Multi-source fusion deep learning for electric vehicle charging station load forecasting and risk early warning

doi:10.16265/j.cnki.issn1003-3033.2026.04.0114

Abstract

Abstract:

With the continuous expansion of electric vehicle charging stations, the power grid faces increasing risks such as power overload, load fluctuations, and uneven demand distribution. To address these issues, this paper proposes an Hybrid Deep Fusion(HDF)-Long Short-Term Memory(LSTM)-based method for load forecasting and graded early warning. The method integrates LSTM, Gated Recurrent Unit(GRU), and Transformer architectures with multi-source data, including historical load, meteorological conditions, and traffic flow. Pearson correlation analysis and a dynamic weight allocation mechanism are employed to improve nonlinear feature representation. Based on the transformer capacity and simultaneity factor specified in the Code for Design of Electric Vehicle Charging Stations, a three-level early warning mechanism is developed for rapid alerting near critical thresholds. Results show that the proposed model outperforms eXtreme Gradient Boosting(XGBoost), GRU, LSTM, and Transformer models, with an Mean Squared Error(MSE) of 0.185 2, an Mean Absolute Error(MAE) of 0.2682, and an R² of 0.985 7. The model also shows good computational efficiency and application potential in charging station load forecasting and operational risk warning.

Key words: multi-source data fusion, deep learning, electric vehicle charging station, load forecasting, risk early warning

CLC Number:

X924.2安全监测技术与设备

Li Fu, Lyu Wei, Cheng Wenyan. Multi-source fusion deep learning for electric vehicle charging station load forecasting and risk early warning[J]. China Safety Science Journal, 2026, 36(4): 28-37.

Figures/Tables 18

Table 1

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Fig.8

Table 2

Fig.9

Fig.10

Fig.11

Fig.12

Fig.13

Fig.14

Table 3

Table 4

References 28

[1]	International Energy Agency. Global electric vehicle outlook 2025[R/OL]. [2025-09-15]. https://www.iea.org/reports/global-ev-outlook-2025.
[2]	黄匀飞, 魏志文, 余江盛, 等. 基于配置协调优化的多站融合供电调峰运行方法研究[J]. 电力系统自动化, 2024, 48(23): 46-53.
	Huang Yunfei, Wei Zhiwen, Yu Jiangsheng, et al. Research on the peak shaving operation method of multi-station Integrated power supply based on configuration coordination optimization[J]. Automation of Electric Power Systems, 2024, 48(23): 46-53.
[3]	GB/T 50966—2024 电动汽车充电站设计标准[S].
	GB/T 50966-2024 Design code for electric vehicle charging stations[S].
[4]	何淑波, 项薇, 石钟淼. 基于机器学习的电动汽车电池系统的风险预警[J]. 中国安全科学学报, 2023, 33(2): 159-165. doi: 10.16265/j.cnki.issn1003-3033.2023.02.1289
	He Shubo, Xiang Wei, Shi Zhongmiao. Risk early warning of electric vehicle battery system based on machine learning[J]. China Safety Science Journal, 2023, 33(2): 159-165. doi: 10.16265/j.cnki.issn1003-3033.2023.02.1289
[5]	Xie Tuo, Su Yang, Zhang Gang, et al. Optimizing peak-shaving cooperation among electric vehicle charging stations: a two-tier optimal dispatch strategy considering load demand response potential[J]. International Journal of Electrical Power & Energy Systems, 2024, 162: DOI: 10.1016/j.ijepes.2024.110228.
[6]	Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780. doi: 10.1162/neco.1997.9.8.1735 pmid: 9377276
[7]	He Xin, Zhao Wenlu, Gao Zhijun, et al. Short-term load forecasting by GRU neural network and DDPG algorithm for adaptive optimization of hyperparameters[J]. Electric Power Systems Research, 2025, 238: DOI: 10.1016/j.epsr.2024.111119.
[8]	张洪财, 胡泽春, 宋永华, 等. 考虑时空分布的电动汽车充电负荷预测方法[J]. 电力系统自动化, 2014, 38(1): 13-20.
	Zhang Hongcai, Hu Zechun, Song Yonghua, et al. A prediction method for electric vehicle charging load considering spatial and temporal distribution[J]. Automation of Electric Power Systems, 2014, 38(1): 13-20.
[9]	陆继翔, 张琪培, 杨志宏, 等. 基于CNN-LSTM混合神经网络模型的短期负荷预测方法[J]. 电力系统自动化, 2019, 43(13): 131-137.
	Lu Jixiang, Zhang Qipei, Yang Zhihong, et al. Short-term load forecasting method based on cnn-lstm hybrid neural network model[J]. Automation of Electric Power Systems, 2019, 43(13): 131-137.
[10]	Wei Ruizeng, Hu Siyu, Fan Yang, et al. New energy power prediction and warning based on multi-source prediction and scene classification recognition[J]. Procedia Computer Science, 2023, 224: 401-406. doi: 10.1016/j.procs.2023.09.055
[11]	Wu Sipei, Xiao Yao, Fu Shengxiang, et al. A hybrid deep learning model for load forecasting of electric vehicle charging stations using time series decomposition[J]. Journal of Power Sources, 2025, 655: DOI: 10.1016/j.jpowsour.2025.237882.
[12]	Aduama P, Zhang Zhibo, Al-sumaiti A S. Multi-feature data fusion-based load forecasting of electric vehicle charging stations using a deep learning model[J]. Energies, 2023, 16(3): DOI:10.3390/en16031309.
[13]	李存斌, 李庆良, 王庆林, 等. 基于多重分形去趋势波动分析的电力负荷风险预警阈值[J]. 电力系统自动化, 2016, 40(5): 1437-1441.
	Li Cunbin, Li Qingliang, Wang Qinglin, et al. Research on electric power load risk warning threshold based on multi-fractal detrended fluctuation analysis[J]. Automation of Electric Power Systems, 2016, 40(5): 1437-1441.
[14]	刘俊峰, 陈剑龙, 王晓生, 等. 基于深度强化学习的微能源网能量管理与优化策略研究[J]. 电力系统自动化, 2020, 44(14): 3794-3803.
	Liu Junfeng, Chen Jianlong, Wang Xiaosheng, et al. Energy management and optimization of micro-energy grid based on deep reinforcement learning[J]. Automation of Electric Power Systems, 2020, 44(14): 3794-3803.
[15]	Gao Dexin, Wang Yi, Zheng Xiaoyu, et al. A fault warning method for electric vehicle charging process based on adaptive deep belief network[J]. World Electric Vehicle Journal, 2021, 12(4): DOI:10.3390/wevj12040265.
[16]	Tang Minan, Guo Xi, Qiu Jiandong, et al. Electric vehicle charging load demand forecasting in different functional areas of cities with weighted measurement fusion UKF algorithm[J]. Energies, 2024, 17(17): DOI:10.3390/en17174505.
[17]	Yi Qiang, Zhou Yifei, Gebreyesus G D, et al. Multi-agent trips simulation-based charging demand forecasting[J]. IEEJ Transactions on Electrical and Electronic Engineering, 2025, 20(8): 1219-1228. doi: 10.1002/tee.v20.8
[18]	Shen Yu, Hu Wei, Dong Mingqi, et al. A risk warning method for steady-state power quality based on VMD-LSTM and fuzzy model[J]. Heliyon, 2024, 10(9): DOI:10.1016/j.heliyon.2024.e30597.
[19]	Zhou Haoyi, Zhang Shanghang, Peng Jieqi, et al. Informer: beyond efficient transformer for long sequence time-series forecasting[C]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021: 11 106-11 115.
[20]	Hogue J. EV charging station usage of California city[DB/OL]. [2025-09-15]. https://www.kaggle.com/datasets/venkatsairo4899/ev-charging-station-usage-of-california-city.
[21]	UCI. Metro interstate traffic volume[DB/OL].[2025-09-15]. https://archive.ics.uci.edu/dataset/492/metro+interstate+traffic+volume.
[22]	Cui Xuyang, Zhu Junda, Jia Lifu, et al. A novel heat load prediction model of district heating system based on hybrid whale optimization algorithm (WOA) and CNN-LSTM with attention mechanism[J]. Energy, 2024, 312: DOI: 10.1016/j.energy.2024.133536.
[23]	Madaram V G, Biswas P K, Sain C, et al. Advancement of electric vehicle technologies, classification of charging methodologies, and optimization strategies for sustainable development:a comprehensive review[J]. Heliyon, 2024, 10(20): DOI:10.1016/j.heliyon.2024.e39299.
[24]	熊智, 钟少波, 宋敦江, 等. 城市轨道交通客流量时间序列分段拟合方法[J]. 中国安全科学学报, 2018, 28(11): 35-41. doi: 10.16265/j.cnki.issn1003-3033.2018.11.006
	Xiong Zhi, Zhong Shaobo, Song Dunjiang, et al. A method of fitting urban rail transit passenger flow time series[J]. China Safety Science Journal, 2018, 28(11): 35-41. doi: 10.16265/j.cnki.issn1003-3033.2018.11.006
[25]	Xu Xianfeng, Wu Jiahao, Lu Yong, et al. A spatio-temporal prediction approach for charging load of clustered electric vehicles in dynamic traffic flow environment of highway[J]. Sustainable Energy, Grids and Networks, 2024:DOI:10.1016/j.segan.2024.101593..
[26]	Li Benxin, Chang Xuanming. EV Charging station load prediction in coupled urban transportation and distribution networks[J]. Energy Engineering, 2024, 121(10): 3001-3018. doi: 10.32604/ee.2024.051332
[27]	Li Xinli, Zhang Kui, Luo Zhenglong, et al. Two-stage dual-attention spatiotemporal joint network model for multi-energy load prediction of integrated energy system[J]. Sustainable Energy Technologies and Assessments, 2024, 72: DOI: 10.1016/j.seta.2024.104085.
[28]	Ba J L, Kiros J R, Hinton G E. Layer normalization[C]. Advances in Neural Information Processing Systems, 2016: 1-9.

类别	相关特征	特征作用分析
电量	交通流量, 交通密度	流量密度决定进出频率,影响负荷波动
充电时间	事件持续时间, 事件类型	事件属性影响停留时长与功率需求
位置	纬度,经度	地理位置反映区域差异,影响负荷分布
时区	时间,交通流量	时段与流量变化揭示负荷高低峰
天气	温度,降雨量, 降雪量,云量	天气条件影响出行充电,致负荷波动

负荷比例	预警等级	区域	响应机制
< 60% P_max	正常运行	绿色	正常运行, 储能保持备用
60%~80% P_max	二级预警	黄色	进入预备调度, 启用储能放电
≥ 80% P_max	三级预警	红色	启动限流或停电保护,并上报调度中心

模型	MSE	RMSE	MAE	R²	训练时间/s	预测时间/s
HDF-LSTM	0.185 2	0.427 2	0.268 2	0.985 7	94.72	0.31
XGBoost	0.265 1	0.514 8	0.384 9	0.979 3	0.169 0	0.003 6
CNN-LSTM	0.384 2	0.619 8	0.442 6	0.970 0	14.511 8	0.591 3
GRU	1.163 8	1.078 8	0.640 9	0.909 1	16.065 3	1.431 2
LSTM	4.223 7	2.055 2	1.455 3	0.670 2	17.913 4	2.262 8
Transformer	1.275 9	1.129 6	0.856 8	0.900 4	11.994 3	0.450 8

模型	精度得分	效率得分	总体得分
HDF-LSTM	1.75	0.50	1.38
XGBoost	1.40	1.00	1.28
CNN-LSTM	1.05	0.75	0.96
GRU	0.70	0.55	0.66
LSTM	0.35	0.45	0.38
Transformer	0.70	0.75	0.72