基于LSTM神经网络的致密气井产水量预测方法研究
Research on the Prediction Method of Water Production of Tight Gas Wells Based on LSTM Neural Network

近年来，机器学习在石油领域中得到了广泛应用，现采用LSTM神经网络模型对致密气井产水量进行预测研究。以某气藏P101-2H井组为例，选取了油管压力、套管压力、套管温度和日产气量4种影响因素作为特征输入变量，以日产水量作为目标输出变量，对数据集进行预处理和特征选择。基于数据预处理的结果将数据集划分为训练集和验证集，结合超参数和训练集数据，进行LSTM模型的训练，模型在验证集上的准确性和可靠性表现较佳。该方法具有一定的实用价值，能够降低致密气井产水量预测的工作难度，从而对产水气井进行优化配产以控制或减缓气藏水侵。
In recent years, machine learning has been widely used in the petroleum field, and now the LSTM neural network model is used to predict the water production of tight gas wells. Taking the P101-2H well group in a gas reservoir as an example, four influencing factors, namely, tubing pressure, casing pressure, casing temperature and daily gas production, were selected as feature input variables, and daily water production was taken as the target output variable, so that the data set was preprocessed and feature selection was carried out. Based on the results of data preprocessing, the dataset is divided into training set and validation set, combining the hyperparameters and training set data, the LSTM model is trained, and the model performs better in terms of accuracy and reliability on the validation set. The method has some practical value and can reduce the work difficulty of water production prediction in tight gas wells, so as to optimize the production allocation of water-producing gas wells to control or slow down the reservoir water intrusion.