For solving the problem of state estimation in nonlinear systems with bounded but unknown noise, a hyperparallel space set-membership filtering based state estimation algorithm is proposed. The Stirling matrix is used to expand the model into the one dimension, and the linearization error boundary is calculated based on convex difference programming. Then, the hyperparallel space is used to represent the error boundary and the feasible state set, before finding the predictive parallelotope incorporating true states in the next time. In the update step, the observation is decomposed into multiple stripe, and the updating feasible state set described by the hyperparallel space is obtained by integrating the linearization error of the observed values into stripe and intersecting them with the parallelotope in turn. The proposed algorithm avoids the volume increase caused by sets bounding remainder in the process of linearization error, thus can reduce the conservatism of nonlinear set-membership filtering algorithm. The simulation example shows the feasibility and effectiveness of the algorithm.