Neural network-based forecasting of energy consumption due to electric lighting in office buildings

This paper presents a novel method to predict the energy consumption due to electric lighting in office buildings, knowing the external daylight. For each month an ideal reference irradiation curve is calculated based on the actual irradiation curves of the days of that month. The office (business) operation hours are seen as a sequence of time intervals (of a few hours) based on the usual office use. The system uses as input parameters the day, the time (hour and minutes), the month, the average difference between the actual irradiation curve and the ideal reference irradiation curve in the considered interval, the instantaneous difference between the two curves at that time, and the average actual irradiation in the considered interval. The system predicts the average active power in the following interval. In this way the electric energy consumption is essentially influenced by the quality of the irradiation curve in the considered day. The system was developed as a feed-forward artificial neural network, which was applied to a case study concerning a small office located in Italy. In the experiments, made on the data pertinent to six months, we achieved an average RMSE error which represents 17.25% of the monthly average electric power.