Prediction Of Single Well Substrata Production On Hadoop Platform

Prediction of oil and natural gas production is an important task in oil and natural gas field development.In the process of oil and gas field development,one of the key issues is production.Only by grasping the change regulation of oil and gas field production,can oil and natural gas field development have reasonable planning and management.Traditional prediction often uses Arps diminishing method to predict oil and gas production.However,this traditional method only use historical production data,sample attribute is too simplex.It leads to the prediction accuracy become low.Besides,with the increase of using years of business system,historical data accumulates ceaselessly and substrata data is more complex than single well data,the size of substrata data is larger than single well data.Thus,this situation causes efficiency of database operation and traditional prediction method in business system become low.For these problems,this thesis uses Hadoop platform and combination of multiple machine learning algorithms to predict production of natural gas reservoir substrata.The prediction plan is divided into the following parts: firstly,creating data index by using MapReduce on Hadoop and store data index in MongoDB database.The substrata data is divided into two types: static and dynamic.Then,preprocessing data and analyzing correlation.At last,cluster analysis of substrata data and building prediction model which is based on KNN algorithm,multiple linear regression and particle swarm optimization algorithm,completing prediction of production.This thesis verifies the accuracy rate of production prediction model which is based on machine learning algorithm combination is higher than traditional method and proves prediction plan has better usability than traditional method.