In this paper, we treat the problem of online nonlinear system identification with parameter constraints. This approach is based upon our prior work on nonlinear system identification that exploits evolving Spatial-Temporal Filters (STF) to dynamically decompose system’s input/output space into a nonlinear combination of weighted local models. We extend the nonlinear system identification framework with the capability of dealing with linear equality and inequality parameter constraints. We leverage the gradient projection method in the local model parameter estimation process to inherently enforce the parameter constraints while retaining optimality. We apply the proposed algorithm to a turbo-charged gasoline engine system and promising results are demonstrated by experimental data.