Research And Application Of Oil Production Prediction Based On Data Mining In Spark

In recent years,with the rapid development of oilfield information,the traditional reservoir data analysis methods and tools are no longer in keeping with the requirements of current data processing.The analysis of reservoir characteristic properties and oilfield recoverable reserves and production trend forecast are the important prerequisite work for oilfield exploration and development.At present,the domestic and foreign production forecasting model mainly focuses on the prediction of output with time or the use of empirical model to evaluate reservoir reserves.This paper mainly research the effective data processing methods and the establishment of accurate and effective yield prediction model.The research shows that the classification of reservoir stratigraphy has a guiding role in the study of reservoir data characteristics and oilfield performance analysis.In this paper,it put forward a reservoir stratigraphic classification model based on local anomaly data detection of the information entropy and the KK-SMOTE(k-means and k-neighbors on SMOTE)method to oversampling the unbalanced data to pretreated reservoirs data,which provides the prerequisite for data mining accurately of the reservoir characteristic attribute,and establishes the reservoir stratigraphic classification model based on the decision tree,through the different geological parameters to analyzes and predict of whether the formation is an oil layer reservoir and prepare for yield prediction analysis.BP neural network(Backpropagation Neural Network)is an efficient and simple predictive model,which can analyze multidimensional nonlinear problems.Based on the characteristics of reservoir data,this paper establishes the method of determining the number of neurons in network input layer based on PCA(Principal Component Analysis).And the new oil well production index is used to predict the overall production performance of the oil well.According to the characteristics of the particle swarm algorithm,an adaptive inertia factor adjustment method based on the adaptability value is established,and the BP neural network yield prediction model based on the particle swarm optimization,which solves the problem of predictive instability caused by the stochastic initialization of weights and thresholds in the BP standard neural networks.Finally,the method is established for reservoir data processing and analysis based on Spark,and the method based on Spark's particle swarm optimization and BP neural network parallelization is proposed.The experimental results show that the improved particle swarm optimization algorithm can improve the training speed and prediction accuracy of the BP neural network model effectively,and the stability and efficiency of the model are have a good performance.The Spark-based data processing method proposed in this paper can be applied to the reservoir data processing and analysis effectively,which is easy and simple to use.In addition,reservoir formation classification and BP production prediction model based on particle swarm optimization can do well in analyzing the production performance and forecast the production of oil wells,which provide decision-making basis for oil exploration and mining.