An Adaptive data driven approach to single unit residential air-conditioning prediction and forecasting using regression trees

Residential Air Conditioning (AC) load has a huge role to play in Demand Response (DR) Programs as it is one of the power intensive and interruptible load in a home. Due to the variety of ACs types and the different sizes of residences, modelling the power consumption of AC load individually is non-Trivial. Here, an adaptive framework based on Regression Trees is proposed to model and forecast the power consumption of different AC units in different environments by taking in just 6 basic variables. The framework consists of an automatic feature selection process, a load prediction module, an indoor temperature forecasting module, and is capped off by a load forecasting module. The effectiveness of the proposed approach is evaluated using data set from an ongoing research project on air-conditioning system control for energy management in a residential test bed in Singapore. Experiments on highly dynamic loads gave a maximum Mean Absolute Percentage Error (MAPE) of 21.35% for 30min ahead forecasting and 27.96% for day ahead forecasting.