Soft Measurement Modeling Of Efluent Total Phosphorus Based On Improved Echo State Network

Vigorous social progress and economical boosting cause increasingly serious environmental pollution and ecological deterioration, such as the highlighted water pollution. Realistically, it is significant to step up efforts to monitor water quality, both for nationally economic development and for individual citizens’ health. Currently, eutrophication is nearly unmanageable resulting in its high incidence rate, as numerous factors simultaneously affects its complex mechanism. Online water quality monitoring is limited by the difficulty of achieving the key parameters in eutrophication, such as Total Phosphorus(TP). Water quality monitoring, the precondition of water evaluation, is capable of early warning water pollution.In recent years, soft sensing modelling based on artificial neural networks(ANNs) have been widely applied into precise construction of models for complex nonlinear systems. Targeting to complicated nonlinear dynamic features with noises in the sewage treatment system, this study constructed a data-derived soft-sensors for predicting the effluent total phosphorus(TP) concentration on the basis of echo state neural network(ESN). Due to their ability of approximating nonlinear functions to any degrees of accuracy and of treating dynamic information excellently, this type of models is competent to simulate nonlinear dynamic change processes in sewage treatment systems so as to predict the effluent TP concentration online.This research mainly involves the following parts:1. This study designed a scheme for constructing a soft-sensing model for prediction of the effluent TP concentration. The construction process involves collection of related data to the sewage treatment process and pretreatment of sample data;selection of TP-related secondary variables using principal component analysis(PCA); construction of soft sensors based on ANN.2. A Particle Swarm Optimization Algorithm with Adaptive Mutation(AMPSO) was developed in this study. Wiener-Hopf equation, which is usually employed to train output connect weights in ESN, has a negative effect on the stability of ESN and its prediction accuracy yet. On the basis of ESN combining with standard particle swarm algorithm, the developed optimization algorithm in this research, AMPSO, employs the adaptive mutation strategy.3. Combining the features of effluent TP concentration with the soft sensing technology, this research constructed a soft-sensor for predicting the effluent TP concentration on the basis of the improved ESN. Effective prediction of MackeyGlass time series demonstrates high prediction accuracy for nonlinear system, providing theoretical foundation for its practical application. When this soft sensor was applied into predicting the effluent TP concentration, the result similarly shows its effectiveness.