车载直流电源的故障预测与健康评估系统开发
Fault Prediction and Health Assessment of Automotive DC Power Supply
余云加 1彭子健 2吴来杰 2文国军 2余叶威1
作者信息
- 1. 广电计量检测集团股份有限公司,武汉 430074
- 2. 中国地质大学(武汉),武汉 430074
- 折叠
摘要
车载电源作为一种移动电站,是野外作业的主要电能来源.其工作环境恶劣且情况复杂,因此各类元器件可能会出现不同程度的故障,若在故障出现早期能及时予以准确预测并诊断,会极大提高车载电源的安全性和可靠性.由于车载电源系统复杂度高难以准确建模,且实际监测数据存在冗余、含有噪声且样本不平衡,传统的故障诊断方法难以对系统进行准确预测与健康评估.鉴于此本文采用数据驱动预测法,通过将机器学习算法与健康评估算法相结合,开展车载直流电源的故障预测与健康评估方法研究.主要内容包括:首先,基于直流电源的组成结构和工作原理,构建健康状态评估体系.针对目前设备健康状态估计的研究中,通常只将采集到的直流电源的电压或电流作为单一故障特征参数,容易导致估计结果不准确.此外,目前的评价方法很少考虑数据噪声对评价结果的影响,容易导致评价结果异常.为了解决以上两个问题,本文首先将输出电压、纹波电压、输出电流、外壳温度等状态指标引入到评价指标中.在此基础上,采用层次分析结合熵权法的组合赋权法确定各评价指标权重.最后采用基于加权马氏距离的多参数健康评估方法对电源进行健康状态评估.其次,研究基于Bi-LSTM模型的直流电源故障预测,将历史输出电压、纹波电压、输出电流、外壳温度等状态指标进行Min-MaxScaler归一化,将处理后的数据作为网络输入,利用Bi-LSTM加LSTM网络模型进行故障预测,并通过对预测曲线和实验数据的对比分析,证实此预测方法的有效性.
Abstract
As a mobile power station used in field operations,the vehicle-mounted power supply is a primary energy source under harsh and complex working environments.Various components of the system may experience different levels of faults due to these conditions.Early and accurate prediction and diagnosis of faults can significantly enhance the safety and reliability of vehicle-mounted power supplies.Given the high complexity of vehicle-mounted power supply systems,which are challenging to model accurately,and considering the presence of redundant,noisy,and imbalanced monitoring data,traditional fault diagnosis methods struggle to accurately predict and assess system health.Therefore,this study adopts a data-driven predictive approach by integrating machine learning algorithms with health assessment methods to investigate fault prediction and health evaluation of vehicle-mounted direct current(DC)power supplies.The main contents include:Firstly,based on the structure and working principles of the DC power supply,we es-tablish a health assessment framework.In current research on equipment health estimation,typically only voltage or current collected from the DC power supply is used as a single fault feature parameter,which can lead to inaccurate estimation results.Additionally,existing eval-uation methods often overlook the impact of data noise on assessment outcomes,resulting in abnormal evaluation results.To address these issues,this study incorporates state indicators such as output voltage,ripple voltage,output current,and casing temperature into the evalu-ation criteria.Subsequently,a combined weighting method using Analytic Hierarchy Process(AHP)combined with Entropy Weight Method(EWM)is employed to determine the weights of each evaluation indicator.Finally,a multi-parameter health assessment method based on weighted Mahalanobis distance is applied to evaluate the health status of the power supply.Next,we study DC power supply fault prediction based on the Bi-LSTM model.Histor-ical state indicators such as output voltage,ripple voltage,output current,and casing tem-perature are normalized using Min-MaxScaler.The processed data is then used as input to a Bi-LSTM combined with LSTM network model for fault prediction.Through comparative analysis of predicted curves and experimental data,we validate the effectiveness of this predic-tion method.
关键词
直流电源/健康状态评估/故障预测Key words
DC Power Supply/Health Status Assessment/Failure Prediction引用本文复制引用
出版年
2024
NETL
NSTL
万方数据