To study the identical parallel machine scheduling problem with total tardiness objective, an improved discrete differential evolution DDE_VND is presented. The proposed algorithm hybridizes discrete differential evolution (DDE) with variable neighborhood descent (VND) to enhance its local search ability. Moreover, a constructive heuristic MDD is employed to generate an initial solution in the algorithm to accelerate the convergence of the algorithm. At last, the experimental results based on 50 benchmark instances validate the effectiveness of the DDE_VND algorithm