Abstract: The widely used high-precision analytical models make the computational cost of interceptor aerodynamic shape design increasing.Therefore,the method of rapid performance evaluation based on surrogate models has received extensive attention in recent years.When the interceptor is affected by jet disturbance factors,the parameter mapping relationship is complicated,the data acquisition cost is high,and the prediction accuracy of the surrogate model is low.For this reason,a surrogate model construction method that integrates engineering experience and data was proposed.The engineering experience accumulated by designers was mathematically characterized,a loss function was constructed by combining engineering data,and an aerodynamic performance prediction surrogate model was established for a hypersonic interceptor in the near space.The experimental results showed that the proposed surrogate model construction algorithm had higher prediction accuracy and better generalization capability than the traditional methods,which provided a new idea for aerodynamic performance evaluation model construction.

Key words: hypersonic interceptor, aerodynamic performance, surrogate model, Bayesian neural network, engineering experience

摘要： 广泛应用的高精度分析模型使得拦截器气动外形设计的计算成本不断增加。因此,基于代理模型的性能快速评估方法近年来得到了广泛关注。当拦截器受喷流干扰因素影响时,参数映射关系复杂、数据获取成本高昂,代理模型预测精度低。为此,提出一种融合工程经验与数据的代理模型构建方法。该方法将设计人员积累的工程经验进行数学化表征,结合仿真数据构建损失函数,并以某临近空间高超声速拦截器为对象构建了气动性能预测代理模型。实验结果表明,所提代理模型构建算法相较于传统方法有更高的预测精度和更好的泛化能力,该方法为气动性能评估模型构建提供了新的思路。

关键词: 高超声速拦截器, 气动性能, 代理模型, 贝叶斯神经网络, 工程经验