Carbon-Constrained Optimization of Oilfield Production and Heavy-Duty Transport Energy Management with Surrogate Models and Reinforcement Learning

Abstract

The transition toward sustainable industrial practices requires the implementation of stringent carbon emission constraints across highly energy-intensive sectors. Two of the most critical sectors in this context are oilfield production and heavy-duty logistics, which traditionally operate as isolated systems despite their inherent interdependencies. This paper presents a comprehensive framework for the carbon-constrained optimization of coupled oilfield production and heavy-duty transport energy management. Because high-fidelity physics-based simulators for these systems are computationally prohibitive for real-time decision-making, we develop data-driven surrogate models to capture the complex, non-linear dynamics of reservoir multiphase flow and transportation fleet fuel consumption. These surrogate models serve as highly efficient simulation environments for a deep reinforcement learning agent. The reinforcement learning framework is designed to determine optimal operational policies, including well injection rates, choke valve settings, and fleet dispatch schedules, while strictly adhering to dynamic carbon emission caps. By defining a comprehensive reward function that penalizes excessive carbon output and rewards energy efficiency and production yield, the agent learns to balance economic objectives with environmental constraints. Extensive simulation experiments demonstrate that the proposed integrated optimization approach significantly reduces overall carbon emissions and operational energy consumption compared to traditional decoupled and rule-based management strategies, while maintaining targeted production and delivery thresholds.