电工技术学报  2024, Vol. 39 Issue (3): 714-730    DOI: 10.19595/j.cnki.1000-6753.tces.222009
电力系统与综合能源 |
基于校正条件生成对抗网络的风电场群绿氢储能系统容量配置
朱玲1,2, 李威1,2, 王骞3, 张学广3
1.智能电网保护和运行控制国家重点实验室 南京 211106;
2.南瑞集团有限公司(国网电力科学研究院有限公司) 南京 211106;
3.哈尔滨工业大学电气工程及自动化学院 哈尔滨 150001
Wind Farms-Green Hydrogen Energy Storage System Capacity Sizing Method Based on Corrected-Conditional Generative Adversarial Network
Zhu Ling1,2, Li Wei1,2, Wang Qian3, Zhang Xueguang3
1. State Key Laboratory of Smart Grid Protection and Control Nanjing 211106 China;
2. Nari Group Corporation (State Grid Electric Power Research Institute) Nanjing 211106 China;
3. Department of Electrical Engineering Harbin Institute of Technology Harbin 150001 China
全文: PDF (5782 KB)   HTML
输出: BibTeX | EndNote (RIS)      
摘要 以条件生成对抗网络(CGAN)为代表的半监督学习可计及风电波动并生成出力场景集合,生成的数据可输入氢气储能容量配置模型以支撑优化求解。为此,该文首先设计一种校正条件生成对抗网络(CCGAN),并基于风电预测误差构建条件校正器,对预测失准事件和风电爬坡事件下输入生成器的标签信息进行识别和校正;然后,以储能定容的综合成本和各风场弃风成本为目标函数,构建绿氢储能容量配置的多目标优化模型,并引入基于切比雪夫距离的膝区域数学概念,以指导多目标优化算法设计;最后,以新英格兰39节点系统为例进行算例分析,结果表明未经校正的条件信息将导致定容决策偏离实际,而CCGAN能生成计及风电不确定性的高置信出力场景,使得容量配置结果兼顾鲁棒性和经济性。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
朱玲
李威
王骞
张学广
关键词 校正条件生成对抗网络绿氢储能容量配置场景生成风电预测误差    
Abstract:The wind power penetration rate in China is increasing. In order to realize the efficient consumption of wind power, the green hydrogen energy storage system (ESS) is becoming an essential carrier of energy storage. However, the intermittence, randomness, and volatility of wind intraday power make it difficult to produce and store green hydrogen. Providing scenario data for green hydrogen ESS planning and displaying the intraday output fluctuation of wind power is a critical challenge. To address these issues, this paper designs a green hydrogen ESS capacity planning framework based on wind power scenario generation. The corrected-conditional generative adversarial network (CCGAN) is constructed to generate scenario data.
Firstly, a corrector based on the improved revolving door algorithm is designed in the CCGAN to identify and clean the prediction data under the prediction misaim and wind power ramp events to ensure the high reference of conditional information. Secondly, the dilated convolution is embedded in the generator and discriminator to overcome the problem of the insufficient receptive field. Thirdly, by imposing constraints on the production and storage of green hydrogen, a multi-objective optimization model for green hydrogen ESS sizing is developed. The definition of knee region under Chebyshev distance is introduced. The non-dominated sorting genetic algorithm based on knee region (kr-NSGA-Ⅲ) is designed to solve the capacity sizing model. Finally, the simulation based on the New England 39-bus system verifies the generated scenario’s accuracy and the planning model’s effectiveness.
Simulation results on the actual wind farm data show that, when predictions are accurate, both conditional generative adversarial network (CGAN) and CCGAN have higher coverage for measured data. At the same confidence level, the power fluctuation range given by CCGAN is smaller, which reflects that the method has higher accuracy for random fluctuation. Under the prediction misaim event, the coverage of CCGAN is much higher than that of CGAN, and the fluctuation range is reduced by 12.4 MW at a 100% confidence level. Under the wind power ramp event, the mean absolute percentage error of the scenario data generated by CCGAN is kept within 30%. At a 90% confidence level, CCGAN can still have a coverage of more than 80%, reflecting the ability to characterize the short-term fluctuation of wind power. In the four generative scenarios, the sizing results of CCGAN are compared with the measured values, CGAN, robust optimization, and Markov chain Monte Carlo (MCMC). The sizing results of CCGAN are generally smaller than CGAN and robust optimization but larger than the measured values, and the total economic cost is also lower than MCMC. Robust optimization and MCMC are easy to overestimate the uncertainty of wind power because these methods do not use the statistical value of prediction data. The scenario generated by the proposed method reasonably considers wind power fluctuation based on measured data, and the sizing result is a tradeoff between robustness and economy.
The following conclusions can be drawn from the simulation analysis: (1) By employing the improved revolving door algorithm, the CCGAN’s corrector can accurately identify the prediction misaim and wind power ramp events and perform targeted correction on the conditional information of the input generator, providing a reliable label for data generation. (2) Compared with the robust optimization, MCMC, and CGAN, the wind power scenario generation based on CCGAN can capture the fluctuation characteristics of wind power, and the generated data can consider both high confidence and narrow fluctuation range, providing a high-quality scenario set for green hydrogen ESS planning. (3) The results of green hydrogen ESS sizing show that under the samples with frequent prediction misaim events, the scenario of CGAN will lead to the deviation of planning results from reality. Compared with robust optimization, the planning results of the proposed method can reduce investment costs.
Key wordsCorrected-conditional generative adversarial network    green hydrogen energy storage    capacity sizing    scenario generation    wind power forecasting error   
收稿日期: 2022-10-26     
PACS: TM614  
基金资助:甘肃省科技计划(21ZD8JA001)、智能电网保护和运行控制国家重点实验室开放课题(考虑风功率波动和调频能力的风电场站储能配置与运行优化技术)资助项目
通讯作者: 朱 玲 女,1986年生,硕士,高级工程师,研究方向为新能源电力系统稳定及控制,储能规划与运行。E-mail:zhuling@sgepri.sgcc.com.cn   
作者简介: 李 威 男,1976年生,博士,研究员级高级工程师,研究方向为电力系统安全稳定分析与控制。E-mail:liwei10@sgepri.sgcc.com.cn
引用本文:   
朱玲, 李威, 王骞, 张学广. 基于校正条件生成对抗网络的风电场群绿氢储能系统容量配置[J]. 电工技术学报, 2024, 39(3): 714-730. Zhu Ling, Li Wei, Wang Qian, Zhang Xueguang. Wind Farms-Green Hydrogen Energy Storage System Capacity Sizing Method Based on Corrected-Conditional Generative Adversarial Network. Transactions of China Electrotechnical Society, 2024, 39(3): 714-730.
链接本文:  
https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.222009          https://dgjsxb.ces-transaction.com/CN/Y2024/V39/I3/714