Dynamic flexible job shop scheduling (DFJSS), as a more realistic extension of job shop scheduling (JSS), has received a great deal of attention from academics and industry researchers due to its theoretical and applied research value. The challenge of DFJSS is how to capture both the machine assignment (routing) decision and operation sequencing (sequencing) decision simultaneously along with the new arrival jobs over time. In the previously proposed methods, dispatching rules come to the most state-of-the-art because of their low time complexity, ease of implementation and the ability to cope with both static and dynamic situations in the job shop floor. However, the dispatching rules are normally designed manually, which is very time-consuming and domain-dependent. Genetic programming (GP) has been proven to be a promising hyper-heuristic method to automatically design dispatching rules for JSS. The overall goal of this thesis is to improve the effectiveness and efficiency of GP to evolve promising and interpretable rules for solving DFJSS problems. This will be achieved by investigating new representations for GP and incorporating feature manipulation, surrogate and multi-objective technologies with different strategies.