Electric Load Forecasting Based on Deep Ensemble Learning

Short-to-medium-term electric load forecasting is crucial for grid planning, transformation, and scheduling power supply departments. Various complex ever-changing factors such as weather, seasons, regional economic structures, enterprise production cycles exert uncontrollable effects on the load. While causal convolutional neural network can significantly enhance long-term sequence prediction, it may suffer from problems vanishing gradients overfitting due to extended time series. To address this issue, paper introduces a new data anomaly detection method, which leverages (CNN) extract temporal spatial information data. The features extracted are then processed using bidirectional long short-term memory (BiLSTM) capture dependencies in more adeptly. An enhanced random forest (RF) classifier employed Furthermore, proposes model framework electricity that combines dilated with ensemble learning. This combination addresses issues encountered networks Extreme gradient boosting (XGBoost), category (CATBoost), light machine (LightGBM) models act base learners modeling comprehend deep cross-features, prediction results generated by learning serve feature set secondary modeling. broadens receptive field of kernel. All acquired values concatenated input into training, achieving short-to-medium-term forecasting. Experimental indicate compared existing models, its root mean squared error (RMSE) (MSE) mid-term reduced 4.96% 12.31%, respectively, underscoring efficacy proposed framework.