基于BP神经网络的飞行绩效评价模型

为实现飞行绩效的客观评价,利用眼动数据,建立了拓扑结构为6-14-3型的BP(Back Propagation)神经网络模型.已有的实验数据和随机插值法得到的数据作为建模的数据来源,数据分为训练集和测试集并归一化.基于Matlab的神经网络工具箱,利用经验公式和实验比较法确定了BP网络模型的隐含层节点数;对BP算法的各种改进算法进行了优化选择;将训练集数据和测试集数据先后输入网络模型进行学习训练和仿真测试;对3个技术水平的飞行员的飞行绩效进行了预测和评价.研究表明,眼动数据的BP神经网络模型可以较为准确地评价飞行绩效,评价方法可以为飞行训练提供参考.

Pilot performance evaluation model based on BP neural networks

In order to evaluate pilot performance objectively, back propagation(BP) neural network model of 6-14-3 form in topology with eye movement data was established. Data source of BP neural networks that came from former experiment and random interpolation was divided into training set and test set and normalized. Based on neural networks toolbox in Matlab, hidden layer nodes of BP networks were determined with empirical formula and experimental comparision; BP algorithms in the toolbox were optimized; The training set data and test data were input into model for training and simulation; Pilot performance of the three skill levels was predicated and evaluated. The research shows that pilot performance can be accurately evaluated by setting up BP neural networks model with eye movement data and the evaluation method can provide a reference for flight training.