Prediction model for coal chemical wastewater quality after catalytic ozonation process treatment based on deep learning algorithm: Performance evaluation and model comparisons

Abstract

In this study, we tested different models for predicting the effluent water quality of coal chemical wastewater treated by the catalytic ozonation process. The optimal model was further modified using the norm feature vector selection methods and intelligent algorithms. The average removal amount and average removal rate of COD in coal chemical wastewater treated by catalytic ozonation process were 223.62 mg/L and 57.98 %, respectively. AAO-pH, AAO-ammonia nitrogen, AAO-total phosphorus, AAO-COD, AAO-total phenol, SST- COD, SST-total phenol, and SST-turbidity were inputs for each model, while the catalytic ozonation-COD was used as the output for each model. We collected 500 samples of one year for the training, validation, and testing of various models. Compared to other models, the RF model had better predictive performance. Additionally, L1 method had a stronger optimization ability for the RF model than L2 method. Compared with BOA and SSA, IGWO had a larger contribution to improving the predictive ability. The L1-RF-IGWO model showed the best predictive ability (test set R² of 0.8717). The present results can aid coal chemical plants achieve data-driven management of water quality prediction and monitoring, providing ideas for the construction of models for predicting water quality in engineering.