Predictive Modelling of Kaliti Wastewater Treatment Plant Performance Using Artificial Neural Networks

Artificial neural networks are a form of artificial intelligence that have the capability of learning, growing, and adapting within dynamic environments. A reliable model for any wastewater treatment plant is essential in order to provide a tool for predicting its performance and to form a basis for controlling the operation of the process. This would enable to assess the stability of environmental balance at minimized operational costs. Wastewater treatment process is complex and attains a high degree of nonlinearity due to the presence of bio-organic constituents that are difficult to model using mechanistic approaches. Predicting the plant operational parameters using conventional experimental techniques is also a time consuming step and is an obstacle in the way of efficient control of such processes. In this work, a soft computing approach based on back propagation artificial neural networks, which employed genetic algorithm and partial mutual information to perform input selection, was used to acquire the knowledge base of Kaliti wastewater treatment plant and has been applied for predicting and optimizing selected plant performance variables viz. pH, BOD5, COD, NH3, and TDS effluent concentration of the plant. In the model structure of the treatment plant performance, two different functional structures in the configuration of the network are constructed and compared. In the first configuration Multiple-Input-Single-Output (MISO), structures differing in the type of data used, raw operational data and outlier removed data in the input layer, are used to build models for each of the five performance indicators selected in this work. Partial Mutual Information-based Input Selection (PMIS) and Genetic Algorithm (GA) based input selection algorithms are applied for both above mentioned MISO configuration based models. In the second configuration Multiple-Input-Multiple-Output (MIMO), GA based input selection is applied for both the raw and outlier removed data. The model performance was evaluated with statistical parameters and the simulation results indicates that the MISO modelling approach achieves much more accurate predictions as compared with the MIMO modelling approach. Optimum model architecture of 14-43-1 for pH, 16-29-1 for BOD5, 14-50-1 for COD, 6-28-1 for NH3, and 10-48-1 for TDS were selected for predicting the performance of Kaliti wastewater treatment plant using MISO configuration. The linear correlation between predicted outputs and target outputs for the optimum model architecture described above are 0.97 for pH, 0.94 for BOD5, 0.98 for COD, 0.94 for NH3, and 0.98 for TDS. Keywords: Artificial neural networks; Back propagation; Genetic algorithm; Partial mutual information; Wastewater treatment