RNN based Disturbance Rejection Model Predictive Control for Waste Heat LTTD Process

Thermal Desalination is one of the most preferable desalination techniques for the island peoples to achieve high quality fresh water yield from sea water. However, it needs high energy requirements and capital cost. A Waste Heat Low Temperature Thermal Desalination (WHLTTD) is an energy efficient desalination technique for producing freshwater from seawater using waste heat from condenser reject water of coastal thermal power plant. The main factors affecting the quantity and quality of fresh water yield in WHLTTD are variations in temperature gradient and salinity of sea water due to thermal power plant load demand and climatic variations. In this paper, novel disturbance prediction based control schemes with Feed Forward (FF) controller are proposed to minimize the effect of temperature variations on fresh water yield using Long Short-Term Memory (LSTM) deep Recurrent Neural Network (RNN). Firstly, Real time data are collected from 5 m
³
/day capacity laboratory based WHLTTD experimental setup and data driven models are developed for process and temperature disturbance. Secondly, two predictive models are developed to predict the Sea Surface Temperature (SST) using ERA5 monthly averaged historical dataset and waste heat water temperature using 525 MW thermal power plant surface condenser data set. Thirdly, Predictive model based Feed Back (FB) controllers namely Proportional Integral(PI) and Model Predictive Controller(MPC) with Feed Forward (FF) controller are developed to control freshwater flow rate by manipulating waste heat warm water flow rate for known temperature disturbance. Finally, the closed loop performances of the proposed control schemes are investigated. The results show the feasibility of the proposed control schemes to improve the quantity of fresh water yield in terms of Integral Square Error (ISE) and unit cost of production in terms of Control Effort(CE).