集成特征选择在VRF系统制冷剂充注量故障诊断中的应用

Forward search ensemble feature selection strategy for multi-line system fault diagnosis

针对VRF系统故障诊断存在的数据量大、特征冗余等问题,本文提出一种前向搜索优化的集成特征选择方法,该方法将单一特征选择方法得到的特征子集进行整合,以获得预测性能更好的特征变量。首先分别利用mRMR算法、ReliefF算法、随机森林算法、Adaboost.M1算法和Boruta特征选择算法对实验数据进行特征选择,然后利用前向搜索策略得到集成后的特征排序,并与算术平均、众数投票两种集成方法形成对比。最后,分别采用上述特征选择方法获得的关键特征变量作为模型的输入变量建立相应的故障诊断模型,通过对比发现前向搜索集成获得了最高的预测准确率,选出了最具代表的特征变量。

Aiming at the problems of large amount of data and feature redundancy in fault diagnosis of VRF systems,this paper proposes an integrated feature selection method for forward search optimization.This method integrates a subset of features obtained by a single feature selection method to obtain predictive performance.Better characteristic variables.First use the mRMR algorithm,the ReliefF algorithm,the random forest algorithm,the Adaboost.M1 algorithm and the Boruta feature selection algorithm to perform feature selection on the experimental data,and then use the forward search strategy to obtain the integrated feature ranking,and vote with the arithmetic average and mode The two integration methods contrast.Finally,the key feature variables obtained by the above feature selection method are used as input variables of the model to establish corresponding fault diagnosis models.Through comparison and discovery,forward search integration has obtained the highest prediction accuracy rate,and the most representative feature variables are selected.