加勁擋土結構物的設計上必須滿足各項內部、外部與整體的穩定性分析，其中，在內部穩定性分析上，以如何評估加勁材的設計強度與張力分佈狀況為主要關鍵。現今已發展出許多預測方式用來評估加勁材受力後的張力發展。但經過去文獻指出，傳統所使用的側向土壓法因未考量牆面的影響，因此普遍呈現高估的趨勢；而K勁度法在大載重下明顯低估加勁材張力值；有限元素法為較能準確預測加勁材實際的張力發展，但在大土壤應變下會有計算收斂的問題。
有鑑於此，本研究擬以人工智慧方式，透過演化式高斯過程推論模式(Evolutionary Gaussian Process Inference Model-EGPIM)發展建立一套加勁材張力預測模式，利用模式內高斯過程(Gaussian Process-GP)釐清資料中輸入及輸出值間的映射關係，並利用貝氏推論，結合粒子群演算法(particle swarm optimization ,PSO)優化GP內共變異函數的超參數，以獲得最佳的加勁材張力預測能力。
本研究首先整理相關文獻及加勁擋土牆實際案例資料，接著應用EGPIM，透過學習過去相似案例的加勁材張力影響因子，找出加勁材張力影響因子與張力之映射關係，進而計算出加勁材受力後張力預測值。再藉此預測結果與實際量測值進行比較，以瞭解預測結果的準確性與實際值之差異及發展趨勢。
最後本研究藉由模式實際案例應用，與經驗預測方式進行綜合性評估
，並根據評估結果探討可能造成誤差的原因，根據EGPIM與經驗公式之預測值與實際值進行誤差比較，結果驗證所建立的預測模式在預測加勁材張力上有良好的表現，可提供方便且準確的預測方式。並可協助分析者改善以往經驗預測方式呈現高估及低估加勁材張力的問題，提供穩定的加勁材張力發展趨勢。

Geosynthetic-reinforced soil structure must be designed to meet the various internal, external and overall stability of the analysis, in which the primary key of internal stability analysis is to assess the design strength and tension distribution of reinforcement. Today many prediction methods have been developed to assess tension developmentof reinforcementafter the force. However, the previous literature indicated that the traditional earth pressure method appears overestimated trend without considering the impact of the wall; While the K-stiffness method significantly underestimated the tensionofthe reinforcement under the big loading; The finite element method has better prediction to the actual tension development of the reinforced material, but the strain calculation has convergence under large soil strain problem.
For this reason, this study intends to utilize artificial intelligence method, through the development of case studies to establish an evolutionary mode (Evolutionary Gaussian Process Inference Model-EGPIM), utilizingthe Gaussian process-GP to clarify the data input and output value mapping relationship and use Bayesian inference, combined with particle swarm optimization (PSO),to optimize GP covariance function within the hyper-parameters, in order to get the best predictive ability.
First, the study compiles the reinforced retaining wall relevant literature and actual case data and then apply EGPIM, learning through the previousre inforced similar cases of tension affecting factors to identify the mappings relationship between it and the tension, and then calculate the tension force predicted value of reinforced material. Comparing predicted results and actual measured values to understand the result’s accuracy and the differences and trendsof the actual value.
Finally, this study apply actual case with GP mode, and comprehensively estimate the experienced prediction methods, and based on the assessment results to explore possible reason for error, according to the empirical formula and EGPIM predicted value and the actual value to conduct the error, the results of predicting reinforced material tension demonstrate that the developed forecasting mode has a good performance, can provide convenient and accurate prediction method. It can assist in the analysis to improve on past forecast experience over estimate and under estimate the tension of reinforced material, and provide a stable development trend of the reinforced material’s tension.

[1]黃勝郎，「演化式高斯過程推論模式於專案現金流量之應用」，碩士論文，國立台灣科技大學營建工程系，台北，2011。
[2]劉玳伶，「柔性與剛性牆面加勁擋土牆加勁材張力發展預測方法之評估」，碩士論文，國立台灣科技大學營建工程系，台北，2012。
[3]吳明峰，「營造工程專案現金流量之模擬分析」，碩士論文，國立台灣科技大學營建工程系，2005。
[4]楊國鑫、劉家男、周南山,(2012), “加勁擋土牆加勁材張力發展預測方法之評估”,地工技術, 第134期, 15-24.
[5]陳景文、吳宗欣、Claybourn, A.F. (1993),“林三賢,(1999),“地工合成材五種設計規範之比較”,地工技術, 第71期, 43-46.
[6]周南山,(1993),“地工合成物加勁牆分析設計之探討與評估”,地工技術, 第43期, 32-42.
[7]范嘉程與周南山,(1998),“加勁擋土結構物設計及施工規範準則之探討”,地工技術, 第70期, 93-106.
[8]林三賢,(1999),“地工合成材五種設計規範之比較”,地工技術, 第71期, 43-46.
[9]台北市土木技師公會(2003)，“結合生態與景觀之加勁擋土結構設計及施工規範”，詹氏出版社。
[10]Allen, T.M.; Bathurst, R.J. (2003)“Prediction of Reinforcement Loads in Reinforced Soil Walls”, Prepared for Washington State Department of Transportation and in cooperation with US Department of Transportation Federal Highway Administration。
[11]Bathurst, R.J., Miyata, Y., Nernheim, A. and Allen, T.M. (2008)“Refinement of K-Stiffness Method for Geosynthetic ReinforcedSoil Walls”, Geosynthetics International, v15, n4, p 269–295。
[12]RASMUSSEN C.E. and WILLIAMS C.K.I.．Gaussian Processes for Machine Learning[M]．Cambridge：the MIT Press，2006。
[13]M. N. Gibbs.“Bayesian Gaussian Processes for Regression and Classification”. PhD thesis, Department of Physics, University of Cambridge, 1997。
[14]Kennedy, J. & Eberhart, R.，”Particle swarm optimization”，Proceedings of the 1995 IEEE International Conference on Neural Networks, 1942-1948，1955。
[15]楊亦東（I-Tung Yang），「Computer-aided Decision and Simulation Analysis—Course Handout」，國立台灣科技大學營建所課程講義，2007。
[16]Huang, C. C., Cheng, M. Y., 2011, “Estimate at completion for construction projects using evolutionary gaussian process inference model,” Proceeding of The International Workshop on Civil Engineering and Urban Planning (WCEUP 2011), Hangzhou, China.
[17]Huang, C. C.,“Evolutionary Gaussian Process Model (EGPIM) for Decision-Making in Construction Engineering and Management”, Ph.D, Department of Constructuon Engineering,National Taiwan University of Science and Technology,2013.
[18]Elias, V., Christopher, B.R., and Berg, R.R., (2001), “Mechanically Stabilized Earth Walls and Reinforced Soil Slopes Design andConstruction Guidelines” Report No. FHWA-NHI-00-043, NationalHighway Institute, Federal Highway Administration, Washington,D.C. March
[19]Bathurst, R. J., Allen, T. M. & Walters, D. L.(2005).“Reinforcement Loads in Geosynthetic Walls and the Case for a New Working StressDesign Method”.Geotextiles and Geomembranes, 23, No. 4,287–322.
[20]AASHTO, (2002), “Standard Specifications for Highway Bridges”, American Association of State Highway and TransportationOfficials, 17th Edition, Washington, DC, USA, 689p
[21]Bathurst, R.J. (2007) “Geosynthetics Classification”. IGS Leaflets on Geosynthetics Applications, IGS Education Committee, available at at www.geosyntheticssociety.org
[22]National Concrete Masonry Association, (2010), “Design Manual for Segmental Retaining Walls”, Bathurst, R.J., Editor, 3rd Edition, Herndon, Virginia, USA, 282p.