基于GAN的负荷数据修复及其在EV短期负荷预测中的应用

随着电动汽车(EV)充电设施规模的不断扩大,EV充电数据可以更方便地获得.某些非人为因素会导致数据集中存在数据缺失和数据异常的问题,阻碍了EV负荷预测精度的提高.因此,文中在生成对抗网络(GAN)中采用用于插补的门控循环单元神经网络(GRUI)细胞来处理不完整负荷数据集中前后观测值间的不规则时滞关系,提出了基于GRUI-GAN的数据插补模型来实现EV负荷数据修复.然后,以带有Mogrifier门控机制的长短期记忆网络进行EV负荷预测.最后,实验结果表明了所提方法可以生成精度较高的新数据对缺失值进行插补,并且经所提方法修复之后的数据有效提高了EV负荷预测精度.

Load Data Restoration Based on Generative Adversarial Network and Its Application in Short-term Load Forecasting of Electric Vehicle

As the scale of electric vehicle (EV) charging facilities keeps expanding, EV charging data can be obtained more conveniently. Some non-human factors will lead to missing data and abnormal data problems in the data set, which hinders the improvement of EV load forecasting accuracy. Therefore, this paper uses the gated recurrent unit neural network cells for imputation (GRUI) in the generative adversarial network (GAN) to deal with the irregular time-delay relationship between the previous and later observation data in the incomplete load data set, and a data imputation model based on GRUI-GAN is proposed to restore the EV load data. Then, the long short-term memory network with the Mogrifier gating mechanism is used for EV load forecasting. Finally, the experimental results show that the proposed method can generate new data with high accuracy to interpolate missing values, and the data restored by the proposed method can effectively improve the accuracy of EV load forecasting.