Determination of operating conditions in activated sludge process using fuzzy neural network and genetic algorithm

In order to realize control of activated sludge process, a simulation model for effluent chemical oxygen demand (COD) was constructed using the time series data of three months. Here, the recursive fuzzy neural network (RFNN) was applied for the simulation. The simulation model could estimate effluent COD value with relatively high accuracy (average error: 0.68 mg/l). Next, to control effluent COD value to the desirable level, the search system for the values of the control variables, dissolved oxygen concentration (DO) and mixed liquor suspended solid (MLSS), was constructed using the genetic algorithm (GA) and GA with the reliability index (R1), called as RIGA. In search for DO and MLSS values, accuracy of GA search system was high (average error: 0.16 mg/l for DO and 214 mg/l for MLSS) and accuracy of RIGA search system was higher than GA (average error: 0.11 mg/l for DO and 144 mg/l for MLSS). Then, the search using RIGA was further extended for one-year data to check the ability of this system. As a result, the constructed system could search DO and MLSS values with the average errors of 0.10 mg/l and 162 mg/l, respectively.