基于神经网络的旋转肋化强弯曲U型通道内压力分布特性预测

依据部分实验数据．利用神经网络对旋转肋化强弯曲U型通道的压力分布特性进行了预测。计算结果表明，当积累了一定的实验测量数据后，经过训练的人工神经网络能够对更高转动数下的通道内压力分布特性进行预测。但是由于U型通道内压力分布的强烈非线性特性，当实验数据较少时，这种预测能力将显著下降，甚至完全丧失。因此，积累相应的实验数据对于基于神经网络的U型通道内压力分布特性的预测是必要的。

Pressure Distribution Estimation of a Rotating U-bend Duct with Rib Based on Neural Network Method

By means of experimental data, pressure distribution of a rotating U-bend duct with rib is estimated based on neural network method. Predicted results demonstrate that the artificial neural network with enough experimental data is able to estimate the pressure distribution in the duct at a higher rotation number. However, because of the strong non-linear feature of the pressure distribution in an U-bend duct, the estimation ability will change if there are not enough samples to the trained networks. Therefore, it is necessary to accumulate enough experimental data so that the networks will be trained up to an acceptable estimation accuracy.