In present time there is very active research in the field of multiprotocol label switching (MPLS), and lots of networks are supporting MPLS. In this paper, we verify a possibility of a heuristic LSP setup policy algorithm, which uses optimization procedure based on multi-objective model with Pareto ranking and Genetic Algorithm. Adaptation to variable and bursty traffic is achieved by using Neural Network learning capabilities, and decision making under uncertainty is made up with Fuzzy Logical approach. This algorithm functions in two phases – learning and operating, which are accomplished consecutive. In this paper algorithm is described and depicted. Experimental data are depicted and future research subjects are pointed.