Application Research On Neural Network Model For Water Quality Assessment

Water quality evaluation is an important issue on protection of the water environment. In recent years, more and more researchers devote to this research field.Scientists have put forward a variety of methods and models for water quality evaluation.We illustrate some of the traditional water quality evaluation methods and analyzes several limitations of these methods. On the basis of previous researches, we also put forward several kinds of evaluation models which based on Neural Network and used for the evaluation of surface water. We studies three kinds of models which are established by BP neural network, probabilistic neural network and neural network based on genetic algorithm, and work for classification and evaluation of the same sample data about water quality.Main contents of this research: The research subject is the quality of surface water in Xindian section, Dunhua city, Jilin province. Research sample consisting of four indexes of COD, CODMn, BOD5, petroleum. Random interpolation method is used to extend the sample data in order to have a better training effect, which according to the standard of those indexes and standard of water quality. We first describes the basic concepts and principles of BP neural network, probabilistic neural network and genetic algorithm, and analyzes the characteristics of each neural network as well as the limitations. The feasibility of neural network for water quality evaluation is indicated by the water sample simulation results through the BP model and the PNN model. As there are certain limitations exist on the BP neural network when dealing with some problems,combined with the characteristics of BP network and genetic algorithm, a stable, fast convergence and high robustness water quality assessment model is constructed by combining global search capability with BP. Combining with the advantage of genetic algorithm will make adjustments on BP's weights and threshold. From each of the simulation results, we can see that improved BP network can not only reduce the number of iterations, but also improve the convergence speed and the accuracy of the network.This study shows that the evaluation of neural network on water quality classification is feasible on theoretic research. And further research needs to be done in the practical application. This thesis has made some contribution to the development of neural network.