Predictive Analytic for Estimating Electric Consumption of Smart Grid Platform in Residential Single-Family Building using Support Vector Regression Approach

Abstract

This present paper attempts to corroborate the theoretical expert knowledge model-based data with the data-driven method in order to optimize energy efficiency use by high fluctuate occupants’ behavior level in typical residential single-family building at tropical country (i.e. Indonesia). Existing data set gathered from field observation and finished computed with Building Performance Simulation (BPS) software tools. The methodology for data-driven knowledge is using supervised machine learning with support vector regression technique, hence the physical engineering data become training data for learning purpose. Predicting algorithm is done in sequential minimal optimization for regression task (SMOreg), a library for support vector machine integrated in WEKA software with using radial basis functions (RBF) as the kernel. Moreover, electric bills are included to forecast future economic value of using smart grid technology.