Physics-driven And Data-driven Pore Pressure Prediction

With the development of oil and gas exploration,more and more abnormal pore pressure areas have been found.Pore pressure is a key parameter for drilling,wellbore stability analysis and mud scheme design.Accurate prediction of pore pressure,especially abnormal high pressure,is of great significance in the field of oil and gas exploration and development.The traditional pore pressure prediction method is mainly based on physical models or empirical models.However,due to the influence of many factors such as formation properties,structural history and sedimentary characteristics in complex areas,it is difficult to fully describe the complex changes of pore pressure only by traditional physical models.With the development of artificial intelligence and big data technology,scholars gradually begin to apply machine learning algorithm to solve geophysical problems.Therefore,the research about physics-driven and data-driven pore pressure prediction is conducted based on well logs reconstruction experiment and machine learning algorithms.The main research results include:(1)Well logs reconstruction experiment including well logs imputation and generation based on machine learning algorithms is completed.The results are analyzed combined with geological background and the XGBoost algorithm shows strong generalization ability.The experiment can provide higher quality well logs data for subsequent pore pressure prediction(2)A new method to distinguish abnormal pore pressure section and establish pore pressure discrimination plate based on machine learning algorithms is proposed.The interpretability of machine learning model is studied,and the importance of factors affecting the distribution of abnormal pressure section is analyzed.(3)Based on the traditional and optimized pore pressure forming mechanism,the causes of abnormal high pressure are analyzed,and the pore pressure is predicted based on various physical models.The results show that the pore pressure calculated by Eberhart Phillips method achieves the highest stability and accuracy.(4)A new data-driven multivariable pore pressure prediction model based on machine learning algorithm is proposed.The single well pore pressure prediction results show that the proposed method captures the variation trend of pore pressure more accurately,and also achieves higher accuracy in the abnormal pressure section.(5)Based on gated recurrent neural network,a new physics-driven neural network(PD-GRU)is proposed.The proposed model changes the GRU network structure to simulate the physical mechanism,and increases the dimension of data through gradient feature enhancement,which makes the model more suitable for the actual situation of the formation and greatly improves the accuracy and stability of multi wells pore pressure prediction.(6)A web application for pore pressure prediction is developed based on the above research,which provides great convenience for the follow-up work.Based on the above research results,this thesis establishes a set of standard digital process for predicting pore pressure by using well logs,which promotes the digital and intelligent development in related fields of pore pressure prediction and has high industrial application value.