对现有运营类充电桩电能表开展强检及监管面临着成本高、效率低、任务重、人员短缺等问题，且目前基于数据驱动的电能表性能分析精度有待提高。为此，文章提出一种基于卷积双向长短期记忆网络的时间切片方法来估计电能表的测量不确定度。针对智能电能表采集到的充电设施运行数据存在的时空特性，设计卷积网络提取变量间空间特征，并将提取的特征输入双向长短期记忆网络中，用以进一步捕捉数据时序特征,考虑到电能表运行过程受环境和充电需求影响而长期处于非平稳状态，因此采用时间切片方案实时计算局部时段内的测量不确定度。所提方案在某地新能源汽车充电站中进行了验证，并与粒子群优化（particle swarm optimization, PSO）算法结合反向传播神经网络(back propagation neural network, BPNN)的PSO-BPNN、动态回归扩展与混合（dynamic regressor extension and mixing, DREM）结合折息最小二乘(recursive least squares with discount factor, DRLS)法的DREM-DRLS和扩展卡尔曼滤波(extended Kalman filter, EKF)结合限定记忆递推最小二乘（limited memory recursive least square，LMRLS）算法的EKF-LMRLS模型进行对比，实验结果表明文章所提方法在直流充电桩电能表测量不确定度预测精度上有较大优势，设计的三种模型性能评价指标至少有26.81%以上的提升。
英文摘要:
The inspection and regulation of existing operational charging pile electricity meters face problems suchas high cost, low efficiency, heavy workload, and shortage of personnel. Additionally, the current performance analysis accuracy of data-driven electricity meters needs to be improved. To solve this problem, a time-slicing method based on convolutional bidirectional long short-term memory(C-BiLSTM-TS) network is proposed to estimate the measurement uncertainty of electricity meters. Considering the spatio-temporal characteristics of the charging facility operation data collected by smart electricity meters, convolutional networks are employed to extract spatio-temporal features between variables. The extracted features are then fed into the BiLSTM networks to further capture temporal characteristics of the data. Furthermore, since the operation of electricity meters is influenced by the environment and charging demands, which results in long-term non-stationary states, a time-slicing method is used to calculate the measurement uncertainty in real-time within time intervals. The proposed scheme has been validated in a new energy vehicle charging station and compared with PSO-BPNN, DREM-DRLS, and EKF-LMRLS models. The experimental results show that the proposed method has a significant advantage in predicting the measurement uncertainty of DC charging pile electricity meters, and three performance evaluation indicators of the model have improved by at least 26.81%.