Diversity-driven Knowledge Transfer for GPHH to Solve Uncertain Capacitated Arc Routing Problem

摘要: Uncertain Capacitated Arc Routing Problem (UCARP) is a dynamic combinatorial optimisation problem which can model many real-world logistic systems. Currently, the best available method for solving UCARP approach of using Genetic Programing as hyper-heuristic to evolve routing policies vehicles automatically. An open challenge in this area that any change features solved instance will make trained ineffective new problem. As result, whenever such changes happen, it required train from scratch. The process training generally expensive. It desirable utilise transfer learning methods reduce retraining cost. However, earlier studies have identified performing handling challenging task. Lack diversity and possible convergence poor local optima are some issues contribute challenge. To address these issues, work, we propose with hyper-mutation GPHH tackle issue insufficient transferred knowledge. Our experiments demonstrate newly proposed increase effectiveness knowledge allow better scenario through learning.