Abstract:

By analyzing the meteorological data and apple annual yield data in Yuncheng City from 2005 to 2018, an early prediction model of apple yield was constructed. Firstly, the annual yield of apple was divided into trend yield and meteorological yield by using HP filtering method. Secondly, multiple linear regression models were established respectively for apple phenology stages: germination stage, flowering stage, young fruit stage, expanding stage and mature stage, to study the influence of each phenology period on apple meteorological yield. Finally, the early prediction model of BP neural network was established and verified in terms of the young fruit stage which had the strongest influence on meteorological yield of apple. The relative average error of early yield prediction model based on BP neural network was 7.08%, and the accuracy of BP neural network early prediction model was verified with the relevant data in 2019, which was 89.6%. The model could accurately predict apple yield and provide theoretical support for early crop yield prediction.

Key words: meteorological data, apple yield, early prediction, HP filtering method, multiple linear regression, BP neural network