基于混合分位数回归长短期记忆神经网络的风电功率短期区间预测  

SHORT-TERM WIND POWER INTERVAL PREDICTION BASED ON MIXED QUANTILE REGRESSION LONG AND SHORT-TERM MEMORY NEURAL NETWORK

作　　者：杨茂[1] 张书天 王勃 于欣楠 Yang Mao;Zhang Shutian;Wang Bo;Yu Xinnan(Key Laboratory of Modern Power System Simulation and Control&Renewable Energy Technology,Ministry of Education,Northeast Electric Power University,Jilin 132012,China;State Key Laboratory of Operation and Control of Renewable Energy&Storage Systems,China Electric Power Research Institute,Beijing 100192,China)

机构地区：[1]现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学),吉林132012 [2]新能源与储能运行控制国家重点实验室,中国电力科学研究院有限公司,北京100192

出　　处：《太阳能学报》2025年第2期582-590,共9页Acta Energiae Solaris Sinica

基　　金：国家重点研发计划(2022YFB2403000)。

摘　　要：为进一步提升风电功率区间预测精度,提出一种基于混合分位数回归长短期记忆神经网络的风电功率短期区间预测方法。通过同时考虑复合、平滑和非交叉3个特点对传统分位数回归模型进行改进,首先使用平滑函数代替弹球损失函数,使长短期记忆神经网络更易于拟合分位数回归模型。然后构建复合目标函数,使其能在给出多个分位数的条件下不重复训练多个独立模型。接着利用ReLU罚函数进行非交叉约束来避免分位数交叉现象的发生。最后将改进后的分位数回归与长短期记忆神经网络相结合并应用于中国甘肃省某风电场,运行结果表明所提模型在不同置信水平下对应PICP和PIAW分别提高了4.17个百分点和降低了2.31 MW,验证了方法的有效性。In order to further improve the accuracy of wind power interval prediction,a short-term interval prediction method of wind power based on mixed quantile regression long-term and short-term memory neural network is proposed.The traditional quantile regression model is improved by considering the three characteristics of compound,smoothing and non-crossing.Firstly,the smoothing function is used instead of the pinball loss function,which makes it easier for long-term and short-term memory neural networks to fit the quantile regression model.Then,a composite objective function is constructed so that it does not train multiple independent models repeatedly under the condition of giving multiple quantiles.thirdly,ReLU penalty function is used for non-cross constraint to avoid the occurrence of quantile crossing.Finally,the improved quantile regression is combined with the long-term and short-term memory neural network and applied to a wind farm in Gansu Province,China.The operation results show that the PICP and PIAW corresponding to thg proposed model increases the PICP increase by 4.17 percentage points and decrease by 2.31 MW respectively at different confidence levels,which verifies the effectiveness of the method.

关 键 词：风电功率 深度学习 区间预测 复合非交叉 分位数回归 ReLU罚函数 

分 类 号：TM614[电气工程—电力系统及自动化]