摘要
BUCK电路作为一种典型电力电子电路,是电子设备中电源模块的重要组成部分,在高速铁路列控车载设备中有广泛的应用。对电子设备的组成电路进行健康度评估与预测是提升电子设备可靠性的有效手段,因此,对BUCK电路开展健康度评估与预测的研究有助于提升车载设备的安全性与可靠性。提出一种基于敏感权重加权马氏距离的BUCK电路健康状态评估方法,可实现对电路健康状态的量化并得到电路的健康度,在得到电路健康度的基础上,构建基于NARX神经网络的BUCK电路健康度预测模型。首先建立BUCK电路的PSpice仿真模型,提取负载端响应输出的四种时域特征,得到BUCK电路健康状态特征集;针对不同特征对电路元件参数变化情况敏感度不同,计算不同特征的敏感度作为衡量该特征权重的指标;根据健康样本加权马氏距离计算结果和“3σ”原则划分健康阈值界限,实现对BUCK电路的健康度评估;利用待测样本健康度,构建基于NARX神经网络健康度预测模型,预测结果的均方根差为0.0071,验证了所提模型的可用性与有效性。
As a typical power electronic circuit as well as an important part of the power module in electronic equipment,BUCK circuit is widely used in on-board control equipment of high-speed trains.The health evaluation and prediction of the component circuits of electronic equipment is an effective means to improve the reliability of electronic equipment.Therefore,the research on health evaluation and prediction of BUCK circuit is helpful to improve the safety and reliability of on-board equipment.In this paper,a BUCK circuit health state evaluation method based on sensitive weight weighted Mahalanobis distance was proposed to quantify the circuit health state and obtain the circuit health degree.Based on the circuit health,a BUCK circuit health prediction model based on NARX neural network was constructed.Firstly,the Pspice simulation model of the BUCK circuit was established to extract four time-domain characteristics of the response output of the load end,and obtain the health state feature set of the BUCK circuit.According to the different sensitivity of different features to the changes of circuit component parameters,the sensitivity of different features was calculated as the index to measure the weight of the feature.According to the calculation results of weighted Mahalanobis distance of healthy samples and the“3σ”principle,the health threshold limit was divided to realize the health evaluation of the BUCK circuit.Using the health of the samples to be tested,a health prediction model based on NARX neural network was constructed.The root mean square difference of the prediction results is 0.0071,which verifies the availability and effectiveness of the proposed model.
作者
上官伟
师泽斌
彭聪
王宗耀
SHANGGUAN Wei;SHI Zebin;PENG Cong;WANG Zongyao(School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China;State Key Laboratory of Rail Traffic Control and Safety,Beijing Jiaotong University,Beijing 100044,China)
出处
《铁道学报》
EI
CAS
CSCD
北大核心
2023年第12期103-111,共9页
Journal of the China Railway Society
基金
北京市自然科学基金(L191013)
国家自然科学基金(61773049)。
