关键词: ECMWF细网格, BP神经网络, MOS方法, 气温预报

Abstract:

Based on the high resolution prediction field data of ECMWF (European centre for medium-range weather forecasts), an optimized BP-MOS model was used to forecast the maximum and minimum temperatures in the next 1-7 days in different seasons, and the results  were compared with the forecast effects of ECMWF 2 m temperature products and conventional linear MOS method. The results show that it could effectively improve the accuracy of prediction effect when the process effects of cloud cover, wind and humidity on temperature were considered. The root mean square errors of ECMWF 2 m temperature product in the short term (less than three days) were 2 ℃ below, however the forecast effect for the medium-term was significantly poorer than that of the two MOS methods. The linear MOS model had some unstable phenomena, and the nonlinear mapping relation of BP neural network made it better in fault tolerance. Although there were seasonal differences, the accuracy of the optimized BP-MOS model could meet the needs of the businesses.

Key words:  ECMWF fine mesh, BP neural network, MOS method, temperature forecast