Application of machine learning and big data for smart energy management in manufacturing

Abstract

Abstract Within the Industry 4.0 context, platforms such as cyber-physical production system (CPPS) offer numerous opportunities for smart energy management in manufacturing. In this study, we demonstrate the application of big data and machine learning (ML) to foster such practices for real manufacturing environments by taking the Model Factory (MF) in Singapore as a testbed. We first used supervised learning algorithms to predict machine-specific load profiles via energy disaggregation at the MF shop floor. Here, the light gradient boosting machines had the best predictive performance with a mean absolute error and root mean squared error of 0.035 and 0.106 (units in Watts). We then coupled unsupervised learning with mathematical optimization to devise an optimal energy scheduling plan for facility management at the MF. When applied for day-ahead scheduling, the data-driven optimizer showed cost benefits of 14% in comparison to the current existing conditions. The study successfully demonstrated the application of big data and ML in the drive towards smart manufacturing practices.