<p>In the cotton spinning manufacturing system, cotton blending plays a decisive role in determining both product quality and cotton cost, making it a critical factor affecting overall spinning efficiency. For the long-term sequential cotton blending optimization problem (L-SCBO) faced by spinning enterprises under large-scale orders and dynamic inventory conditions, traditional optimization algorithms struggle to balance solution quality and computational efficiency when the solution space expands dramatically. Inspired by the “less-change, slow-change” experience in manual batch scheduling, and integrating an inventory recursion mechanism, this study formulates a multi-stage combinatorial optimization model oriented toward sequential decision-making. Cost, yarn quality deviation, and batch-transition penalties are jointly incorporated into a unified multi-objective sequential decision-making framework. On this basis, we propose a Cascaded Ant Colony Optimization System (CACOS). It uses a two‑level cooperative strategy: System 1 searches for the initial cotton blending table, and System 2 performs progressive replacements. A hierarchical pheromone matrix and tensorized parallel computation are applied to improve search efficiency and solution quality. Across six test scenarios of varying scales and complexities, CACOS demonstrates stable convergence and strong robustness. Comparative experiments are conducted against both the classical Ant Colony Optimization (ACO) and its variants, the Deep Reinforcement Learning Enhanced Ant Colony Optimization (DeepACO), as well as other advanced heuristic algorithms. Results show that none of the comparative algorithms can simultaneously satisfy both yarn quality compliance and batch-transition stability requirements for all blending plans. In contrast, CACOS consistently outperforms the comparative algorithm in solution quality, stability, and convergence speed, achieving up to 66.3% cost reduction with an average runtime of only about 20% of the fastest comparative algorithm. The integration of multi-level pheromones and the cascaded decision chain markedly enhances search efficiency and solution quality in large-scale, strongly coupled solution spaces, demonstrating that CACOS can serve as an intelligent and efficient decision-support tool for continuous industrial production with significant engineering application value.</p>