Prediction and optimization of indoor thermal environment and energy consumption based on artificial neural network

Abstract

Abstract Commercial software and simulation tools frequently used by researchers in predicting indoor thermal condition and energy consumption of a green building can ensure high precision, however, the computational process are usually time-consuming and cannot clearly and directly give suggestions for a ‘greener’ design. This paper aimed to develop a novel model based on artificial neural network (ANN) to speed up the simulation and provide optimal design solutions. Training and testing data representing design scenarios of different insulation thickness, shading coefficient, ventilation rate and their corresponding Annual Energy Demand (AED) and Uncomfortable Degree Hours (UDH) value were obtained from a numerical analysis model at first. The deduced ANN model was tested and validated, showing very high accuracy as a predictor for broad range of inputs. Relationship between design parameters and outputs were analysed and presented intuitively. The ANN results directly offered suggestions of minimizing AED and UDH, the optimum solution was to reduce ventilation and increase insulation to the limit, and reduce shading coefficient form base-case 0.5 to approximately 0.2. UDH could be lessened by 4% and AED could be reduced by nearly 29% compared to the base case design in this way.