论文部分内容阅读
在风洞试验中,马赫数的稳定性和快速性对风洞流场品质有着重要影响。为了实现马赫数的精确控制,必须对马赫数进行快速、准确的预测。风洞试验积累了大量数据,大数据集包含了更多的有益信息,为实现马赫数的精确预测提拱了可能性,但也增加了建模的复杂度。通常高度复杂的模型会加重其在实际使用时的计算负担。针对大数据集问题,本文将随机森林方法应用于风洞马赫数建模。随机森林是一种集成模型建模方法,它从3方面降低模型的复杂度:产生多个样本子集,减少了子模型的训练样本个数;具有并行集成结构,子模型可在不同的CPU上运行,提高了运行速度;以简单学习算法回归树作为基学习机,降低了子模型的复杂度。试验证明基于随机森林的马赫数预测模型能够有效利用试验积累的大数据,满足工程上预测速度及精度的要求。
In the wind tunnel test, the stability and speed of Mach number have an important influence on the flow field quality in wind tunnel. In order to achieve the precise control of the Mach number, the Mach number must be quickly and accurately predicted. Wind tunnel tests have accumulated a large amount of data, and larger data sets contain more useful information, raising the possibility of accurate Mach-number prediction, but also increasing the complexity of modeling. Often highly complex models add to the computational burden of their actual use. For the problem of large data sets, this paper applies stochastic forest method to wind tunnel Mach number. Random forest is an integrated model modeling method, which reduces the complexity of the model from three aspects: generating multiple sample subsets and reducing the number of training samples for the sub-model; with parallel integration structure, the sub-model can be implemented in different CPUs And improve the running speed. Using the simple learning algorithm to return the tree as a learning machine, the complexity of the sub-model is reduced. Experiments show that the Mach-based prediction model based on random forest can effectively utilize the big data accumulated in the experiment to meet the engineering requirements of predicting the speed and accuracy.