Neural network model for short-term and very-short-term load forecasting in district buildings

Abstract

Load forecasting plays an important role in energy management in smart buildings. It is expected that precise prediction of loads can bring significant economic benefits to smart buildings by enabling accurate demand response strategies for peak load reduction, reducing electricity use and integrating distributed energy resources. The complexity increases for load prediction in case of a district of buildings having much functional diversity and including several heterogenous buildings-blocks. The main goal of this paper is to present a comprehensive and detailed study for very-short term and short-term load forecasting in a district building using artificial neural network (ANN). The main objectives of this paper are: (a) Evaluate the performance of the ANN considering two back-propagation learning algorithms, namely Bayesian regularization (BR) and Levenberg-Marquardt (LM); (b) Analyse the relative performance of the model for hour-ahead and day-ahead load forecasting for different types of buildings; (c) Investigate how the network design parameters such as number of hidden layers, hidden neurons, number of inputs and training data affect the model’s ability to accurately forecast loads. In order to demonstrate the efficiency of the proposed approach, it is examined on real-world data of a Campus in downtown Montreal that includes many types of buildings.