Energy storage sizing for office buildings based on short-term load forecasting

This paper presents a three-layer Artificial Neural Network as the short-term load forecasting model adopting the fastest back-propagation algorithm with robustness, i.e., Levenberg-Marquardt optimization, and moreover, the momentum factor is considered during the learning process. Based on predicted data by aforementioned model, size determination of energy storage system in terms of power rating and capacity is undertaken according to the desired level of shaving peak demand. The illustrative example in reference to the weather and power load data of office building from July to August in 2011 gets the results that the average relative error -0.7% and the root-mean-square error 2.79% which show aforementioned forecasting model can work effectively with the attractive percentage, i.e. 87.5%, of error within the acceptable one 2.79%; Furthermore, size determination of energy storage system adopting battery energy storage technology, i.e. 7.03kW/36.42kWh, is carried out to meet the desired peak shaving demand.