模糊-類神經混合網路之颱風波浪推算模式

Abstract

本研究以類神經網路與模糊規則為基礎，經過分析、測試、比較不同複合結構的模式，結果是以利用類神經網路、模糊規則與複合轉換函數所建構出的颱風波浪推算模式（FN2）推算的颱風波浪有最佳表現。在經驗與數值方法之外，本研究提供了另一個可靠、快速的颱風波浪推算模式建構方法。 FN2模式採用JTWC發佈的颱風規模與路徑，配合Holland風場模式建立模式輸入資料，輸入參數包括：推算點上10公尺之風速、颱風中心與推算點之距離、颱風中心與推算點連線之方位角、颱風行進方向與觀測站夾角。模式藉由輸入參數24小時延時資料，推算颱風因行進路徑不同對颱風波浪的影響程度。再利用可調整的複合轉換函數，降低類神經網路因學習資料不足可能產生的誤差，並提高模式學習的效率與推算的準確性。最後，配合模糊規則與陸地特徵建立模糊歸屬函數，以修正颱風波浪所受到陸地的影響。 本研究以台灣東岸花蓮港海域為例，以FN2 結構所建立的花蓮港海域颱風波浪推算模式，經配合實測資料進行檢驗，結果證實對預測颱風最大波之時間與大小具有令人滿意的推算能力。此外，FN2模式亦可廣泛應用在與推算點的位置地理環境相近的鄰近海域。

This study analyzes, tests, and compares different models with hybrid structures based on neural networks and fuzzy rules. It is found that the forecasting results of the FN2 typhoon wave model, which combines neural network, fuzzy rules and multi-trend transfer functions, are the best. In addition to experiential and numerical methods, this study provides another fast and reliable method to construct the typhoon wave forecasting model. FN2 model uses the typhoon category and route data issued by JTWC and the Holland wind model to build up the input data. Suitable parameters selected include the wind velocity at 10m high above the sea surface, the distance from typhoon center to the interesting point, the azimuth between typhoon center and the interesting point, the angle of an interesting point in a typhoon between the typhoon moving direction and the radial direction from the typhoon’s eye to that point.This model evaluates the influence generated by different typhoon routes through the 24-hour time-delay input data. Then, it applies the adjusted multi-trend transfer function to decrease the errors resulted from the shortage of learning data and to enhance its learning efficiency and accuracy. Finally, FN2 applies fuzzy rules and land attributes to build up the fuzzy membership functions so as to adjust the environment influence on typhoon waves. This study takes Hua-Lian Harbor in the eastern waters of Taiwan as the subject and applies its related observed data for verification. The construction of Hua-Lian Harbor typhoon wave forecasting model is based on the FN2 structure. The results show that the model’s capability for simulating the most height and scale of the typhoon waves is quite satisfied. Besides, FN2 can also be extended to the points with similar land attributes neighboring to Hua-Lien Harbor.