多點風速輸入的FCM-ANFIS波浪推算模式發展

Development of the FCM-ANFIS wave forecast model with an input of wind velocity at multi points

張家羲(Jia-Si Jhang)

指導教授 ： 張憲國

國立交通大學/工學院/土木工程系所/碩士(2016年)

摘要

為了提高海上活動及工程作業安全，本文發展一套可精準且快速地推算局部波高的FCM-ANFIS模式。本模式乃利用ARW大氣預報模式的計算風速或交通部運輸研究所港灣技術研究中心在臺北港2010年量測的風速的FCM分類，並以適應性類神經建立風速與波高關係。由本模式與往昔研究的ANFIS模式及迴歸模式的推算波浪結果比較，證實FCM-ANFIS波浪推算模式比他兩種有較高的推算能力。
本文分析實測風速及計算風速的偏差，二者有低的偏差、均方誤差及高的向量相關係數，0.7922，並比較使用這兩種風速分別建立模式的波浪推算能力是相近的，此種結果都證實以計算風速來建立波浪推算模式是可行的。
本文增加數值計算的外圍八方位上的風速於模式的輸入變數，建立一個多點複合波模式。此模式的推算波浪能力優於相同輸入參數的單點風速模式。評估所有發展的各種模式的推算能力，以數值計算的外圍風速所建立FCM-ANFIS(u,v)多點複合波浪推算模式為最佳。其推算能力由原本單點模式的R2=0.70提高至0.77，而RMSE從0.35 m降至0.31m，R2提升7.0%，RMSE降低11%。以西北方位置的外圍風場風速與臺北港波高關係強，而增加此方位的風速有利於模式的推算能力。

關鍵字

適應性類神經模式 ； 波浪推算 ； 多點風速 ； FCM群集分析

並列摘要

A FCM-ANFIS model with fast and concise wave calculation was developed for the safety of marine actions and coastal engineering. The algorithm of fuzzy C-means (FCM) was applied to data cluster analysis on the calculated data of wind velocity using the ARW wind model and on the observed data at the offshore observation pole of the Taipei harbor for 2012. Adaptive neuro-fuzzy inference systems (ANFIS) was used to establish the relationship between the clustered wind data and the corresponding wave heights. The proposed FCM-ANFIS model was verified to have better simulation on wave heights than the ANFIS model or the empirical model.
The small deviations of the calculated data of wind velocity from the observations were evaluated by low bias and mean root of squared error and high vector correlation of 0.7922. The models with an input of calculated wind data or observations have equivalent simulation on wave heights. The results indicate that the calculated data of wind velocity are applicable for the input of FCM-ANFIS model instead of observed wind velocity.
The calculated data of wind velocity at outer positions in the semi-cardinal directions around the observation pole by a distance of 25 km and 50 km were added in the inputs of the original FCM-ANFIS model to make a new multi-input FCM-ANFIS model. The multi-point FCM-ANFIS model has better simulation than the single-point FCM-FNFIS model. Specially, the best multi-input FCM-ANFIS (u,v) model has the highest R2 of 0.70 and the lowest RMSE of 0.31m among all assessed models. The promotion is 7% in R2 and 11% in RMSE compared with the single-point FCM-ANFIS (u,v) model. The calculated north western winds have stronger relation with the wave heights at the observation pole than other directions and give an advantage over the FCM-ANFIS model for wave calculations.

並列關鍵字

adaptive neuro-fuzzy models ； wave forecasting ； multi-point winds ； fuzzy c means

參考文獻

2.張憲國、張家羲、劉勁成、陳志弘，考慮單點或多點風速輸入的適應性類神經波浪推算模式 ，海洋工程學刊，台北，2016。

4.賴瑩嬛，應用群集分析探討侵臺颱風之分類特性，國立交通大學土木工程學系碩士論文，2012。

7.張高瑋，考慮海陸風及湧浪特性的波高與風速之迴歸分析，國立交通大學土木工程系碩士論文，2015。

6.翁瑞嘉，應用混合式類神經網路模式推算季節風波浪之研究，國立交通大學土木工程系碩士論文，2009。

12.簡國基，海棠颱風登陸台灣前內核結構演變之研究，大氣科學期刊，2011。