Application And Improvement Of Machine Learning Algorithm In Well Logging Reservoir Parameter Prediction

Geophysical logging is a method to measure geophysical parameters by using the geophysical characteristics of rock strata.Reservoir refers to the rock strata with connected pores in which oil and gas can be stored and percolated.Permeability and porosity of reservoir parameters are important links to reflect reservoir characteristics,and porosity can determine whether a reservoir can store fluid.Permeability is a comprehensive reflection of various geological characteristics and an important basis for determining the existence of fluids.These two parameters are usually derived from the simulation of logging curves.Most of the methods are based on linear equations,including regression analysis,empirical formula and Archie’s formula.An emerging approach in recent years is to introduce machine learning algorithms to predict reservoir parameters.In actual logging,due to complex reservoir conditions,anisotropy characteristics and heterogeneity,both traditional methods and general machine learning algorithms have limited prediction accuracy and insufficient prediction range.In order to solve the above problems,this paper analyzes the principles of conventional logging methods and general machine learning methods,and synthesizes the prediction results of porosity parameters and permeability parameters predicted by five common machine learning algorithms.By combining the five machine learning algorithms,a committee machine model based on genetic algorithm optimization is established.The main work of this paper is as follows:Firstly,the evaluation criteria of model regression performance were determined according to logging data,and the porosity and permeability parameters were predicted by using support vector machine,random forest,XGBoost,CNN and LSTM methods.The emphases of different methods had compared and analyzed,and it has found that the parameter values of the whole stage could not be accurately predicted.Therefore,this paper has forward the genetic algorithm,constructs and searches the optimization method in the artificial form by simulating the Mendelian laws of heredity in nature and Darwinian biological evolution,uses the committee model to predict the porosity and permeability regression,and combines the output results of these five machine learning models according to the weight coefficient as the input of the committee machine.The prediction results of each model are processed and outputted by genetic algorithm.By comprehensive comparison,it is found that the committee machine model optimized based on genetic algorithm has achieved good results in predicting porosity and permeability.Compared with a single machine learning model,the prediction performance of GA-CM model in both porosity and permeability has been improved to some extent.Finally,in order to further verify the practicability of the model,the model is applied to single well treatment and interpretation,and it is found that the prediction effect of GACM model is the best,realizing the advantages of integrating different machine learning models and achieving the purpose of relatively accurate prediction of logging reservoir parameters(porosity and permeability),which is a relatively reliable method in a certain range and conditions.It can further provide help for oilfield logging interpretation and processing.