For the problem that online learning path optimization methods don't have a high degree of matching between learning path and learners, this paper firstly constructs a multi-dimensional information feature mapping model(MIFMM) of online learning path, which is based on the multi-dimensional information characteristics of learners and learning resources, and integrates kolb learning style and learning resource type information. Then, a dual mapping binary particle swarm optimization(DMBPSO) algorithm is designed. According to the evolution factor ef, the DMBPSO algorithm divides the learning path recommendation process into two evolutionary states: convergent and out of local optimism, adopts a mapping function selection strategy which matches the evolutionary state features, and dynamically adjusts the inertia weight to improve the learning path recommendation performance. Futhermore, this paper combines the MIFMM with the DMBPSO algorithm to propose an online learning path optimization method based on the MIFMM model (MIFMM-POA). Finally, the MIFMM-POA method is compared with the learning path optimization methods based on the other four particle swarm algorithms, and the analysis is carried out from the three perspectives of optimization accuracy, optimization process and optimization time. The experimental results show that the MIFMM-POA method is an effective method to optimize the learning path.