Construction of COD Simulation Model for Activated Sludge Process by Recursive Fuzzy Neural Network

Using a fuzzy neural network (FNN), we constructed a simulation model which estimates the effluent chemical oxygen demand (COD) value from daily routine measurements. Since the water quality of wastewater is changing day by day, an FNN model with a recursively renewing method of learning data (R-FNN) is proposed. With this R-FNN, learning data used to construct an FNN model are renewed with elapsed time so as to estimate the effluent COD value with good accuracy. The estimation results for 9 weeks data using R-FNN were compared with those using a conventional FNN. The average error using the R-FNN model was 0.36 mg/l, while that using the conventional FNN was 1.50 mg/l. Moreover, estimation of the effluent COD throughout one year was carried out, and the average error was only 0.40 mg/l. This result can show the usefulness of the R-FNN for the simulation model of the activated sludge process.