Abstract: To improve the accuracy of crop pest forecasting, this paper introduced and applied radial basis wavelet network into the area of crop pest forecasting for the first time. The author modified the learning algorithms of radial basis wavelet network for application in pest forecasting. The scale and translation parameters were determined by the theory that timefrequency support of analyzed data sequence is covered with time-frequency support of radial basis wavelet functions. Based on the Euclidean distance between central vectors, the hidden-layer neurons are selected preliminarily. At case study, the investigation data of Helicoverpa armigera in Huimin, Shandong between 1966 and 1995 were used to establish the forecasting model of oviposition peak day in the second generation of H. armigera based on radial basis wavelet network, while the investigation data between 1996 and 2000 were used to test the model. The test results showed that the forecasting deviation of four years was less than three days and the forecasting deviation of one year was four days. The forecasting results proved satisfactory. This paper developed a new studying method for crop pest forecasting.

Key words: Helicoverpa armigera, oviposition peak day, forecasting, frame, radial basis wavelet network, gram-schmidt orthogonalization