An innovative integrated optimization strategy for gasoline blend planning is proposed, and an improved method to achieve online optimization of real-time blend recipes is described. The proposed strategy can calculate a rough blend and delivery sequence of gasoline and then adapt to process changes by using a three-level discrete-time algorithm. Only one blender is considered in this study. A single-period nonlinear model (NLP) is solved at the top level of the algorithm to check the feasibility of a long-term production plan. A multi-period mixed-integer nonlinear model is formulated and solved at the middle level of the algorithm to compute a short-term blend plan. Finally, a single-period NLP is solved circularly at the lowest level of the algorithm to optimize blend recipes to consider the changes in the quality of blend components. The initial plan is modified if the top-level model is not feasible. The middle-level model is resolved if an unexpected event occurs during blending. The proposed approach is advantageous because the initial planning and blending recipes can be modified online, remarkably minimizing quality giveaway and increasing the blending success rate. The performance of the proposed strategy is illustrated through its industrial application in real-world gasoline blending.