考虑激励特性的汽轮机做功模型辨识数据优选方法

Selection method for identification data of steam turbine work model considering excitation characteristics

针对历史运行数据中难以选择合适样本辨识汽轮机做功模型问题,提出一种考虑激励特性的辨识数据优选方法。首先,采用费歇尔(Fisher)信息矩阵条件数提取历史运行数据的激励特性,与数据的趋势特性和参数间相关性共同构成特征变量集。其次,以特征变量作为输入,基于标准汽轮机做功模型生成的标识结果作为输出,采用随机森林分类算法生成辨识数据分类规则模型,实现辨识数据的在线选择。最后,对模型分类结果的准确性与所选数据的辨识效果进行验证。结果表明,分类规则模型的准确度为97.561%,可准确选出历史运行数据中含有充分激励的样本段,其汽轮机做功模型辨识结果与标准模型具有较高的一致性。

A method of identifying data by considering the excitation characteristics is proposed to solve the problem that it is difficult to select suitable samples from the historical operation data to identify the turbine work model.Firstly,Fisher’s information matrix condition number is applied to extract the excitation characteristics of the historical operating data,which together with the trend characteristics and the correlation between parameters constitute the set of feature variables.Secondly,by using the feature variables as inputs and the identification results generated based on the standard turbine work model as outputs,the Random Forest classification algorithm is used to generate a classification rule model for the identification data to realize the online selection of identification data.Finally,the accuracy of the model classification results and the identification effect of the selected data are verified.The result proves that the accuracy of the classification rule model is 97.561%,which can accurately select the sample segments containing sufficient incentives in the historical operation data,and the identification results of the turbine work model are in high consistency with that of the the standard model.