Data Synthesis Guided Multi-Scale Optimization: A Novel Heuristic Approach

In complex high-dimensional optimization problems, traditional heuristic optimization algorithms face problems such as local convergence, low search efficiency, and insufficient global exploration capability. This paper proposes a novel multi-scale adaptive optimization algorithm (Data Synthesis Guided Multi-Scale Optimization, DSGMSO) based on the idea of data synthesis and augmentation. This algorithm draws on the core concept of data augmentation in machine learning, generating new solutions by multi-scale perturbation and weighted synthesis of candidate solutions, and introducing adaptive augmentation coefficients and diversity preservation mechanisms to achieve a balance between local search and global exploration. This paper systematically constructs the algorithm framework theoretically and elaborates on the algorithm mechanism in detail through mathematical formulas, including the multi-scale perturbation generation formula, the weighted synthesis formula, the adaptive perturbation adjustment formula, and the diversity fitness function formula. This method provides a novel approach that deeply integrates data augmentation mechanisms with heuristic optimization, providing theoretical support for high-dimensional, multimodal, and black-box optimization problems.