Intelligent optimization strategy for electrochemical removal of ammonia nitrogen by neural network embedded in a non-dominated sorting genetic algorithm

Abstract

Electrochemical is a promising approach for the removal of ammonia nitrogen, but the challenge is to achieve better performance under lower energy consumption. In this study, a new electrochemical system hybrid with intelligent optimization algorithms was developed for the efficient removal of ammonia nitrogen and energy saving. Ammonia removal performance and energy consumption were recorded when the traditional electrochemical system operated at various parameters. As the premise of model training, the data were processed through Scatter diagram matrix, Box plot, Principal Component Analysis, Spearman correlation and Shapley Additive Explanations to evaluate the redundancy and independence of parameters. Backpropagation neural network based on deep learning was used as surrogate model of non-dominated sorted genetic algorithm-II, meanwhile, the range of electrochemical parameters was used as the constraint for multi-objective optimization. The optimized result is the Pareto front and the optimal solution was obtained by combining the Technique for Order Preference by Similarity to an Ideal Solution. This new hybrid system achieved an increase of 11.75 % ~ 13.61 % in ammonia removal and a reduction of 21.31 % ~ 36.84 % in energy consumption. The optimal solution represents a better ammonia removal performance at low energy consumption, which is meaningful for the concept of a real-time controlled electrochemical system.