Considering the situation that facilities are distributed into multilayers space，a double layer corridor allocation problem was proposed，and the mixedinteger nonlinear programming model was built for this problem． Based on the discrepancy of feasible solutions and complexity of the issue，a discrete flower pollination algorithm was proposed． By redefining the pollination process，the random search process that sets the problem size as the search depth was taken as the global search． In the local optimization stage， the individual updated itself via the exchange pair by following the optimal solution． In order to improve the performance of the algorithm， the thresholds were set in the global search stage to actualize variable neighborhood search by mutating the individual that was in the local optimum，and another threshold was set to improve the efficiency of the solution． The effectiveness of the algorithm was validated by comparing the results that are gotten from two algorithms via solving the 38 benchmark instances． Finally， the improved flower pollination algorithm was applied to solve the corridor allocation problem， and the contrast experiments illustrate that the proposed algorithm possesses more advantages both in quality and efficiency．