ADAPTING RUNGE - KUTTA LEARNING ALGORITHM IN ANFIS FOR THE PREDICTION OF COD FROM AN UP-FLOW ANAEROBIC FILTER

Water consumes vast area in the earth’s surface and safe drinking water is essential for humans and other organisms to survive in the world. Eliminating waste matters from water is the necessary requirement nowadays. The ultimate purpose of wastewater treatment is the protection of good quality water which is the most priceless resource. Use of Artificial Neural Network (ANN) models is gradually increasing to predict wastewater treatment plant variables. This detection helps the operators to take proper action and manage the process accordingly as per the norms. Anaerobic processes are often preferred to aerobic processes for treating waste streams that contain high Chemical Oxygen Demand (COD) concentrations. Up-flow Anaerobic Filter (UAF) is a common process used for various anaerobic wastewater treatments. COD is used to measure the strength (in terms of pollution) of waste water. COD level in the effluents of the UAF determines the pollutants in the wastewater. The proposed method uses cheese whey as an influent. It is tested in the anaerobic reactor using COD test to predict the level of oxygen requirement of the effluent. Predicting the effluent parameters is a time consuming process when using Classical Models as it involves complexity and high non-linearity. Hence the proposed method uses an efficient technique namely Z-Score Normalization technique as a preprocessing step, Particle Swarm Optimization (PSO) for feature selection process and Adaptive Neuro-Fuzzy Inference System (ANFIS) with RungeKutta Learning Method (RKLM) as a learning algorithm is used for prediction of COD. Experiments conducted on a real data indicates that the application of Z-Score normalization schemes followed by a PSO feature selection and ANFIS with RKLM prediction results in better performance compared to other methods.