Prediction Model of Flight Operation Risk Based on Fuzzy Inference and Back-Propagation Neural Network

This paper proposes a fuzzy inference and back propagation neural network (FI-BPNN) model to predict flight operation risk quantitatively. Initial risk values of human error are quantified by fuzzy inference, and then the quantified data is trained by back propagation neural network. Through adjusting parameters dynamically, the model is used to optimize and correct the sample data. Flight operation sequences are obtained by HTA (Hierarchical Task Analysis) based on SOPs (Standard Operation Procedures), and modified HET (Human Error Template) analysis is performed subsequently. The results of using fuzzy inference only are compared with that of FI-BPNN, which proves that the latter can improve the prediction accuracy significantly. Aircraft landing process is studied to validate the model. The operation risk during landing stage is predicted by using 2-norm of the risk vector to synthesize all human error modes in each operation. Relative risk coefficient is proposed to measure the degree of operation risk compared with the maximum operation risk value. Simulation results suggest that FI-BPNN model has the advantages of accurate quantification, comprehensive integration and intuitive visualization.