The Application Of Neural Network In Logging Lithology Identification

Lithology identification is not only the key elements in reservoir evaluation and reservoir description, but also very important foundation for obtaining reservoir parameters. Accurate results of lithology identification can provide reliable basis for the exploration of oil and gas, for more it has played an enormous role in searching oil and gas resources and evaluating the oil. Because of the heterogeneity of actual reservoir, the traditional lithology identification methods are difficult to express the true characteristics of the reservoir.Neural network has the characteristics of distributed processing,self-study,self-organization,highly nonlinear and fault tolerance capabilities, so it is a new effective lithology identification method that taking advantage of neural network to process logging information.This paper includes three parts, using Principal Component Analysis to process logging data; setting up BP and PCA-BP network lithology identification models and using them to identify lithologies from logging data; setting up SOM network lithology identification model for logging data clustering.Firstly, on one hand, in order to identify lithology effectively, we should be taking advantage of logging parameters as more as possible, but the lithology information, which logging parameters carries, have a certain degree of overlap; on the other hand, for complex issues and high-dimensional input variables, neural network's direct prediction will bring dramatic increase in network size, increase in computing time, decrease in the convergence of network and the ability of generalization. In order to solve the above problems, using PCA to process logging data and the PCA scores could instead the mostly lithology information what the original logging data carries. This not only reduces the dimension of original data and simplifies calculation, but also eliminates the correlation of original logging data.Secondly, for the supervision learning samples, setting up PCA-BP lithology identification model which is a method based on PCA and BP network. First using PCA to pretreat logging data, then putting the PCA results to BP network for training, last using PCA-BP model to identify the lithology of testing samples. Compared with common BP neural network model in lithology identification, the results indicate that the PCA-BP network model could not only predigest the network structure (the number of input neurons reduces from five to three) and accelerate the network convergence speed by 21 percent, but also increase the precision of recognition by 25 percent.Thirdly, for unsupervised learning samples, designing SOM network lithology identification model.SOM network is a kind of competitive learning networks and has strong clustering and fault tolerance capacity. An example is used to show SOM model's application to logging lithology identification. The results indicate that SOM network can be able to automatically cluster learning samples and has higher recognition accuracy.To sum up, it is a new effective lithology identification method that using neural networks to process logging data and identify lithology.This method has certain practical significance and good prospects in exploring and identifying the accuracy of oil and gas layers and the field of oil and gas resource development.