Modeling of urban wastewater treatment plant efficiency with artificial neural network: The case of Konya province

With the climate crisis, the necessity of using our natural resources efficiently and effectively has become a topic of global discussion. The reality of drought we are facing has brought the development of suitable technologies for the reuse of wastewater to the forefront. Predicting the parameter values of the effluent from existing wastewater treatment plants in advance will help prevent the potential harmful effects of the effluent on the receiving environment and increase the efficiency of process management. For this purpose, studies are being conducted to predict the performance of urban wastewater treatment plants using methods such as artificial neural networks, which are among the advanced computer technologies. In this thesis, the values of the influent water parameters pH, temperature, conductivity, BOD, COD, and TSS for the Konya Urban Wastewater Treatment Plant for the years 2021-2022, provided by the Konya Water and Sewerage Administration (KOSKI) General Directorate, were used to predict the values of the effluent water parameters BOD, TSS, and COD. These predictions were made using the Neural Net Fitting neural network tool of the Artificial Neural Network (ANN) module in the MATLAB(2023b) program. To ensure the model works more effectively, normalization was applied to the raw data to keep the values within a specific range (0-1). After normalization, 70% of the dataset was used for training, 15% for testing, and 15% for validation. The best learning algorithm was sought by varying the number of neurons, the number of input data, and the number of hidden layers in the modeling performed using the Levenberg-Marquardt (LM) learning algorithm. For estimation of BOD value; In the [3 20 1] network structure created in the 1st scenario, learning was achieved with R2 = 1 values in the training data set, and considering all data sets, the overall performance of the model was MSE = 0.0194, R2 = 0.7271. It was observed that the overall performance of the model in the [6 8 1] network structure, where the number of hidden layer neurons was determined as 8, was MSE = 0.014, R2 = 0.7522.