【摘要】 智能制造的高效发展亟需解决多目标多任务协同优化中的知识迁移瓶颈问题.尽管前沿信息技术与工业场景深度融合显著提升了生产效能,但任务间知识负向迁移导致的优化效率衰减,仍是制约协同优化算法实际应用的核心挑战.为解决上述问题,本文提出一种基于知识分类转移与负向检测的多目标多任务优化框架.首先,本文构建一种由支持向量机驱动的动态知识遗传分类模型用于识别和提取不同任务之间的相关转移知识;其次,在知识转移早期,通过遗传分类模型开展跨任务知识分类转移克服因种群分布不均难以正确捕捉不同任务相关性的难题;然后,在知识转移中期,本文提出一种基于知识转移量的自适应随机交配概率,通过动态调整个体之间的知识流动程度,保证知识传递的相关性和有效性;继而在知识转移后期,利用知识遗传分类模型进行知识负向检测,排除无效或有害的知识进而提高知识转移质量;最后,本文给出基于知识分类转移与负向检测的多目标多任务优化算法的完整框架.实验结果表明,该算法在一系列基准测试问题和实际铝电解工艺参数优化案例中表现优异.因此该研究为推动协同优化智能制造快速发展提供了一种有效的理论与技术支撑.更多还原

【Abstract】 The efficient development of intelligent manufacturing urgently requires solving the bottleneck of knowledge transfer in multiobjective and multi-task collaborative optimization. Although the deep integration of advanced information technologies with industrial scenarios has significantly improved production efficiency, the decay in optimization efficiency caused by negative knowledge transfer between tasks remains the core challenge restricting the practical application of collaborative optimization algorithms. To address this issue, this paper proposes a multi-objective multi-task optimization framework based on knowledge classification transfer and negative detection. First, a dynamic knowledge genetic classification model driven by a support vector machine is constructed to identify and extract relevant transfer knowledge between tasks. Next, in the early stage of knowledge transfer, cross-task knowledge classification transfer is carried out through the genetic classification model to overcome the difficulty of capturing task-relatedness due to uneven population distribution. In the mid-stage of knowledge transfer, an adaptive random mating probability based on the amount of knowledge transfer is proposed, dynamically adjusting the degree of knowledge flow between individuals to ensure the relevance and effectiveness of knowledge transfer. Subsequently, in the later stage of knowledge transfer, the knowledge genetic classification model is used for negative knowledge detection to exclude ineffective or harmful knowledge, thereby improving the quality of knowledge transfer. Finally, the complete framework of the multi-objective multi-task optimization algorithm based on knowledge classification transfer and negative detection is presented. Experimental results show that the proposed algorithm performs excellently in a series of benchmark test problems and practical aluminum electrolysis process parameter optimization cases. Therefore, this study provides effective theoretical and technical support for promoting the rapid development of collaborative optimization in intelligent manufacturing.更多还原