Experimental investigation and bio-inspired optimization of fixed bed photocatalytic reactor system for chlortoluron removal from water

This work aims to predict and optimize the efficiency and energy requirements of a continuous photocatalytic fixed bed reactor towards removal of chlortoluron, an organic herbicide of emerging concern, from synthetic wastewater. The photo-degradation process is optimized by integrating a multi-objective genetic algorithm with a machine learning model. The experimental study using ultraviolet irradiation and titanium dioxide catalyst revealed that maximum degradation of 94% was reached at optimum conditions with an irradiation time of 420 min, a chlortoluron concentration of 10 mg L⁻¹, a recirculating flowrate of 91.1 mL min⁻¹, a distance between the lamp and reactor of 8 cm, and a free pH of 6.5. The performance of two machine learning models namely, artificial neural network and support vector machines, was investigated for forecasting the herbicide removal yield and the energy requirements evaluated in terms of electric energy per order. The performance metrics showed that both models were capable of producing accurate predictions, with the neural network results being slightly superior. To search the optimal values of the degradation process parameters, the neural networks was selected as objective function for the genetic algorithm. Among the thirty-five Pareto solutions, one optimal solution is selected using the Technique for Order Preference of Similarity to Ideal Solution and the recommended values of the objective functions are 94% for removal efficiency and 588 KWh m⁻³ order⁻¹ for energy. These values were in satisfactory agreement with the experimental results. Thus, the proposed approach appears to be effective for predicting and optimizing the performance of photo-catalytic reactors.