短期风电出力预测的融合模型比较  

Comparison of Fusion Algorithms for Short-term Wind Power Forecasts

作　　者：杨炎昆 李玉玲[1] 杨仕友[1] YANG Yankun;LI Yuling;YANG Shiyou(College of Electrical Engineering,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]浙江大学电气工程学院,浙江杭州310027

出　　处：《电工技术》2024年第24期75-78,88,共5页Electric Engineering

摘　　要：风力发电具有间歇性、波动性和强随机性等特点,增加了电力系统调峰、调度的压力,因此对风电出力进行准确预测具有显著技术、经济收益。为解决传统风电预测学习方法难以消除风电随机性的问题,提出了一种基于Stacking算法的短期风电功率预测集成学习方法,以降低风电出力预测的误差。首先依据Bagging和Boosting算法训练基学习器,然后应用等权重融合模型等5种线性融合模型和Stacking融合模型将其组合得到确定性预测结果。计算结果表明,6种融合模型的预测结果均有较满意的提升效果。此外,由于Stacking算法可更灵活地优化第二层的预测,其性能稍优于其他组合策略。Wind power generation is characterized by intermittency,volatility and strong stochasticity,which can lead to increased pressure on power system peaking and dispatching.In this respect,an accurate prediction of wind power has significant economic benefits.To address the difficulty of traditional learning methods to eliminate the stochasticity,this paper proposes an integrated learning method for short-term wind power forecasts based on a Stacking algorithm,which is used to reduce the prediction error of the wind power output.The base learners are trained based on Bagging and Boosting algorithms,and then combined with five linear fusion models such as equal weight fusion model and Stacking fusion model to obtain deterministic forecast results.The numerical results show that performances of the six fusion models under three evaluation criteria are all improved.Moreover,the Stacking algorithm performs slightly better than the other combination strategies because it can optimize the prediction of the second layer more flexibly.

关 键 词：集成学习 风电预测 Stacking算法 融合模型 

分 类 号：TM71[电气工程—电力系统及自动化]