An Attention-BiGRU Based Power Generation Forecasting Method for Multi-Energy Systems

The high complexity, nonlinearity, and data heterogeneity of multi-energy systems (MES) pose significant challenges in power generation forecasting. To solve these problems, a novel power generation forecasting method is proposed, leveraging a bidirectional Gated Recurrent Unit (BiGRU) integrated with an attention mechanism. First, considering the heterogeneity of environment information data, one-hot encoding is employed to convert all environment information data into numerical features, supplemented by a feature weighting mechanism to assess the significance of each feature. Secondly, considering the intricate temporal correlations within the features, a multi-layer attention mechanism is proposed to facilitate the multidimensional extraction of key features, thereby enhancing the model's ability to capture contextual relationships in time-series sequences. Finally, a BiGRU based encoder-decoder architecture is utilized to predict the power generation value in MES. Numerical results demonstrate that the proposed method significantly improves the prediction accuracy compared to baseline schemes.