Artificial Neural Network Optimization in Prediction Baseline Energy Consumption to Quantify Energy Savings in Commercial Building

Abstract

This paper presents a proper baseline energy model of a chiller system for the measurement and verification activity was developed. In measurement and verification, the baseline energy has been modelled using linear regression in finding the correlation between input and output variables. Baseline energy model was proposed applying the Artificial Neural Network. Three optimization methods, Evolutionary Programming, Particle Swarm Optimization and Artificial Bee Colony are hybridized with Artificial Neural Network. These methods were used to optimize the training process and selecting the optimal values of Artificial Neural Network initial weights and biases. The coefficient of correlation, Mean Square Error, Mean Absolute Percentage Error and Standard Error were used to measure the model's accuracy. The dataset composed of three input variables that were affecting the energy consumption of a chiller system were selected namely operating time, refrigerant tonnage and differential temperature. Meanwhile, the output was energy consumption of the building's chiller system. These three Hybrid Artificial Neural Network techniques were then compared with Linear Regression and Artificial Neural Network. The results revealed that Artificial Bee Colony Hybrid with Artificial Neural Network outperforms other methods. This selected method was further used to quantify the chiller system retrofitting energy saving. The energy saving obtained from this model was 165,478.46 kWh.