Key Words: SISO; power plant; Machine Learning; PID control; error assessments.
Abstract. The research is focused on possibility of applying the Machine Learning (ML) approach, which is using previous data knowledge for the energy system and monitored process values for control behaviour. The main focus of the research is to make assessments for the quality of the control process by comparing the classical PID Control algorithm and Machine Learning Control (ML) one. The investigation is based on possibility to apply supervisor Machine Learning Control to Single Input Single Output (SISO) Power Plant. In the current research the previous information about the system data (manipulated, control, output variables and system states) are used for forming the training sets which is used for performing the Machine Learning Control. Some simulation results illustrating the behaviour of the power system, which is controlled via ML and PID have been presented. The research gives some assessments about the quality of the control by assumption for presence of low and big system disturbances and uncertainties, which may occur in energy system. In current paper applying the supervisor ML algorithm is only under consideration. The steady describes the basic architecture of developed software system implementing the SISO Plant Control. The research gives assessments for four cases – PID and ML control with low disturbances, PID and ML control with big disturbances. Some assessments about system behavior have been made with possibility to overfeed the training set applying big uncorrelated data. The software system is released with java and weka ML library. Realization of system program modules and architecture of the system has been presented.
