In this work we consider the problem of detection of stationary intervals for random waves using clustering techniques for time series. The detection of changes in time series from an spectral point of view has been studied in several areas, but the methods proposed usually focus on detecting instantaneous changes, and the results are not satisfactory when changes are slow, which is frequently the case for waves. In the method used in this work there is a change of viewpoint: instead of looking for change-points, the method looks for intervals having similar oscillatory behavior as candidates for stationary periods. The procedure is based on the comparison of normalized estimated spectral densities using the total variation distance. This method is applied to a long data series of buoy measurements and the results are statistically analyzed. This analysis gives statistical characteristics for the duration of stationary intervals, which may vary for different periods of the year.