基于谱聚类和LSTM神经网络的电动公交车充电负荷预测方法
Electric Bus Charging Load Forecasting Method Based on Spectral Clustering and LSTM Neural Network
-
摘要: 目前电动公交车的渗透率较大,且充电频率和充电量较高,故而其充电负荷对电网运行与调度产生着不可忽略的影响。因此,电动公交车的充电负荷预测研究具有重要的理论与现实意义,但由于公交车间歇性与随机性的充电行为在时间上给充电负荷预测增加了难度。为此,提出基于谱聚类和长短期记忆(long short-term memory, LSTM)神经网络的电动公交车充电负荷预测方法。首先,利用考虑距离与形态的谱聚类,对充电负荷曲线进行聚类;其次,综合考虑影响充电负荷的关键因素,如温度、日类型等,利用不同簇的总充电负荷,分别训练LSTM神经网络的模型参数,并预测每簇的充电负荷;接着,对不同簇的预测结果求和即可得到预测日的总充电负荷;最后,通过利用某市实际数据,验证本文所提方法。结果表明,所提方法充电负荷预测结果的平均绝对百分误差(mean absolute percentage error, MAPE)在11%以下,预测准确度有所提升。Abstract: At present, the penetration rate, charging frequency and charging capacity of electric buses are relatively high, so the charging load has a non-negligible impact on the operation and dispatch of the power grid. So, the charging load forecasting research has important theoretical and practical significance, but the intermittent and random charging behavior increase the spatial forecasting difficulty. Therefore, the charging load forecasting method of electric buses is proposed on the basis of spectral clustering and long short-term memory（LSTM） neural network. First of all, the charging load curve is clustered according to spectral clustering considering the distance and the shape. And then, considering the key factors that affect the charging load, such as historical load, temperature and day type, the model parameter of LSTM neural network is trained using each cluster charging load, and the charging load of each cluster is predicted. Then, the total charging load of the forecasting day is to sum the forecasting results of different clusters. Finally, on the basis of the historical real data in a certain city, the proposed method is verified. The result shows the mean absolute percentage error（MAPE） of charging load prediction result of the proposed method is below 11%, and the accuracy of load forecasting is improved.
下载:
