運用倒傳遞類神經網路法預測航空業安全績效

Using the Backpropagation Neural Network to Predict the Flight Safety Performance of Airlines

鄭羽辰(Yu-Chen Cheng)

指導教授 ： 蕭育霖

中原大學/電機資訊學院/工業與系統工程研究所/碩士(2019年)

https://doi.org/10.6840/cycu201900824

摘要

航空運輸業在台灣仍持續發展中，每年飛航架次和載客數皆穩定增長，因此「飛航安全」持續維持更顯重要。歷來飛航事故的發生有50%以上與人為因素相關，舉凡判斷失誤、操作失誤到管理不良都屬於人為疏失的一環。而飛航事故大多是由數個人為疏失環環相扣，最終才會導致嚴重意外的發生。近年來，航空界對於飛安的管理已逐漸由事後改善提升為事前預防，從組織管理層面探討影響安全的根本原因，有效降低人為疏失發生。本研究即企圖透過人為因素指標建立一套飛航安全績效的預測模型。首先運用人為因素分析歸類系統-維修查核（Human Factors Analysis and Classification System-Maintenance Audit, HFACS-MA）系統，將民航業者日常安全查核報告進行人為疏失的量化分類，計算人因指標疏失率，作為預測模型之自變數。使用各業者每月意外事件做為各組織日常安全績效指標（應變數），並使用管制界限與意外事件率上升／下降／持平作為預測分類基準，據以建立預測模型。本研究透過使用倒傳遞類神經網路法，以八種人為疏失種類作為自變數時，安全績效預測模型之分類準確率可達到六成。研究成果驗證了人為疏失與安全績效（意外事件）之間的因果關係，證明可透過人為因素建立飛安領先指標，以提高航空業預測式安全管理能量，進一步落實飛航安全管理，讓民眾享有更安全的航空運輸。

關鍵字

人為疏失 ； 意外事件 ； 飛航安全 ； 倒傳遞類神經網路

並列摘要

Following the fast growth of air transportation industry, the concern of “flight safety” remains crucial at Taiwan. About 50% of aircraft accidents are relevant to human factors. Although human error is inevitable, the management of flight safety has transformed from passive correction to proactive prevention to eliminate the occurrence of human error by exploring the root causes from organization and management perspectives. This study focus on establishing a predictive model of flight safety performance by setting human factors indicators. We used HFACS-MA （Human Factors Analysis and Classification System-Maintenance Audit）to quantify daily safety audit reports into human error categories, then calculated the happening rates of each specific error per month as the independent variables of the prediction model. For the dependent variable, we took the monthly incident records of local airlines per 1,000 departures as the safety performance indicator of operators. The prediction model is used to foresee the change of the incident rate next month where the rate change would be classified by the control upper/lower limits and increase /decrease/retain patterns. Through the utilization of Backpropagation method, the accuracy of the prediction model could reach 60% while using eight human error categories as the independent variables. The results validated the causal relationship between human error and safety performance （incidents）again. We could develop leading human factors indicators to enhance the predictive capability of safety management system in the aviation industry.

並列關鍵字

Human Error ； Incident ； Fight Safety ； Backpropagation

參考文獻

外文部分：

Pacelli, V., & Azzollini, M. （2011）. An Artificial Neural Network Approach for Credit Risk Management. Journal of Intelligent Learning Systems and Applications, 103-112.

2. Hsiao, Y.-L., Druryb, C., Wub, C., & Paquetb, V. （2013）. Predictive models of safety based on audit findings： Part 2： Measurement of model validity. Applied Ergonomics, 44（2）.

3. Shappell，& Wiegmann. （2000）. The Human Factors Analysis and Classification System–HFACS. Technical Report Documentation.

4. Ung, S. T., Williams, V., Bonsall, S., & Wang, J. （2006）. Test case based risk predictions using artificial neural network. 37（3）, 245-260.