基于实时监测参数的民用飞机重着陆预警方法

针对目前民用飞机重着陆事件的识别只能通过飞行员事后上报和维修人员被动检查的问题，提出了一种基于实时监测参数的民用飞机重着陆预警方法；分析了飞机重着陆的影响因素，在对快速存取记录器数据预处理的基础上，采用灰色关联度分析方法，从飞机重着陆相关的52个监测参数中提取了26个特征监测参数；以着陆质量、垂直加速度、垂直下降率和俯仰率等4个重着陆评价参数作为预测参数，26个特征监测参数作为输入，建立了基于长短期记忆网络的飞机重着陆预测模型；采用重着陆案例数据对预测模型进行训练，分析了飞行高度区间、输入输出步长对模型预测精度的影响，进而对模型进行了优化；在案例验证中引入混淆矩阵验证了模型的预测效果。研究结果表明:利用长短期记忆网络所建立的民用飞机重着陆预警方法有效利用了实时监测参数中反映重着陆趋势的信息，实现了飞机的重着陆预警，在提前8 s预警的情况下，预测精度达到了98%，平均绝对误差仅为0.018 3，可为飞行员提供足够的时间裕度采取措施，避免重着陆的发生。 

Early warning method for heavy landing of civil aircraft based on real-time monitoring parameters

It was considered that the heavy landing events of civil aircraft can only be reported by pilots or checked passively by the maintenance personnels afterward at present, an early warning method for the heavy landing of civil aircraft based on real-time monitoring parameters was proposed. The influencing factors in heavy landing were analyzed, and on the basis of the preprocessed data of a quick access recorder (QAR), the grey relational analysis (GRA) was employed to extract 26 feature monitoring parameters from 52 monitoring parameters related to the heavy landing of aircraft. Taking the landing weight, vertical acceleration, vertical decreasing rate, and pitch rate as the prediction parameters and the 26 feature monitoring parameters as the inputs, a prediction model for the heavy landing of aircraft was built based on the long short-term memory (LSTM). The prediction model was trained with heavy landing cases, and the influence of the flight height range and the input/output step size on the prediction accuracy was analyzed to optimize the prediction model. The confusion matrix was introduced into the case verification to verify the prediction results of the model. Research results indicate that the LSTM-based prediction model can make use of the information that reflects the trend of heavy landing in the real-time monitoring data to realize early warning of heavy landing, the prediction accuracy of the model can reach 98% for 8 seconds of warning, and the average absolute error is only 0.018 3, which means the model can provide pilots adequate time margin to take measures to avoid the heavy landing. 6 tabs, 17 figs, 29 refs.