Research On Reservoir Prediction Automatic Optimization Model Based On Seismic Data

The seismic data contains abundant information of stratigraphic lithology,fluid physical properties and geological structure.The information about reservoir property and lithology which is effectively extracted from seismic data has very strong practicability.However,with the development of oil-gas exploration development in our country and the constant extension of exploration fields.The number and kinds of seismic attributes increase rapidly.It makes reservoir prediction much more difficult.Data mining is an effective method for reservoir prediction.In this dissertation,the reservoir prediction model is constructed based on feature dimension reduction and clustering analysis.The reservoir prediction model based on data mining often needs some parameters to be set manually.If a user does not understand data mining technology,parameters setting is very difficult for him.Reservoir prediction model parameters Selection has great impact on its predicted results.Reservoir prediction model.Some of them vary over a wide range.There are many combination of model parameters values.It takes time to find the optimal parameters of reservoir prediction manually.This dissertation presents a method of automatically optimizing reservoir prediction model parameters.The model parameters input is not required.The method reduces the burden of users.It can converge to global optimal solution rapidly.Firstly,aiming at the problem of clustering algorithm parameters optimization in reservoir prediction model,in this dissertation,a hybrid particle swarm algorithm is presented to be used to optimize the clustering algorithm parameters.According to clustering objective-the maximum inner-class similarity and the minimum between-class similarity,the fitness function is designed to improve the quality of clustering results.Secondly,the parameters optimization of the reservoir prediction model is studied.The reservoir prediction model includes two parts:feature reduction and cluster analysis.These two parts are usually optimized separately.But the obtained results are local optimal,and not overall optimal.This dissertation applies hybrid particle swarm algorithm to optimize all parameters of the model simultaneously,which jointly optimizes feature dimension reduction and clustering analysis.Finally,the vast majority of seismic data is unlabeled data,the number of seismic labels is small.Therefore,this dissertation proposes a method of label data selection based on clustering results to obtain more seismic labels.It makes supervised classification algorithms which depends on a certain number of training data can generate more accurate classifiers.This dissertation uses the hybrid particle swarm algorithm to optimize the parameters of reservoir prediction model based classification algorithms,which can improve classification accuracy.Experiments indicate that Hybrid particle swarm optimization method can automatically optimize the parameters of reservoir prediction model.The accuracy of model prediction has been improved.Optimizing all the parameters of Reservoir prediction model at the same time is better than optimizing them separately.Classification results based on automatic labels is not worse than clustering results.This shows that the automatic labels selection method based on clustering results is effective.