Prediction of Sewage Wastewater Quality Based on PSO-LIBSVM

Aiming at the problems of nonlinearity, time-varying and big lagging in an activated sludge wastewater treatment process, the forecast modeling of COD can be established according to the historical data of chemical oxygen demand(COD) collected from sewage plant, and using the LIBSVM toolbox to determine the model structure and parameters. With the use of the output error data and the particle swarm algorithm, we can optimize the parameters of support vector machine(SVM) and correct model, until the output error is minimum. The results on simulation show that the more simple modeling process, the prediction effect will be much better . Compared with the BP neural network, the standard SVM model, it can reflect the characteristics of COD distribution in the future time.