Abstract: It is necessary to predict the spectrum occupancy so as to provide high-frequency users with reliably available band. This paper proposes a prediction method based on adaptive Volterra filter theory, which reconstructs the congestion time series based on State-space Reconstruction theory and modifies the kernel coefficients of Volterra model in real-time with RLS algorithm. Experiments are implemented based on real measurements and the results demonstrate that Volterra prediction method can effectively capture nonlinear variations of congestion, as well as it has the advantages of a small forecasting error and a low computational complexity in the training process.