利用循环网络的降落伞开伞载荷补偿方法

为了对降落伞充气展开过程中的开伞载荷进行更加准确的预测，提出一种基于循环神经网络的开伞载荷补偿计算方法，包括模型架构和数据处理方式。该方法将充气时间法计算的预测值代入循环网络进行二次计算，使最终结果能够更加贴近试验真值。使用多层前馈网络、标准循环网络与长短时记忆网络三种网络进行比较，验证了所提模型预测结果的适用性和准确性，研究了学习率、输入层维度和隐层维度等超参数对模型性能的影响，并给出了基于长短时记忆网络的补偿模型最优训练条件。实验结果表明，利用循环网络进行开伞载荷预测具有较好的拟合结果，为机器学习与降落伞工业的学科交叉研究提供了一定的参考方向。

Compensation method of parachute deployment load based on recurrent neural networks

Focusing on predicting the parachute deployment load in the process of inflation accurately, a compensation calculation method of parachute deployment load with RNN (recurrent neural networks) was proposed, including the model architecture and data processing. The predicted value calculated by inflation time method was brought into the RNN for the secondary calculation, so that the final result could be close to the airdrop experiment data. The feedforward neural networks, standard recurrent networks and long short-term memory networks were used to compare the model characteristic. The research verified the applicability and accuracy of the prediction results and analyzed the effects of hyperparameters such as learning rate, input layer dimension and hidden layer dimension on the performance. The optimal training condition for reference to the compensation model was developed through the test. The results show that the utilization of RNN for parachute deployment load prediction is effective and provides a referential significance for the interdisciplinary research of machine learning and parachute industry.