Fluid Loss Risk Prediction And Optimization Design Based On Machine Learning

Oil and natural gas resources are the basis for human society development,and drilling engineering is the necessary means of exploration and development of oil and gas resources.However,frequent well leakage accidents in the process of drilling construction lead to a lot of time and capital losses,which seriously restricts the efficient development of oil and gas resources.The traditional well loss prevention method starts from the theoretical model and has a solid physical foundation,but the model is difficult to comprehensively consider the coupling between formation factors and construction parameters.In addition,the uncertainty of formation information also makes it difficult for traditional well loss prediction methods to get good prediction performance.Therefore,it is of great significance to build an efficient well loss prediction method.In view of the above problems,this paper uses the method of machine learning to build a set of data-driven well loss prediction model,and designs the methods to assess the fluid loss risk and optimize the operation parameter.The main work is as follows:Based on the data of 13 wells in halfaya oil field,Iraq,16 data features are determined through data cleaning,normalization and importance evaluation of data features,which are used to build a data-driven well loss risk prediction model.Firstly,the traditional deterministic neural network model is established,and the inherent defects and potential risks of the "point estimation" model in machine learning are described by using the prediction results of the model,which embodies the importance of adopting the Bayesian neural network model.Then the theoretical basis of three popular Bayesian neural networks is analyzed.Through the construction and optimization of the models,the prediction performance of the three models in the data set of this paper is evaluated.Finally,this paper chooses the mixed density network model to design the method of well loss risk prediction.First,the source and essence of the cognitive uncertainty information in the mixed density network model are explained,and the evaluation method of well loss risk is established by using the uncertainty information predicted by the model.Based on the uncertainty information,the risk index of lost circulation is constructed,and the real-time fluid loss risk prediction system is designed.Taking the constructed risk index as the optimization objective,the Bayesian optimization method is used to optimize the design of construction parameters,so as to avoid well loss accidents caused by human factors in the drilling process as much as possible.Through the simulation experiment of this paper,it shows that the construction parameter design based on Bayesian optimization can meet the requirements of real-time optimization.In this paper,the application of Bayesian neural network in petroleum engineering is described in detail.It is a typical application of Bayesian neural network to deal with the problem of fluid loss.In addition,the model introduced in this paper has a very practical application for many typical problems in petroleum engineering,such as enhancing rate of penetration,blowout prevention,drilling tool stuck prevention and so on.