RNN在降落伞开伞特性研究中的应用

Application of RNN in the Study of Parachute Opening Characteristics

为了对降落伞充气展开过程中的拉力、速度等关键特性进行预测,结合当下机器学习的研究热点,使用循环神经网络对空投试验数据进行学习和训练。文章对原始数据进行了归一化预处理并且采用批量梯度下降的训练方式;试验验证了循环网络方法对充气过程中重要参数进行计算的可行性;并且说明了时间序数索引项在试验中需带入网络参与计算的特点。试验结果表明,使用循环神经网络对训练集数据的开伞速度曲线拟合效果很好;计算结果能够准确反映开伞过程中的拉力变化趋势,峰值误差小于1%,且峰值时刻误差小于2%。

To predict the essential characteristics such as strain of suspension-line and velocity of descent in the parachute deployment process,this paper uses recurrent neural networks to train the airdrop experiment data through the combination of machine learning and parachute research method.The normalization of original data and batch gradient descent is applied to verify the feasibility of RNN to calculate the important parameters in the inflation process.The research result gives the explanation of the input characteristic about the time series index items that need to participate in the calculation of the networks.Meanwhile,the conclusion shows that the method of using the RNN is appropriate to the curve fitting of the velocity of descent.The calculation results can accurately reflect the trend of the suspension-line strain in the parachute deployment process,the peak error is less than 1%and the peak time error is below 2%.