Computational Intelligence And Its Application In Water Conservancy And Hydropower Engineering

COMPUTATIONAL INTELLIGENCE AND ITS APPLICATION IN WATER CONSERVANCY AND HYDROPOWERENGINEERING(Ph.D. Dissertation)Doctoral Candidate: Xu Shigang Supervisor: Suo LishengChen Shoulun (Hohai University, Nanjing, 210098)ABSTRACTComputational Intelligence (CI) which includes Evolutionary Computation (EC), Neural Networks (NN) and Fuzzy system is a new subject developing rapidly. In this paper, Genetic algorithm (GA), Feedforward Neural Network (FNN), Self-Organizing Feature Map (SOFM) neural networks and hybrid system are studied. The result is applied in water conservancy and hydropower engineering. The integrated system of theory, method and application is established. The content of the paper is as follows:1. The advance of GA and NN is discussed. The headway of CI being used in water conservancy and hydropower engineering is overviewed in the meantime.2. Based on study to essential theory of GA, the weakness of Simple GA (SGA) in optimization is indicated, and the mend to SGA is analyzed. A new Genetic Algorithm based on Diffluent Mechanism (DMGA) is put forward. Its strategy is that excellent seeds are set limit to reproduce, certified seeds are crossed and bad seeds are mutated. The crossover probability and mutation probability is adjusted by the evolutionary equality. DMGA changed the classical structure of SGA. The global convergence and efficiency are improved. Test demonstrates its fine performance. The implementation process of DMGA in hydraulic optimization is studied in detail. An engineering example indicates that it doesn't demand initial solution, its quality and efficiency are batter than Complex Algorithm.3. After thinking over the essence of the hidden layer of FNN, fuzzy cluster method is put forward for extracting the feature of sample data, moreover, the input and outputinformation are comprehensively considered. So the number of hidden nodes is determined. The example proved its feasibility. The optimal networks is triumphantly applied in runoff forecast, it is more efficient and precise.4. A new method to determine the structure of SOFM neural networks is put forward. Singular Values Decomposition (SVD) is performed on competitive layer's output. Based on the distribution of the singular values, the number of neural nodes of competitive layer is chose.5. ES-SOFM hybrid model is set up. SOFM neural networks is embedded into Evolutionary Strategy (ES). Fitness function is constructed based on the state of SOFM neural networks. The sensitivity of SOFM neural networks to initial weight matrix and sequence of input exemplars is overcome by the strong global optimum of ES. The result of its application in water quality assessment shows that the method is excellent.6. In the end, conclusion of the paper and prospect are provided.