Optimization Models and Algorithms for the Petroleum Industry Supply Chain under the Dual‑Carbon Goal

Introduction

Dong Fenglian, senior technical expert at China Petroleum Planning Institute, shares insights from more than a decade of experience in optimizing production and operation of the petroleum industry chain, focusing on the challenges and optimization needs under the "dual‑carbon" (carbon‑peak and carbon‑neutral) background.

Challenges and Optimization Targets

Continuously reduce raw‑material procurement cost.

Increase the proportion of high‑value‑added products.

Optimize carbon‑emission reduction and carbon‑footprint tracking.

Key challenges include volatile crude‑oil prices, market fluctuations, stricter product‑quality requirements, excess refining capacity, rising labor and transportation costs, and increasing regulatory pressure on emissions.

Optimization Models

The institute has developed two major models:

Production‑planning optimization model for individual refineries, maximizing economic benefit while considering raw‑material purchase, processing routes, product blending, and sales.

Integrated upstream‑downstream oil‑chain model covering exploration, import, transportation, refining, chemical production, and distribution, aiming at overall group‑level profit maximization.

Both models are highly detailed, incorporating thousands of variables, constraints, and non‑linear relationships such as material flow, physical properties, and blending equations.

Large‑Scale Non‑Convex Non‑Linear Problem Solving

The models involve over 400,000 variables and more than 100,000 equations, with thousands of non‑linear constraints, especially in the refining section (mixing, property‑dependent yields). To solve these, the institute has researched and implemented:

Improved variable‑step distribution‑recursive algorithm with adaptive property updating to enhance convergence.

Trust‑region based Sequential Linear Programming (SLP) algorithm, combined with the recursive method for fast initialization, refined step‑acceptance criteria, and business‑rule‑driven trust‑region settings.

These algorithms can obtain high‑quality solutions for a refinery model (≈8,000 variables, 1,400 non‑linear constraints) within 30 seconds and solve the group‑level model (≈10 million equations) in a few minutes.

Application Scenarios

Practical uses include:

Annual/seasonal refinery production‑plan optimization.

Crude‑oil purchase optimization (selection, blending, pricing).

Ethylene raw‑material optimization.

Unit‑level process route, maintenance, start‑stop, and capacity planning.

Multi‑period inventory and product‑mix optimization.

Carbon‑emission reduction and carbon‑footprint tracking, providing product‑level carbon labels.

Case studies such as Guangdong Petrochemical’s pre‑startup crude‑oil package optimization and carbon‑footprint‑driven adjustments demonstrate significant cost savings and emission reductions.

Impact

The integrated models have been deployed across more than 27 CNPC refineries, enabling data‑driven decision making, improving planning efficiency, and supporting China’s carbon‑peak and carbon‑neutral objectives.