• Peer Review
  • Non-profit
  • Global Open Access
  • Green Channel for Rising Stars
Volume 8 Issue 1
Mar.  2021
Turn off MathJax
Article Contents

ZHENG Kan,WEI Yufeng,WEN Zhisheng,et al.Analysis and Application of Wind Speed Interpolation in Wind Farm Based on BP Neural Network Method[J].Southern Energy Construction,2021,08(01):51-55. doi:  10.16516/j.gedi.issn2095-8676.2021.01.007
Citation: ZHENG Kan,WEI Yufeng,WEN Zhisheng,et al.Analysis and Application of Wind Speed Interpolation in Wind Farm Based on BP Neural Network Method[J].Southern Energy Construction,2021,08(01):51-55. doi:  10.16516/j.gedi.issn2095-8676.2021.01.007

Analysis and Application of Wind Speed Interpolation in Wind Farm Based on BP Neural Network Method

doi: 10.16516/j.gedi.issn2095-8676.2021.01.007
  • Received Date: 2020-08-25
  • Rev Recd Date: 2020-09-16
  • Publish Date: 2021-03-25
  •   Introduction  Accurate wind resource data is of great significance to wind resource evaluation and power generation calculation of wind farm. Due to mechanical failure, weather factors and human influence, there are many problems in wind speed data acquisition, such as short collection time, many discontinuities,data distortion and so on,which bring a lot of trouble to the evaluation of wind resources.  Method  At present,the traditional interpolation method based on MCP method for discontinuous data interpolation and fitting in the wind power industry is not accurate enough. In this paper, the wind resource data prediction scheme based on neural network algorithm was proposed for wind speed interpolation of wind turbine and wind speed interpolation of wind measurement mast.  Result  The results show that the interpolation effect of BP neural network is better than the traditional interpolation method,and the wind speed interpolation of anemometer tower in flat terrain is better than that in complex terrain.  Conclusion  The research shows that the wind speed interpolation technology based on BP neural network method is suitable for wind speed interpolation application of wind farm, and the accuracy of wind resource assessment is significantly improved.
  • [1] 杨靖文,张静. 不同MCP方法在平原及山地风电项目中的插补应用 [J]. 风能,2020(6):60-66.

    YANGJ W,ZHANGJ. Interpolation application of different MCP methods in plain and mountain wind power projects [J]. Wind Energy,2020(6):60-66.
    [2] 王远,钟华. 风电场缺失测风数据插补方法的分析 [J]. 可再生能源,2012,30(3):14-17+21.

    WANGY,ZHONGH. Analysis of interpolation method for missing wind measurement data in wind farm [J]. Renewable Energy,2012,30(3):14-17+21.
    [3] 朱晓玲,董金凤,王朝辉. 基于BP神经网络方法的测风塔缺测数据插补分析 [J]. 风能,2018(12):78-80.

    ZHUX L,DONGJ F,WANGC H. Interpolation analysis of missing measurement data of anemometer tower based on BP neural network method [J]. Wind Energy,2018(12):78-80.
    [4] 罗恩博,苏适,陆海,等. BP神经网络在风电场功率预测中的应用 [J]. 云南电力技术,2018,46(4):32-34+37.

    LUOE B, SUS, LUH, et al. Based on improved BP neural network in wind farm Application of power prediction [J]. Yunnan Electric Power, 2018, 46(4):32-34+37.
    [5] 向健平,凌永志,詹俊,等. 基于SCADA系统的风电机组主轴承故障预警方法 [J]. 电力科学与技术学报,2019,34(3):223-228.

    XIANGJ P, LINGY Z, ZHANJ, et al. Fault early warning method of wind turbine main bearing based on SCADA system [J]. Journal of Electric Power Science and Technology,2019,34(3):223-228.
    [6] Taiyuan,China,June 24,2017.Taiyuan:Research Institute of Management Science and Industrial Engineering,2017:564-567.
    [7] HONGY Y, SATRIANIT R A. Day-ahead spatiotemporal wind speed forecasting using robust design-based deep learning neural network [J]. Energy, 2020,209(C):118441.
    [8] LIUM S,CAOZ M,ZHANGJ,et al. Short-term wind speed forecasting based on the jaya-SVM model [J]. International Journal of Electrical Power and Energy Systems,2020,121(8),106056.
    [9] LIUX J,ZHANGH,KONGX B,et al. Wind speed forecasting using deep neural network with feature selection [J]. Neurocomputing,2020,397(7):393-403.
    [10] 韩庆生. TensorFlow与Pytorch环境的搭建 [J]. 计算机产品与流通,2020(5):124.

    HANQ S. Tensorflow and pytoch environment construction [J]. Computer Products and Circulation,2020(5):124.
    [11] EZZATA A. Turbine-specific short-term wind speed forecasting considering within-farm wind field dependencies and fluctuations [J]. Applied Energy,2020,269(C),115034.
    [12] 王红光,赵海军,段晓辉. 复杂地形风电场微观选址实践探讨 [J]. 南方能源建设,2019,6(2):59-63.

    WANGH G,ZHAOH J,DUANX H. Discussion on micro site selection of wind farm in complex terrain [J]. Southern Power System Technology,2019,6(2):59-63.
  • 通讯作者: 陈斌, bchen63@163.com
    • 1. 

      沈阳化工大学材料科学与工程学院 沈阳 110142

    1. 本站搜索
    2. 百度学术搜索
    3. 万方数据库搜索
    4. CNKI搜索

Figures(7)  / Tables(2)

Article Metrics

Article views(426) PDF downloads(54) Cited by()

Related

Analysis and Application of Wind Speed Interpolation in Wind Farm Based on BP Neural Network Method

doi: 10.16516/j.gedi.issn2095-8676.2021.01.007

Abstract:   Introduction  Accurate wind resource data is of great significance to wind resource evaluation and power generation calculation of wind farm. Due to mechanical failure, weather factors and human influence, there are many problems in wind speed data acquisition, such as short collection time, many discontinuities,data distortion and so on,which bring a lot of trouble to the evaluation of wind resources.  Method  At present,the traditional interpolation method based on MCP method for discontinuous data interpolation and fitting in the wind power industry is not accurate enough. In this paper, the wind resource data prediction scheme based on neural network algorithm was proposed for wind speed interpolation of wind turbine and wind speed interpolation of wind measurement mast.  Result  The results show that the interpolation effect of BP neural network is better than the traditional interpolation method,and the wind speed interpolation of anemometer tower in flat terrain is better than that in complex terrain.  Conclusion  The research shows that the wind speed interpolation technology based on BP neural network method is suitable for wind speed interpolation application of wind farm, and the accuracy of wind resource assessment is significantly improved.

ZHENG Kan,WEI Yufeng,WEN Zhisheng,et al.Analysis and Application of Wind Speed Interpolation in Wind Farm Based on BP Neural Network Method[J].Southern Energy Construction,2021,08(01):51-55. doi:  10.16516/j.gedi.issn2095-8676.2021.01.007
Citation: ZHENG Kan,WEI Yufeng,WEN Zhisheng,et al.Analysis and Application of Wind Speed Interpolation in Wind Farm Based on BP Neural Network Method[J].Southern Energy Construction,2021,08(01):51-55. doi:  10.16516/j.gedi.issn2095-8676.2021.01.007
  • 风机的平稳运行及发电量的计算对风资源的准确性要求高,准确及时的风资源数据对风场的风资源评估和发电量计算有着重大意义。目前,风资源评估中对风速缺失值的处理多采用传统的数值插补方式,即通过对历史风速数据的比对,选取相关性强的风速数据进行数值插补。传统风速插补采用Windographer软件,该软件提供了8种MCP算法:线性最小二乘法、正交回归法、矩阵时间序列法、快速排序法、方差比法、风速比法、垂直分层算法以及威布尔拟合算法1。由于上述算法插补前均假设是风速存在线性特定关系1-2,而真实关系并非线性,导致插补出的风速失真,准确性略显不足。

    随着人工智能技术的发展,神经网络的应用被引入风电行业。朱晓玲3等人通过BP神经网络对单台测风塔不同高度层风速进行训练,并完成单台测风塔指定高度层风速的插补,效果较好。罗恩博4等人通过改进型的BP神经网络进行风功率预测,并验证了LM-BP神经网络的准确性。向健平5等人通过粒子群化BP神经网络,分析风机SCADA系统的历史数据,建立主轴承故障预警模型,实现主轴承故障预警。针对传统方法风速插补的诸多不足,本文引入神经网络方法,采用BP神经网络系统进行风速插补6-7,对历史的风速数据进行训练建模,并根据模型对未来缺失的风速值进行插补。基于此方法,本文将其应用于测风塔风速插补和风机风速插补两种应用场景,以插补和比对风电场内的实测风速。

  • BP(Back Propagation)网络是一种多层感知器,包括输入层、中间层和输出层,层与层之间的神经元采用全互连的连接方式,通过相应的网络权函数相互连接;每层内的神经元之间没有连接8。BP算法包括信号的正向传播和误差的反向传播两个过程,正向传播时,输入信号通过隐含层作用于输出节点,通过每个节点的权重参数进行非线性变换产生输出信号,比较输出值与期望值的大小,若实际输出与期望输出不相符,则转入误差的反向传播过程9。反向传播由输出层向输入层方向逐层反转,并将误差分摊到各层所有单元并调整各单元的权重值。当下一次训练开始后,信号再次经过神经单元,并正向传播,重复上述过程,直至输出值与期望值误差满足要求为止,BP神经网络风速插补示意图如图1所示。

    Figure 1.  Wind speed interpolation diagram of BP neural network

    本文基于pytorch环境搭建BP神经网络6,以历史风速相关量作为输入参数,输出量为预测插补风速值10-11,BP神经网络插补流程如图2所示。

    Figure 2.  Flow chart of wind speed interpolation based on BP neural network method

    具体流程为:

    1)选取风电场风速相关变量,并进行归一化处理。

    2)将相关变量由输入层传入神经网络,通过神经网络的前向转播,计算传播过程中的每个神经元的权重值和误差值。

    3)比较最后的风速误差,并反向传递误差值,调整权重,直到最后符合精度为止。

  • 选取明阳智能西北地区某平坦地形风电场,该风电场平均海拔1 340 m,地势最大高差为2 m,地形图如图3所示。

    Figure 3.  Flat terrain map

    选取场内测风塔实测风速相关数据,包括风速、风向、温度、湿度、气压数据12,该数据为每10 min采样一次。选用风电场内3座测风塔10 m、40 m、60 m、80 m和100 m高度处的测风数据作为数据集,并根据需要将历史数据划分为数据训练集,需要插补的风速时间段划分为测试集。训练集选取2017年7月10日0点至2018年2月15日9时的3座测风塔风速相关历时数据;通过BP神经网络对训练集的数据进行训练建模,将生成的模型用于对目标测风塔风速数据进行预测插补。测试集选取2018年2月15日9时20分至2018年4月15日7时30的风速相关历时数据。通过前两座测风塔所有高度层的风速相关有效数据,通过模型插补出第3座测风塔100 m高度处的风速数据。平坦地形神经网络插补与实测数据对比结果如图4所示。

    Figure 4.  Comparison between BP neural network interpolation and actually measured values for flat terrain

    选取明阳智能西南地区某复杂地形风电场,该风电场平均海拔281 m,场内地势高差达到了50 m,复杂地形示意图如图5所示。

    Figure 5.  Complex terrain map

    选取场内5个测风塔10 m、40 m、60 m、80 m和100 m高度处的测风数据(包括风速、风向、温度、湿度、气压数据)作为数据集,分别采用BP神经网络和传统插补方式对目标风速进行插补。其中,复杂地形训练集选取2017年3月18日0时至2019年1月1日0时风电场测风塔数据,测试集选取2019年1月1日0时至2019年5月4日0时风电场测风塔,计算结果如图6所示。

    Figure 6.  Comparison between BP neural network interpolation and actually measured values for complex terrain

    为对比神经网络的插补效果,另通过传统MCP方法对同样的数据集进行插补并验证。文献1指出,通过Windographer中的MCP方法进行风速插值与地形关系不大。本文采用MCP方法分别对实际数据分别采用插值验证,通过均方根误差(MSE)、决定系数(R2)等指标的插补比对,本文选用传统方法中插补效果最优的线性最小二乘法(Linear Least Squares Algorithm)。

    线性最小二乘法是一种将目标速度和参考速度数据关联起来的方法,其基础是将线性最小二乘法直接应用于目标速度与参考速度的散点图,得到的线性曲线拟合为线性y=mx+b的关系,斜率m和截距b值如公式(1)公式(2)所示:

    m=SxySXX ((1))
    b=y¯-mx¯ ((2))

    式中:SxxSxy分别如公式(3)公式(4)所示:

    Sxx=(xi-x¯) ((3))
    Sxy=xi-x¯(yi-y¯) ((4))

    式中:x为参考风速(m/s);y为目标风速(m/s)。

    Windograph软件通过该组数据线性最小二乘法拟合,求出相关参数,并在需要预测的数据列进行线性插值,得出插补风速。平坦地形和复杂地形的风速插补具体结果表1所示。

    指标平坦地形测风塔风速插补复杂地形测风塔风速插补
    评价指标神经网络传统插补提升效果神经网络传统插补提升效果
    MSE0.6081.0843.7%↓0.6120.6911.3%↓
    MAE0.5710.76325.2%↓0.5880.6276.2% ↓
    COR0.9730.9522.2% ↑0.9390.9281.1% ↑
    R20.9460.9054.5% ↑0.8770.8611.9% ↑

    Table 1.  Improvement statistics of neural network interpolation compared to traditional interpolation for different terrains

    测风塔数据插补结果表明:通过神经网络插补方法得出的风速值与传统方法相比,平坦地形风速平均绝对误差下降了25.2%,复杂地形下降了6.2%,平坦地形相关性系数和决定系数分别上升了2.2%和4.5%,复杂地形相关性系数和决定系数分别上升了1.1%和1.9%。比较可知,神经网络风速插补相较于传统风速插补具有明显优势,且神经网络对风速的插补在平坦地形下更加有效。

  • 风电场内单台风机的运行状况与风电场内的发电量紧密相关。通过单台风电机组的风速数据分析,能够更加精确的评估风电场的实际发电功率和预测未来风电场发电量。由于停机检修、测试等状态下风机启停的影响,风电场内个别机组的风速出现资料缺失的状况。为解决风速资料缺失对风电机组数据分析的影响,采用BP神经网络的方法对缺失风速值进行插补,满足数据分析所需。

    通过提取明阳智能某风电场10台风机的主控SCADA数据,主控数据包括风机实时风速、风向、温度、压强、桨距角实时数据,数据时段为2019年1月1日0时1分0秒至2019年12月31日23时59分0秒全年主控数据。其中前9台风机的主控数据正常完整,第10台风机部分数据缺失,缺失时段分布在2019年11月1日0时0分0秒至2019年12月31日23时59分0秒区间。根据插补要求,将缺失时间段划分为测试集,将主控数据中2019年1月1日0时1分0秒至2019年10月31日23时59分0秒数据作为训练集,通过BP神经网络训练生成模型。

    通过完整时间段的数据训练生成的BP神经网络模型,对第10台风机的风速进行预测插补。为验证插补效果,通过传统的线性最小二乘法进行了相同数据的预测插补,对比结果如图7所示。

    Figure 7.  Comparison among wind turbine’s wind speed interpolation based on neural method, traditional interpolation and actually values

    通过神经网络方法的风机风速插补与传统方法的风速插补效果如表2所示。对比神经网络对风机风速插补和传统方法插补,测试集中风机风速预测的平均绝对误差下降了14.4%,相关性系数和决定系数分别提升了1.7%和2.8%,结果表明,神经网络风速插补相较于传统风速插补方法准确性有较大的提升。

    评价指标神经网络传统插补提升效果
    MSE1.2451.90134.5% ↓
    MAE0.8971.04814.4% ↓
    COR0.9790.9631.7% ↑
    R20.9520.9262.8% ↑

    Table 2.  Improvement statistics of wind turbine’s wind speed interpolation based on BP neural network method compared to traditional interpolation

  • 基于pytorch框架搭建BP神经网络,对测风塔风速和风机风速缺失数据进行插补,与传统的插补方法进行了对比分析,得出结论如下:

    1)神经网络对平坦地形的插补效果优于对复杂地形风速插补,原因可能为复杂地形的风速变化大,对神经网络提出了挑战。

    2)神经网络风速插补对测风塔风速插补和风机风速插补的应用,均优于传统方法,可应用与测风塔风速插补和风场内风机风速插补,进而为风资源评估提供精准的数据。

Reference (12)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return