多种风电场风速预报订正方法的适用性研究与集成应用

doi:10.12118/j.issn.1000-6060.2021.547

摘要/Abstract

摘要：

以内蒙古中部某风电场为实验风电场,采用随机森林（Random forest,RF）方法、相似误差订正（Analogue correction of errors,ACE）方法以及概率密度匹配方法（Probability density function matching method,PDF）分别对风电场风速预报进行订正及适用性研究。结果表明：3种方法在各季均对中尺度天气预报模式（Weather research and forecasting model, WRF）风速预报具有不同程度的订正效果,RF方法可以有效改善WRF误差较大的问题,但兼具误差过分放大情况,ACE方法和PDF虽然对较大误差的改善能力不及RF方法,但是能够较好地控制误差过分放大问题。此外,3种方法针对小于5 m·s^-1的小风速段,订正效果不理想,随着风速的增加,订正能力逐渐增强。参照预报模型各自的优势,尝试开展多种预报模型的分风速等级集成应用,可以对不同风速等级下的WRF预报起到较好的改善作用。

关键词: 风速预报, 随机森林, 相似误差订正, 概率密度匹配方法, 集成预报

Abstract:

The method used for evaluating the correction effect of a model is very important for improving the short-term wind speed forecast. At present, the comparison of the correlation coefficient and error between the corrected and forecasted wind speeds in the research period is generally adopted. This evaluation method often ignores the limited correction ability of most statistical correction methods in low and high wind speed sections. Taking a wind farm in central Inner Mongolia, China as the experimental site, this study uses the three methods of random forest (RF), analog correction of errors (ACE), and probability density function matching method (PDF) to correct the wind speed forecast in the experimental wind farm. A correlation coefficient and error analysis of the whole research period is performed. Wind speed is classified into cut-in wind speed, low, high, and full generating wind speeds, and cut-out wind speed. The applicability study of the three models in each grade is then conducted. The results show that the three methods have different degrees of correction effects on the weather research forecast (WRF) of wind speed in each season. The average absolute error promotion rates in sequence are 6.7%-36.7%, 11.1%-34.7%, and 3.7%-20.6% after correction via the RF, ACE, and PDF methods, respectively, in the model validation period. The RF method can effectively improve the large error problem of the WRF, but it also has the error overamplification problem. Although the ACE and PDF methods have less ability for improving the large error compared with the RF method, they can better control the error overamplification problem. For the small wind speed section of less than 5 m·s^-1, the correction effect of the three methods is not ideal. The original WRF forecast and PDF method are better than the RF and ACE methods. With the wind speed increase, the WRF forecast error significantly increases, and the correction ability of the three methods gradually increases. The RF and ACE methods are better than the PDF methods in the wind speed level of [5,16). The correction ability of the ACE method begins to decline, whereas that of the PDF method begins to improve in the wind speed level of [16,+∞). According to the respective advantages of the forecast models, the integrated application of various forecast models considering the wind speed forecast level is attempted to improve the WRF forecast under different wind speed levels.

Key words: wind speed forecast, random forest, analogue correction of errors, probability density function matching method, integrating forecast

徐丽娜,申彦波,冯震,叶虎. 多种风电场风速预报订正方法的适用性研究与集成应用[J]. 干旱区地理, 2022, 45(4): 1114-1124.

XU Li’na,SHEN Yanbo,FENG Zhen,YE Hu. Applicability research and integrated application of various correction methods for wind speed forecast in a wind farm[J]. Arid Land Geography, 2022, 45(4): 1114-1124.

图/表 12

图1

表1

图2

表2

表3

图3

图4

图5

表4

图6

表5

图7

参考文献 29

[1]	易跃春. 中国可再生能源发展报告2019[R]. 北京: 水电水利规划设计总院, 2020.
	[ Yi Yuechun. China renewable energy development report 2019[R]. Beijing: China Renewable Energy Engineering Institute, 2020. ]
[2]	Xu J Z, He D X, Zhao X L. Status and prospects of Chinese wind energy[J]. Energy, 2010, 35(11): 4439-4444. doi: 10.1016/j.energy.2009.06.058
[3]	Feng Y, Lin H Y, Ho S L, et al. Overview of wind power generation in China: Status and development[J]. Renewable and Sustainable Energy Reviews, 2015, 50: 847-858. doi: 10.1016/j.rser.2015.05.005
[4]	顾洪宾. 中国可再生能源发展报告2020[R]. 北京: 水电水利规划设计总院, 2021.
	[ Gu Hongbin. China renewable energy development report 2020[R]. Beijing: China Renewable Energy Engineering Institute, 2021. ]
[5]	程兴宏, 陶树旺, 魏磊, 等. 基于WRF模式和自适应最小二乘回归法的风能预报试验研究[J]. 高原气象, 2012, 31(5): 1461-1469.
	[ Cheng Xinghong, Tao Shuwang, Wei Lei, et al. Short-term wind power forecasting experiment based on WRF model and adapting partial least square regression method[J]. Plateau Meteorology, 2012, 31(5): 1461-1469. ]
[6]	石岚, 徐丽娜, 郝玉珠. 多模式风速融合预报应用研究[J]. 高原气象, 2017, 36(4): 1022-1028. doi: 10.7522/j.issn.1000-0534.2017.00021
	[ Shi Lan, Xu Lina, Hao Yuzhu. Application research on the multi-model fusion forecast of wind speed[J]. Plateau Meteorology, 2017, 36(4): 1022-1028. ] doi: 10.7522/j.issn.1000-0534.2017.00021
[7]	石岚, 徐丽娜, 郝玉珠. 基于风速高相关分区的风电场风速预报订正[J]. 应用气象学报, 2016, 27(4): 506-512.
	[ Shi Lan, Xu Li’na, Hao Yuzhu. The correction of forecast wind speed in a wind farm based on partitioning of the high correlation of wind speed[J]. Journal of Applied Meteorological Science, 2016, 27(4): 506-512. ]
[8]	Wang J J, Wang Y F, Li Y N. A novel hybrid strategy using three-phase feature extraction and a weighted regularized extreme learning machine for multi-step ahead wind speed prediction[J]. Energies, 2018, 11(2): 321, doi: 10.3390/en11020321. doi: 10.3390/en11020321
[9]	肖建辉. 基于机器学习的风电场短期风速预测混合模型研究[D]. 上海: 东华大学, 2020.
	[ Xiao Jianhui. Research on hybrid model of wind farm short-term wind speed prediction based on machine learning[D]. Shanghai: Donghua University, 2020. ]
[10]	李慧, 陈湘萍. 基于相空间重构和长短期记忆网络的风电预测[J]. 新型工业化, 2020, 10(3): 1-6.
	[ Li Hui, Chen Xiangping. Wind power prediction based on phase space reconstruction and long sort-term memory networks[J]. The Journal of New Industrialization, 2020, 10(3): 1-6. ]
[11]	王丹, 高红燕, 杨艳超, 等. 基于最优训练期的PP与MOS的风电功率趋势预报对比[J]. 干旱区地理, 2021, 44(3): 819-829.
	[ Wang Dan, Gao Hongyan, Yang Yanchao, et al. Comparative study of wind power trend forecast by using methods of PP and MOS with optimal training period[J]. Arid Land Geography, 2021, 44(3): 819-829. ]
[12]	Ren Y, Suganthan P N, Srikanth N. A novel empirical mode decomposition with support vector regression for wind speed forecasting[J]. IEEE Transactions on Neural Networks and Learning Systems, 2016, 27(8): 1793-1798. doi: 10.1109/TNNLS.2014.2351391 pmid: 25222957
[13]	张铁军, 颜鹏程, 李照荣, 等. 基于短期历史资料的风能预报中风速误差循环订正新方法[J]. 干旱气象, 2017, 35(6): 1042-1052.
	[ Zhang Tiejun, Yan Pengcheng, Li Zhaorong, et al. A new cycle correction method for wind speed error in wind energy forecast based on short-term historical data[J]. Journal of Arid Meteorology, 2017, 35(6): 1042-1052. ]
[14]	邓华, 张颖超, 顾荣, 等. 基于PCA-RBF的风电场短期风速订正方法研究[J]. 气象科技, 2018, 46(1): 10-15.
	[ Deng Hua, Zhang Yingchao, Gu Rong, et al. Correction method of short-term wind speed in wind farm research based on PCA and RBF neural network[J]. Meteorological Science and Technology, 2018, 46(1): 10-15. ]
[15]	张颖超, 肖寅, 邓华. 基于ELM的风电场短期风速订正技术研究[J]. 气象, 2016, 42(4): 466-471.
	[ Zhang Yingchao, Xiao Ying, Deng Hua. Modification technology research of short-term wind speed in wind farm based on ELM method[J]. Meteorological Monthly, 2016, 42(4): 466-471. ]
[16]	杨培宏, 胡庆林, 胡忠林, 等. 风能特性对风电场电能输出预测研究[J]. 计算机仿真, 2016, 33(12): 122-127.
	[ Yang Peihong, Hu Qinglin, Hu Zhonglin, et al. The predictive study of wind energy characteristics for the power energy production of the wind farm[J]. Computer Simulation, 2016, 33(12): 122-127. ]
[17]	中华人民共和国国家质量监督检验检疫总局, 中国国家标准化管理委员会. 地面气象观测规范自动观测(GB/T 35237-2017)[S]. 北京: 中国标准出版社, 2017.
	[General Administration of Quality Supervision, Inspection and Quarantine of the People’s Republic of China, Standardization Administration of China. Specification for surface meteorological obversation:Automatic observation(GB/T 35237-2017)[S]. Beijing: China Standards Press, 2017. ]
[18]	刘鸿波, 董理, 严若婧, 等. ERA5再分析资料对中国大陆区域近地层风速气候特征及变化趋势再现能力的评估[J]. 气候与环境研究, 2021, 26(3): 299-311.
	[ Liu Hongbo, Dong Li, Yan Ruojing, et al. Evaluation of near-surface wind speed climatology and long-term trend over China’ s mainland region based on ERA5 reanalysis[J]. Climatic and Environmental Research, 2021, 26(3): 299-311. ]
[19]	刑丽珠, 张方敏, 黄进, 等. 1961-2018年内蒙古6级及以上大风日数时空变化特征[J]. 干旱区地理, 2021, 44(5): 1290-1298.
	[ Xing Lizhu, Zhang Fangmin, Huang Jin, et al. Spatial and temporal changes of high wind days over category 6 and above in Inner Mongolia from 1961 to 2018[J]. Arid Land Geography, 2021, 44(5): 1290-1298. ]
[20]	Lin Y J, Kruger U, Zhang J P, et al. Seasonal analysis and prediction of wind energy using random forests and arx model structures[J]. IEEE Transactions on Control Systems Technology, 2015, 23(5): 1994-2002. doi: 10.1109/TCST.2015.2389031
[21]	孙全德, 焦瑞莉, 夏江江, 等. 基于机器学习的数值天气预报风速订正研究[J]. 气象, 2019, 45(3): 426-436.
	[ Sun Quande, Jiao Ruili, Xia Jiangjiang, et al. Adjusting wind speed prediction of numerical weather forecast model based machine learning methods[J]. Meteorological Monthly, 2019, 45(3): 426-436. ]
[22]	徐晶晶, 胡非, 肖子牛, 等. 风能模式预报的相似误差订正[J]. 应用气象学报, 2013, 24(6): 731-740.
	[ Xu Jingjing, Hu Fei, Xiao Ziniu, et al. Analog bias correction of numerical model on wind power prediction[J]. Journal of Applied Meteorological Science, 2013, 24(6): 731-740. ]
[23]	钱磊, 邱学兴, 郑淋淋. 基于概率密度匹配方法的WRF模式阵风风速误差订正[J]. 气象科技, 2019, 47(6): 916-926.
	[ Qian Lei, Qiu Xuexing, Zheng Linlin. Error correction of WRF model gust speed based on probability density function matching method[J]. Meteorological Science and Technology, 2019, 47(6): 916-926. ]
[24]	Wang W Q, Xie P P. A multiplatform-merged (MPM) SST analysis[J]. Journal of Climate, 2007, 20(9): 1662-1679. doi: 10.1175/JCLI4097.1
[25]	徐丽娜, 申彦波, 李忠, 等. 基于概率密度匹配方法的FY-4A地表入射太阳辐射订正[J]. 高原气象, 2021, 40(4): 932-942. doi: 10.7522/j.issn.1000-0534.2020.00080
	[ Xu Li’na, Shen Yanbo, Li Zhong, et al. Correction of FY-4A surface solar irradiance based on probability density function matching method[J]. Plateau Meteorology, 2021, 40(4): 932-942. ] doi: 10.7522/j.issn.1000-0534.2020.00080
[26]	中国气象局. 风电场风速预报准确率评判方法(QX/T 243-2014)[S]. 北京: 气象出版社, 2014.
	[China Meteorological Administration. Evaluation of wind speed forecast accuracy in wind farm (GB/T 243-2014)[S]. Beijing: China Meteorological Press, 2014. ]
[27]	任宏利, 丑纪范. 统计-动力相结合的相似误差订正方法[J]. 气象学报, 2005, 63(6): 988-992.
	[ Ren Hongli, Chou Jifan. Analogue correction method of errors by combining both statistical and dynamical methods together[J]. Acta Meteorologica Sinica, 2005, 63(6): 988-992. ]
[28]	疏杏胜, 王子茹, 李福威, 等. 基于机器学习模型的短期降雨多模式集成预报[J]. 南水北调与水利科技, 2020, 18(1): 42-50.
	[ Shu Xingsheng, Wang Ziru, Li Fuwei, et al. Short-term rainfall multi-mode integrated forecasting based on machine learning models[J]. South-to-North Water Transfers and Water Science & Technology, 2020, 18(1): 42-50. ]
[29]	熊聪聪, 潘璇, 赵奇, 等. 多模式集成的RBF神经网络天气预报[J]. 天津科技大学学报, 2014, 29(1): 75-78.
	[ Xiong Congcong, Pan Xuan, Zhao Qi, et al. RBF neural network for weather forecast based on multi-model integration[J]. Journal of Tianjin University of Science & Technology, 2014, 29(1): 75-78. ]

近地sigma层取值	模式输出高度/m
1.0000	0.00
0.9985	11.32
0.9970	22.65
0.9950	37.78
0.9935	49.14
0.9928	54.44
0.9920	60.51
0.9890	82.29
0.9850	113.74
0.9780	167.25

季节	相关系数				平均绝对误差/m·s^-1				平均绝对误差提升率/%
	订正前	订正后			订正前	订正后			平均绝对误差提升率/%
	订正前	RF	ACE	PDF	订正前	RF	ACE	PDF	RF	ACE	PDF
春季	0.72	0.69	0.75	0.72	3.0	2.8	2.5	2.7	6.7	16.7	10.0
夏季	0.43	0.40	0.40	0.43	2.7	2.5	2.4	2.6	7.4	11.1	3.7
秋季	0.75	0.68	0.76	0.74	3.4	2.8	2.6	2.7	17.6	23.5	20.6
冬季	0.65	0.64	0.65	0.66	4.9	3.1	3.2	3.9	36.7	34.7	20.4

风速区间/m·s^-1	样本数量
[0, 3)	206
[3, 5)	317
[5, 8)	748
[8, 12)	888
[12, 16)	392
[16, 20)	187
[20, +∞)	70

季节	集成权重			相关系数		平均绝对误差/m·s^-1
季节	RF	ACE	PDF	集成前最优	集成后	集成前最优	集成后
春季	0.30	0.40	0.30	0.75	0.76	2.5	2.4
夏季	0.30	0.40	0.30	0.43	0.46	2.4	2.3
秋季	0.33	0.34	0.33	0.75	0.76	2.6	2.5
冬季	0.40	0.40	0.20	0.66	0.66	3.1	3.1