基于神经网络的多参数气膜有效度模型研究

Study on Multi-parameter Cooling Effectiveness Model based on Neural Network

气膜冷却是燃气轮机高温透平重要的冷却方式,其冷却效果受多个参数的影响。本文采用BP及LSTM神经网络方法,建立了兼顾气膜孔的孔型、孔倾角、侧倾角、相对曲率强度、吹风比、孔位置以及相对流向位置等多参数影响的单排孔曲面气膜冷却系统侧向平均绝热气膜冷却效率(气膜有效度)预测模型。采用文献中的试验数据对CFD方法的准确性进行验证,基于验证后的CFD方法以燃气轮机透平叶片气膜冷却的实际运行工况为基准建立了数据库。结果表明:采用LSTM神经网络训练得到的模型在拟合精度上优于BP神经网络,但训练时间成本较高;基于LSTM神经网络的多参数气膜有效度模型具有较强的预测能力和泛化能力,可为气膜冷却优化设计提供有效的工具。

Film cooling is an important cooling method in high temperature gas turbine,and its cooling effect is affected by many parameters.In this paper,BP and LSTM neural network methods were used to establish a prediction model for the lateral average adiabatic cooling efficiency(film effectiveness)of a single-row curved air film cooling system,taking into account the influence of many parameters,such as hole type,hole inclination,side inclination,relative curvature intensity,blow ratio,hole position and relative flow direction position.The CFD numerical simulation method based on the test data verification was used to establish a database based on the actual operating conditions of gas turbine blade film cool-ing.The results show that the model trained by LSTM neural network is better than BP neural network in fitting accuracy,but the training time cost is higher.The multi-parameter cooling effectiveness model based on LSTM neural network has strong prediction ability and generalization ability,which can provide an effective tool for the optimal design of film cooling.