Forecast Of Typhoon Rainfall Based On B Neural Network

Because of its unique geographic location and diverse topography, Hong Kong is subject to an annual typhoon season that often results in serious damage to Hong Kong residents’ lives and property. However, in China’s coastal areas, most of the damage associated with typhoons is the result of typhoon rainfall, so typhoon rainfall prediction research has become an important key for mitigating typhoon disasters.In this paper, typhoon rainfall prediction research for the Hong Kong area is examined in terms of spatial distribution and dynamic changes in typhoon rainfall.First, according to22typhoon materials and data from48weather stations typhoon rainfall data were provided by Hong Kong Observatory covering the period from1997through June,2012. The analysis categorized each typhoon according to their corresponding complete typhoon rainfall data sets. At the same time, specific analysis of typhoon rainfall characteristics in the Hong Kong region is based on typhoon rainfall data sets.Second, the study selected a more accurate method than the traditional meteorological forecast BP neural network method to predict the rainfall brought by each typhoon. After analysis of BP neural network method theory and steps, the researchers selected the network topology structure and the related parameters of the BP neural network, and used GWBASIC software to write the neural network prediction program, which learns to improve the accuracy of its predictions based on input of the training samples.Third,22typhoons were divided into six types according to their different movement paths, i.e. to the N, NNW, NW, WNW, W and SW, and researchers selected10stations from the48stations to provide typhoon rainfall data, including Shau Kei Wan, High Island, Sha Tin, Yuen Long, Tung Chung, Shek Kong, Kwai Chung, Hong Kong Observatory, Tai Mei Tuk and Tap Shek Kok. When examining the forecast typhoon rainfall spatial distribution, one type of typhoon was selected from the six direction types:Maggie (9903), Imbudo (0307), Kompasu (0409), Pabuk (0706), Fengshen (0806) and Molave (0906). Rainfall data from Hong Kong Observatory, Tai Mei Tuk and Tap Shek Kok were used as test samples; data from the other seven stations were used as training samples; when the forecast typhoon rainfall dynamics changed, Hong Kong Observatory, selected Tai Mei Tuk and Tap Shek Kok from10stations to study, and selected typhoons of the No. IV path from the six identified typhoon paths, among typhoon rainfall data of York (9915), Utor (0104), Yutu (0107), Hagupit (0218), Kammuri (0809), Hagupit (0814) and Molave (0906) as training samples, others were used as test samples.Forth, the training samples were input into the program written for training and building the BP neural network forecast model of typhoon rainfall in the Hong Kong region. After many iterations, and constant adjustment of the neural network parameters to get more accurate predictive values, researchers compared the forecasts generated by the test samples, the summary forecast results, and the analysis of the BP neural network forecast model of typhoon rainfall to the actual typhoon rainstorm using historical typhoon data.Finally, based on statistical analysis of predictive results of the BP neural network prediction model of typhoon rainfall, the study shows that in terms of typhoon rainfall prediction for one typhoon in the Hong Kong region, the prediction accuracy rate for typhoon rainfall spatial distribution is greater than70%, the prediction accuracy rate for typhoon rainfall dynamic changes is greater than80%, therefore, prediction results are relatively accurate in term of typhoon rainfall prediction for one station, The maximum relative error for the typhoon rainfall spatial distribution prediction reached75.1%at Hong Kong Observatory, Imbudo (0307). The maximum relative error for the typhoon rainfall dynamic change prediction was up to71.3%at Hong Kong Observatory, Doksuri (1206), indicating that the prediction results error is large. Therefore, typhoon rainfall prediction models established using the BP neural network method, in terms of typhoon rainfall prediction for a single typhoon, is more practical. However, in terms of typhoon rainfall prediction for a single station further research still needs to be done.