Voltage Sag Classification And Recognition Based On Fusion Of Knowledge Graph And Convolutional Neural Network

With the rapid development of new energy and the wide use of sensitive power electronic equipment,the power quality problem has been paid more and more attention.Because of the high frequency of voltage sag and the strong destructive to sensitive power electronic equipment,it becomes the most prominent problem in power quality.Voltage sag mainly includes short circuit fault,large motor start,transformer switching and other types.Different types of Voltage Sag have different characteristics and different harm to power and electronic equipment.Accurate identification of voltage sag type is an important premise for the formulation of targeted governance scheme and the definition of the responsibilities of both parties to reduce unnecessary disputes.Firstly,this paper analyzes and summarizes the background and significance of voltage sag classification and the existing methods.The classification and identification methods of voltage sag are mainly divided into two types:mechanism analysis and data driven.The method based on mechanism analysis has many disadvantages because it is unable to construct a physical model describing complex disturbance and the feature loss in the feature extraction stage manually.Although the classification and identification method of voltage sag based on data-driven avoids the limitation of the physical model based on mechanism analysis,it has some problems such as relying on large-scale annotation data and model incomprehensibility.Secondly,according to the above problems,this paper proposes a classification and recognition method of voltage sag which combines knowledge graph and convolutional neural network.(1)The large amount of voltage sag data stored in the power quality data center is used as the data source to construct an interpretable knowledge graph of voltage sag for regional power grid,(2)the entity,relationship and relationship plane of triple structure in knowledge graph are represented and studied by TransH model,and the discrete knowledge in knowledge graph is converted into low-dimensional real value vector,(3)convolutional neural network is used to replace TransH The score function of the model can enhance the learning ability of the model,(4)determine the loss function and take it as the model objective to guide the model learning process.Finally,the experiment shows that the convolution neural network can be used to replace the score function of TransH,which can effectively solve the problem of low accuracy in voltage sag classification recognition by using the score function in TransH.Meanwhile,by adding the knowledge graph of voltage sag,which has a lot of expert experience,convolutional neural network can learn more comprehensive data characteristics of voltage sag events during training,thus improving the accuracy of classification and identification of different voltage sag events and effectively reducing the dependence of convolutional neural network mode on large-scale annotation data.In addition,because of the characteristics of rich entity semantics,excellent knowledge and interpretability,the network model is well interpretable by combining it with deep learning technology.