摘要
对区域多风电场功率的概率预测有利于应对大规模风电并网现状下风电功率的随机性和波动性。通过建立基于径向基神经网络的分位数回归模型,实现了气象预报变量和风电功率间的非线性映射,得到了分位数形式的短期概率预测结果。针对区域风电数据维度高的问题,提出用交替方向乘子算法进行模型参数优化,从而提高了高维度模型的计算效率。提出了基于区域风向聚类的机制转换模型,对不同风向特征的样本进行独立建模,进一步提高了预测效果。以华东28处风电场为例,通过对可靠度、锐度和Pinball分数3个评价指标进行对比分析,标明所提预测方法相比传统预测方法取得了更好的概率预测效果。
For large-scale integration of wind power, probabilistic forecasts for regional wind farms are needed to deal with randomness and volatility of wind power. A quantile regression model based on radial basis function(RBF) network is proposed to fit the nonlinear mapping between predicted weather variables and wind power. The achieved short-term probabilistic forecasting results are in the form of quantiles. To improve efficiency, the alternating direction method of multipliers(ADMM) is applied to estimate the parameters of the high-dimensional regional forecasting model. A regimeswitching method based on the clusters of regional wind directions is proposed to further improve forecasting performance. Compared with traditional forecasting methods, the proposed model performs better in reliability, sharpness and pinball skill score for the case of 28 wind farms in East China.
作者
王钊
王勃
冯双磊
王伟胜
WANG Zhao;WANG Bo;FENG Shuanglei;WANG Weisheng(State Key Laboratory of Operation and Control of Renewable Energy&Storage Systems(China Electric Power Research Institute),Haidian District,Beijing 100192,China)
出处
《电网技术》
EI
CSCD
北大核心
2020年第4期1368-1375,共8页
Power System Technology
基金
国家电网公司总部科技项目(基于天气过程差异化分析的风电功率预测及调度方法研究)。
关键词
概率预测
区域多风电场
分位数回归
交替方向乘子算法
机制转换
probabilistic forecasts
regional wind farms
quantile regression
alternating direction method of multipliers(ADMM)
regime-switching
