The Soft-computing System Of Sludge Bulking Based On PSO-RBF Neural Network

Activated sludge method is one of the most extensive biological wastewatertreatment method currently. Activated sludge can remove resolvable and colloidalstate organics and suspended solid which can be adsorbed by activated sludge. Andactivated sludge also can dislodge a part of phosphorus and nitrogen. However, theremay be the phenomena of sludge bulking in every activated sludge process. Becausethe rate of sludge bulking is very high and it can give serious harm to the watertreatment. So sludge bulking has brought the huge influence of sludge bulking tosewage treatment process, but also to the city sewage treatment has brought the hugeeconomic loss. Therefore, prediction of sludge bulking is particularly important.The important parameter of sludge bulking is sludge volume index(SVI). Dueto it is very complex of wastewater treatment and it is a highly nonlinear system, andit also has the characteristics of time-varying, uncertainty. The traditionalmathematics model cannot online calculate SVI. Therefore a soft sensor modelbased on PSO-RBF is put forward through analyzing the characteristic ofwastewater deeply in this paper. The soft-measuring model can develop therelationship between SVI and other parameters in the system. It has reached thehigh-accuracy prediction of SVI with its self-learning and adaptive ability. Themodel has created a condition for SVI online detection in wastewater and automaticcontrol.The main research works of this paper are as follows:(1) A intelligent soft measurement model of SVI is put forward. We determinethe factors which influence SVI through analyzing the characteristics of wastewater.Then we reduce the influencing factors by the rough set. And we set a softmeasurement model of SVI based on RBF neural network.(2) A SVI soft measurement method based on PSO-RBF neural network isproposed. In order to improve the soft measurement accuracy of SVI, a PSO-RBFneural network is proposed. We use PSO whose weight changing by nonlinear tooptimize the weight, center and width of RBF neural network. At the same time, inorder to avoid the particles fall into local optimum, according to characteristics ofthe neural network, combining the variation thought, an improved particle swarmoptimization algorithm is proposed. We also make the learning factor variety according to the nonlinear.(3) A set of SVI soft measurement simulation software is made by VisualStudio (VS) and MATLAB development tools. With the same programming model,language and framework of WPF in VS, we design and implement a graphical userinterface which has multimedia interactive characteristics. Thus we provide aneffective tool for SVI online prediction.