Analysis of Machine Learning algorithms for prediction of Condenser Vacuum in Thermal Power Plant

The demand from the grid requires a flexible operation approach by the generating stations. It requires the generator to vary load as per the requirements of the grid which changes rapidly. One of the parameters crucial for power generation is condenser vacuum. The operator needs to keep a constant watch on condenser pressure for safe and reliable generation. At low load, to save Auxiliary Power Consumption (APC) operator takes several steps and makes a trade off with condenser vacuum. In this context, it becomes essential to have a tool which can assimilate changes in the system and predict condenser vacuum. Parameters on which Condenser Vacuum largely depends upon, is the Cooling Water (CW) inlet temperature, CW Outlet Temperature, CW flow to Condenser, Exhaust Hood Temperature and instant power generated by the unit. A guiding tool can be handy for the operator to switch off CW pumps, Cooling Tower (CT) fans and other auxiliaries which affect Condenser Vacuum in order to save APC. An advance prediction tool can also assist the operator to analyze any unwanted changes in the operating parameters of the unit. In this paper we will be using several regression algorithms on sample plant data to train our predictive model and to adjudge the prediction given by each algorithm for condenser vacuum. The data used for training and testing of model is obtained from the 210 MW Russian Turbine based Thermal Power Station.