A Multi-task Learning Approach to Short-Term Load-Forecasting for Multiple Energy Loads in an Educational Building

Accurate forecast of end use energy in smart buildings play an important role in the building's systems which can release the energy savings potential of the buildings sector. Therefore, energy load forecasting in buildings can be used for control purposes, and more efficient energy use by reduce the cost of unnecessary energy usage in smart buildings. The increased quantity of energy related-data that generated from Building Energy Management Systems facialists the utilization of Machine Learning and Deep Learning approaches in the area of building energy consumption and performance. Once there are multiple historical time-series datasets associated with different types of energy loads in building, multiple tasks of time-series forecasting should be solved concurrently. Therefore, this study propose a novel Multi-task Learning approach for the day-ahead Short-Term-Load-Forecasting for Heat pump, Air Handling Unit and Lighting loads in an educational building. The proposed approach has construed by utilizing Long-Short-Term-Memory with Encoder-Decoder architecture. The proposed approach has been implemented on a real-time datasets that have been generated from an educational building in the city of Newcastle upon Tyne in the UK. The dataset was 11 month long with one-hour interval, where 10 months used for training purposes, and 1 month for the final testing. The simulation results indicate that the proposed approach proves itself to has a satisfactory forecasting accuracy for the task of Short-Term-Load-Forecasting. The proposed framework can be applied to other real-world case studies, e.g., a group of domestic or commercial buildings.