Multi-input Multi-output Self-learning-Based Control System

Abstract

We have proposed a machine learning-based approach for automation of control room operations. The central idea is to introduce machine learning in SCADA-based control rooms, to assist the operator. We are making use of gradient boosted decision tree (GBDT) to construct a learning model for decision making. We tested the performance of the GBDT against various decision tree (DT) algorithms using the classification accuracy on an industrial plant dataset provided by the Technology Development Division of Nuclear Recycle Group of Bhabha Atomic Research Center (BARC). The predictions given by the GBDT were then sent to a SCADA operator using MODBUS communication protocol over TCP/IP. The experimental results have proved that the proposed method can be useful in real-life large-sized plants where the data to be handled is very large and there is immense work pressure on the operator.