以調適性網路模糊推論系統推估西北太平洋颱風強度預報

Applying Adaptive Neuro-Fuzzy Inference System to estimate typhoon intensity forecast in the Northwest Pacific

宋政樺(Jheng-Hua Sung)

指導教授 ： 林旭信

中原大學/工學院/土木工程研究所/碩士(2019年)

https://doi.org/10.6840/cycu201900093

摘要

本研究利用機器學習法評估颱風強度預報，颱風強度預報模式以調適性網路模糊推論系統(Adaptive Neuro-Fuzzy Inference Systems, ANFIS)架構為基礎，建立未來五天每十二小時之ANFIS颱風強度預報模式，並以多元線性回歸(Multiple Linear Regression, MLR)建立MLR基準模式，分析比較颱風強度預報之效能改進。
本研究以西北太平洋為研究區域，蒐集西元2000年~2012年西北太平洋颱風未登陸之SHIPS資料(SHIPS Developmental Data)，利用逐步線性迴歸法(Stepwise Regression Procedure, SRP)與主成分分析(Principal Components Analysis，PCA)簡化模式輸入資料，另外也採用減法聚類法減少輸入和輸出變量之間的關係，降低模式複雜度，本研究設計MI指標選用法做為減法聚類法的聚類標準，決定ANFIS颱風強度預報模式網路設定。
模式預報結果顯示：藉由MI指標選用法使得模式測試效能較佳。ANFIS颱風強度預報模式整體表現比MLR颱風強度基準模式低，所有的提前預報時刻最佳模式都是以SRP篩選輸入因子為佳，其中以加入海洋環境因子表現明顯較佳。另外本研究以JTWC(Joint Typhoon Warning Center，JTWC)颱風強度分級標準分類，在超級颱風的強度標準預報中，ANFIS明顯優於MLR，使用PCA轉換的輸入因子組合於超級颱風等級的預報效能是最好的。

關鍵字

颱風強度預報 ； 調適性網路模糊推論系統 ； 逐步線性迴歸法 ； 主成份分析 ； 減法聚類法

並列摘要

This study employs machine learning to estimate typhoon intensity prediction. The prediction model for typhoon intensity is based on the adaptive neuro-fuzzy inference systems (ANFIS). The ANFIS typhoon intensity prediction model is built every 12 hours for the next five days; the improvement of typhoon intensity forecasts is compared to a baseline model with multiple linear regression (MLR).
This study uses the Northwest Pacific basin as a case area and collects 2000~2012 typhoon non-landing data in SHIPS Developmental Data. The stepwise regression procedure (SRP) and principal components analysis (PCA) are applied to simplify the model input variables. In addition, subtractive clustering is used to reduce the relationship between input and output variables, reducing the complexity of the model. The MI index as the selection criteria for the ANFIS typhoon intensity prediction model is proposed to determine the optimal model.
The simulated results show that the model test performance is better by using the MI indicator. The overall performance of the ANFIS typhoon intensity prediction model is better than the MLR typhoon intensity baseline model. The best model for all forecasting time is to select the input factors by SRP, and the performance is better by adding marine environmental factors. In addition, the typhoon data are classified according to the Joint Typhoon Warning Center (JTWC) typhoon intensity grading standard. In the super-typhoon intensity standard, ANFIS is significantly better than MLR, and the input factors using PCA is the best in the super typhoon level.

並列關鍵字

typhoon intensity forecast ； adaptive neuro-fuzzy inference system ； stepwise regression procedure ； principal components analysis ； subtractive clustering

參考文獻

1.Alvisi, S., Franchini, M., 2011. Fuzzy neural networks for water level and discharge forecasting with uncertainty. Environmental Modelling & Software, 26(4): 523-537.

2.Chang, F.-J. et al., 2014a. Watershed rainfall forecasting using neuro-fuzzy networks with the assimilation of multi-sensor information. Journal of Hydrology, 508: 374-384.

3.Chang, F.-J. et al., 2014b. Assessment of arsenic concentration in stream water using neuro fuzzy networks with factor analysis. Science of the Total Environment, 494: 202-210.

4.Chang, F.-J., Lai, H.-C., 2014. Adaptive neuro-fuzzy inference system for the prediction of monthly shoreline changes in northeastern Taiwan. Ocean Engineering, 84: 145-156.

5.Chang, L.C., Chang, F.J., 2001. Intelligent control for modelling of real‐time reservoir operation. Hydrological processes, 15(9): 1621-1634.