Short and Medium-term Power Load Anomaly Detection Method Based on Convolutional Neural Network and EL-DCC

Power load anomaly detection is critical to grid stability and security. With the increasing complexity of load patterns, it is difficult for traditional methods to meet the requirements of accurate detection. Therefore, this study proposes a short and medium-term power load anomaly detection method that combines convolutional neural networks and ensemble learning with deep convolutional classifiers to improve detection accuracy and robustness. The model first extracts local features using convolutional neural network, then captures the time series dependencies by bi-directional long short-term memory, and finally the classification judgment is made by random forest. The experimental results indicated that on the Global Energy Forecasting Competition 2012 Dataset, the convergence speed of the power load anomaly detection model proposed by the research was faster, and the loss value was stabilized at about 0.2 after about 40 iterations. Moreover, the detection accuracy gradually increased and stabilized with the increase of the quantity of samples, and the maximum value was about 0.98. The accuracy of the power load forecasting detection model proposed by the study was maximum about 0.97 when the quantity was 1000. Its accuracy gradually increased with the increase of the iterations to reach the maximum value of about 0.95. The study shows that the proposed model can effectively detect and predict power load anomalies with good precision and accuracy. The proposed method provides an efficient anomaly detection and forecasting scheme for the power system and improves the intelligent management level of the power grid.