摘要
数值天气预报(Numerical Weather Prediction,NWP)对风电功率短期预测起着重要作用,考虑风电场周围多个位置的NWP信息,提出利用聚类分析与机器学习相结合的方法对海上风力发电短期功率预测进行研究。同时考虑风电场内部空间差异,通过层次聚类分析将机组内多个风机风速划分为若干组代表风速,基于优选机器学习方法进行订正后处理,获得更加准确反映风电场特性的若干组初始预报风速。在此基础上,将该若干组风速作为输入特征,基于正则化极限学习机(Regularized Extreme Learning Machine,RELM)对风电功率预测结果进行校正,获得风电场最终预测功率。通过对实例电场的发电功率进行预测,相比基于风电场单一风速的功率预测模型,该模型日准确率最大可提升3.59%,改善了极端误报情况,考核电量降低了19.24%,有效减轻电场考核。
Numerical weather prediction(NWP)plays an important role in the short⁃term prediction of wind power.Considering the NWP information at multiple locations around the wind power farm,a method combining clustering analysis and machine learning was proposed to study the short⁃term power prediction of offshore wind power.Moreover,considering the spatial differences within the wind farm,the wind speeds of multiple wind turbines in the unit were divided into several groups of representative wind speeds by hierarchical clustering analysis.Based on the optimized machine learning method,the post⁃processing was carried out to obtain several groups of initial forecast wind speeds that more accurately reflect the characteristics of the wind farm.On this basis,by using several groups of wind speeds as input features,the wind power prediction results were corrected based on the regularized extreme learning machine(RELM)to obtain the final predicted wind power.Compared with the power prediction model based on single wind speed of wind farm,the daily accuracy of the model can be increased by 3.59%,the extreme false alarm situation is improved,and the assessment power is reduced by 19.24%,which effectively reduces the assessment cost of wind farm.
作者
祁乐
唐健
江平
郑芳雯
刘德志
QI Le;TANG Jian;JIANG Ping;ZHENG Fangwen;LIU Dezhi(Guangxi Power Grid Dispatch and Control Center,Nanning 530023,China;Seniverse Technology Center,Beijing 100102,China)
出处
《山东电力技术》
2021年第5期16-22,共7页
Shandong Electric Power
基金
中国南方电网有限责任公司科技项目“基于广西电网新能源运行特性的调度综合管理应用研究”(GXKJXM20181020)。
