滤波在数据驱动的燃气轮机参数预测中的应用

Application of Filter in Data Driven Prediction of Gas Turbine Parameters

在直接使用神经网络模型对燃气轮机的状态参数进行预测时,发现预测结果往往不够精确。通过观察采集到的机组实际运行数据发现,误差很可能是由实时数据的噪声引起的。结合其他学者的研究经验,提出了一种先对原始数据进行滤波处理,再将其输入神经网络进行计算的方法。对同一组数据的预测情况的比较结果证明了该滤波环节对提高神经网络预测精度十分有效。

The prediction results of state parameters in gas turbines are not accurate enough with direct application of neural network models.By observing the actual collected data during unit operation,it was found that the error was likely caused by the noise of real-time data.Combined with the research experience of other scholars,it was proposed to carry out the filtering of original data firstly and then put it in neural network to calculate.For the filtering procedure,its effectiveness of improving the prediction accuracy of neural network has been verified by comparison of prediction of the same group of data.