當飛安事件發生時,雖然可能牽涉到多種狀況或複雜因素,但造成事故的主要原因,大致可簡單分為人為、機器及其他因素三種;其中,又以人為因素為飛航事故的主要肇因,也是所有因素當中最難以掌控及管理的部分。當代先進的安全管理理念企圖在工作前即預知風險,以便在突如其來的災害中把傷害損失減到最低,據此,建立主動預防式的安全管理系統使飛航事件發生率下降,正是現代航空安全管理的重要課題之一。 本研究將兩間國內航空業者的歷年查核資料透過公式量化後,嘗試建立最佳的分析模式,根據每月的人為疏失率,預測未來飛安事件的發生率。研究中選擇類神經網路法中的倒傳遞(BP)及輻狀基底函數(RBF)法,找出兩種方法的最佳參數配置,並比較兩種方法在最佳參數配置下的預測成效。結果發現在此研究課題中,倒傳遞(BP)比輻狀基底函數(RBF)有更好的預測表現。 此研究成果不僅進一步驗證了人為疏失與飛安事件率之間的因果關係,倒傳遞法所建立的飛航安全預測模式,應足可做為未來航空安全量化管理在大資料(Big Data)時代的建模參考。未來期能逐步提升預測模型對於飛安事件率的預測能力,並進而找出顯著影響飛安事件的因素類型,協助航空業者改善事件預防的能力及提升飛航安全。
With the high-tech industry is constantly promoted, we concern at the safe management whether the capability is also more intact. The security is important and we should be aware of danger during work. We usually do a good safe supervision, when the incident will be happen, we can minimize the injury to loss. It is the most prior attention of the topic in the high-risk industry. In the air transportation, the safety should not be underestimated. When the accident will be occurred, it may involve a variety of conditions or factors. The major reason would be divided into human, machine and other of three factors. Among the human factor is a main cause in the aviation accident, as well as the human factor is not be easy to control at the all factors. How to establish an active preventive security management system enables the accident rate has dropped, it is an advance concept in the safe management. In this study, the audit data of two airlines were quantified. After transforming the audit data, it obtain per month of the human error rate. We utilize human error rate, in order to predict the incident rate in the future and try to establish an appropriate analysis model. We want to find out the potential factors of human error, which the reason will be caused the incident. The study adopted two neural network methods, Back-propagation (BP) and Radial basis function (RBF). We also compare the predictive performance under the different parameters in the two neural network methods and find the appropriate parameter configuration mode in the study. After using the configuration of different parameters, we found that not only the back-propagation (BP) has a good predictive performance than the radial basis function (RBF), but also search the best prediction mode of the back-propagation (BP). The result of the study is validated the causal relationship between the human factor and incident rate. By comparing the two methods, we consider that the back-propagation (BP) can produce appropriate flight safe prediction mode. It also can apply to construct the model with aviation safety quantitative management in the period of big data future. We hope that increase gradually predictive capability of the flight safe incident rate and thus improve flight accidents continue to occur. Finally, we also hope to assist airline industry to think about how to prevent accidents and enhance flight safety of our country.