🤖 AI Summary
Accurately forecasting building-level short-term thermal loads is crucial for enhancing the efficiency of district heating systems, yet it is significantly influenced by exogenous factors such as outdoor temperature and occupant behavior. This study leverages hourly data from 25 buildings in Germany spanning 2017–2025 to systematically compare, for the first time in a multi-building setting, the performance, cross-building generalization capability, and computational overhead of xLSTM, Temporal Fusion Transformer (TFT), and fully connected networks across 3-hour and 24-hour forecasting horizons. Results show that xLSTM achieves the lowest RMSE (3h: 19.88 kWh; 24h: 21.47 kWh), while TFT yields the best MAE (3h: 9.16 kWh). Notably, lightweight models attain comparable accuracy, revealing a pronounced imbalance between the marginal gains of high-parameter models and their associated computational costs.
📝 Abstract
Obtaining an accurate short-term forecasting for heat demand is an essential part of operating district heating networks cost-efficient and reliable. Heat consumption time series at the building level are highly dependent on exogenous variables such as outdoor temperature and individual usage patterns, making forecasting in this context a challenging task. Thus, this paper benchmarks novel Transformer-based and xLSTM architectures for short-term heat-demand forecasting. Using hourly data from 25 German buildings (2017-2025), we compare three-hour and 24-hour forecasting horizons relevant for intraday control and day-ahead scheduling. We establish a multi-building benchmark that tests whether models trained on pooled, heterogeneous building data are able to generalize across diverse building stock. The results show that the xLSTM achieves the lowest RMSE (19.88 kWh for three-hour, 21.47 kWh for 24-hour forecasts), while the Temporal Fusion Transformer attains the best MAE (9.16 kWh for three-hour forecasts). As xLSTMs and Transformers require long training times and have a huge number of trainable parameters, their sustainability remains questionable. Therefore, this paper further investigates the trade-off between predictive accuracy and computational resource demand of the evaluated forecasting models. The findings indicate that also low-parameter models like a traditional fully-connected network achieve good predictive results, highlighting that marginal accuracy gains of the novel prediction models come at substantial resource expense for this use case.