由於台灣經濟快速發展，都會區人口密集，為了有效利用有限的都市空間，超高層建築因應而生，相對基礎結構也愈挖愈深，因此如何防止基礎工程因開挖所造成之鄰房損害或地層下陷等施工災害發生，已為重要研究課題。
目前都會區高層建築深開挖工程之擋土壁選擇，在考量地質條件、鄰近建物損害、地下水位及深開挖的條件下，以連續壁最常為業主及設計者採用，連續壁具有低噪音、低震動、剛性佳、止水性佳、施工快捷、水中施工、且可作為地下結構物外牆壁使用，被廣泛使用在深開挖工程之擋土壁體。
因此本研究藉由文獻回顧探討，初步確認連續壁施工品質影響因子，再應用3個統計關聯性分析方法篩選出最後影響因子，據此蒐集歷史案例，建立案例資料庫，然後應用所發展之演化式深度機器學習Symbiotic Bidirectional Gated Recurrent Unit (SBiGRU)進行案例學習訓練，進而建立「連續壁施工品質預測模式」，本模式可輔助監造及施工單位，提早於開挖前預測連續壁施工品質是否良好，若預測結果為品質不佳，則應於開挖前或開挖中提前採取適當的處置措施，以有效預防災害發生及減少人命財產損失。
本研究利用「SBiGRU」進行案例訓練與測試後，以誤差衡量指標之結果顯示「SBiGRU」優於其他預測模式，擁有最佳的預測能力，後續將以「SBiGRU」做為連續壁施工品質預測模式。
本研究應用「SBiGRU」於實例之兩處案例工地，預測連續壁施工品質與實際值比較，並利用平均絕對百分比誤差(MAPE)、平均絕對誤差(MAE)、均方根誤差(RMSE)及相關係數(R)做誤差衡量指標，研究結果顯示MAPE均低於0.68%、MAE均低於0.1、RMSE均低於0.109、R均高於0.997，皆屬於高精確性及高度相關。
分析其原因為「SBiGRU」可針對時間序列因子與非時間序列因子分別訓練並對序列因子，作自動權重調整，可增加預測準確度。「SBiGRU」結合SOS2.0與BiGRU兩者，BiGRU透過前進與後退雙向GRU之計算以改進加強GRU之學習能力，然後應用SOS2.0搜尋BiGRU之最佳參數，進而提升模式之預測準確度。
因此「SBiGRU」應用於預測連續壁施工品質有使用簡便性、高精確準度及可靠度高等優點。本預測模式可輔助監造及施工單位，提早於開挖前預測連續壁施工品質是否良好，若預測結果為品質不佳，則應於開挖前或開挖中提前採取適當的處置措施，可避免道路塌陷、鄰近建物受損及傾斜、地下管線受損甚至危及人員性命等嚴重災害，做有效的風險管控。

Due to the rapid economic development in Taiwan and the dense population in the metropolitan area, in order to effectively utilize the limited urban space, super high-rise buildings have been built, and the relative infrastructure has been dug deeper. The occurrence of construction disasters such as stratum subsidence has become an important research topic.
At present, in the selection of retaining walls for deep excavation projects of high-rise buildings in metropolitan areas, considering geological conditions, damage to adjacent buildings, groundwater level and deep excavation, diaphragm walls are most commonly used by owners and designers. Noise, low vibration, good rigidity, good water resistance, fast construction, underwater construction, and can be used as the outer wall of underground structures, it is widely used in the retaining wall of deep excavation projects.
Therefore, through literature review and discussion, this study initially confirmed the influencing factors of diaphragm wall construction quality, and then used three statistical correlation analysis methods to screen out the final influencing factors, collected historical cases, established a case database, and then applied the developed evolution. The deep machine learning Symbiotic Bidirectional Gated Recurrent Unit (SBiGRU) is used to conduct case study training, and then establish a " diaphragm wall construction quality prediction model". This model can assist the supervision and construction units to predict whether the diaphragm wall construction quality is good before excavation. If the predicted result is poor quality, appropriate disposal measures should be taken before or during excavation to effectively prevent disasters and reduce loss of life and property.
This study uses "SBiGRU" for case training and testing. The results show that "SBiGRU" has better error metrics than other prediction models and has the best prediction ability. In the future, "SBiGRU" will be used as the diaphragm wall construction quality prediction model.
In this study, "SBiGRU" was applied to the two case sites of the example to predict the construction quality of the continuous wall compared with the actual value, and used the mean absolute percentage error (MAPE), mean absolute error (MAE), root mean square error (RMSE) and correlation The coefficient (R) is used as an error measure. The results show that MAPE is lower than 0.68%, MAE is lower than 0.1, RMSE is lower than 0.109, and R is higher than 0.997, all of which belong to high accuracy and high correlation.
The reason for this is that "SBiGRU" can separately train time-series factors and non-time-series factors and perform automatic weight adjustment for sequence factors, which can increase the prediction accuracy. "SBiGRU" combines both SOS2.0 and BiGRU. BiGRU improves and strengthens the learning ability of GRU through the calculation of forward and backward bidirectional GRU, and then applies SOS2.0 to search for the best parameters of BiGRU, thereby improving the prediction accuracy of the model.
Therefore, the application of "SBiGRU" to predict the construction quality of diaphragm walls has the advantages of ease of use, high accuracy and reliability. This prediction model can assist the construction supervision and construction units to predict whether the construction quality of the diaphragm wall is good before excavation. Avoid serious disasters such as road collapse, damage and inclination of adjacent buildings, damage to underground pipelines and even endangering people's lives, and do effective risk management and control.

[1] 倪至寬，2017，連續壁工程實務，臺北市：詹氏書局。
[2] 交通部鐵道局，C213標林森站路段地下化工程，資料引自https://www.rb.gov.tw/web/p_worksinfo.php?lmenuid=52&smenuid=301&tmenuid=0&infoid=677。
[3] 大陸工程，地下連續壁工法施工技術，資料引自https://www.continental-engineering.com/TechnicalDetail.php?id=157&start=5。
[4] 竣盛工程有限公司，連續壁開挖成果，資料引自https://www.asianmaterials.net/company/139190/project-detail-69348.html。
[5] 富國技術工程股份有限公司，2020，連續壁包砂之問題
[6] 臺北市政府捷運工程局中區工程處，2006，CL700B區段標工程CL802標徐匯中學站0908意外事故處理報告。
[7] 自由時報，長興路塌陷一旁7樓建物傾斜，資料引自https://news.ltn.com.tw/news/local/paper/597408。
[8] 民視新聞，施工連續壁破損?永和路面塌陷原因曝光，資料引自https://www.ftvnews.com.tw/news/detail/2020711W0071。
[9] 中央通訊社，板橋建築掏空地基搶修安全無虞才能復工，資料引自https://www.cna.com.tw/news/asoc/202008190373.aspx。
[10] 自由時報，彰基醫療工程報工安意外!鄰驚:1個月4次塌陷、工人跌落，資料引自坑洞https://news.ltn.com.tw/news/life/breakingnews/3283207。
[11] 新新聞，高市光復三路道路坍塌建案工地連續壁坍孔惹禍，資料引自https://www.storm.mg/localarticle/2551985。
[12] 自由時報，高雄建案「鄰損」害25房倒塌建商獲不起訴處分，資料引自https://news.ltn.com.tw/news/society/breakingnews/1714296。
[13] 陳澤修，2019，深開挖基礎工法及觀測系統介紹，中華民國全國建築師公會。
[14] 郭晉榮、呂芳熾、盧怡志，地下連續壁施工實務，財團法人地工技術研究發展基金會，資料引自2020https://docsplayer.com/109898414-%E5%9C%B0%E4%B8%8B%E9%80%A3%E7%BA%8C%E5%A3%81%E6%96%BD%E5%B7%A5%E5%AF%A6%E5%8B%99.html 。
[15] 倪志寬、邱仕礽，2006，「連續壁工程品質管制與研判分析機制」，現代營建，321期。
[16] 蘇國隆，2006，「連續壁工程施工經驗與台灣工銀總部扇圓形連續壁案例」，地工技術，107期，第53-64頁。
[17] 蕭江碧，2005，建築工程地下連續壁施工技術參考規範與解說，內政部建築研究所。
[18] 余烈，2004，「高雄捷運連續壁漏水之災害原因探討與防範對策」，資料引自http://www.twce.org.tw/info/%E6%8A%80%E5%B8%AB%E5%A0%B1/403-3-1.htm
[19] 李特、林四川、宋明生，1999，「連續壁施工技術及災害搶救對策探討」由三重市「中正百分百」談起，資料引自http://www.twce.org.tw/info/%E6%8A%80%E5%B8%AB%E5%A0%B1/131-2-1.html
[20] 宋隆斌，1999，地下連續壁完整性評估，國立臺北科技大學，碩士論文。
[21] 歐晉德，1984，地下連續壁-施工品質控制及其規範，資料引自http://www.maa.com.tw/common/publications/1981/1981-022.pdf。
[22] Clough, G.W., and O’Rourke , T.D., “Construction–induced Movements of Insitu Walls,” Desing and Performance of Earth Retaining, ASCE Special Conference, Ithaca, New York,(1990).
[23] Moh, Z. C. and Hwang,R.N.(1993),”Earth pressures on walls of a deep excavation”,Proc.,3rdInt.Conf.on Case Histories in Geotechnical Engineering,St. Louis,Missouri,June 1-6.
[24] John Chien-Chung LI; Kung-Tai CHOU; Kweishr LI; Nientzu YU “A Case History of Diaphragm Wall Trench Failure Counter Measures Applied in Taipei Loose Ground＂, 2011.05.22, Hong Kong.
[25] K. Cho, B.V. Merrienboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk,Y. Bengio, Learning Phrase Representations Using RNN Encoder-Decoder for Statistical Machine Translation, arXiv preprint arXiv:1406.1078 (2014).
[26] Natural Language Computing Group (2017). R-NET: Machine Reading Comprehension with Self-matching Networks. https://www.microsoft.com/en-us/research/wp-content/uploads/2017/05/r-net.pdf
[27] Schuster, M., & Paliwal, K. K. (1997). Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 45(11), 2673-2681.
[28] G.G. Tejani, V.J. Savsani, Patel V. K., 2016, Adaptive symbiotic organisms search (SOS) algorithm for structural design optimization. Journal of Computational Design and Engineering.
[29] M.-Y. Cheng, D. Prayogo, D.-H. Tran, 2015, Optimizing Multiple-Resources Leveling in Multiple Projects Using Discrete Symbiotic Organisms Search. Journal of Computing in Civil Engineering: 04015036.
[30] D.-H. Tran, M.-Y. Cheng, D. Prayogo, 2016, A novel Multiple Objective Symbiotic Organisms Search (MOSOS) for time–cost–labor utilization tradeoff problem. Knowledge-Based Systems, 94: 132-145.
[31] M.-Y. Cheng, C.-K. Chiu, Y.-F. Chiu, Y.-W. Wu, Z.-L. Syu, D. Prayogo, C.-H. Lin, 2014, SOS optimization model for bridge life cycle risk evaluation and maintenance strategies. Journal of the Chinese Institute of Civil and Hydraulic Engineering, 26(4): 293-308.
[32] S. Duman, 2016, Symbiotic organisms search algorithm for optimal power flow problem based on valve-point effect and prohibited zones. Neural Computing and Applications: 1-15.
[33] H. Kamankesh, V.G. Agelidis, A. Kavousi-Fard, 2016, Optimal scheduling of renewable micro-grids considering plug-in hybrid electric vehicle charging demand. Energy, 100: 285-297.
[34] E. Ruskartina, V.F. Yu, B. Santosa, A.A.N.P. Redi, 2015, Symbiotic Organism Search (SOS) for Solving the Capacitated Vehicle Routing Problem. International Journal of Mechanical, Aerospace, Industrial, Mechatronic and Manufacturing Engineering, 101: 857-861.
[35] A. Panda, S. Pani, 2016, A Symbiotic Organisms Search algorithm with adaptive penalty function to solve multi-objective constrained optimization problems. Applied Soft Computing, 46: 344-360.
[36] 廖仕元，2018，鄰近施工引致捷運潛盾隧道徑向變位預測模式之研究，國立臺灣科技大學，碩士論文。
[37] 王南傑，2017，自調適潛盾隧道沉陷量預測模式之建立-以臺北市捷運松山線為例，碩士論文。
[38] Lewis, C. D., 1982, International and Business Forecasting Methods. London: Butterwo
[39] 林明傑、董子毅，危險評估中ROC曲線在預測2x2表上與敏感度及特異度之關係，亞洲家庭暴力與性侵害期刊，第四卷第二期，第64-74頁，2008。
[40] 葉宗諺，利用ROC曲線評估色差量尺，大同大學，光電工程研究所，碩士論文，2014。
[41] Wikipedia-the free encyclopedia, Receiver operating characteristic(ROC curve),internet, https://en. Wikipedia.org/wiki/Receiver_operating_characteristic.。
[42] Fawcett, Tom;An introduction to ROC analysis,Pattern Recognition Letters,27,861-874.,2006。