Prediction model of short-term electrical load in an air conditioning environment

In a building office, an air-conditioning system is one of the systems that contributes most to the electrical energy expense. The ability to predict the short-term electrical energy consumption in an air-conditioning environment can provide valuable information in controlling electrical appliance usages so that the overall energy consumption can be kept at an acceptable level for most of the time. In this paper, we apply data mining techniques to the short-term prediction of energy consumption in air-conditioning rooms typically found in an office building. Energy consumption data and related variables in actual air-conditioning environments are collected, preprocessed, and fitted to three different models, including Multiple Linear Regression (MLR), Artificial Neural Network (ANN), and Bagged Decision Tree (BDT). Unlike previous works that use only temperature and humidity as predictors, we include additional factors such as room size and BTU of air-conditioning units to improve the prediction accuracy. Our results show that the highest accuracy is achieved by using the ANN model with all the predictors included.