To reduce the random drift of a laser Doppler velocimeter effectively and improve its measurement accuracy, a metabolic double time series model based on the traditional time series model is put forward for filtering drift data of a laser Doppler velocimeter. The model consists of a cascade of two metabolic time series models, each of which models 13 data points using a time series model in turn. The static and dynamic drift data of a laser Doppler velocimeter are modeled and filtered based on the model respectively. The variance analysis method and the Allan variance method are used to analyze the static drift data before and after being modeled and filtered. The dynamic drift data is also compared by the spectrum analysis method. The results show that this method reduces the standard deviation of the static drift data to 44% of the original data, and reduces the angular random walk to 41%. This method can not only reduce the static random drift error in real time, but can also suppress the dynamic output noise effectively.