Dynamic job shop scheduling has a wide range of applications and has attracted a lot of research interest from both academic and industrial communities. To solve this type of problems effectively and efficiently, genetic programming based hyper heuristic method is one of the most popular methods. However, existing studies of genetic programming-based hyper heuristic method are mainly designed based on tree-like structures, and some limitations of the tree-like structures also limit the further improvement of existing methods. To evolve more effective dispatching rules for dynamic job shop scheduling problems, this thesis intends to design hyper-heuristic methods based on linear genetic programming, which has been shown to have different advantages over tree-based genetic programming. Specifically, this thesis will improve the training efficiency bydesigningnewsurrogatemodels,improvetheeffectivenessbynewgrammarguided techniques, improve the effectiveness in multiple tasks by linear genetic programming specific multitasking techniques, and improve the robustness by coevolution paradigms