As the dominant computing paradigm, cloud computing has become an ideal solution for processing large-scale computing applications, which are often decomposed into massive workflows. However, efficiently scheduling workflow tasks with complex dependencies and allocating appropriate virtual execution units on dynamically fluctuating resources remain significant challenges. Traditional heuristic algorithms often struggle to balance convergence speed and solution diversity when addressing such large-scale multi-objective optimization problems. In this paper, we propose a genetically-modified multi-population particle swarm optimization approach (GMPSO) for workflow scheduling in cloud environments, aiming to achieve a comprehensive optimization of makespan and energy consumption. GMPSO divides the swarm into fitness-based sub-populations and applies distinct genetic operators to each, combining the global search ability of genetic algorithms with the local refinement strength of particle swarm optimization. The (1) load-aware population initialization method, (2) differentiated multi-population search mechanism, and (3) adaptive genetic operator perturbation strategy significantly improve the search quality of GMPSO. We evaluate GMPSO on eight different scales of four types of workflows, and compare its performance with four state-of-the-art algorithms. Experimental results show that GMPSO achieves average improvements of 23.88% in inverted generational distance (IGD), 15.77% in hypervolume (HV) ratio , and 26.01% in spread, demonstrating the excellent overall performance in makespan-energy optimization.