摘要
为克服非均衡样本影响风电功率预测的精度,提出了一种基于最小二乘生成对抗网络(LSGAN)和基于遗传算法的极限学习机(GA-ELM)的风电功率短期预测方法。以数值天气预报(NWP)数据为研究对象。首先,采用模糊C均值(FCM)算法将风电场NWP数据聚类成若干个天气类型;其次,采用LSGAN算法生成少数类样本,将具备原始特征的生成数据扩充至NWP数据集并使各类天气的样本均衡;最后,将均衡的数据集代入GA-ELM模型中建立预测模型。仿真结果表明:LSGAN算法缓解了GAN算法中梯度消失、训练不平稳和样本品质差等缺陷,在生成数据时具有收敛速度快、稳定性高和易达到纳什平衡等优点,有效地提高了风电功率的短期预测精度。
To overcome the influence of unbalanced samples on the accuracy of wind power prediction,a short-term wind power prediction method based on least squares generative adversarial network(LSGAN)and genetic algorithm extreme learning machine(GA-ELM)is proposed.Numerical weather prediction(NWP)data is the research object.First,the FCM algorithm is used to cluster the NWP data of the wind farm into several weather types.Second,the LSGAN algorithm is used to generate a few types of samples.The generated data with original characteristics are expanded to the NWP data set and the samples of various weather are balanced.Finally,the balanced data set is substituted into the GAELM model to establish a prediction model.The results of the simulation experiments show that the LSGAN algorithm improves the gradient disappearance,unstable training and poor sample quality in the GAN algorithm,and has the advantages of fast convergence speed,high stability and easy access to Nash balance when the data is generated.The proposed method effectively improves the short-term prediction accuracy of wind power.
作者
赵睿智
丁云飞(指导)
ZHAO Ruizhi;DING Yunfei(State Grid Shanghai Changxing Power Supply Company,Shanghai 201913,China;School of Elctrical Engineering,Shanghai Dianji University,Shanghai 201306,China)
出处
《上海电机学院学报》
2022年第6期311-317,共7页
Journal of Shanghai Dianji University
基金
国家自然科学基金项目(11302123)
上海市浦江人才计划项目(15PJ1402500)
航空科学基金项目(20200001012015)。
关键词
风电功率
聚类
短期预测
最小二乘生成对抗网络
基于遗传算法的极限学习机模型
wind power
clustering
short-term forecasting
least squares generative adversarial network
genetic algorithm extreme learning machine
