結合非線性主成分分析之神經網路降尺度模式評估臺中與花蓮未來降雨

Integrating Nonlinear Principal Component Analysis and Neural Networks to Develop a Downscaling Model Evaluating Future Rainfall of Taichung and Hualien

胡鈞甯(Jun-Ning Hu) ； 林旭信(Shiu-Shin Lin) ； 朱楷洋(Kai-Yang Zhu)

《農業工程學報》 67卷4期 (2021/12) Pp. 78-90

https://doi.org/10.29974/JTAE.202112_67(4).0005

摘要

本研究提出一結合主成份分析（Linear Principal Components Analysis，PCA）、非線性主成分分析（Non-Linear Principal Components Analysis，NLPCA）及輻狀基底函數類神經網路（Radial Basis Function Neural Network，RBFNN）之非線性主成分分析統計降尺度模式（Nonlinear Principal Component for Genetic Algorithm based-Radial Basis Function Neural Network，NGRB）。首先使用PCA及NLPCA分析氣象站與GCM（General Circulation Model）模式資料，其次透過基因演算法（Genetic Algorithm，GA）優化NGRB模式參數，最後再以4種GCM模式的未來情境資料預測臺中與花蓮測站之中期（2050/01～2069/12）及長期（2080/01～2099/12）之降雨趨勢。研究顯示以NLPCA較PCA能解釋原資料之特徵，並且能有效減少NGRB模式運算時間，GA世代數為25、族群數為25時NGRB能合理地收斂；預測2個測站中及長期未來降雨趨勢，花蓮測站長期降雨變異性較臺中測站大，兩測站之長期降雨相比中期降雨有減少之趨勢。臺中測站於測試結果顯示中及長期之豐／枯水期約提前1～2個月，然而未來模擬結果並無明顯趨勢。

關鍵字

非線性主成份分析 ； 輻狀基底函數類神經網路 ； 降雨變異性 ； 氣候變遷

並列摘要

The study combines principal component analysis (PCA), nonlinear principal component analysis (NLPCA), and genetic algorithm (GA) based radial basis function neural network (RBFNN) to develop a statistical downscaling model (NGRB). Firstly, the PCA and NLPCA are used to analyze data from meteorological stations and GCM model. GA is then employed to optimize parameters of NGRB models. Finally, four GCM models outputs from A1B scenario are applied to predict near (2050/01 to 2069/12) and far future (2080/01 to 2099/12) monthly rainfall of Taichung and Hualien stations. The simulated results show that NLPCA can extract features of data better than PCA and can reduce much computation time of NGRB; GA can converge effectively with 25 generations and populations. Predicting near and far future rainfall trends at two stations. Long-term rainfall variability at the Hualien station is greater than that at the Taichung station. The long-term rainfall at the two stations tends to decrease compared to the medium-term rainfall. It reveals that the wet and dry season may happen one to two months ahead of schedule for the near and far future rainfall in Taichung station. The annual average monthly rainfall of Hualien station may small increase for the near future. Rainfall trends for the far future in Hualien station has a gentle bell curve.

並列關鍵字

Nonlinear principal component analysis ； Radial basis neural networks ； Rainfall variability ； Climate change

參考文獻

Canchala, T.,Alfonso-Morales, W.,Cerón, W. L.,Carvajal-Escobar, Y.,Caicedo-Bravo, E.(2020).Teleconnections between Monthly Rainfall Variability and Large-Scale Climate Indices in Southwestern Colombia.Water.12(7),1863-1863.

Chen, S.,Cowan, C. f. n.,Grant, P. m.(1991).Orthogonal least squares learning algorithm for radial basis function networks.IEEE Transactions on Neural Networks.2(2),302-309.

Cheng, C. T.,Zhao, M. Y.,Chau, K. W.,Wu, X. Y.(2006).Using genetic algorithm and TOPSIS for Xinanjiang model calibration with a single procedure.Journal of Hydrology.316(1-4),129-140.

Elanayar V. T., S.,Shin, Y. C.(1994).Radial basis function neural network for approximation and estimation of nonlinear stochastic dynamic systems.IEEE Transactions on Neural Networks.5(4),594-603.

Chng, E. S.,Yang, H. h.,Bos, S.(1996).Orthogonal least-squares learning algorithm with local adaptation process for the radial basis function networks.IEEE Signal Processing Letters.3(8),253-255.