基于卷积神经网络的机舱风速修正

doi:10.3969/j.issn.1674-1951.2021.05.012

摘要/Abstract

摘要：

风电机组机舱风速计受到风机尾流和叶片扰动影响,国际电工委员会（IEC）提出的机舱传递函数无法准确描述机舱实测风速与来流风速的复杂关系。提出一种基于卷积神经网络的机舱风速修正模型,该模型采用多层卷积池化,可有效过滤风机尾流和叶片扰动的影响,高度抽象特征变量,提高修正风速的精度。工程实例表明：平原风电场中基于卷积神经网络法的计算风速与实测风速拟合精度R²达到0.844 7,平均绝对误差（MAE）仅为0.071,该方法计算的功率曲线与实测功率曲线评估电量差异仅为4.07%,各项偏差指标均优于IEC机舱传递函数,充分反映出该模型在机舱风速修正方面的优越性。

关键词: 风电机组, 卷积神经网络, IEC机舱传递函数, 机舱风速, 来流风速, 自由流风速, 功率曲线, 风速修正

Abstract:

The wind turbine nacelle anemometer is affected by the wake of turbines and disturbance of blade.International Electrotechnical Commission（IEC） indicated that nacelle transfer function cannot accurately describe the complex relationship between measured wind velocity and inflowing wind velocity.A nacelle wind speed correction model based on convolution neural network is proposed.The model adopting multi-layer convolution pooling can effectively filter the influence brought by turbine wake and blade disturbance,abstract feature variables,and improve the accuracy of the corrected wind speed.The engineering example shows that the fitting accuracy R² of the convolution neural network method is 0.844 7,and its mean absolute error （MAE） is only 0.071.The difference between the power calculated by this method and the one measured by anemometers is only 4.07%.All the deviation indexes are better than those made by IEC nacelle transfer function,which fully reflects the advantages of the wind speed correction model.

Key words: wind turbine, convolutional neural network, IEC nacelle transfer function, nacelle wind speed, inflowing wind velocity, free flow velocity, power curve, wind speed correction

中图分类号:

TK29

杨明明. 基于卷积神经网络的机舱风速修正[J]. 华电技术, 2021, 43(5): 75-79.

YANG Mingming. Wind speed correction for wind turbine based on convolutional neural network[J]. Huadian Technology, 2021, 43(5): 75-79.

图/表 12

图1

图2

表1

图3

图4

图5

图6

图7

图8

表2

图9

表3

参考文献 19

[1]	王晓宇, 丁同光, 摆念宗, 等. 风电机组风向仪测量误差分析与修正方法[J]. 分布式能源, 2019,4(6):57-62.
	WANG Xiaoyu, DING Tongguang, BAI Nianzong, et al. Error analysis and correction method of wind direction meter for wind turbine[J]. Distributed Energy, 2019,4(6):57-62.
[2]	姜全越. 我国风电后市场运营探究[J]. 经贸实践, 2018(18):204-205.
[3]	付立, 刘晓光. 基于激光雷达测风仪的风电场风电机组性能评估研究[J]. 华电技术, 2017,39(6):14-16,40.
	FU Li, LIU Xiaoguang. Wind power plant wind turbine unit performance evaluation research based on laser radar anemoscope[J]. Huadian Technology, 2017,39(6):14-16,40.
[4]	张岩, 王冬冬. 激光雷达测风系统在风电机组功率曲线测试中的应用研究[J]. 太阳能学报, 2017,38(6):1489-1494.
	ZHANG Yan, WANG Dongdong. Application of lidar wind measurement system for power curve measurement of wind turbine[J]. Acta Energiae Solaris Sinica, 2017,38(6):1489-1494.
[5]	王晓宇. 风力机尾流对测风仪数据精度的影响研究[D]. 西安:西安理工大学, 2018.
[6]	风力发电机组基于机舱风速计法的功率特性测试:GB/T 33225-2016[S]. 北京:中国机械工业联合会, 2016.
[7]	刘永前, 高小力, 韩爽, 等. 基于修正风速的风电场等效功率特性模型研究[J]. 华北电力大学学报, 2014,41(3):49-54.
	LIU Yongqian, GAO Xiaoli, HAN Shuang, et al. Study on equivalent power characteristic model of a wind farm based on corrected wind speed[J]. Journal of North China Electric Power University, 2014,41(3):49-54.
[8]	张新房, 徐大平, 柳亦兵, 等. 大型变速风力发电机组的风速软测量[J]. 太阳能学报, 2006(4):321-325.
	ZHANG Xinfang, XU Daping, LIU Yibing, et al. Wind speed soft sensor for large-scale variable speed wind turbine[J]. Acta Energiae Solaris Sinica, 2006(4):321-325.
[9]	高昆仑, 杨帅, 刘思言, 等. 基于一维卷积神经网络的电力系统暂态稳定评估[J]. 电力系统自动化, 2019,43(12):18-26.
	GAO Kunlun, YANG Shuai, LIU Siyan, et al. Transient stability assessment for power system based on one-dimensional convolutional neural network[J]. Automation of Electric Power Systems, 2019,43(12):18-26.
[10]	李东东, 王浩, 杨帆, 等. 基于一维卷积神经网络和Soft-Max分类器的风电机组行星齿轮箱故障检测[J]. 电机与控制应用, 2018,45(6):80-87,108.
	LI Dongdong, WANG Hao, YANG Fan, et al. Fault detection of wind turbine planetary gear box using 1D convolution neural networks and Soft-Max classifier[J]. Electric Machines & Control Application, 2018,45(6):80-87,108.
[11]	沈小军, 付雪姣, 周冲成, 等. 风电机组风速-功率异常运行数据特征及清洗方法[J]. 电工技术学报, 2018,33(14):3353-3361.
	SHEN Xiaojun, FU Xuejiao, ZHOU Chongcheng, et al. Characteristics of outliers in wind speed-power operation data of wind turbines and its cleaning method[J]. Transactions of China Electrotechnical Society, 2018,33(14):3353-3361.
[12]	KHOURY E, CAZACLIU B, REMOND S. Control of effective water in recycled aggregate concrete using power curves of the mixer[J]. Materials Today Communications, 2019,21.
[13]	袁红亮, 胡义. 基于运行数据的风电机组功率曲线可靠性评估[J]. 西北水电, 2018(5):80-83.
	YUAN Hongliang, HU Yi. Operation-data-based assessment of reliability of WTG power curve[J]. Northwest Hydropower, 2018(5):80-83.
[14]	贾锋, 李征, 蔡旭. 提高大型风电机组恒转速段发电量的变桨策略[J]. 电工技术学报, 2017,32(1):58-68.
	JIA Feng, LI Zheng, CAI Xu. Advanced pitch control for improving power production for large scale wind energy conversion system under constant speed region[J]. Transactions of China Electrotechnical Society, 2017,32(1):58-68.
[15]	周志超, 王成山, 郭力, 等. 变速变桨距风电机组的全风速限功率优化控制[J]. 中国电机工程学报, 2015,35(8):1837-1844.
	ZHOU Zhichao, WANG Chengshan, GUO Li, et al. Output power curtailment control of variable-speed variable-pitch wind turbine generator at all wind speed regions[J]. Proceedings of the CSEE, 2015,35(8):1837-1844.
[16]	张树鑫. 双馈风力发电机变桨与变速协调控制[D]. 兰州:兰州理工大学, 2011.
[17]	姜寅, 任荭葳, 陈凯, 等. 基于BP神经网络的风电系统控制器I/O硬件故障自诊断方法[J]. 华电技术, 2020,42(5):55-60.
	JIANG Yin, REN Hongwei, CHEN Kai, et al. Self-detecting method for faults in wind turbine controller I/O hardware based on BP neural network algorithm[J]. Huadian Technology, 2020,42(5):55-60.
[18]	解加盈, 郭鹏. 基于多变量选择的深度神经网络功率曲线建模[J]. 华电技术, 2019,41(8):27-31,52.
	XIE Jiaying, GUO Peng. Deep neural network modeling on power curve based on multi-variable selection[J]. Huadian Technology, 2019,41(8):27-31,52.
[19]	杨明明. 基于机舱传递函数的风机功率曲线研究[J]. 华电技术, 2020,42(5):50-54.
	YANG Mingming. Research on wind turbine power curve based on nacelle transfer function[J]. Huadian Technology, 2020,42(5):50-54.

项目	数据
叶轮直径/m	104
轮毂高度/m	80
RIX	≤0.04
地形倾角	符合附录B
最大脊台阶/m	4

计算方法	回归平方和（ESS）	平均绝对误差（MAE）	拟合优度R²
IEC机舱传递函数法	1 266.54	0.098	0.669 2
卷积神经网络法	757.12	0.071	0.844 7

风电场模型	发电量/（kW⋅h）	误差率/%
测试功率曲线模型	4 939 085.471	-
卷积神经网络功率曲线模型	5 139 934.554	4.07
IEC机舱传递函数功率曲线模型	5 225 233.242	5.79