A knowledge-assisted integrated optimization framework for planning-scheduling coordination and quality giveaway elimination in refinery gasoline blending

The decoupling of long-term planning and short-term scheduling in large-scale petroleum refining engenders a severe synergy gap, manifesting as systematic resource waste, pseudo-feasible production directives and costly quality giveaway during multi-component gasoline blending.To mitigate these industrial inefficiencies, this paper proposes a knowledge-assisted integrated optimization framework prioritizing resource efficiency and dynamic operational executability. Specifically, Knowledge-Driven Feasibility Projection (KDFP) is introduced as a parameterized knowledge-formalization mechanism that converts documented formulation rules and discrete scheduling constraints into continuous feasibility boundaries within the planning model. To navigate the highly constrained, non-convex thermodynamic feasible regions inherent to precise fuel blending, a Sequential Hybrid Evolutionary Algorithm (SHEA) is deployed to solve the resulting high-dimensional, multi-batch scheduling problem. SHEA autonomously detects optimization bottlenecks via search stagnation gradients, triggering a dynamic transition from rapid swarm-based exploration to rigorous Pareto-front exploitation. Validation using operational data from an Eastern China refinery with a 20,000-ton daily capacity confirms the framework's practical utility. The coordinated approach ensures continuous multi-batch executability strictly within dynamic tank safety limits and systematically eliminates the historical 0.2–0.4 research octane number (RON) quality giveaway per batch. Supported by robust convergence metrics, including a 167.2 s runtime and an optimal inverted generational distance (IGD) of 0.14, this strategy successfully translates advanced optimization mechanics into tangible industrial material efficiency, providing a practical technical pathway for sustainable refinery operations.