Research on Intelligent Prediction of Formation Pressure and Log Curve Completion Based on Machine Learning Algorithms

Aiming at the problem of low accuracy in formation pressure prediction caused by missing logging data and severe noise interference in oil and gas fields, this paper integrates deep learning with geological prior knowledge and proposes an integrated method for log curve completion and intelligent formation pressure prediction. The method implements multi-dimensional data preprocessing through wavelet threshold denoising, adaptive hierarchical normalization and SMOTE oversampling, constructs a Cascaded LSTM model to accurately complete missing segments of logging curves, designs a PC-LSTM model for intelligent formation pressure prediction, and introduces an EnLSTM model to solve the problems of small-sample training and prediction uncertainty quantification. Verified by field data from the Daqing Oilfield, Williston Basin (USA) and Eagle Ford Shale Reservoir: the signal-to-noise ratio of data reaches 21.36 dB after wavelet denoising; the coefficient of determination (R²) of Cascaded LSTM completion is 0.931; the mean absolute error of PC-LSTM pressure prediction is only 0.82 MPa; EnLSTM achieves an accuracy of 88.5% and a 95% confidence interval coverage rate of 92.3% with only 10% training samples. This method realizes the collaborative optimization of logging data restoration and formation pressure prediction, and can provide technical support for the safe and efficient exploration and development of complex oil and gas reservoirs.