Machine Learning in Electricity Load Forecasting of Prosumer Buildings
Abstract: Short-term load forecasting plays a key role in energy optimizations such as peaking shaving and cost arbitrage. Forecasting the aggregated load of a city or region has been researched for years and produced accurate results with time lead ranging from an hour to a week. However, little attention has been paid to the building level due to the fact that its dynamics are considerably different from those of a utility or other middle or large-scale customers. This thesis work focuses on short-term load forecasting at a building level, which is more challenging and will be taken as the pre-work for peak shaving optimization by employing Battery Energy Storage System (BESS). The forecast method is based on Artificial Intelligence (AI) and Machine Learning which is the most flourishing field in the present time. Extreme Gradient Boosting (XG-Boost) and Correlation are used to filter redundant features, and Auto-correlation is used to find similar hours in the past. Support Vector Machines (SVM) with different kernel functions and Long short-term memory (LSTM) are applied for day-ahead and hour-ahead load forecasting. The data used in the modeling are hourly electricity load and meteorological data collected by University of Central Florida and the nearest climatological data station. The forecasting performances are estimated and show that LSTM works better in hour-ahead, day-ahead and peak hours prediction at the cost of longer training time.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)