A Forecast Based Load Management Approach For Commercial Buildings -- Comparing LSTM And Standardized Load Profile Techniques

Abstract

Load-forecasting problems have already been widely addressed with different approaches, granularities and objectives. Recent studies focus not only on deep learning methods but also on forecasting loads on single building level. This study aims to research problems and possibilities arising by using different load forecasting techniques to manage loads. For that the behaviour of two neural networks, Long Short-Term Memory and Feed Forward Neural Network and two statistical methods, standardized load profiles and personalized standardized load profiles are analysed and assessed by using a sliding-window forecast approach. The results show that machine learning algorithms have the benefit of being able to adapt to new patterns, whereas the personalized standardized load profile performs similar to the tested deep learning algorithms on the metrics. As a case study for evaluating the support of load-forecasting for applications in Energy management systems, the integration of charging stations into an existing building is simulated by using load forecasts to schedule the charging procedures. It shows that such a system can lead to significantly lower load peaks, exceeding a defined grid limit, and to a lower number of overloads compared to uncontrolled charging.