Author:

Keywords:

Energy prediction, SVM-regression, FFD, Energy Management

Abstract:

Considerable energy savings in industrial environment are possible in an industrial environment by detecting installations not working at their optimum operating point. The present paper proposes a new generalized data driven FDD method capable of automatically detecting the abnormal energy demand of different types of installations or machines based on process data. The paper contains a comprehensive overview of the research, focusing on a trade-off between performance and computing time together with minimizing the human input. The proposed method contains an automated feature selection, a hyper-parameter optimization of the chosen SVM regression algorithm and a residual control algorithm. The method was tested in several industrial installations and two case studies are presented to demonstrate the performance of the proposed method, while underlining the significance of a decent number of relevant features.