Sequential dynamic artificial neural network modeling of a full-scale coking wastewater treatment plant with fluidized bed reactors

This study proposed a sequential modeling approach using an artificial neural network (ANN) to develop four independent models which were able to predict biotreatment effluent variables of a full-scale coking wastewater treatment plant (CWWTP). Suitable structure and transfer function of ANN were optimized by genetic algorithm. The sequential approach, which included two parts, an influent estimator and an effluent predictor, was used to develop dynamic models. The former parts of models estimated the variations of influent COD, volatile phenol, cyanide, and NH4+-N. The later parts of models predicted effluent COD, volatile phenol, cyanide, and NH4+-N using the estimated values and other parameters. The performance of these models was evaluated by statistical parameters (such as coefficient of determination (R2), etc.). Obtained results indicated that the estimator developed dynamic models for influent COD (R2 = 0.871), volatile phenol (R2 = 0.904), cyanide (R2 = 0.846), and NH4+-N (R2 = 0.777), while the predictor developed feasible models for effluent COD (R2 = 0.852) and cyanide (R2 = 0.844), with slightly worse models for effluent volatile phenol (R2 = 0.752) and NH4+-N (R2 = 0.764). Thus, the proposed modeling processes can be used as a tool for the prediction of CWWTP performance.