Optimizing scheduling and resource allocation in oil and gas storage and transportation instruction-to-scheduling translation using deep reinforcement learning

View at publisher Summarise with VIRAT
Authors: Lei Lv, Ren Huang, J. M. Zhang
Cited by: 0
References: 34

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Translating instruction to schedule in the context of oil and gas storage and transport involves taking operational instruction to develop a schedule and allocate resources. The traditional approaches are inefficient in dealing with complex or constrained environments due to multiple variables and limits. The application of rule-based scheduling algorithms results in local optima being identified, while those based on predictive modeling are subject to disruption caused by environmental change, reducing the scheduling efficiency and increasing the time to respond. This paper proposes a methodology to solve these issues using improved deep reinforcement learning through deep deterministic policy gradients for optimality of scheduling. First, a complicated state space that considers significant components associated with the storage and transportation of oil and gas is created. This includes elements such as the capability of the pipeline, the volume capacity of the tanks, the time sequence of transport, and the load of the equipment. High-dimensional temporal feature extraction is achieved with a convolution neural network that enhances the amount of decision-making information available for the policy network. A prioritized experience replay technique is introduced to increase sample efficiency, while an adaptive soft update strategy is implemented to provide greater stability and quicker convergence time during the training of the algorithms. It is now possible to generate both scheduling and resource allocation decisions efficiently in real-time and evaluate these with the assistance of a value network. Through experimental results, we show that when used under normal conditions of operation, this method provides an overall scheduling efficiency of 95%, with a maximum instruction response delay of 1.3 s. In addition, this method has a high degree of robustness when faced with variable conditions and unexpected failures. As such, this method is highly adaptable to changes to these variables. The intelligent optimization technique proposed by this work enhances the scheduling efficiency and resource usage of oil and gas storage and transportation; it also represents an accessible technical approach to digital transformation of the oil and gas sector under uncertainty.

Fields of Study & Subjects

Scheduling and Optimization Algorithms Reinforcement Learning in Robotics Process Optimization and Integration

Publication details

Journal / source
Journal of Renewable and Sustainable Energy
Volume
18
Issue
2
Publication date
1 March 2026
Language
EN