Modeling, simulation & control of photovoltaic reverse osmosis desalination syste

Desalination is the process of removal of salts from seawater or brackish water to produce fresh water. There are various desalination techniques that are available but the most popular is the reverse osmosis desalination which is based on membrane separation. This work presents the performance evaluation, system identification and control of a small-scale reverse osmosis desalination system which is intended to be used with solar panels but the solar energy is utilized to charge the batteries with MPPT (Maximum power point tracking). The RO system then operates on batteries at night, at steady pressure to ensure smooth and trouble-free operation of the system which is not guaranteed when the system is operated on solar energy. For the purpose of performance evaluation, the system is run at steady state for certain duration of time and corresponding current and voltage is measured every second. Other performance evaluation parameters namely the salt rejection, the recovery ratio and the specific energy consumption in kWh/m3 are also studied. Three discrete time models namely ARX, ARMAX and Output Error model are estimated using system identification and compared against each other. The data for identification is collected for the product water flow rate considering the pump motor control voltage as the manipulated variable. The models are also validated with a separate experimental dataset. Two state-space models are also developed using subspace identification and prediction error method and their performance is evaluated using optimal linear quadratic regulator. The best model was further investigated for different control penalties. Model predictive controller is also evaluated for the state-space model obtained from subspace identification. The model predictive control was investigated for the unconstrained and the constrained case and the robustness was also studied by considering disturbance in the input variable. Simulation results show that the constrained model predictive control is more suited for the given system. The optimal and model predictive control algorithms were developed and simulated using MATLAB.