Intelligent Prediction Of Formation Pressure And Parameter Optimization Based On Machine Learning Algorithm

Formation pressure is a critical parameter in the exploration and development of oil and gas.Accurate prediction of formation pressure can not only effectively determine the density of drilling fluid,but also reduce the probability of drilling accidents such as well blowouts and well leaks or even prevent these accidents,as well as guarantee the safety of drilling operations during construction,and improve efficiency and reduce costs,ultimately achieving the goal of protecting oil and gas formations.In Shunbei oilfield,due to the development of faults,complex geological structure,large reservoir depth(up to 8000m),high temperature and high pressure,leakage and collapse are easy to occur in the drilling process,so it is difficult to accurately predict the formation pore pressure.Traditional formation pressure prediction methods often require specific formation parameters(such as density,acoustic time,seismic layer velocity,etc.),and some well sections in the Shunbei area lack the parameters required by traditional prediction methods,and cannot solve the problem of formation pressure prediction.Combined with the advantages of machine learning algorithm,a new hybrid decomposition integrated prediction algorithm(HP-PSO-RF-GB)is proposed,and particle swarm optimization algorithm(PSO)is used to optimize the parameters of random forest(RF)and gradient lifting algorithm(GB)simultaneously,so as to obtain the optimal model.Finally,the formation pressure prediction model which is more reliable and can meet the engineering requirements is obtained,which solves the problem that the traditional method cannot be used in the absence of a specific curve,and achieves the purpose of using other parameters to predict the formation pressure.This thesis takes Shunbei oilfield as the study object,selects 11 types of drilling logging data with well depth,spontaneous potential,natural gamma,acoustic time et al.of three wells in Shunbei oilfield as research data.And uses the hybrid decomposition integrated prediction algorithm(HP-PSO-RF-GB)proposed in this paper to train the intelligent prediction model of formation pore pressure.The formation pressure label is obtained based on the traditional calculation model and local modification by experts after drilling analysis.Seven machine learning algorithms including decision tree,support vector machine,random forest,linear regression,elastic net,gradient boosting and naive bayesian are used as the comparison modle.The prediction accuracy of the proposed model is verified by comparing the performance indexes such as R~2,explained variance,mean absolute error,mean square error,median absolute error and mean relative error.The results show that the decomposition and integration prediction algorithm proposed in this thesis has the highest prediction accuracy and stronger generalization ability.In addition,the impact of the amount of training sets on prediction results is also investigated,and prediction accuracy is found to be positively connected with the proportion of training data sets.In the prediction of adjacent wells,the proposed algorithm also achieves the best prediction effect,indicating that the method is suitable for the prediction of pore pressure in the Shunbei area.The optimal combination of logging parameters with different number of logging parameters is obtained by changing the input logging parameters,and it is proved that the well depth is the main parameter affecting the prediction accuracy.The results show that the algorithm proposed in this thesis has the best prediction effect,which can effectively predict formation pore pressure with high efficiency,and can achieve the purpose of formation pressure prediction without specific parameters,and meet the engineering requirements.