基于集成学习的电动汽车充电站超短期负荷预测

doi:10.16183/j.cnki.jsjtu.2021.486

摘要/Abstract

摘要：

精确的电动汽车充电站充电负荷预测是提高充电站安全经济运行的重要措施,也是支撑充电基础设施新建、扩容规划决策的重要基础.为提高电动汽车充电站超短期充电负荷预测的精度,提出一种基于集成学习的充电站超短期充电负荷预测方法.首先,以预测精确度与响应速度为主要目标,使用轻量级梯度提升框架构建基础回归器模型;其次,通过自适应提升方法对基础回归器群进行集成;最后,通过超参数调整与优化,建立基于能量集成轻量梯度提升框架(EEB-LGBM)的双层充电站超短期充电负荷预测模型.算例结果表明,相较于反向传播神经网络、卷积神经长短期记忆网络、差分自回归移动平均模型等预测模型,所提出的基于EEB-LGBM的超短期充电负荷预测模型具有更高的精确度,同时可以大幅度缩短训练时间和降低计算资源需求.

关键词: 电动汽车充电站, 充电负荷, 超短期预测, 集成学习, 经济性

Abstract:

Accurate electric vehicle load forecasting is the basis for maintaining the safe and economical operation of charging stations, and for supporting the planning and decision-making of new and expanded charging infrastructure. In order to improve the accuracy of the ultra-short-term load forecasting of charging stations, an ultra-short-term load forecasting method based on ensemble learning is proposed. First, aimed at the prediction accuracy and the response speed, the light gradient boosting machine (LightGBM) framework is utilized to build several basic regressors. Next, the basic regressors are integrated by using the adaptive boosting (Adaboost) method. Finally, by using hyperparameter adjustment and optimization, a dual-system for ultra-short-term load forecasting of charging stations named energy ensemble boosting-light gradient boosting machine (EEB-LGBM) is generated. The analysis of the numerical examples shows that the proposed model has a higher accuracy than the back propagation neural network (BPNN), convolutional neural networks-long short term memory (CNN-LSTM), autoregressive integrated moving average (ARIMA), and other load forecasting methods, which can greatly reduce the training time and the computing power requirements of the training platform.

Key words: electric vehicle charging station, charging load, ultra-short-term forecasting, ensemble learning, economy

中图分类号:

TM734

李恒杰, 朱江皓, 傅晓飞, 方陈, 梁达明, 周云. 基于集成学习的电动汽车充电站超短期负荷预测[J]. 上海交通大学学报, 2022, 56(8): 1004-1013.

LI Hengjie, ZHU Jianghao, FU Xiaofei, FANG Chen, LIANG Daming, ZHOU Yun. Ultra-Short-Term Load Forecasting of Electric Vehicle Charging Stations Based on Ensemble Learning[J]. Journal of Shanghai Jiao Tong University, 2022, 56(8): 1004-1013.

图/表 13

图1

表1

图2

图3

图4

图5

表2

表3

图6

图7

图8

图9

表4

参考文献 23

[1]	陈文彬, 徐大勇, 郭瑞鹏. 负荷预测对新能源电网多目标优化调度的影响规律研究[J]. 电力系统保护与控制, 2020, 48(10): 46-51.
	CHEN Wenbin, XU Dayong, GUO Ruipeng. A study on the influence rule of load forecasting on multi-objective optimal dispatching of a new-energy grid[J]. Power System Protection and Control, 2020, 48(10): 46-51.
[2]	孙超, 吕奇, 朱思曈, 等. 基于双层XGBoost算法考虑多特征影响的超短期电力负荷预测[J]. 高电压技术, 2021, 47(8): 2885-2898.
	SUN Chao, LÜ Qi, ZHU Sitong, et al. Ultra-short-term power load forecasting based on two-layer XGBoost algorithm considering the influence of multiple features[J]. High Voltage Engineering, 2021, 47(8): 2885-2898.
[3]	李玉志, 刘晓亮, 邢方方, 等. 基于Bi-LSTM和特征关联性分析的日尖峰负荷预测[J]. 电网技术, 2021, 45(7): 2719-2730.
	LI Yuzhi, LIU Xiaoliang, XING Fangfang, et al. Daily peak load prediction based on correlation analysis and Bi-directional long short-term memory network[J]. Power System Technology, 2021, 45(7): 2719-2730.
[4]	罗凤章, 张旭, 杨欣, 等. 基于深度学习的综合能源配电系统负荷分析预测[J]. 高电压技术, 2021, 47(1): 23-32.
	LUO Fengzhang, ZHANG Xu, YANG Xin, et al. Load analysis and prediction of integrated energy distribution system based on deep learning[J]. High Voltage Engineering, 2021, 47(1): 23-32.
[5]	谭嘉, 李知艺, 杨欢, 等. 基于分布式优化思想的配电网用电负荷多层协同预测方法[J]. 上海交通大学学报, 2021, 55(12): 1544-1553.
	TAN Jia, LI Zhiyi, YANG Huan, et al. A multi-level collaborative load forecasting method for distribution networks based on distributed optimization[J]. Journal of Shanghai Jiao Tong University, 2021, 55(12): 1544-1553.
[6]	钟光耀, 邰能灵, 黄文焘, 等. 基于多维聚类的配变负荷注意力短期预测方法[J]. 上海交通大学学报, 2021, 55(12): 1532-1543.
	ZHONG Guangyao, TAI Nengling, HUANG Wentao, et al. Attention short-term forecasting method of distribution load based on multi-dimensional clustering[J]. Journal of Shanghai Jiao Tong University, 2021, 55(12): 1532-1543.
[7]	林珊, 王红, 齐林海, 等. 基于条件生成对抗网络的短期负荷预测[J]. 电力系统自动化, 2021, 45(11): 52-60.
	LIN Shan, WANG Hong, QI Linhai, et al. Short-term load forecasting based on conditional generative adversarial network[J]. Automation of Electric Power Systems, 2021, 45(11): 52-60.
[8]	肖白, 肖志峰, 姜卓, 等. 基于降噪自编码器、奇异谱分析和长短期记忆神经网络的空间电力负荷态势感知[J]. 中国电机工程学报, 2021, 41(14): 4858-4867.
	XIAO Bai, XIAO Zhifeng, JIANG Zhuo, et al. Spatial load situation awareness based on denoising autoencoder, singular spectrum analysis and long short-term memory neural networks[J]. Proceedings of the CSEE, 2021, 41(14): 4858-4867.
[9]	范士雄, 刘幸蔚, 於益军, 等. 基于多源数据和模型融合的超短期母线负荷预测方法[J]. 电网技术, 2021, 45(1): 243-250.
	FAN Shixiong, LIU Xingwei, YU Yijun, et al. Multi-source data and hybrid neural network based ultra-short-term bus load forecasting[J]. Power System Technology, 2021, 45(1): 243-250.
[10]	张金金, 张倩, 马愿, 等. 基于改进的随机森林和密度聚类的短期负荷频域预测方法[J]. 控制理论与应用, 2020, 37(10): 2257-2265.
	ZHANG Jinjin, ZHANG Qian, MA Yuan, et al. Short-term load frequency domain prediction method based on improved random forest and density-based spatial clustering of applications with noise[J]. Control Theory & Applications, 2020, 37(10): 2257-2265.
[11]	罗平, 程晟, 王严, 等. 考虑用户出行特性的电动汽车充电负荷预测[J]. 华中科技大学学报(自然科学版), 2020, 48(11): 126-132.
	LUO Ping, CHENG Sheng, WANG Yan, et al. Electric vehicle charging load forecast considering user's travel characteristics[J]. Journal of Huazhong University of Science and Technology (Natural Science Edition), 2020, 48(11): 126-132.
[12]	MA X J, SHA J L, WANG D H, et al. Study on a prediction of P2P network loan default based on the machine learning LightGBM and XGboost algorithms according to different high dimensional data cleaning[J]. Electronic Commerce Research and Applications, 2018, 31: 24-39. doi: 10.1016/j.elerap.2018.08.002 URL
[13]	SHAKER B, YU M S, SONG J S, et al. LightBBB: computational prediction model of blood-brain-barrier penetration based on LightGBM[J]. Bioinformatics, 2020, 37(8): 1135-1139. doi: 10.1093/bioinformatics/btaa918 URL
[14]	WANG Y Y, CHEN J, CHEN X Q, et al. Short-term load forecasting for industrial customers based on TCN-LightGBM[J]. IEEE Transactions on Power Systems, 2021, 36(3): 1984-1997. doi: 10.1109/TPWRS.2020.3028133 URL
[15]	KE G, MENG Q, FINLEY T, et al. Lightgbm: A highly efficient gradient boosting decision tree[J]. Advances in Neural Information Processing Systems, 2017, 30: 3146-3154.
[16]	MEHTA M, AGRAWAL R, RISSANEN J. SLIQ: A fast scalable classifier for data mining[C]// Advances in Database Technology - EDBT '96. Avignon, France: Springer, 1996: 18-32.
[17]	ZHOU Z H. Ensemble Methods[M]. New York, USA: Chapman and Hall (CRC Press), 2012.
[18]	REDDY L N, BUTAKOV S, ZAVARSKY P. Applying ML algorithms to improve traffic classification in intrusion detection systems[C]// 2020 IEEE 19th International Conference on Cognitive Informatics & Cognitive Computing. Beijing, China: IEEE, 2020: 1-6.
[19]	COOK R D. Detection of influential observation in linear regression[J]. Technometrics, 1977, 19(1): 15-18.
[20]	LING R F. Residuals and influence in regression[J]. Technometrics, 2012, 26(4): 413-415. doi: 10.1080/00401706.1984.10487996 URL
[21]	李汉巨. 考虑气象因素累积效应的电力系统96点日负荷预测方法[J]. 电气技术, 2018, 19(4): 28-32.
	LI Hanju. The 96 point daily load forecasting method for power system considering the cumulative effect of meteorological factors[J]. Electrical Engineering, 2018, 19(4): 28-32.
[22]	CHICCO D, WARRENS M J, JURMAN G. The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation[J]. PeerJ Computer Science, 2021, 7: e623. doi: 10.7717/peerj-cs.623 URL
[23]	BECTA. Elaadnl[EB/OL]. (2020-01-15)[2021-08-23]. https://platform.elaad.io/.

参数	名称	意义
X	真实样本	实际负荷数据
Y	条件样本	充电负荷影响因素
Z	噪声样本	随机噪声数据
r	学习率	收敛权重
d_max	最大深度	树结构的深度
w_min	最小支点量	分支节点的权重
λ	收缩步长	新特征权重
n	迭代次数	自我迭代的次数
w_n	基础回归器群	迭代中阶段性回归器群
F_n(x)	集成预测模型	迭代中阶段性预测模型

序号	算法名称	简称
1	Light Gradient Boosting Machin	LGBM
2	Extra Trees Regressor	Et
3	Random Forest	Rf
4	Gradient Boosting Regressor	Gbr
5	Decision Tree	Dt
6	Bayesian Ridge	Br
7	Ridge Regression	Ridge
8	AdaBoost Regressor	Ada
9	Lasso Regression	Lasso
10	Linear Regression	Lr
11	Elastic Net	En
12	Orthogonal Matching Pursuit	Omp
13	Lasso Least Angle Regression	Llar
14	K Neighbors Regressor	Knr
15	Huber Regressor	Huber
16	Passive Aggressive Regressor	Par

模型	MAE	MSE	RMSD	R²	MAPE	RMSE	TT/s
LGBM	1.4835	9.9652	3.1472	0.9271	1.4090	0.1603	0.9520
Et	1.4767	10.9566	3.2791	0.9166	1.1199	0.1542	19.9650
Rf	1.4707	11.0504	3.2915	0.9161	1.2410	0.1519	10.6240
Gbr	1.7289	11.1057	3.3130	0.9156	1.9650	0.1876	37.1450
Dt	1.7881	16.4038	3.9911	0.8900	1.4830	0.1817	1.2130
Br	3.6100	34.2069	5.8318	0.8016	7.4220	0.4518	247.2700
Ridge	3.6205	34.3932	5.8494	0.8005	7.3350	0.4535	1.9980
Ada	4.8668	37.4468	6.0925	0.7160	11.5790	0.5769	56.5100
Lasso	3.5851	38.7354	6.2008	0.7797	6.7480	0.4169	0.4880
Lr	4.0792	38.8846	6.2222	0.7782	8.4950	0.5090	29.1220
En	3.5522	39.6849	6.2760	0.7750	6.2390	0.4039	0.6320
Omp	3.4045	42.6831	6.5088	0.7602	4.9530	0.3717	1.5520
Llar	10.6004	202.9634	14.2317	-0.0014	26.0870	0.9852	0.8980
Knr	10.7520	226.0024	15.0233	-0.1179	25.1830	0.9985	9.7160
Huber	9.3039	234.1162	15.2823	-0.1545	14.4480	0.9020	10.5580
Par	14.9551	318.1632	17.6986	-0.5662	20.0780	1.3277	1.7940

模型	MAPE	R²/%	TT/s
BPNN	6.65	88.63	29.60
CNN-LSTM	4.21	94.67	334.27
ARIMA	6.98	84.71	197.00
EEB-LGBM	1.21	97.23	3.92