With A New Hybrid Algorithm Based On Neural Networks And Applications

The research of artificial neural network, simulated annealing algorithm and genetic algorithm are the focus of information technology. It has important value in theory and application for identification and prediction on nonlinear systems. In the article, it is studied neural network, simulated annealing algorithm and genetic algorithm, with which combining identification principle and optimization technology it is proposed three kinds of new hybrid algorithms. In order to solve the problem of the prediction of deep volcanic rocks reservoir, by using above three hybrid algorithms, they are proposed radial basis function neural network with new hybrid algorithms, neural network with new hybrid simulated annealing algorithm and neural network with new hybrid genetic algorithm. So, it is finished the modeling and oil gas recognition of complex nonlinear volcanic rocks reservoir. The main research has been finished as follows.It is considered a kind of second-order algorithm, which has been given by N B Karayiannis. Some predigestions are done when deducing algorithm, so some valuable information is lost, especially, in predigesting Hessian Matrix. Because the deduction of algorithm is not thorough, its calculation quantity is still big and its properties are badly affected. So, it is proposed a new second-order back-propagation learning algorithm to multilayer feedforward'neural network. It is proved that it is equivalent to Newton iterative algorithm and has second-order convergence rate. It is clearly stated that this new algorithm is superior to Karayiannis' second-order algorithm according to analysis of their properties and applications.In the article, it is systematically studied the principle of radial basis function genetic algorithm and it is proposed a kind of optimal selection cluster algorithm. Combining this algorithm with orthogonal least squares and gradient algorithm in optimization technology, it is proposed a kind of new hybrid algorithm, by which they are identified the constructions, parameters and weights of radial basis function neural network. It is systematically studied the principle and design of simulated annealing algorithm, by which it is proposed a kind of simulated annealing algorithm with adaptive cooling schedule. Combining this algorithm with Powell algorithm, it is proposed a kind of new hybrid algorithm, by which they are identified the weights of multilayer feedforward neural network. It is systematically studied the principle and design of genetic algorithm. It is considered the convergence rate of error back propagation algorithm is slow and it can ultimately get local minimum value by using error back propagation algorithm, and it can get the approximate value of global optimal value by using exclusively genetic algorithm, so it is proposed genetic BP algorithm combining genetic algorithm and error back propagation algorithm. By using genetic BP algorithm, they are identified the weights of multilayer feedforward neural network.It is applied neural network with new hybrid algorithms to the oil gas recognition of deep volcanic rocks reservoir, and it is proved the validity of the scheme.