基于粒子群优化和长短期记忆神经网络的气井生产动态预测

Production dynamic prediction of gas well based on particle swarm optimization and long short-term memory

薛亮 顾少华 王嘉宝 刘月田 涂彬

XUE Liang GU Shaohua WANG Jiabao LIU Yuetian TU Bin

中国石油大学 ( 北京 ) 石油工程学院 中国石油化工股份有限公司石油勘探开发研究院

College of Petroleum Engineering, China University of Petroleum (Beijing), Beijing 102249, China SINOPEC Exploration & Production Research Institute, Beijing 100083, China

气井生产动态预测是气藏产量规划、开发方案编制及生产制度动态调整的重要依据，对天然气藏开发有着极其重要的指导意义。建立了基于长短期记忆深度神经网络的生产动态预测模型，并采用了粒子群优化算法对神经网络模型超参数进行优化，提高长短期记忆深度神经网络的预测效果。研究结果表明，基于粒子群优化和长短期记忆神经网络的气井生产动态预测模型能够实现对气井生产动态的准确预测和神经网络超参数的自动优化，使预测结果的平均绝对误差均小于10%，大幅度简化了神经网络模型的优化过程。

Production performance prediction of gas well is an important basis for gas reservoir production planning, development scheme preparation and dynamic production system adjustment and is of extremely important guiding significance to the development of gas reservoirs. In this paper, the production performance prediction model of long short-term memory (LSTM) deep neural network was established based on the machine learning method. In addition, the super parameters of neural network model were optimized by means of the particle swarm optimization algorithm, so as to improve the prediction effect of LSTM deep neural network. It is indicated that the production performance prediction model of gas well based on particle swarm optimization and LSTM can accurately predict the production performance of gas well and automatically optimize the super parameters of neural network and the average absolute error of its prediction result is less than 10%. What’s more, it greatly simplifies the optimization process of neural network model.

10.13639/j.odpt.2021.04.017