Abstract: Aiming at the performance bottlenecks and slow convergence of fireworks algorithm (Fireworks Algorithm, FWA), By putting reverse learning and maneuver explode into FWA, this paper proposes fireworks algorithm based on reverse learning and maneuver explode (Fireworks Algorithm based on Reverse learning and Maneuver explode, RLMEFWA). In the algorithm, reverse learning strategy was introduced to generate initial population, which strengthened the diversity of population, then each firework applies the position in the current group to choose a different explode way. Fireworks in the better position to select maneuvering explosion mode whose random orbit closer to the optimum position. Fireworks in an inferior position explode select non-motorized mode with random orbit. In our simulation, we compare FWA, SPSO, EFWA and RLMEFWA with 10 typical benchmark functions. The results show that RLMEFWA is better than FWA, SPSO, and EFWA in terms of convergence speed and accuracy and stability.