Modelling Adjusted Baseline Energy in an Office Building Using Artificial Neural Network
This paper proposes Artificial Neural Network (ANN) to determine adjusted baseline energy for quantifying energy savings from an energy efficiency program implemented in an office building. The input data to the ANN includes number of working days and cooling degree days (CDD) each month for one year period before implementation of the retrofitting program. On the other hand, output data is baseline energy use (i.e. energy use before retrofit). Since the input data to the network encompasses of 36 months set of data only, Bootstrap method is used to generate more input data without changing the input and output trend of the original data set. This is performed to increase validity of the training process. Once the optimum training parameters have been obtained, adjusted baseline energy is determined by feeding the number of working days and CDDs in the post-retrofit period (i.e. 12 months set of data) to the network. Energy savings is then calculated by comparing the adjusted baseline energy with the energy use after implementing the retrofit program. The performances of the ANN model are then compared with Multi-regression technique in term of R2, Mean Absolute Percentage Error (MAPE), Mean Square Error (MSE) and Mean Absolute Deviation (MAD). Results show that the proposed ANN model has smaller errors and R2 closer to one compare to Multi-regression technique.