Optimization Strategy of the Hyena Algorithm With Multi-Directional Variable Weights in Complex Dyeing Workshop Scheduling

This paper focuses on the dyeing process and scheduling model, conducting an in-depth analysis of the dyeing production process and its characteristics. The core of fabric dyeing involves the systematic transformation of fiber raw materials through multiple precise processes to produce textile products. The industry features tightly interconnected and continuous operations, requiring strict material balance control and the establishment of dynamic monitoring mechanisms to ensure stable quality. Traditional production processes are divided into pre-treatment (including fabric preparation, scorching, degumming, and boiling) and post-treatment (bleaching, mercerization, dyeing, and finishing) stages, each of which significantly impacts fabric quality. However, current market demand changes have posed challenges such as diversified product varieties, rapid style iterations, small order quantities but large total volumes, making traditional fixed production line models difficult to adapt to. Frequent equipment switching leads to increased costs and reduced efficiency. To tackle this issue, a scheduling model covering four product categories was established. Drawing on sales-side order clustering, production notices are created with a batch consolidation strategy for the same product variety; resources are dynamically scheduled by machine unit per the "first idle" principle, with defined process sequences, equipment configurations and scheduling logic rules to balance production continuity and multi-variety flexibility. Moreover, the Hyena Algorithm was improved to solve scheduling problems in dyeing workshops, as the original algorithm had three key technical flaws: insufficient population diversity (caused by initial solution generation ignoring process constraints and leading to concentrated individual distribution), unbalanced local/global search capabilities (due to fixed exploration step sizes without dynamic adjustment during iteration), and premature convergence (resulting from the lack of adaptive mutation mechanisms). The improved algorithm added exploration strategies to expand individual movement ranges, optimized position updates through weight coefficients, and set safety distances to avoid local optima. Experimental comparisons showed the improved algorithm significantly outperformed the original in terms of Average Relative Deviation Value (ARDV) and Minimum Relative Deviation (MRD) metrics, converging faster and escaping local optima. It demonstrated robust performance across various working conditions, particularly excelling in complex data scenarios. Under the fixed production line mode, the GP2 algorithm achieved a fitness of 0.005 after 554 iterations, compared to 0.08 for the GP1 algorithm. In the non-fixed production line mode, the Good Programs 2 (GP2) algorithm converged to a fitness of 0.01 after 523 iterations, while the Good Programs 1 (GP1) algorithm reached 0.102 after 578 iterations. This study is expected to promote the transition of the dyeing industry toward efficient and flexible production models, helping enterprises enhance competitiveness in the era of consumer sovereignty and achieve sustainable development.