An Efficient Differential Evolution for Truss Sizing Optimization Using AdaBoost Classifier

Design constraints verification is the most computationally expensive task in evolutionary structural optimization due to a large number of analyses that must be conducted. Building surrogate model approximate behavior structures instead exact possible solution tackle this problem. However, existing models have been designed based on regression techniques. This paper proposes novel method, called CaDE, which adopts machine learning classification technique for enhancing performance Differential Evolution (DE) optimization. The proposed method separated into two stages. During first stage, original DE implemented as usual, but all individuals produced phase are stored inputs training data. Based design verification, these labeled “safe” or “unsafe” and their labels saved outputs When collecting enough data, an AdaBoost trained evaluate safety state structures. then used second stage preliminarily assess new individuals, unpromising ones rejected without checking constraints. reduces unnecessary analyses, thereby shortens process. Five benchmark truss sizing problems solved using demonstrate its effectiveness. obtained results show CaDE finds good optimal designs with less comparison four other variants. reduction rate five examples ranges from 18 over 50%. Moreover, applied real-size transmission tower problem exhibit applicability practice.