A method of visibility forecast based on hierarchical sparse representation

摘要

This paper proposes a visibility forecast method based on hierarchical sparse representations. Firstly, it selects meteorological factors from the data of 138 ground stations located in Beijing, Tianjin and Hebei during the months (Oct-to-Dec. and January) of years 2002-2016. Then, it uses fuzzy C means algorithm (FCM) to construct historical databases containing 5000 samples. Finally, it takes the meteorological factors corresponding to visibility as the sample of historical databases, and uses a hierarchical sparse representation to predict the visibility of new inputs. Experiment, conducted with the data of European Centre for Medium-Range Weather Forecasts (ECMWF), indicates a better performance of the hierarchical sparse representation in contrast to a sparse representation. And the visibility forecast based on hierarchical sparse representation is better than Beijing Regional Environmental Meteorology Prediction System (BREMPS) and BP neural network. The hierarchical sparse representation is simple and easy to expand, which improves the accuracy and reduce the absolute error, which is convenience for other meteorological analysis. (C) 2018 Elsevier Inc. All rights reserved.