基于机器学习的结冰风洞温度场预测

结冰风洞是开展飞行器结冰与防除冰研究的重要基础设施，其制冷系统通过调节压缩机吸气压力实现风洞内气流温度的精确控制，吸气压力控制及降温方式影响着风洞的试验效率。为实现压缩机吸气压力的准确预测，本文采用自适应粒子群算法优化后的支持向量回归（APSO–SVR）建立预测模型；在此基础上，利用多层感知机（MLP）神经网络建立分析模型，研究试验工况参数对风洞降温速率的影响。结果表明:压缩机吸气压力的预测值与试验值的平均绝对百分比误差（E_MAP）低于4%，均方误差（E_MS）低于0.003；影响风洞降温速率的工况参数主要有气流压力、试验风速、压缩机吸气压力和换热器出口初始温度，其中，压缩机吸气压力对降温速率的影响是最显著的。

Prediction of icing wind tunnel temperature field with machine learning

Icing wind tunnel is an important infrastructure for research on aircraft icing and anti-deicing in which refrigeration system realizes precise control of the airflow temperature in the wind tunnel by adjusting the suction pressure of the compressor unit. The suction pressure control and cooling methods affect the wind tunnel test efficiency. In this paper, aiming at accurate prediction of compressor suction pressure, the support vector regression （APSO–SVR） optimized by adaptive particle swarm algorithm is used to establish pressure prediction model to conduct pressure prediction and experimental research. In order to further improve the efficiency of icing wind tunnel testing, multi-layer perceptron （MLP） neural network is used to establish an analysis model to analyze the influence of test parameters on the cooling rate of wind tunnel. The results show that the average absolute percentage error （E_MAP） between the predicted and test value of the compressor suction pressure is less than 4%, and the mean square error （E_MS） is less than 0.003; the parameters affecting the wind tunnel cooling rate are mainly airflow density, test wind speed, compressor suction pressure and the initial temperature of the heat exchanger outlet. Among them, the compressor suction pressure has the most significant effect on it.