Deep learning based power risk prevention and control tracing method

Electric power risk information has concealment and is difficult to directly utilize its temporal data characteristics. The ROC curve convergence speed of the electric power risk tracing process is low, and the TPR value during convergence is also low. Therefore, a deep learning based electric power risk prevention and control tracing method is proposed. We use a deep learning architecture that includes one-dimensional and two-dimensional convolutions to extract features from power time series data and power image data. When fitting the features using the Concatenate function, we use the Sigmoid function to correct the fitting results and improve convergence speed. Combined with the mean value of the fitted features of the power data, we set a dynamic risk assessment index - MESCM. Data with fitted feature values greater than MESCM is used as the output result of risk prevention and control tracking. In the experimental results, the F1 scores were all above 0.8, and the ROC curve converged the fastest. The TPR value at convergence was significantly higher than that of the control group, enabling accurate tracking of abnormal risk load data.