A Study On Application Of Neural Network Based On Hybrid Intelligence Learning Algorithms In Reservoir And Hydrocarbon Prediction

The artificial neural network is a newly risen nonlinear science. In recent years, it has been applied to many fields according as the development of theory of artificial neural network, besides the technical requisition of artificial neural network has been becoming more and more rigorous in these fields. Seeing that the method of reservoir and hydrocarbon prediction adopting artificial neural network has stronger ability of resisting interferences and enduring wrongs, not only sets less strict demands on the independence of parameters, but also may simultaneously study a large quantity seismic characteristic parameters, it raises the level of reservoir and hydrocarbon horizontal prediction. Since the 1990's, neural network has widely been applied in the field of reservoir description and hydrocarbon prediction. In this paper, I study the present state of application of reservoir and hydrocarbon prediction adopting artificial neural network, lead into the multiscale chaos optimization algorithm and make the multiscale chaos optimization algorithm and the genetic algorithms combined to bring about the multiscale chaos-genetic algorithms against the shortcoming of the BP network used in reservoir and hydrocarbon prediction and the limit of genetic algorithm used in optimizing the connection weight of the feedforward neural networks. Chaotic searching is introduced to improve the initial community and it is reintroduced to optimized genetic process of genetic algorithm which is taken as study algorithm of feedforward neural network. Improved genetic study algorithm is applied to predict reservoir and hydrocarbon. Better result is obtained based on improved feedforward neural network.Moreover, in this paper I study the methods of seismic attribute optimization. It is not sure good way to use the much more attribute parameters in reservoir prediction. The optimal dimension of seismic attribute depends upon the effect of reservoir prediction. And only when the samples selected widely can cover the whole dimension of seismic attribute in the process of training, the neural network which has been trained can give fairly good results. On this condition that the specimen has few samples in this paper, I make use of clustering method to choose seismic characteristic parameters. The method is getting together the characters which Euclidean distance is further or which similarity coefficient is more close to zero. Using these seismic parameters selected by clustering method to predict reservoir and hydrocarbon, good effect is obtained in the practice.