The Application Of Fuzzy Neural Network On Water Quality Evaluation

Water quality evaluation is the key foundation of calculating water environmental capacity and implementing the planning of water pollution control. On the basis of a comprehensive exposition and analysis of the basic principles, the algorithm and the varied pattern features of Artificial neural network (ANN), especially Back-Propagation (BP) network, and on the basis of a exposition of the basic principles of Fuzzy Mathematics, according to the demand and trait of water quality evaluation, this thesis combines Fuzzy Mathematics and ANN and develops Back-Propagation network in series with degree of membership (BPDM) model and Fuzzy neural network (FNN) model. The models are tested after learning and case studies show that the evaluation results are objective and accuracy. Chapter 3 introduces the principles, the algorithm and the pattern features of BPDM model that is composed by ANN in series with degree of membership. BP network does not have to prescribe the weights of every evaluation parameters artificially. It only need to learn five kinds of water quality standards and is able to master the rational rules among the water quality parameters automatically. The evaluation results are objective. After learning of the network, the values of weight and threshold can be used to evaluate the other samples besides the training samples. It is simple and convenient to calculate and has better practicability. Different from a normal BP network, BPDM model regards the output of Bp network as the input of Fuzzy System and calculates the degrees of membership of which the testing samples belong to each water quality standards. The final output results of BPDM are accurate and specific water quality categories of testing samples. Chapter 4 emphasizes the principles, the algorithm and the pattern features of FNN model that is composed by ANN and Fuzzy System according to learning integrated. In FNN model, ANN is not a black box any more and its all nodes and parameters have physical meaning, and it overcomes the shortcoming which choosing ANN configuration is short of sufficient theoretical analysis. FNN model can not only direct expresses the logic meaning of people's customs and be fit for direct or advanced expression of knowledge but also have the merits of ANN self-adaptation learning and non-linear expression. Researches on FNN application in the water quality evaluation are preliminary exploration of author's. Case studies show that FNN is able to precisely evaluate othersamples besides the training samples after learning, thus has better objectiveness,reliability and expression.The having learning models of BPDM and FNN are tested with the data from Changjiang river and Jialingjiang river in Chongqing city. Case studies show that the evaluation results of BPDM model and FNN model have very high precision and accuracy, and in the latter's configuration, each node and all parameters have definite physical meaning and it is easy to understand. In this thesis, the evaluation results of different data transformation to the same samples are compared and, some effective strategies are brought forward to improve the algorithm of BP considering model operation. The effect is better. The applications of BPDM and FNN in water quality evaluation have a bright future and it is not only feasible in theory but also has huge significance in practice. Especially FNN, if more researches can be done in the ascertaining of its fuzzy partition and rules, the better effect of evaluation can be gained. It is well worth exploiting and researching for researchers.