Applying machine learning based on multilayer perceptron on building energy demand in presence of phase change material to drop cooling load

Abstract

In this study, by developing a multi-layer neural network, energy consumption in buildings with phase change material (PCM) was investigated. Three input parameters including transition temperature, setpoint, and PCM thickness were combined in a way that covers 324 different cases. The results of the neural network indicated that it is possible to predict the amount of cooling load with an accuracy of more than 95%. In a situation where the temperature is set at 20°C, if PCM with a melting point of 26°C is added with a thickness of 10 cm, the cooling demand experience a sharp decrease by 373.8 kWh/m2. Considering energy saving of 59.1 kWh/m2 for heating load, the annual energy demand of the building declined by 432.9 kWh/m2. As the setpoint increases, the effect of the phase change becomes more intense in heating demand than in cooling one. At a setpoint of 28°C, only the PCM with a phase change temperature of 29°C is useful and in this case, the cooling/heating loads are reduced by 242.7 and 218 kWh/m2 which resulted in annual energy saving of 460.7 kWh/m2. The neural network showed that if the internal temperature is adjusted between 22 and 28°C, the phase transition of 29°C is the most suitable case.