Multi-path acquiring methods based on multi-population parallel genetic algorithm

To avoid the negative effect of guidance and to decrease congestion shifting caused because the drivers choose the optimum path simultaneously, more reasonable paths should be recommended, but there is no quick and effective algorithm to seek the k-shortest path. Genetic algorithm, characterized by overall optimization and potential parallel, is suitable for k-shortest path seeking. When the ordered nodes are applied, premature convergence and low searching efficiency at later stage of evolutionary often appear because a large number of invalid paths are produced during the genetic operation, and the path is more blindly generated. This paper will improve the algorithm in two ways: firstly, convergence speed is improved by the introduction of multi-population parallel algorithm which can maintain the population diversities, prevent premature convergence and improve later searching efficiency; secondly, a new code method is used to reduce algorithm complexity. The code method based on the turn actions is adopted to decrease the invalid path number during the genetic operation while the loop punishment factor is introduced in the fitness calculation to avoid the operations of eliminating loops. Numerical research shows that the improved algorithm has the advantages on fast convergence and high searching efficiency.