基于深度学习算法的直流充电桩相对误差预测研究

Research on Relative Error Prediction of Direct Current Charging Pile Based on Deep Learning Algorithm

针对充电桩建设越来越多、设备易出现偏差的现状,提出了一种基于深度学习算法的直流充电桩相对误差预测的方法。首先将直流充电桩采集的数据集进行预处理,然后搭建了LightGBM、N-Linear以及CNN模型进行相对误差预测,并采用MAE以及MSE作为评估指标进行评估。结果表明,LightGBM模型效果最理想,MAE较N-Linear模型降低了57.91%,较CNN降低了30.16%,MSE较N-Linear模型降低了82.85%,较CNN降低了约47.32%。

In response to the increasing number of charging pile constructions and the tendency of equipment deviations,a method for relative error prediction of direct current charging piles based on deep learning algorithms is proposed in this paper.Firstly,the dataset collected from the direct current charging piles is preprocessed.Then,LightGBM,N-Linear and CNN models are constructed for relative error prediction,and MAE and MSE are adopted as evaluation metrics.The results indicate that the LightGBM model performs the best,with a decrease of 57.91%in MAE compared to the N-Linear model and a decrease of 30.16%compared to the CNN model.The MSE is reduced by 82.85%compared to the N-Linear model and approximately 47.32%compared to the CNN model.