Civil aviation safety evaluation based on deep belief network and principal component analysis. (February 2019)

• Record Type: Journal Article
• Authors: Ni, Xiaomei; Wang, Huawei; Che, Changchang; Hong, Jiyu; Sun, Zhongdong
• Is Part Of: Safety science. Volume 112(2019)
• Page Start: 90
• Page End: 95
• Publication Date: 2019-02
• Subjects: Unsafe events -- Serious flight incident-rate -- PCA -- DBN -- Prediction; Industrial accidents -- Periodicals; Accident Prevention -- Periodicals; Safety -- Periodicals; Travail -- Accidents -- Périodiques; 363.11
• DOI: 10.1016/j.ssci.2018.10.012
• Languages: English
• ISSNs: 0925-7535

Abstract

Highlights: Forecasting civil aviation safety with unsafe events. Establishing the forecast network by combining Deep Belief Network and Principal Component Analysis. Providing the superiority by comparing the new network with Deep Belief Network, Support Vector Regression and Grey Network. Providing solutions to long-term predictions that take all unsafe incidents into account. Abstract: Civil aviation in modern industries is becoming increasingly automatic, precise, and efficient. Serious accidents and unsafe incidents are used to describe and investigate the safety level. Therefore, effectively predicting big data from these features and accurately identifying the safety level with advanced theories are new issues in civil aviation. The prediction of serious flight incident rate for unsafe events is proposed on the basis of deep learning considering the characteristics of big data. In this method, deep belief network (DBN) is combined with Principal Component Analysis (PCA). The deep architecture is beneficial for safety prediction because each layer learns more complex features than the layers before. Compared with the previous prediction based on historical accident data, The DBN predicts the serious flight incident rate based on the results from PCA. Results indicate that the prediction data of PAC-DBN is consistent with the actual data of serious flight incident rate. The proposed method is superior to forecasting the serious flight incident rate compared with the gray neural network method, support vector regression, DBN. Simultaneously, the main influencing factors can be extracted to reduce flight incident rate.