A Study On Theory And Application Of Nonlinear Inverse Based On BP-GA-ANFIS

In recent years, Nonlinear seismic inverse has made an encouraging improvement in both the theory and the application. Neural network nonlinear inverse is a important part of seismic nonlinear inverse. In general, we can select the three-layer forward neural network to establish inverse model, using seismic data and logging data to make up the sampler. We usually use BP or GA to train the network. In addition, we can alse select ANFIS neural network using minimum sqare method and speedest descent method. But the efficiency of these method is usually low. It is difficult to converge, or even can not converge. In the theses,we propose some combined algorithms to promote the computing speed of the networks.Firstly, we analyze the development and the current situation of BP, the fuzzy neural networks and Genetic Algorithms, the related theory of genetic algorithms, such as the basic concept, components, learning rule and simple genetic algorithms, and apply genetic algorithms to problems optimizing the connection weight of the fuzzy neural networks. Secondly, we have researched the related theory of nerve cell, character and learning rule of artificial neural networks, the related theory and performance of the algorithm. Thirdly, considering the character of the feature of GA neural networks which is good at global optimization, we propose a new algorithm which combines GA with TS method. Finally, we have designed the program of Fuzzy neural networks which based on the new algorithm that combines GA with TS method in VC++, and applied those algorithms to the some other problems such as curve fitting, function approximating and explaining high resolence seismic data. The experiment results proved that the learning algorithms for training neural network was better, faster and more accurate than BP algorithms and genetic algorithms.