基于粗糙集与神经网络的水质评价模型分析-analysis of water quality evaluation model based on rough set and neural network.docx

ABSTRACTThe water resources pollution problem in the Three Gorges area of the Yangtze River has drawn great attention from the outside and inside. The water environment is complicated, it also contains many factors and redundancy data in its evaluation. Both set up correct evaluation model and decrease complication of the model are what we supposed.The work of this dissertation is derived from a research project named Three Gorges Dam Area Water Pollution and Counter. Through the analyses of the dam area data, we set up the water evaluation model in a combined method which contains rough set theory and artificial neural networks theory.Rough sets is a kind of mathematical tool that is based upon math’s conception methods, which can be used to select the right evaluation factors set without any preliminary expert knowledge. Artificial neural networks have been applied in water evaluation area successfully because of its abilities of self-learning, self-organization, fault-tolerant and nonlinear- approximation. There are several combination strategies based on rough sets and artificial neural networks, two of them are discussed in this dissertation, one is using rough set method to preprocess the data and the other is a rough neural networks which contains rough neuron. At last, we also discussed the evaluation model combined of above two combination strategies, rough set method is used to preprocess the data and rough neural networks is used to evaluate the water.The main purpose of the data preprocessing by using rough sets is clean the noisy data and reduce the evaluation factors. After preprocessing, the reduced data will be as the input of artificial neural networks, and then the evaluation model will be set up by using BP neural networks. The advantages of this combination strategy are not only clean the noisy data and decrease the probability of over-fitting in trained neural networks, but also reduce the training data which save the training time and improve t