本研究以相關分析檢定臺灣貨櫃吞吐量的影響變數，透過灰色理論來預測變數未來值，並且運用類神經網路進行臺灣海運貨櫃吞吐量之預測。研究結果顯示在未考慮時間落後原始變數之最佳模式預測誤差為 3.25%，至於有考慮時間落後變數之誤差降為 1.54%，加入時間落後期數之變數的確能明顯提升預測能力之效果，預測精準度高達 97% 以上。至於本模式的預測結果，臺灣海運貨櫃吞吐量從 2011年第四季的 3,367,765 TEUs 到 2014 年第四季的 3,427,264 TEUs 微幅波動，呈現穩定成長的趨勢。

The study applies the correlation analysis to derive the variables infuencing container throughput in Taiwan. The gray theory is adopted to forecast the future values of the influential variables and the artificial neural network is adopted to predict the container throughput in Taiwan. The results reveal that the error of the best forecasting model could signifcantly drops from 3.25% to 1.54% with more than 97% of the precise prediction rate whenever considering the lagged dependent variables. The predicted container throughput in Taiwan varied from 3,367,765 TEUs in the 4th quarter of 2011 to 3,427,264 TEUs in the 4th quarter of 2014 and showed a stably growing trend.