting a limited Boltzmann machine; And obtaining a wind power plant fault discrimination result according to the abnormality score.
10. The knowledge-based enhanced grid new energy operation fault discrimination system of claim 9, wherein the anomaly detection module is further configured to: Setting a self-adaptive dynamic abnormality judgment threshold according to the historical detection result; and comparing the abnormality score with a dynamic threshold value to obtain a wind power plant fault discrimination result.

Description

Power grid new energy operation fault discrimination method and system based on knowledge enhancement Technical Field The invention relates to the technical field of wind farm fault diagnosis, in particular to a method and a system for discriminating new energy operation faults of a power grid based on knowledge enhancement. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. With the popularization of wind farm intelligent monitoring technology, SCADA is widely applied to data acquisition of wind turbine running states. Based on analysis of the multi-parameter time sequence data, early fault early warning and diagnosis of the wind turbine generator can be achieved. However, the conventional fault diagnosis method (such as a method based on a physical model) is highly dependent on expert experience, and has problems of poor adaptability, difficult modeling and the like. To introduce an intelligent means, early artificial intelligence methods represented by expert systems were applied. According to the method, logic reasoning is carried out according to manually summarized rules and priori knowledge, and although experience dependence is partially reduced, a complete knowledge base is still required to be constructed, and unknown faults are difficult to process. With the progress of technology, a purer data driving method represented by deep learning is gradually becoming the main research flow by virtue of the strong self-learning and nonlinear fitting capabilities. Common techniques include automatic encoders, long and short term memory networks, and graphic neural networks, among others. The method comprises the steps of detecting abnormality through reconstruction errors in an automatic encoder and other unsupervised methods, wherein the method is sensitive to dynamic working condition changes, a long-term and short-term memory network can capture time sequence dependence and often neglect spatial association among variables, and a graph neural network can model topological relations among variables, but most of the methods assume static and fixed graph structures, are difficult to adapt to high dynamic characteristics of wind power data, and have the limitations of complex calculation, weak noise resistance and the like. Under the background that the scale of a wind farm is continuously enlarged and the running data is continuously increased, the accuracy of fault diagnosis is seriously dependent on the feature extraction and associated mining capability of an algorithm on high-dimensional, nonlinear and strong-coupling SCADA data. However, the existing method cannot effectively fuse multi-source information, cannot well adapt to dynamic changes and has efficiency and robustness. In addition, most of the existing artificial intelligent diagnosis schemes often adopt a single model or an isolated learning mechanism, and lack of multi-level knowledge fusion and self-adaptive optimization capability, so that generalization performance is limited in a complex fault scene, and diagnosis precision still has room for improvement. Disclosure of Invention Aiming at the defects existing in the prior art, the invention aims to provide a new power grid energy operation fault judging method and system based on knowledge enhancement, which can realize the unsupervised and accurate identification of complex fault scenes such as gear box faults, abnormal pitch system and the like. In order to achieve the above object, the present invention is realized by the following technical scheme: The invention provides a power grid new energy operation fault judging method based on knowledge enhancement, which comprises the following steps: Acquiring time sequence data of the operation of the new energy equipment of the power grid, and constructing a graph structure according to the time sequence data of the operation of the new energy equipment of the power grid to obtain graph data representation; Expert knowledge is encoded into a layering rule set according to the characteristic of the operation data of the new energy equipment of the power grid, and a knowledge rule base is constructed according to the layering rule set; Constructing a fault diagnosis model by using a graph neural network with a knowledge enhancement mechanism, taking a layering rule set in a knowledge rule base as a teacher, taking the graph neural network as a student, and training the fault diagnosis model; and carrying out anomaly detection on the graph data representation by using the trained fault diagnosis model to obtain a wind power plant fault discrimination result. Further, the specific steps of constructing the graph structure according to the time sequence data of the operation of the new energy equipment of the power grid are as follows: Node characteristics are designed according to the physical characteristics of the operation of the new