Short-term forecast model of cooling load using load component disaggregation

摘要

Data-driven approaches are widely applied in predicting the cooling load of buildings. Among these approaches, modelling the decomposed components of the cooling load can best capture data characteristics to enhance prediction performance. To date, however, no conventional decomposition technique has extracted physically meaningful components which consequently limits their capacity for improving prediction accuracy. This paper proposes a short-term forecast model of cooling load using load component disaggregation (LCD). First, dictionary learning and sparse representation algorithms are applied to extract four sub-loads: conduction, solar, fresh air and internal. Subsequently, a back propagation neural network and auto-regressive integrated moving average algorithm are adopted to construct forecasting models for these four loads, and a predicted cooling load is obtained by aggregating the sub-load results. The results of this simulation case study of a typical civilian building in Tianjin show that the proposed forecasting method has high accuracy. The paper then explores the influence of disaggregation and prediction techniques on forecasting accuracy, indicating that LCD improves prediction performance. The proposed method could illuminate current practice and bring more effective solutions for predicting building energy consumption.