A hierarchical multi-innovation stochastic gradient identification algorithm is proposed for Hammerstein-Wiener(H-W) nonlinear systems with non-uniformly sampling. The corresponding state space models of H-W are derived by using the lifting technique. Considering the causality constraints, the H-W system is decomposed into two subsystems firstly. Then the model parameters are identified by using the multi-innovation based stochastic gradient algorithm with forgetting factors. In order to improve the convergent rate and the disturbance rejection, a new kind of variable forgetting factor algorithm is also presented. Simulation examples demonstrate that the proposed algorithm has fast convergence speed and is robust to the noise.