Smart Grid Power Load Prediction Method Based on VMD-TCN-BiGRU-TSAM

To address the challenges of high volatility and low prediction accuracy in smart grid power load data, we propose a novel power load prediction method that integrates Variational Mode Decomposition (VMD), Temporal Convolutional Networks (TCN), Bi-directional Gated Recurrent Units (BiGRU), and a Temporal Self-Attention Mechanism (TSAM). Our approach begins with the application of VMD to decompose the smart grid power load data, resulting in subsequences characterized by stronger regularity. We then utilize the Maximum Information Coefficient (MIC) to identify weather factors that exhibit a strong correlation with power load variations. These selected weather factors, along with historical load data and the decomposed subsequences, are reorganized to form a comprehensive new load dataset. Each subsequence is subsequently input into the combined TCN-BiGRU model, where the feature vectors derived from this process are classified using an enhanced BiGRU model that incorporates the Temporal Self- Attention Mechanism (TSAM) to enhance prediction performance. Finally, we aggregate the predicted values from all subsequences to obtain a complete set of predicted load values. We validate our proposed method using real power load data from a station area in a city in China. The results demonstrate that our algorithm achieves superior prediction accuracy compared to traditional forecasting methods, thereby contributing to more effective optimization of power dispatch in smart grids.