基于知识挖掘的 HDMR 优化方法与工程应用

Despite the wealth of experimental, simulation, and design experience in engineering, traditional design methods struggle with low knowledge utilization. To address this, a high dimensional model representation (HDMR) optimization method grounded in knowledge mining is presented. The approach employs an improved multivariate model screening strategy for enhanced efficiency and prediction accuracy of HDMR subcomponents. The optimization strategy integrates a global surrogate model, utilizing optimal samples to construct HDMR sub-items and identifying local advantages in each dimension. Confidence comparisons expedite global potential advantage discovery, accelerating algorithmic optimization. The proposed method is applied to shape optimization for a blended-wing-body underwater glider (BWBUG). Under volume constraints, the glider's lift-to-drag ratio increases by 5.04%, surpassing the 2.93% increase without knowledge assistance. This validates the impactful role of knowledge mining in the proposed methodology, providing a novel perspective and method for high-dimensional optimization problems while contributing to the advancement and application of optimization algorithms.