基于数据和神经网络的压气机性能预测研究

Prediction of a Compressor's Performance Based on Data and Neural Networks

由于缺乏相关实验数据,压气机的全转速性能曲线往往难以获得。从有限的数据点出发,搭建了不同的神经网络预测模型并做了相关讨论,同时提出了能有效处理特性线的新建模思路。结果表明,RBF神经网络能够更好地吻合实验样本,而BP神经网络则预测性能更佳。最后提出了考虑IGV的处理方法,并利用燃气轮机的实际运行数据对预测模型进行了检验,证明该方法能够很好地描述压气机的动态性能,可为燃气轮机的动态过程模拟提供参考。

Due to the lack of experimental data of compressor performance, it＇s hard to get a full-ranged velocity performance curve. Based on limited data points, the different prediction models of neural networks were established and relevant discussion was conducted. At the same time, a new modeling method which could effectively handle the characteristic line was proposed and validated. The results indicate that while the RBF network is in the best agreement with the experimental data, the BP networks have better performance of prediction. Furthermore, a method to deal with the IGV was proposed and operating data of a real power plant was utilized to certify the prediction results. It is proved that this prediction method can perfectly describe the dynamic performance of compressor, and provide reference for dynamic simulation of gas turbines.