Interpretable Load Patterns of Building District Energy Systems using Attention-based LSTM

Abstract

With the increasing demand for energy and focus on environmental sustainability, district energy systems (DESs) have emerged as a promising solution. To optimize DES operations and energy savings, accurate load forecasting is crucial. This study proposed an LSTM model with an attention mechanism for accurate heating load forecasting within a real DES. By introducing an attention mechanism, the heatmaps generated by weight distribution can reveal the load pattern’s periodicity and building thermal inertia. Research on single buildings and district systems has shown that load forecasting with district systems is more stable regarding forecasting accuracy and load pattern extraction capability under irregular external disturbances. The outcomes illustrate the effectiveness of the proposed framework in accurately predicting heating loads and extracting interpretable load patterns. This can assist building managers in enhancing operational strategies, resulting in energy conservation.