近年來，中國大陸各大城市廣泛進行捷運系統的建設，出現了大量的深開挖工程。深開挖變形一旦超出設計允許值，就可能造成巨大的經濟損失和人員傷亡，因而對深開挖變形進行合理之預測成為了一項重要的議題。傳統的深開挖變形預測方法在預測精度或預測效率等方面存在一定不足，需要尋求一種新的方法。Neural Networks Long Short-Term Memory(NNLSTM)[1](鄭明淵，張于漢於2017年所提出之方法)結合傳統神經網路及長短期記憶網路之特點，可針對序列性因子及非序列性因子，分別輸入到兩個網路中進行訓練學習，再結合兩者學習權重輸出結果。同時在實際應用中，其具有動態調整權重之功能，可針對全新案例適當自動調整權重，以得到較佳預測結果。
本研究之推論模式針對深開挖中最為明顯且突出之兩種變形形式壁體變位及地表沉陷，分別建立兩者之預測模式，對兩種變形形式進行數值預測，可作為預警參數，經由準確的預測結果，施工管理者可以針對變形採取適宜的判斷，決定是否要採取加固基礎或者加快開挖速度，減少開挖時間等措施設法避免災害之產生。
本研究透過案例學習發展建立深開挖變形推論模式，首先整理相關文獻，並利用影響圖和因子篩選表來彙整深開挖變形影響因子，再結合SPSS進行第二輪因子篩選，驗證篩選因子的顯著相關性，確立輸入變數並建立案例資料庫，作為深開挖變形推論模式之基礎。接著應用NNLSTM為推論模式之核心，學習案例機制，分別找出每階層輸入變數與壁體變位及地表沉陷之間的映射關係。藉此作為深開挖過程中變形管控的參考依據，以達到提前預警的的目的。
比較本研究預測之變形最大值與實際變形最大值，其預測結果為高精確、高相關且變異性較小而穩定，且本研究之預測變形趨勢曲線與實際變形趨勢相當吻合。同時經由結果比較後，證實可提供較其他人工智慧更佳的預測準確率，以協助管理者進行變形之管控。

Recently, many cities in China Mainland have been building MRT (Mass Rapid Transportation) system, and occur a lot of deep excavation constructions. Once deep excavation deformation exceeds the design value, it may cause huge economic losses and casualties. Therefore, a reasonable prediction of the deep excavation deformation become an important issue. The traditional predication ways lack of accuracy and efficiency, and the author need to find a new way. NNLSTM combine traditional Neural Networks with Long Short-Term Memory, which combine sequential factors and non-sequential factors，input into two networks to train, and output the result which adjust weights. Besides, in real application, it can adjust the weight dynamically according to new cases, and obtain a better prediction result.
This theory focus on the most obvious and prominent two deformation forms in deep excavation constructions-diaphragm wall deflection and ground movement. Then against two models, predicting these values. And these prediction value can be warning parameters of deep excavation constructions. The manager can use these values to decide reinforcement or not.
The research using case study to build deep excavation construction models. Firstly, arrange some relative literature reviews, using influence diagrams and factor screening tables to organize deep excavation factors, and using SPSS to choose the screening factors, to evaluate significance of screening factors, to evaluate significance of screening factors, establish output variables and build case database, as a fundamental of deep excavation deformation inference models. Next apply Neural Networks Long Short-Term Memory (NNLSTM) as the core inferential models. Using study cases, to find the mapping relation between input variables of each stratum and wall deformation and surface dubs. As a reference frame of deep excavation deformation control, to get the target of early warning.
Compare the prediction maximum value of transformation with the real maximum value, the result is more precision and relevance, less and stable variability. Besides, the prediction of deformation trend curve has highly consistent with actual deformation tendency. After compare with another AI models, the research can get more accurate prediction figures than other AI, to help administrators do deformation controls.

[1]鄭明淵, 張于漢. 橋樑地震損害預測推論模式之研究[J]. 第 21 屆營建工程與管理學術研討會, 2017.
[2]张旷成, 李继民. 杭州地铁湘湖站 “08.11. 15” 基坑坍塌事故分析[J]. 岩土工程学报, 2010, 32(1): 338-342.
[3]胡邵敏，「深開挖工程鄰產保護設計與施工（一）」，地工技術，第40期，pp.35~50，1992。
[4]Goh A T C, Zhang F, Zhang W, et al. A simple estimation model for 3D braced excavation wall deflection[J]. Computers and Geotechnics, 2017, 83: 106-113.
[5]Likitlersuang S, Surarak C, Wanatowski D, et al. Finite element analysis of a deep excavation: A case study from the Bangkok MRT[J]. Soils and Foundations, 2013, 53(5): 756-773.
[6]Zhang W, Goh A T C, Xuan F. A simple prediction model for wall deflection caused by braced excavation in clays[J]. Computers and Geotechnics, 2015, 63: 67-72.
[7]张运良, 聂子云, 李凤翔, 等. 数值分析在基坑变形预测中的应用[J]. 岩土工程学报, 2012, 34(suppl): 113-119.
[8]赵启嘉. 基於动态递归神经网络及相空间重构理论的深基坑工程变形预测研究[D]. 上海: 同济大学, 2008.
[9]朱旭芬. 基於 ICSP 神经网络与遗传算法深基坑监测信息动态预测研究[D]. 硕士论文, 河海大学, 2005.
[10]曹净, 丁文云, 赵党书, 等. 基於 PSO-LSSVM 模型的基坑变形时间序列预测[J]. 控制工程, 2015, 22(3): 475-480.
[11]Kung G T C, Hsiao E C L, Schuster M, et al. A neural network approach to estimating deflection of diaphragm walls caused by excavation in clays[J]. Computers and Geotechnics, 2007, 34(5): 385-396.
[12]田華勳, 2004，類神經網路應用於曼谷地層深開挖連續壁變形之預測, 碩士論文，國立成功大學土木工程研究所
[13]Jan, J. C., et al. "Neural network forecast model in deep excavation." Journal of Computing in Civil Engineering 16.1 (2002): 59-65.
[14]Marsland A. Laboratory and in-situ measurements of the deformation moduli of London Clay[J]. 1973.
[15]Mana A I, Clough G W. Prediction of movements for braced cuts in clay[J]. Journal of Geotechnical and Geoenvironmental Engineering, 1981, 107(ASCE 16312 Proceeding).
[16]Clough G W, Reed M W. Measured behavior of braced wall in very soft clay[J]. Journal of Geotechnical Engineering, 1984, 110(1): 1-19.
[17]Finno R J, Harahap I S. Finite element analyses of HDR-4 excavation[J]. Journal of geotechnical engineering, 1991, 117(10): 1590-1609.
[18]Ou C Y, Hsieh P G, Chiou D C. Characteristics of ground surface settlement during excavation[J]. Canadian Geotechnical Journal, 1993, 30(5): 758-767.
[19]Hashash, Y. M. A. and A. J. Whittle, "Ground Movement Prediction for Deep Excavations in Soft Clay", Journal of Geotechnical Engineering, ASCE, Vol. 122,No. 6, pp. 474-486 (1996).
[20]劉建航、侯學淵，基坑工程手冊，中國建築工業出版社，1997
[21]倪立峰, 李爱群, 韩晓林, 等. 基坑变形的动态神经网络实时建模预报方法[J]. 振动. 测试与诊断, 2002, 22(3): 217-220.
[22]曹红林, 王靖涛. 用小波神经网络预测深基坑周围地表的沉降量[J]. 土工基础, 2003, 17(4): 58-60.
[23]张伟丽, 陈爱云, 李霞. 灰色系统理论在基坑变形预测中的应用[J]. 莱阳农学院学报, 2003, 20(1): 60-61.
[24]邓跃进, 张正禄. 自适应卡尔曼滤波在变形监测动态数据处理中的应用[J]. 武测科技, 1996 (1): 1-4.
[25]沈琛. 深基坑变形机理与控制技术研究[D]. 安徽理工大学, 2014.
[26]白永学. 软土地铁车站深基坑变形的影响因素及其控制措施[D]. 成都: 西南交通大学土木工程学院, 2006.
[27]刘井学, 陈有亮, 陈剑亮. 某深基坑开挖的三维有限元模拟与分析[J]. 建筑结构, 2007, 37(6): 66-68.
[28]钱秋莹, 张柱, 熊中华. 深基坑变形影响因素的正交分析[J]. 河北工程大学学报: 自然科学版, 2014, 31(1): 20-24.
[29]谢玖琪. 深基坑支护结构变形对周围地表沉降的影响研究 [D]. 南京林业大学, 2012.
[30]Peck R B. Deep excavations and tunneling in soft ground[J]. Proc. 7th Int. Con. SMFE, State of the Art, 1969: 225-290.
[31]Goldberg D T, Jaworski W E, Gordon M D. Lateral support systems and underpinning: construction methods[M]. Federal Highway Administration, Offices of Research & Development, 1976.
[32]O'ROURKE T D. Ground movements caused by braced excavations[J]. Journal of Geotechnical and Geoenvironmental Engineering, 1981, 107(ASCE 16511).
[33]Woo S M, Moh Z C. Geotechnical characteristics of soils in the Taipei basin[C]//Proceedings of the 10th Southeast Asian Geotechnical Conference. 1990, 2: 51-65.
[34]Hsiao E C, Schuster M, Juang C H, et al. Reliability analysis and updating of excavation-induced ground settlement for building serviceability assessment[J]. Journal of Geotechnical and Geoenvironmental Engineering, 2008, 134(10): 1448-1458.
[35]Chua C G, Goh A T C. Estimating wall deflections in deep excavations using Bayesian neural networks[J]. Tunnelling and underground space technology, 2005, 20(4): 400-409.
[36]Sou-Sen L, Hsien-Chuang L. Neural-network-based regression model of ground surface settlement induced by deep excavation[J]. Automation in construction, 2004, 13(3): 279-289.
[37]詹君治，冀樹勇，陳錦清，「類神經網路於深開挖壁體變位之預測」，中興工程，第69期，pp21-38頁，2000。
[38]李文通，深開挖壁體變位預測-應用演化式支持向量機推論模式 (ESIM)[D]. , 2007.
[39]吳建逸, 陳俶季. 應用多層函數連結網路預測深開挖壁體變位之研究[J]. 中國土木水利工程學刊, 2012, 24(1): 1-13.
[40]徐炳伟. 基於神经网络的地下连续墙变形预测[D]. 天津大学, 2005.
[41]黄传胜. 地铁深基坑开挖变形预测方法及工程应用研究[J]. 博士学位论文, 2011.
[42]Palmer J H L V, Kenney T C. Analytical study of a braced excavation in weak clay[J]. Canadian Geotechnical Journal, 1972, 9(2): 145-164.
[43]Clough G W, O'Rourke T D. Construction induced movements of insitu walls[C]//Design and performance of earth retaining structures. ASCE, 1990: 439-470.
[44]李云安, 张鸿昌, 糜崇蓉. 深基坑工程变形控制及其影响因素的有限元分析[J]. 水文地质工程地质, 2001, 28(4): 1-5.
[45]刘戈, 仝国柱. 基於灰色 BP 神经网络组合模型的深基坑周围地表沉降预测研究[J]. 天津城建大学学报, 2016, 22(3): 184-189.
[46]欧章煜. 深开挖工程分析设计理论与实务[J]. 台北: 科技图书股份有限公司, 2004Ou Zhang-yu. Design Theory of Deep Pit Engineering and Its Application, 2004.
[47]董永坡，孔进，杨玉玲，基於主成分分析和径向基神经网络的基坑变形预测 [J].城市建设理论研究：电子版，2013（11）
[48]高文华, 杨林德, 沈蒲生. 软土深基坑支护结构内力与变形时空效应的影响因素分析[J]. 土木工程学报, 2001, 34(5): 90-96.
[49]Ko, C. H., 1999, Computer aided decision support system for disaster prevention of hillside residents, MS thesis, National Taiwan University of Science and Technology.
[50]Fukahori, K. and Kubota, Y., 2000, Consistency evaluation of landscape design by a decision support system, Computer-Aided Civil and Infrastructure Engineering, 15(5), 342-354.
[51]Sundin, S and Braban-Ledoux, C., 2001, Artificial intelligence-based decision support technologies in pavement management, Computer-Aided Civil and Infrastructure Engineering, 16(2), 143-157
[52]M.-Y. Cheng, Y.-W. Wu, 2009, Evolutionary support vector machine inference system for construction management, Automation in Construction, 18(5): 597-604.
[53]M.-Y. Cheng, P.M. Firdausi, D. Prayogo, 2014, High-performance concrete compressive strength prediction using Genetic Weighted Pyramid Operation Tree (GWPOT). Engineering Applications of Artificial Intelligence, 29: 104-113.
[54]M.-Y. Cheng, D. Prayogo, Y.-W. Wu, 2014, Novel Genetic Algorithm-Based Evolutionary Support Vector Machine for Optimizing High-Performance Concrete Mixture. Journal of Computing in Civil Engineering, 28(4): 06014003.
[55]M.-Y. Cheng, D.K. Wibowo, D. Prayogo, A.F.V. Roy, 2015, Predicting productivity loss caused by change orders using the evolutionary fuzzy support vector machine inference model. Journal of Civil Engineering and Management, 21(7): 881-892.
[56]M.-Y. Cheng, N.-D. Hoang, Y.-W. Wu, 2013, Hybrid intelligence approach based on LS-SVM and Differential Evolution for construction cost index estimation: A Taiwan case study. Automation in Construction, 35: 306-313.
[57]M.-Y. Cheng, A.F.V. Roy, 2011, Evolutionary fuzzy decision model for cash flow prediction using time-dependent support vector machines. International Journal of Project Management, 29(1): 56-65.
[58]M.-Y. Cheng, M.-T. Cao, 2014, Evolutionary multivariate adaptive regression splines for estimating shear strength in reinforced-concrete deep beams. Engineering Applications of Artificial Intelligence, 28: 86-96.
[59]D. Prayogo, 2012, A Novel Genetic Algorithm-Based Evolutionary Support Vector Machine for Optimizing High-Performance Concrete, Master Thesis, Department of Construction Engineering, National Taiwan University of Science and Technology.
[60]D.E. Goldberg, K. Deb, H. Kaegupta, and G. Harik, 1993, Rapid, accurate optimization of difficult problems using fast messy genetic algorithms.
[61]Suykens, J., et al., 2002, Least Square Support Vector Machines. World Scientific Publishing Co. Pte. Ltd.
[62]Min-Yuan Cheng and Nhat-Duc Hoang, 2012, Evolutionary Least Squares Support Vector Machine – Userguide, Technical Report, CIC Lab, National Taiwan Univ. of Sci. and Tech.
[63]Sepp Hochreiter and Jurgen Schmidhuber(1997), “Long short-term memory. Neural Computation,” 9(8):1735-1780.
[64]Felix A. Gers, Jurgen Schmidhuber, Fred Cummins.(2000), “Learning to forget:Continual prediction with LSTM. Neural computation”, 12(10):2451-2471
[65]Ilya Sutskever, Oriol Vinyals, and Quoc V. Le.(2014), “Sequence to sequence learningwith neural networks.”, In Advances in Neural Information Processing Systems,pages 3104-3112
[66]林宗元. 岩土工程治理手册[J]. 1993.
[67]张戈, 毛海和. 软土地区深基坑围护结构综合刚度研究[J]. 岩土力学, 2016, 37(5): 4.
[68]Lewis, C. D., 1982, International and Business Forecasting Methods. London: Butterwo