Placing multiple controllers in a large-scale software defined network (SDN) is an NP-hard problem. Currently, it is still a challenge to design a solution algorithm with high accuracy, short time consumption, and compliance with engineering standards. To overcome this challenge, we proposed the concept of synthetical delay, and took the controller capacity as a constraint condition, and established a controller placement model. The firefly algorithm (FA) was chosen as the solving algorithm for the controller placement problem. However, the FA has the drawbacks of insufficient search breadth in the early stage and low convergence efficiency in the later stage. To this end, we first introduced the dynamic parameter strategy for the light attractiveness, the light absorption coefficient, and the parameter to control the random walk of the FA. Then, we adopt the particle swarm optimization to improve the initial static parameters of the FA. Experimental results in the real large-scale SDN topologies show that, compared with the latest controller placement algorithms based on cuckoo optimization, chaotic salp swarm algorithm, Varna-based Optimization, and Teaching-Learning-based Optimization, the controller placement scheme obtained by the improved FA has a lower synthetical delay and more balanced controller load. Simultaneously, the time consumption of the improved FA is acceptable.