基于ConvLSTM的飞机发动机喷流时空预测

ConvLSTM Based Spatiotemporal Prediction of Aircraft Engine Jet Flow

为探究起飞飞机发动机喷流对后侧穿越飞机的影响程度,提出一种基于卷积长短时记忆网络(ConvLSTM)的飞机发动机喷流预测模型,旨在预测未来某一时段的流场数据。通过激光测风雷达采集飞机发动机喷流数据并进行预处理。分别采用时间子网和空间子网捕捉飞机发动机喷流的时间和空间结构特征。通过融合时空特征,使用全连接层输出未来流场数据,以此构建卷积长短时记忆网络面向飞机发动机复杂喷流数据的未来帧预测。结果表明:ConvLSTM模型能够准确地预测出飞机发动机喷流的时空分布,取得RMSE12.28和MAE9.26的实验结果,较传统神经网络模型预测结果拥有更稳定的RMSE值及预测精度,有效提高了喷流时空预测的质量和精度,为研究飞机发动机喷流影响范围提供支撑。

To investigate the extent to which the engine jet of a takeoff aircraft affects the rear passing aircraft,a prediction model of aircraft engine jets based on Convolutional Long Short Term Memory Network(ConvLSTM)was proposed to predict the flow field data at a certain time period in the future.Aircraft engine jet data were acquired and pre processedby lidar.The temporal and spatial structure characteristics of jet flow of aircraft engines are captured by temporal and spatial sub networks.By fusing spatial temporal features and using full junction layer to output future flow field data,a convolved short and long term memory network is constructed for future frame prediction of aircraft engine complex jet data.The results show that the ConvLSTM model can accurately predict the spatial temporal distribution of jet flow in aircraft engines,and the experimental results of RMSE 12.28 and MAE 9.26 are obtained.Compared with the traditional neural network model,the prediction results have more stable RMSE values and prediction accuracy,which effectively improve the quality and accuracy of the spatial temporal prediction of jet flow,providing support for the study of jet flow influence ranges in aircraft engines.