Dynamic multi-mode resource-constrained project scheduling problem (DMRCPSP) is crucial for effectively managing complex projects where activities have multiple options of resource demand and durations are uncertain. Efficient solutions to this problem are vital for industrial applications, where optimal scheduling can significantly impact project costs and timelines. Genetic programming hyper-heuristic (GPHH) has been applied to evolve effective scheduling heuristics to make real-time decisions. However, current decision-making procedures for solving DMRCPSP primarily generate non-delay schedules, which are often suboptimal. This paper proposes two new decision-making procedures to generate (semi-)active schedules. The core idea is to expand the decision set and allow activities that cannot immediately start to reserve resources for future execution. We further developed GPHH to evolve scheduling heuristics based on these procedures and a comparison was conducted with existing decision-making procedures. The results demonstrated that the proposed GP approaches managed to evolve rules that generate more effective (semi-)active schedules than the best-known non-delayed schedules in real-time.