Multi-energy Load Forecasting of Integrated Energy Systems Based on MMoE-Trans-LSTM

With the development of Integrated Energy Systems (IES), accurate forecasting of multi-energy loads is crucial for enhancing energy efficiency and ensuring reliable energy supply. However, due to the high complexity and coupled relationships between different load types, traditional forecasting methods struggle to simultaneously handle the diverse characteristics of multi-energy load data. This paper proposes a multi-energy load forecasting model based on MMoE-Trans-LSTM, which integrates Long Short-Term Memory (LSTM) layers, Transformer encoding layers, and Multi-gate Mixture-of-Experts (MMoE) multi-task learning architecture. First, the model uses LSTM layers to extract temporal features, capturing dependencies in loads. Next, the Transformer encoding layer captures long-range dependencies in loads, thereby enhancing the model’s ability to handle complex patterns. Finally, the MMoE layer employs multiple expert networks to handle the features of different load tasks and produce weighted outputs, improving the accuracy and robustness of predictions for each task. Case studies demonstrate that the proposed model significantly outperforms traditional single-task learning models and other benchmark models, especially in electricity and gas load forecasting tasks.