For constrained many-objective optimization problems, a reference point-based constrained dominance principle(RPCDP) is designed, regareding the feasible solutions and infeasible solutions as a whole and considering the convergence, the diversity and the feasibility simultaneously. Then on this basis, an improved NSGA-III algorithm is proposed. The experimental results on CDTLZ test suite show that compared with three state-of-the-art constrained many-objective evolutionary algorithms, the proposed algorithm has better performance on convergence and distribution.