融入时间序列表征的充电桩复杂故障诊断方法OA

With the rapid development of electric vehicles over the past decade,the reliability and safety of charging piles have become key factors in ensuring user safety and enhancing the travel experience.Aiming at existing fault diagnosis methods that model the tasks with single factors and have the poor performances of fault types with few samples in complex fault detection in charging piles,a complex fault diagnosis method was proposed for charging piles that incorporates time-series characterization by taking into account the data of charging pile maintenance work orders,historical operation bills,and charging pile uploaded messages.Firstly,the method completes and normalizes the values of work order,order and message data of the charging pile.Secondly,feature engineering and bidirectional gated recurrent unit(Bi-GRU)are employed to extract the important features of multi-source data and the time series fault representations(temperature,voltage and electricity)during charging.Finally,with the characterized data samples,the methods of machine learning are utilized for training so as to judge whether the charging pile is faulty and the types of complex faults.The experimental results show that the method proposed outperforms the single machine learning and deep learning characterization methods in terms of accuracy,precision,recall and F1 value.