Study On Lost Loss Prediction Of Longmaxi Formation In Chuannan Based On BP Neural Network And Support Vector Machine

At present,our country is the world’s largest oil importer,and energy security is facing huge challenges.In recent years,a large number of shale gas reservoirs have been discovered in my country’s Sichuan,Qaidam and other basins,among which historical breakthroughs have been made in the exploration and development of shale gas in the Sichuan Basin.Shale gas has become an important basic energy source for my country’s economic development.Because the shale formation has the characteristics of loose bedding and low pressure-bearing capacity,lost circulation is very easy to occur during the drilling process,which prolongs the non-productive time of drilling operations and causes serious economic losses.With the de-velopment of intelligent oilfield data,lost circulation prediction technology has become the direction of research in drilling and development.The effective prevention of lost circulation accidents is achieved by establishing a lost circulation prediction model.157 wells in the southern Sichuan block have lost 607 times,with high leakage frequency,and the maximum number of single wells reached 41;the loss is large,and the maximum loss of a single well is 32699.7 m~3;the loss time is long,and the longest loss time of a single well is 1445.57 h.The loss of the Longmaxi formation accounted for 56.8%.Among the 40 miss-ing wells,16 of the wells were lost in the Longmaxi formation.The Longmaxi formation is considered to be the main missing formation in the southern Sichuan block.The reasons for the leakage of the Longmaxi formation in southern Sichuan are low pressure bearing capacity of the formation,cracks encountered in the formation,and strong formation permeability.Mi-cro leakage and small leakage are the main causes.Fracture leakage and permeability leakage account for 73.6%.Combining field data to analyze geological,engineering and drilling pa-rameters,using the analytic hierarchy process to determine the main influencing factors of lost circulation are drilling fluid displacement,well depth,drilling fluid density,ROP and outlet temperature,and the proportions that affect lost circulation weight are respectively0.149,0.128,0.119,0.110,0.109.Using the principal component method to weight and re-ducing the dimensionality of the lost circulation data to obtain the lost circulation sample data.Using the layered sampling method to divide the lost circulation sample data into the training set and the test set,and the lost wells are numbered according to the block.Dividing into four blocks to study the lost circulation prediction model.The method of stochastic gradient descent and additional momentum term is used to op-timize the BP network algorithm.The method of optimizing the error value solution is used to optimize the SVM algorithm.By analyzing the influence of the parameters on the lost circula-tion prediction model,the BP neural network and SVM lost circulation are constructed re-spectively forecast model.Using genetic and particle swarm algorithm to optimize the lost circulation prediction model,add fitness function and range search method to find the optimal solution,study the influence of the parameters in genetic and particle swarm algorithm on the model performance,and select the best solution for each block good leakage prediction model.The research results show that the GA-BP neural network leakage prediction model has the best prediction performance.Among them,the training accuracy of block 1,block 2,and block 4 are 98.46%,98.25%,98.31%,and the prediction accuracy is 97.24%,96.28%and96.84%,after optimization,the training accuracy and prediction accuracy are increased by about 3%and 4%respectively.The lost circulation prediction model was verified and com-pared with field data,and the prediction result was 96.10%,which achieved the expected goal.It proved the feasibility and effectiveness of the lost circulation prediction model,which can bring guarantee for the realization of safe and efficient drilling.A lost circulation prediction software for the Longmaxi formation in southern Sichuan was designed with matlab,which can predict lost circulation based on wellhead data in different regions to achieve the purpose of predicting lost circulation horizons.