Multi-Domain Electric Load Forecasting via Attention-Based Feature Fusion

Electric load forecasting plays a vital role in the operation and planning of power plants and can help utilities and policymakers design stable and reliable energy infrastructure. In recent years, deep learning methods have achieved promising results in electric load forecasting. Various neural networks are employed to enhance electric load forecasting such as CNN or LSTM. However, there are differences in electric load curves of different domains due to different electricity utilization patterns. Existing methods often build different forecasting models for different domains. Such practice increases the complexity of the electric load forecasting system and leads to a waste of computing resources. To solve this problem, this paper proposes an attention mechanism-based electric energy prediction method for multi-domain prediction. By modeling domain information such as cities and industries, the domain information is integrated into feature representations by attention mechanism to achieve multi-domain electric energy prediction with a single model, which can effectively integrate electric load data in different fields through domain information learning to make more accurate predictions. Experiments show that the proposed method outperforms competitive baselines on the real-world dataset.

I. Introduction

Electric load forecasting plays a crucial role in efficiently managing and planning power systems. Accurate predictions of future electricity demand enable utilities and grid operators to optimize resource allocation, plan maintenance schedules, and ensure reliable and cost-effective energy supply. With modern power grids' increasing complexity and interconnectedness, there is a growing demand for accurate load forecasting[1] [2].