Evaluation and improvement of energy consumption prediction models using principal component analysis based feature reduction

Abstract

The building sector is a major source of energy consumption and greenhouse gas emissions in urban regions. Several studies have explored energy consumption prediction, and the value of the knowledge extracted is directly related to the quality of the data used. The massive growth in the scale of data affects data quality and poses a challenge to traditional data mining methods, as these methods have difficulties coping with such large amounts of data. Expanded algorithms need to be utilized to improve prediction performance considering the ever-increasing large data sets.In this paper, a preprocessing method to remove noisy features is coupled with predication methods to improve the performance of the energy consumption prediction models. The proposed preprocessing method is based on the well-known principal component analysis (PCA) and treats the historical meteorological and energy data of buildings. The cleaned and processed data are used in five prediction models including linear regression, support vector regression, regression tree, random forest and K-nearest neighbors.The proposed methodology is applied to four case studies with different climate zones (cold, mild, warm-dry and hot-humid) to study the effect of dataset patterns on the feature reduction and prediction performance. The results show that the proposed method enables practitioners to efficiently acquire a smart dataset from any big dataset for energy consumption prediction problems. In addition, the best prediction model for each climate zones with considering mean square error, R2, residual values and execution time is proposed.