Rate-dependent hysteresis is a strongly nonlinear phenomenon which exists in the giant magnetostrictive actuator (GMA); it has influence in the precision and stability of active vibration control. It is highly important in the control theory and control engineering that the influence of hysteresis is eliminated by the modeling of rate-dependent hysteresis for GMA. So an online intelligent modeling method, which is based on an improved online least squares support vector machines (IOLS-SVM), is presented for identifying rate-dependent hysteresis nonlinearity for GMA, and is used to online real-time training. The data measured in the experiment are used for modeling. The numerical simulation shows the effectiveness of the method.