基于GAIN的数据插补及Bi-GRU在充电桩预警中的应用

Data Imputation Based on GAIN and Application of Bi-GRU in Charging Pile Early Warning

在实际采集充电桩多维运行数据的过程中,会出现非人为因素而导致的数据丢失,因此在生成对抗网络的基础上引入提示机制,研究基于生成对抗插补网络的缺失数据插补方法。同时基于充电桩安全影响因素以及典型故障类型的分析,提出充电桩运行状态评价方法,在此基础上建立基于双向门控循环单元的充电桩安全预警模型,对异常告警的充电桩异常运行状态进行实时故障预测,实现充电桩的安全预警。通过算例分析,结果表明所提的生成对抗插补网络的数据插补方法在数据插补方面精度较高,使用插补后的数据集对充电桩的故障预测精度更高,从而提高充电桩的安全预警准确率。

The safe operation and maintenance of charging piles affect the efficient use of electric vehicles(EVs)and the safe operation of the power grid.In the process of collecting multi-dimensional operation data of charging piles,data loss caused by non-human factors may occur.Therefore,this paper introduces a hint mechanism based on generative adversarial network(GAN)to study the missing data interpolation method based generative adversarial interpolation network(GAIN).Meanwhile,based on the analysis of charging pile safety influencing factors and typical fault types,a charging pile operation state evaluation method is proposed,based on which a bi-directional gated cycle unit(Bi-GRU)based charging pile safety warning model is established to predict the abnormal operation state of charging piles with abnormal alarms in real time and realize the safety warning of charging piles.The results show that the proposed GAIN data interpolation method is more accurate in data interpolation,and the fault prediction accuracy of charging piles is higher using the interpolated data set,thus improving the safety warning accuracy of charging piles.