Integrated Design and Operational Optimization of Gathering Pipeline Networks Using Neural Network Surrogate Models
Yongtu Liang,Shiyuan Pan,Junzhuo Chen,Guangtao Fu,Huanyu Zhao,Ning Xu
DOI: https://doi.org/10.1115/ipc2024-133778
2024-09-23
Abstract:The optimization of the surface gathering pipeline network (GPN) in oil and gas fields encompasses both design and operational improvements, involving economic and environmental benefits. Previous research often tackled these two optimization problems independently, which may lead to local optima. Furthermore, oilfield often involve multiphase flows, necessitating detailed hydraulic and thermodynamic calculations. Although the calculation time of current mechanistic models has reached the level of seconds, the total time required for the tens of thousands or even millions of optimizing calculations needed is still unacceptable. To address the issue above, an MILP method that integrated design and operational optimization is proposed. The proposed method employs the neural network surrogate model for hydraulic and thermodynamic calculations and utilizes the genetic algorithm to solve the optimization model. The obtained results include detailed topological structure, pipeline diameters, locations and types of devices, temperatures and pressures at each station, and the total construction and operational costs. The dataset is generated using PIPESIM software, and the predicted results are evaluated using RMSE, MAPE, and R2 metrics, demonstrating that the proposed neural network agent model meets the accuracy requirements for hydraulic and thermodynamic calculations and achieves a computational speed improvement of three orders of magnitude compared to mechanistic models. Compared with the two-stage algorithm (where design and operational optimizations are solved independently), the optimization method presented in this paper reduces construction costs by 2.7 million CNY and operational costs by 6.6 ten thousand CNY per year, representing relative reductions of 27.83% and 0.93%, respectively.
Engineering,Environmental Science