An accuracy model for on-line prediction of effluent ammonia nitrogen in anammox treatment system based on pca-bp algorithm

Abstract

Anaerobic ammonium oxidation (anammox) process has been recognized as efficient biological nitrogen removal process, which has the advantages of cost-effective and low energy compared to the conventional nitrification and denitrification processes. However, the efficient operation and control is limited due to the complexity of nonlinear and biochemical phenomena involved. This paper proposes an appropriate combinational model based on improved back propagation (BP) neural network to forecast effluent ammonia nitrogen concentration in anammox process, the network is optimized by the principal component analysis algorithm. As a result, the proposed PCA-BP model is a precise and efficient tool for predicting the effluent ammonia nitrogen concentration with determination coefficients (R2) was 0.997, the root mean square normalized error (RMSE) and mean absolute percentage error (MAPE) between the predicted and observed values was 17.47 and 16.07%. Therefore, the integration model can be applied in the actual measurement to timely estimate the effluent ammonia nitrogen concentration from other variables easily measured. Furthermore, the proposed model is promising for future applications of the controller in anammox process and as a tool to help systematically design logic control applications for other biological processes.