Artificial Intelligence-based Digital Fault Diagnosis and Prediction for Power Grids

Abstract

When power grid faults occur, especially complex faults, there are many uncertainties such as switch and protection mis-operation, the power system response will be complicated, which causes many difficulties in power grid fault diagnosis. This paper uses the word2vec model vectorization to process the digitized alarm information during grid faults. The processed fault features are input into the DPCNN model to extract global features of the alarm information. Then, the fully connected layer is used to classify grid faults accurately. Subsequently, a convolution module based on the self-attention mechanism is proposed to achieve accurate prediction of grid faults, and the ReLU function and Dropout strategy are used to realize the optimization of the grid fault diagnosis and prediction model. The simulation model test results reveal that the proposed model can effectively diagnose and predict grid faults, with an average accuracy of 97.05% and 95.93%, respectively. The response time for fault diagnosis in this paper’s model for the empirical application of grid diagnosis is reduced from 6.32 minutes to 0.96 seconds, significantly improving diagnosis efficiency compared to the traditional method. This paper provides an effective method for diagnosing and predicting power grid faults and a solution for improving the management of power grids.

Keywords

• word2vec
• dpcnn model
• self-attention mechanism
• relu function
• grid diagnosis and prediction
• 93c62