基于神经网络的气动热预测与建模研究  

Study on Aerodynamic Aeroheating Identification and Modeling Based on Neural Network

作　　者：司芳芳 汪文凯 秦诗牧 胡汉东 SI Fang-fang;WANG Wen-kai;QIN Shi-mu;HU Han-dong(Beijing Aerohydrodynamic Frontier Research Center,Beijing 100120,China)

机构地区：[1]北京流体动力科学研究中心,北京100120

出　　处：《计算机仿真》2025年第3期50-54,共5页Computer Simulation

摘　　要：为高效、高精度的获取高超声速飞行器表面热流数据,提出一种径向基优化的稀疏自编码器气动热预测模型。首先,利用稀疏自编码器模型对气动热数据进行表征学习,提取数据表示特征;然后将径向基函数网络模型与稀疏自编码器模型进行融合改造,最后利用改造后的模型实现对高超声速飞行器表面热流的预测。对双锥体高超声速武器表面热流预测结果表明,上述模型具有较高的精度和良好的外推性能,热流密度预测结果与多源融合数据偏差在10%以内。To efficiently and accurately obtain heat flux data on the surface of hypersonic vehicles,a predictive model for aerodynamic heating based on a radial basis function(RBF)-optimized sparse autoencoder is proposed.Firstly,the sparse auto-encoder model is used to characterize and learn aerodynamic thermal data and extract data representation features.Then,the radial basis function network model is combined with the sparse auto-encoder model.Finally,the modified model is used to identify the surface heat flux of hypersonic vehicles.The results of identifying the surface heat flux of a dual cone hypersonic weapon show that this model has high accuracy and good extrapolation performance.The deviation between the recognition result of heat flux density and the multi-source fusion data is within 10%.

关 键 词：高超声速飞行器 神经网络 数值模拟 气动热 模型 

分 类 号：TJ765.43[兵器科学与技术—武器系统与运用工程]