Particle Swarm Optimization for NARX-ANN Baseline Energy Modelling in Educational Buildings

Abstract

Multiple Linear Regression (MLR) model and Non-Linear Auto-Regressive with Exogenous Input encompass Artificial Neural Network (NARX-ANN) have been widely used for energy consumption prediction. The main purpose of the prediction is to estimate potential energy saving in the building from various energy management and retrofit programs. Accurate prediction from a precise baseline energy model are important because they ensure energy savings is correctly determined. MLR and NARX-ANN have gained popularity in modeling and prediction due to thier user-friendly and simple approach. Even though NARX-ANN offers simplicity, certain limitations exist within its architecture and often associated with optimization problems. Furthermore, the MLR prediction is suitable for a system that behaves linearly. Thus, the main intention of this paper is to hybrid the NARX-ANN technique with Particle Swarm Optimization (PSO) technique for energy consumption prediction and baseline energy modeling. NARX-ANN architecture is optimized with PSO technique and the predicted outcome is compared with the actual value. The modeling error is compared between MLR and the non-optimized NARX-ANN model. The baseline energy consumption is developed for two buildings in a university compound. It is found that NARX-ANN-PSO optimization performs better in terms of error measurement. From this work, the proposed baseline energy models can be used to predict energy consumption hence determine savings accurately.