本研究利用1993年－2006年間我國全產業與營造業兩類職業災害之職災千人率、傷病千人率、殘廢千人率與死亡千人率等資料，應用非線性柏努力灰預測模型(Nonlinear Grey Bernoulli Model, NGBM)作為本研究方法之工具，有關NGBM其中參數γ值的最佳化選取，利用平均均方根誤差(Root Mean Squared Error, RMSE)、平均絕對誤差(Mean Absolute Error, MAE)、平均相對誤差(Mean Relative Error, MRE）等作為檢驗準則依據，最佳γ值的NGBM預測結果並與灰色模型(Grey Model, GM, γ=0)進行比較。結果顯示在全產業職災千人率、傷病千人率、殘廢千人率與死亡千人率，其等NGBM依RMSE誤差為判斷法則之值γ分別出現在0.5、0.5、0.5和0.4，而營造業之γ值分別出現在0.8、0.8、0.9和0.3。由所得γ值之NGBM模型與GM模型在預測準確度之比較，並檢視三項誤差檢驗準則，在全產業與營造業各職災率的表現上，均顯示最少提升了二倍以上的精度改善。

This paper investigates the application of Grey Model (GM) and Nonlinear Grey Bernoulli Model (NGBM) to predict the occupational disaster rate per thousandth of the whole industry and construction industry by using the data from Council of Labor Affairs database during the period of 1993-2006. Three criteria, namely Mean Squared Error (RMSE), Mean Absolute Error (MAE) and Mean Relative Error (MRE) were used to evaluate the reliability (or the optimized γ values) of these two models. By using NGBM, the optimized γ values of occupational disaster rate per thousandth, sick and wounded rate per thousandth and disabled rate per thousandth for the whole industry appeared at 0.5, 0.5, 0.5 and 0.4 respectively; while for the construction sector γ the values appeared at 0.8, 0.8, 0.9 and 0.3 correspondingly. The results also showed that compared with GM, NGBM had at least two-fold improvement in accuracy in occupational disaster prediction accuracy for both the whole industry and construction sector.