A behavior-orientated prediction method for short-term energy consumption of air-conditioning systems in buildings blocks

Abstract

The short-term prediction of air-conditioning (AC) energy consumption is a crucial part of building operation optimization and demand-response strategies. However, the AC energy consumption in some types of buildings can be highly stochastic due to arbitrary occupants' behavior of AC usage, which make it had to predict. Currently, very few prediction models considered the impacts of stochasticity related to occupants’ behavior of AC, which leads to less accuracy. To predict the AC energy consumption under stochastic energy use scenarios, this study proposes a behavior-orientated prediction method for short-term AC energy consumption with a case study on teaching building blocks. Firstly, cluster analysis is employed to identify typical patterns to quantify stochastic AC energy use. Then, the AC usage rate is predicted based on weighted k-Nearest-Neighbors method, which can provide a precise prediction of stochastic AC usage rate. Based on these, RandomForest method is used to develop a basic prediction model of AC energy consumption. The importance of each variable is also evaluated. Finally, an enhanced part of prediction is implemented by support vector machine to achieve higher accuracy under stochastic AC energy use scenarios. The results show that both proposed models can accurately predict the AC energy consumption, while the proposed enhanced model can reap significant accuracy improvement under the more stochastic scenarios. This study provides a new approach to predict AC energy consumption in buildings with stochastic AC energy use, which can match the real-time amount with good accuracy. The proposed model can provide an accuracy reference for the demand response strategies and optimization of AC systems operation.