Abstract:In the medical field, due to the fact that diseases are often affected by many factors, it is difficult to use a structural causal model, while on the other hand, it would be effective to establish a dynamic model, based on their own time-series changes. To predict the number and incidence of diseases, because meteorological factors, the monthly average atmospheric pressure, monthly mean temperature, monthly mean relative humidity, monthly average wind speed, the monthly average precipitation are strongly correlated between themselves and with very high dimensions, the accuracy of the Radial Basis Function (RBF) neural network prediction may be very low. To solve this problem, this paper proposes the use of the Principal Component Analysis (PCA) method to reconstruct the original input space and, based on the contribution rate of all main components to determine the network structure, which will effectively solve the problem of low prediction accuracy. The incidence data of upper respiratory tract infection, from August 2001 to September 2006, in Wuwei City of Gansu Province are used to validate the method. The clinical characteristics in each time period should be duly considered in order to carry out a more focused health prevention and treatment and to effectively reduce the hazards to human bronchial pneumonia.