剪力強度為混凝土的重要力學性質之一，故在各種建築與橋梁設計規範中均將其納入考量。然而，混凝土於剪力作用下的非線性行為相當複雜，其數理模式不易建立。有鑑於現今實驗資料蒐集的便利及資料分析技術的改善，探討容易使用且準確的混凝土剪力強度預測方法將是一件有意義的事。本研究首先蒐集承受剪力作用之鋼筋混凝土 (Reinforced Concrete，簡稱RC) 深梁之剪力強度資料，以免除繁複的試驗工作；其次，建構預測RC 深梁剪力強度之多層倒傳遞類神經網路 (Multilayer Perceptrons Networks，簡稱MLP)，以分析其極限剪力強度，並將所建構MLP 評估模式之預測值與現有RC 深梁剪力分析模式之預測值作比較。研究結果顯示，應用類神經網路可有效預測RC 深梁之剪力強度，且其預測值的準確度優於既有之解析公式。

Shear strength is one of the major concrete mechanical properties that are indispensably used in different building and bridge design codes. However, the nonlinear behavior of concrete under shear is very complicated; modeling its behavior is a hard task. Thus, it would
be of interest to develop new methods that are easier, convenient, and
accurate than the existing methods in light of the availability of more
experimental data and recent advance in the area of data analysis
techniques. In this study, a database on shear failure of reinforced
concrete deep beams with rectangular section subjected to shear force
was retrieved from the existing literature for analysis instead of the
practical and experimental work. Multilayer perceptrons networks
(MLP) were developed sequentially and the ultimate shear strength of
each beam was determined from the MLP model. Besides, the MLP
model’s predictions were also compared with those obtained using
empirical equations. It was found that the MLP models could infer
solutions from the data presented to them, capturing quite subtle
relationships. In other words, the MLP models give reasonable
predictions of the ultimate shear strength of RC deep beams. The
results also show that the MLP models provide better accuracy than the
existing parametric models.