论文部分内容阅读
针对飞机飞行性能模型的非线性、动态性的特点,分析了影响飞行性能的关键要素。飞行员的操纵量,提出了基于神经网络的数据建模方法。该方法在BP神经网络结构的基础上,加入了外部输入量的延迟和输出量的反馈连接,建立了NARX神经网络预测模型。该模型利用飞行模拟器采集的飞行数据训练网络,并对训练好的网络进行验证和评估。实验结果表明,与BP神经网络以及引入动量因子和自适应调整学习率的改进BP神经网络相比,NARX神经网络预测模型收敛速度和预测结果更好,可以长期准确地预测飞行性能模型。
Aiming at the non-linear and dynamic characteristics of aircraft flight performance model, the key elements affecting the flight performance are analyzed. Pilot manipulation, proposed data modeling method based on neural network. Based on the structure of BP neural network, this method adds feedback connection between delay and output of external input, and establishes NARX neural network prediction model. The model uses the flight data collected by the flight simulator to train the network and validates and evaluates the trained network. Experimental results show that, compared with BP neural network and improved BP neural network with momentum factor and adaptive learning rate adjustment, NARX neural network prediction model has better convergence speed and prediction results, and can predict the flight performance model in a long-term and accurate manner.