Heavy metal removal performance of capacitive deionization technology studied by machine learning

Abstract

Abstract Capacitive deionization (CDI) technology is utilized for efficient treatment of industrial wastewater, characterized by low energy consumption and environmental protection. In order to comprehend the correlation between key experimental parameters and the electrosorption capacity (EC) of heavy metals in CDI technology, this paper employs a genetic algorithm (GA) to optimize a backpropagation artificial neural network (BPANN) for predicting the EC of CDI technology for heavy metal ions, with the characteristics of electrode materials converted into numerical characteristics for further analysis. Compared to the BPANN, the optimized GABPANN model demonstrates superior predictive accuracy. It achieves automatic adjustment of the hidden layer structure, neuron count, and transfer functions. Furthermore, the grey relational analysis indicates that the electrode material and the initial pH value of the solution are pivotal in determining the EC of heavy metal ions. This underscores the efficacy of machine learning (ML) algorithms in forecasting the nonlinear dynamics of CDI systems and elucidates the influence of individual parameters on the efficacy of heavy metal removal.