基于深度强化学习与改进Jensen模型的风电场功率优化

doi:10.11930/j.issn.1004-9649.202410051

中国电力 ›› 2025, Vol. 58 ›› Issue (4): 78-89.DOI: 10.11930/j.issn.1004-9649.202410051

• 风电机组暂态运行控制与试验验证关键技术 • 上一篇下一篇

基于深度强化学习与改进Jensen模型的风电场功率优化

王冠朝¹(), 霍雨翀¹(), 李群²(), 李强²()

1. 南京理工大学电气工程系，江苏南京　210094
2. 国网江苏省电力有限公司电力科学研究院，江苏南京　211103

收稿日期:2024-10-15 录用日期:2025-01-13 发布日期:2025-04-23 出版日期:2025-04-28
作者简介:
王冠朝（2001），男，硕士研究生，从事风电参与调频、新能源并网与控制研究，E-mail：wangguanchao2001@163.com
霍雨翀（1989），男，通信作者，博士，讲师，从事电力系统运行与控制技术、电力经济研究，E-mail：yuchong.huo@njust.edu.cn
李群（1967），男，博士，高级工程师（教授级），从事电力系统运行和可再生能源并网研究，E-mail：qun_li@sina.com
李强（1981），男，博士，高级工程师（教授级），从事远海风电的柔性直流输电研究，E-mail：35830342@qq.com
基金资助:
国家电网有限公司科技项目（攻关团队项目）（含多构网型变流器的中远海风电场经柔直并网主动频率支撑关键技术，5108-202218280A-2-241-XG）。

Power Optimization of Wind Farms Based on Improved Jensen Model and Deep Reinforcement Learning

WANG Guanchao¹(), HUO Yuchong¹(), LI Qun²(), LI Qiang²()

1. Department of Electrical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
2. Electric Power Research Institute of State Grid Jiangsu Electric Power Co., Ltd., Nanjing 211103, China

Received:2024-10-15 Accepted:2025-01-13 Online:2025-04-23 Published:2025-04-28
Supported by:
This work is supported Science and Technology Project of SGCC (Reseach Team Project) (Active Frequency Support for Mid and Long Distance Offshore Wind Farm with Multiple Grid-Forming Converter Connected via VSC-HVDC, No.5108-202218280A-2-241-XG).

摘要/Abstract

摘要：

风电场的功率捕获能力受多种因素的制约。为最大化风电场的功率输出，并应对尾流效应和湍流风速的影响，提出一种基于深度强化学习的风电场控制方案。该方案结合有模型与无模型的控制方法，并整合至基于Actor-Critic架构的深度确定性策略梯度强化学习网络中。在控制精度方面，采用改进的Jensen尾流模型，通过考虑时间延迟，提升了尾流效应的精确性，并有效捕捉了风电场长期功率输出。仿真结果表明，相比于传统单纯的有模型或者无模型方法，所提方法有效提升了风电场的最大功率输出，同时在保证控制精度的基础上，显著降低了训练时间和计算资源消耗，从而提升了控制策略的整体性能。

关键词: 风电场控制, 最大化风能捕获, 深度强化学习, 无模型控制, 有模型控制, 神经网络

Abstract:

The power capture capability of wind farms is constrained by various factors. To maximize the power output of wind farms and address the impacts of wake effects and turbulent wind speeds, this paper proposes a wind farm control scheme based on deep reinforcement learning. This scheme combines both model-based and model-free control methods and integrates them into a deep reinforcement learning deep deterministic policy gradient network with an Actor-Critic architecture. In terms of control accuracy, Jensen wake model consider time delay is adopted to enhance the precision of wake effects and effectively captures the long-term impact on the wind farm's power output. Simulation results show that, compared to traditional model-based or model-free methods, this scheme significantly increases the maximum power output of the wind farm while maintaining control accuracy, and significantly reduces training time and computational resource consumption, thereby improving the overall performance of the control strategy.

Key words: wind farm control, maximizing wind energy capture, deep reinforcement learning, model-free control, model-based control, neural network

王冠朝, 霍雨翀, 李群, 李强. 基于深度强化学习与改进Jensen模型的风电场功率优化[J]. 中国电力, 2025, 58(4): 78-89.

WANG Guanchao, HUO Yuchong, LI Qun, LI Qiang. Power Optimization of Wind Farms Based on Improved Jensen Model and Deep Reinforcement Learning[J]. Electric Power, 2025, 58(4): 78-89.

导出引用管理器 EndNote|Ris|BibTeX

链接本文: https://www.electricpower.com.cn/CN/10.11930/j.issn.1004-9649.202410051

https://www.electricpower.com.cn/CN/Y2025/V58/I4/78

图/表 17

图 1 Jensen尾流模型示意

Fig.1 Illustration of the Jensen wake model

图 2 深度强化学习控制算法框架

Fig.2 Deep reinforcement learning control algorithm framework

图 3 Actor-Critic网络结构

Fig.3 Actor-Critic network structure

图 4 本文提出的结合有模型与无模型方法的深度强化学习框架

Fig.4 The proposed deep reinforcement learning framework combined model-based and model-free methods

图 5 风电场与深度强化学习结合框架

Fig.5 The framework for combining wind farms with deep reinforcement learning

图 6 风电场拓扑

Fig.6 Wind farm topology

图 7 2种模型的输出功率对比

Fig.7 Comparison of output power between two models

表 1 仿真参数设置

Table 1 Simulation parameters setting

参数		数值
$D$/m, ${d_{\rm H}}$/m, $\rho $/(kg·m^–3), $A$/m³		126, 500, 1.225, 7850
$n$, $\tau $, ${\eta _{\rm actor}}$, ${\eta _{\rm critic}}$, $m$, $\gamma $		64, 0.001, 0.01, 0.01, 10⁴, 0.99
迭代轮数, 步数		1000, 10
$C_{\rm T}^{\min }$, $C_{\rm T}^{\max }$		0.1,1

图 8 风速曲线

Fig.8 Wind velocity curve

图 9 不同方法奖励对比

Fig.9 Comparison of rewards using different methods

图 10 风电场输出功率对比

Fig.10 Comparison of wind farm output power

图 11 风电场尾流对比

Fig.11 Comparison of wind farm wake

图 12 归一化推力系数变化曲线

Fig.12 Normalized thrust coefficient variation curve

图 13 风机输出功率曲线

Fig.13 The power curve of wind turbine output

图 14 桨距角曲线

Fig.14 Pitch angle curve

图 15 转速曲线

Fig.15 Rotational speed curve

图 16 风机捕获功率对比

Fig.16 Comparison of wind turbine capture power

参考文献 25

1	BOLINGER M, WISER R. Understanding wind turbine price trends in the U. S. over the past decade[J]. Energy Policy, 2012, 42, 628- 641. DOI
2	王俊, 段斌, 苏永新. 基于尾流效应的海上风电场有功出力优化[J]. 电力系统自动化, 2015, 39 (4): 26- 32, 75. DOI
	WANG Jun, DUAN Bin, SU Yongxin. Optimization of active power output in offshore wind farms based on wake effect[J]. Automation of Electric Power Systems, 2015, 39 (4): 26- 32, 75. DOI
3	刘玉山, 胡阔海, 王灵梅, 等. 考虑尾流效应的风电场输出功率优化[J]. 可再生能源, 2023, 41 (10): 1336- 1342. DOI
	LIU Yushan, HU Kuohai, WANG Lingmei, et al. Optimization of wind farm output power considering wake effect[J]. Renewable Energy Resources, 2023, 41 (10): 1336- 1342. DOI
4	高丙团, 胡正阳, 张磊, 等. 考虑尾流效应的风电场减载出力优化控制[J]. 可再生能源, 2018, 36 (1): 117- 125. DOI
	GAO Bingtuan, HU Zhengyang, ZHANG Lei, et al. De-loading optimal control of wind farm based on wake effect[J]. Renewable Energy Resources, 2018, 36 (1): 117- 125. DOI
5	朱成名, 魏云冰, 蒋成成, 等. 基于PSO-BP神经网络的风电功率备用容量估计模型[J]. 电子科技, 2021, 34 (12): 36- 41.
	ZHU Chengming, WEI Yunbing, JIANG Chengcheng, et al. Estimation model of wind power reserve capacity based on PSO-BP neural network[J]. Electronic Science and Technology, 2021, 34 (12): 36- 41.
6	DAR Z, KAR K, SAHNI O, et al. Windfarm power optimization using yaw angle control[J]. IEEE Transactions on Sustainable Energy, 2017, 8 (1): 104- 116. DOI
7	VALI M, PETROVIĆ V, BOERSMA S, et al. Adjoint-based model predictive control for optimal energy extraction in waked wind farms[J]. Control Engineering Practice, 2019, 84, 48- 62. DOI
8	YIN X X, ZHAO X W. Deep neural learning based distributed predictive control for offshore wind farm using high-fidelity LES data[J]. IEEE Transactions on Industrial Electronics, 2020, 68 (4): 3251- 3261.
9	ZHOU Q, LI W B, LIU C B, et al. Strategy of wind farm participating in power system secondary frequency regulation based on model predictive control[C]//2021 IEEE Sustainable Power and Energy Conference (iSPEC). Nanjing, China. IEEE, 2021: 524-531.
10	AHMAD M, AZUMA S I, SUGIE T. A model-free approach for maximizing power production of wind farm using multi-resolution simultaneous perturbation stochastic approximation[J]. Energies, 2014, 7 (9): 5624- 5646. DOI
11	ZHIWEI X, HUA G, BING C. Model-free power optimization of wind farm based on nelder-mead method [M]. 2020.
12	VIJAYSHANKAR S, STANFEL P, KING J, et al. Deep reinforcement learning for automatic generation control of wind farms[C]//2021 American Control Conference (ACC). New Orleans, LA, USA. IEEE, 2021: 1796–1802.
13	YAO Q, ZHANG Y M, SUN S X, et al. Optimization of energy efficiency for offshore wind farms via wake modeling-free NMPC[J]. Ocean Engineering, 2024, 305, 117923. DOI
14	DONG H Y, ZHAO X W. Data-driven wind farm control via multiplayer deep reinforcement learning[J]. IEEE Transactions on Control Systems Technology, 2022, 31 (3): 1468- 1475.
15	DONG H Y, ZHANG J C, ZHAO X W. Intelligent wind farm control via deep reinforcement learning and high-fidelity simulations[J]. Applied Energy, 2021, 292, 116928. DOI
16	HUANG Y B, LIN S Y, ZHAO X W. Multi-agent reinforcement learning control of a hydrostatic wind turbine-based farm[J]. IEEE Transactions on Sustainable Energy, 2023, 14 (4): 2406- 2416. DOI
17	袁飞. 风电场常用工程尾流模型对比与分析[J]. 能源与节能, 2023, (11): 37- 41,58. DOI
	YUAN Fei. Comparison and analysis of commonly used engineering wake models in wind farms[J]. Energy and Energy Conservation, 2023, (11): 37- 41,58. DOI
18	朱洁, 匡婵, 赵宜婵. 基于Jensen和Gaussian尾流模型的风电场布局优化的比较[J]. 河南科学, 2021, 39 (3): 345- 352. DOI
	ZHU Jie, KUANG Chan, ZHAO Yichan. Comparison of wind farm layout optimizations based on Jensen and Gaussian wake models[J]. Henan Science, 2021, 39 (3): 345- 352. DOI
19	ARCHER C L, VASEL-BE-HAGH A, YAN C, et al. Review and evaluation of wake loss models for wind energy applications[J]. Applied Energy, 2018, 226, 1187- 1207. DOI
20	XU C, YIN M H, LI Q, et al. Wind farm power maximization based on analytical sensitivity model considering wake effect[J]. Electric Power Systems Research, 2023, 224, 109734. DOI
21	SERRANO GONZÁLEZ J, BURGOS PAYÁN M, RIQUELME SANTOS J, et al. Maximizing the overall production of wind farms by setting the individual operating point of wind turbines[J]. Renewable Energy, 2015, 80, 219- 229. DOI
22	ZHANG H Q, MA H B, ZHANG X F, et al. An efficient and lightweight off-policy actor–critic reinforcement learning framework[J]. Applied Soft Computing, 2024, 163, 111814. DOI
23	李强, 邹小明, 任必兴, 等. 基于Actor-Critic框架的风机换流器参数优化策略[J]. 现代电力, 1–10[2025-04-09]. http://doi.org/10.19725/j.cnki.1007-2322.2023.0310.
	LI Qiang, ZOU Xiaoming, REN Bixing, et al. Parameter optimization strategy for wind generator converters based on actor-critic framework[J]. Modern Electric Power, 2024, 1–10[2025-04-09]. http://doi.org/10.19725/j.cnki.1007-2322.2023.0310.
24	BOERSMA S. Towards closed-loop dynamical wind farm control: model development and control applications [D]. Delft: Delft University of Technology, 2019.
25	NREL. NREL 5-MW baseline wind turbine[EB/OL]. https://nrel.github.io/turbine-models/NREL_5MW_126_RWT.html.

[1]	李鹏, 祖文静, 刘一欣, 田春筝, 郝元钊, 李慧璇. 基于不完全量测数据的配电网状态估计方法[J]. 中国电力, 2025, 58(5): 1-10.
[2]	樊会丛, 段志国, 陈志永, 朱士加, 刘航, 李文霄, 杨阳. 基于多智能体深度策略梯度的离网型微电网双层优化调度[J]. 中国电力, 2025, 58(5): 11-20, 32.
[3]	沈鑫, 王钢, 赵毅涛, 骆钊, 李钊, 杨晓华. 融合SENet注意力机制和GA-CNN的非侵入式负荷识别方法[J]. 中国电力, 2025, 58(5): 33-42.
[4]	王力, 蒋宇翔, 曾祥君, 赵斌, 李均昊. 基于深度强化学习的孤岛微电网二次频率控制[J]. 中国电力, 2025, 58(5): 176-188.
[5]	季湛洋, 胡阳, 孔令行, 宋子秋, 邓丹, 刘吉臻. 考虑多领域耦合特性的风电机组一次调频动态建模与仿真[J]. 中国电力, 2025, 58(4): 56-67.
[6]	曾仪, 周毅, 陆继翔, 周良才, 唐宁恺, 李红. 基于多智能体安全深度强化学习的电压控制[J]. 中国电力, 2025, 58(2): 111-117.
[7]	吴小忠, 肖立华, 童超, 夏向阳, 袁翎, 甘星, 江志文, 黄湘源. 基于传递熵与JS-BP神经网络的锂离子电池容量预测模型[J]. 中国电力, 2025, 58(2): 186-192, 215.
[8]	王宇飞, 杜桐, 边伟国, 张钊, 刘慧婷, 杨丽君. 基于DTW K-medoids与VMD-多分支神经网络的多用户短期负荷预测[J]. 中国电力, 2024, 57(6): 121-130.
[9]	吴军英, 路欣, 刘宏, 张彬, 柴守亮, 刘蕴春, 王佳楠. 基于Spearman-GCN-GRU模型的超短期多区域电力负荷预测[J]. 中国电力, 2024, 57(6): 131-140.
[10]	邵必林, 纪丹阳. 基于VMD-SE的电力负荷分量的多特征短期预测[J]. 中国电力, 2024, 57(4): 162-170.
[11]	陈巳阳, 韩利, 方济中, 丁五行, 成诚, 李文, 张源, 钱勇. 基于不确定参数的变电站碳储量预估方法[J]. 中国电力, 2024, 57(4): 200-210.
[12]	娄奇鹤, 李荣盛, 谭捷, 袁铁江. 基于卷积神经网络的暂稳极限功率计算[J]. 中国电力, 2024, 57(4): 211-219.
[13]	马汝括, 董杰, 王雅湉, 伊国鑫, 丁祥浩, 马乐. 基于神经网络的高寒地区CF₄和SF₆/CF₄检测[J]. 中国电力, 2024, 57(3): 103-112.
[14]	阎洁, 杨佳琳, 王航宇, 卢姣阳, 刘永前, 张磊. 基于风况预测误差自适应的海上风电场尾流偏转控制方法[J]. 中国电力, 2024, 57(3): 190-196.
[15]	李卓, 王胤喆, 叶林, 罗雅迪, 宋旭日, 张振宇. 从感知-预测-优化综述图神经网络在电力系统中的应用[J]. 中国电力, 2024, 57(12): 2-16.

基于深度强化学习与改进Jensen模型的风电场功率优化

Power Optimization of Wind Farms Based on Improved Jensen Model and Deep Reinforcement Learning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 17

参考文献 25

相关文章 15

编辑推荐

Metrics

模态框（Modal）标题

基于深度强化学习与改进Jensen模型的风电场功率优化

Power Optimization of Wind Farms Based on Improved Jensen Model and Deep Reinforcement Learning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 17

参考文献 25

相关文章 15

编辑推荐

Metrics