Research On Soft-sensing Method Based On Neural Network For Wastewater Treatment

With the rapidly development of modern industry, the water resource was destroyed more and more seriously, which is human's subsistence, so the process control of wastewater treatment has been the focus of study in recent years.Many methods have been put forward concerning wastewater treatment. The methods of Sequencing Batch Reactor (SBR) is used more and more widely because of its so many advantages. But, because the wastewater quality parameters (such as BOD, COD, N, P and so on) can't be detected on-line or real-time measured worse, its lead to the short of effective control parameters and affects the control effect and economic benefit deeply. Wastewater treatment is controlled by settled-time at present. Thus, how to gain the wastewater quality parameters on-line exactly and economically to carry out close-loop control has been the key to improve the control effectiveness for enterprises.Because SBR is a typical complex dynamic engineering system of biology, it is nonlinear, time-variable, stochastic, uncertain and difficult to establish mathematics model between Primary variable and Secondary variable. This paper applied soft-measurement technology based on neural network to online-measurement of wastewater treatment according to engineering application, made the best of the capability of networks' nonlinear approach and learning and got a good result.Firstly, this paper summed up the principle and treatment of wastewater at present. Secondly, this paper analyzed the relation between treatment specification and the parameters that could be measured on-line systematically, and proposed the soft-sensing modeling method. Thirdly, in terms of the need to modeling the soft-sensing model of wastewater, studied parameters measurement method by instrument and experiment respectively. Fourthly, according to the characteristics of SBR and modeling based on neural network, the soft-sensing method based on BP (back-propagation) and RBF (radial basis function) neural networks is proposed to solve this problem accordingly. The models were trained by the data from the experiments of wastewater treatment; the results of network training coincide with those of wastewater treatment in practice. Therefore, it can be safely said that the soft-sensing system of wastewater treatment based on BP and RBF neural networks is capable of effectively resolving the problem quality parameters real-time estimation in wastewater treatment.