摘要
地铁车辆车厢空调能耗占整车能耗的25%~40%,若长期处于故障运行状态会造成大量的能源浪费,且影响乘客的乘车体验甚至危害乘客身体健康。地铁车辆空调故障中,制冷剂充注量故障发生频繁却难以察觉,短期内不会引发故障报警,但会使空调长期偏离正常工况。针对地铁空调的制冷剂充注量故障诊断,文章采用基于交叉验证的递归特征消除法进行了特征选择,筛选出含18个特征变量的较优特征子集,然后根据随机森林特征重要性度量选择出重要性得分在0.03以上的8个特征作为最优特征子集,并将该子集分别用于构建基于支持向量机、K-最近邻算法和反向传播神经网络的诊断模型,以验证最优特征子集的故障诊断效果。验证结果显示,该特征选择算法经过两步筛选得到最优特征子集,在2种制冷工况和4种制冷剂充注量水平下,3种故障诊断模型的总准确率分别为99.83%、99.98%、99.96%。
The air conditioning energy consumption of metro carriages accounts for 25%~40%of the whole vehicle energy consumption.Long term failure operation will cause a large amount of energy waste,and affect the ride experience of passengers and even endanger the health of passengers.In the air conditioning failure of metro vehicles,the fault of refrigerant charging volume occurs frequently but is difficult to detect.In the short term,the failure alarm will not be triggered,but the air conditioning will deviate from the normal working condition for a long time.Aiming at the fault diagnosis of refrigerant charge of metro air conditioning,this paper uses the recursive feature elimination based on cross-validation to select features,screens out the relatively optimal feature subset containing 18 feature variables,and then select 8 features with an importance score of more than 0.03 as the optimal feature subset according to the random forest feature importance measurement,and utilizes the subset for constructing the diagnosis model based on Support Vector Machine,K-Nearest Neighbor algorithm and Back Propagation Neural Network respectively,so as to verify the fault diagnosis performance of the optimal feature subset.The verified results show that the feature selection algorithm obtains the optimal feature subset through two-step screening.Under two refrigeration conditions and four refrigerant charge,the overall diagnosis accuracy of three fault diagnosis models are:99.83%,99.98%,99.96%respectively.
作者
张丽
鲍超
王钊
陈焕新
程亨达
张鉴心
陈璐瑶
ZHANG Li;BAO Chao;WANG Zhao;CHEN Huanxin;CHENG Hengda;ZHANG Jianxin;CHEN Luyao(School of Energy and Power Engineering of Huazhong University of Science and Technology,Wuhan 430074,China;Guangzhou Metro Group Co.,Ltd.,Guangzhou 510330,China;Guangzhou Dinghan Railway Vehicles Equipment Co.,Ltd.,Guangzhou 510260,China)
出处
《铁道车辆》
2022年第6期115-121,共7页
Rolling Stock
基金
国家自然科学基金(51876070)。
关键词
地铁车辆
空调
制冷剂充注量
RFECV算法
随机森林特征重要性度量
故障诊断
metro vehicle
air conditioning
refrigerant charge
RFECV algorithm
random forest feature importance measurement
fault diagnosis
