Baseline Energy Model Development Using Artificial Neural Network: Small Dataset Approach

Abstract

The population in Malaysia is growing continuously, which is contributed to the increasing in electrical energy consumption. Due to that, there is important to reduce energy consumption and cost. Therefore, Measurement &Verification (M&V) baseline energy model in the baseline period is developed to calculate energy savings in the post-retrofit period for any energy management programs. Recently, Linear Regression (LR) is a method to develop the baseline energy model, but this method is less suitable for non-linear data. Artificial Neural Network (ANN) has been widely used in predicting and forecasting in various fields. This study is to develop baseline energy models for Option C International Performance of Measurement & Verification Protocol (IPMVP) using LR, ANN and ANN with resampling techniques to compare the performance of these three models based on their accuracy. The small dataset was chosen to examine the ability of the ANN model to train its network with less amount of data. Microsoft Excel was used to develop the LR model to correlate the energy consumption with several inputs which were working days, class days and Cooling Degree Days. MATLAB software was used to develop the ANN and ANN with resampling technique models using a single hidden layer with 3, 5 and 7 numbers of neurons in the hidden layer. The model with the highest accuracy were compared and analysed. Results show that ANN with resampling technique model is the best model to choose for a small dataset due to having the highest accuracy amongst the three models.