According to the problem of Electromagnetism Mechanism (EM) algorithm which may easily get into local optimal solution and has poor search capability, this paper combined the Limited memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) with chaotic model into EM. The main idea of the algorithm was using the L-BFGS which has high precision, in the later stage of algorithm, and using the chaotic model through the whole algorithm to keep the diversity of population. The tests suggested that the algorithm could jump out from the local optimal solution, had better solution and converged faster than EM, Particle Swarm Optimization (PSO) and particle swarm optimization with Time-Varying Accelerator Coefficients (TVAC). Tests also showed that it could be used in path planning and had better results than both PSO and Ant Colony Optimization (ACO), so the algorithm can be applied to the discrete domain question.