Prediction of effluent ammonia nitrogen in wastewater treatment plant based on self-organizing hybrid neural network

Timely and accurate prediction of key indicators of sewage is the focus of intelligent sewage treatment research. But traditional deep learning model performs unsteadily in the case of large fluctuations of inlet water quality. Self-organizing map (SOM) is a typical unsupervised learning algorithm, which can automatically classify the input patterns according to its learning rules. Herein, a self- organizing classification integrated model based on SOM algorithm was established to predict effluent ammonia nitrogen (NH3−N) of a wastewater treatment plant in China. Influent chemical oxygen demand (COD), influent total phosphorus (TP), influent NH3-N, influent total nitrogen (TN) and pH were selected as the input indicators. Firstly, SOM algorithm is used to divide the original data set into several different categories according to their characteristics. Secondly, the different categories are used to train different Radial Basis Function models (RBF). Finally, the input data is sorted through SOM to find the corresponding unit (RBF), depending on what to predict effluent NH3-N. After evaluation, the results show that the proposed integrated model accurately predicts the value of effluent NH3-N, with the values of root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) are 0.06, 0.05, 2.19, respectively. Moreover, the integrated model also outperforms single RBF, with 55.61 % lower RMSE, 48.66 % lower MAE, and 48.31.90 % lower MAPE. This study is helpful for the development of intelligent sewage treatment.