安全風險評估一直是航空業者必須面臨的重大課題,如何有效預防失事及保持警惕,不只影響業者的日常營運,也會提高人民的安全保障。飛安事故主要的原因,一般可以歸於人為因素,因此本研究利用過去民航局針對兩家國內業者之安全檢查資料作為分析依據,找出資料內存在的人為疏失,再利用反傳遞模糊類神經網路(Counterpropagation Fuzzy Neural Network, CFNN)進行未來飛安意外事件之預測。研究中使用的安全檢查資料,業經HFACS-MA進行人為疏失分類,包括組織影響(Organizational Influences),不安全管理行為(Unsafe Supervision),不安全行為影響因素(Preconditions For Unsafe Acts)以及不安全行為(Unsafe Acts)四大類,再量化成每個月的人為疏失率,並用來預測下一個月的意外發生事件率,藉此證明人為疏失與意外事件之間存在著因果關係。結果顯示使用CFNN所得預測結果,最佳模式配置下,訓練與測試樣本的R2值分別達到0.98與0.92,與相關文獻進行比較之後,發現CFNN相較BPNN能提供更佳之預測能力。之後再針對各項因素進行敏感度分析,發現其中對未來意外事件率影響較大之因子為當月意外事件率、前一個月的人為疏失率、不安全管理行為以及不安全行為影響因素,證明了人為疏失對於未來意外事件的影響。本研究證明了各項人為因素對於航空安全危害的因果關係,未來可以藉此建立飛安預警機制,告知航空公司在何時應提高警覺,加強內部的安全管理,以免造成無法挽救的災難發生。
Accurate risk assessment of flight operation is always an essential topic to the aviation industry. Preventing accidents from happening is important to both the airliners and the public. Among the root causes of flight accidents, human factors is the most significant one. In this study, we utilized the quantitative data transformed via HFACS-MA form the historical safety inspection records of two domestic airlines from 2002-2008. The data was inputted to investigate the causal relationship between human factors and flight incident rates as the foundation of risk assessment methods. Counterpropagation Fuzzy Neural Network (CFNN) was adopted to develop the prediction model. The results found that CFNN method was better than Backpropagation (BP). The results of sensitivity analysis found that the incident rate of the current month and previous month were vital predictors for safety performance. Among the four kinds of human factors, unsafe supervision and precondition for unsafe acts had more impacts to the safety performance than unsafe acts and organizational influences. The contributions of this study are to support the causality of human factors and safety performance (incident rates), and to identify the impact of the human factors to flight safety. The CFNN method could be considered as one of the various quantitative tools while establishing risk assessment models for airlines under different context.