随着"双碳"目标的推进,清洁能源所占比重大幅度增加,分布式光伏发电在我国农村地区快速发展,但其随机性、间歇性的特点给新能源消纳和电网稳定带来很大的挑战.光伏发电预测可以在一定程度上改善新能源消纳问题,减少光伏发电的不稳定性对电网的冲击.因此,为提高光伏发电功率预测精度,提出一种基于改进向量加权平均算法优化CNN-QRGRU网络的光伏发电概率预测方法.首先采用ReliefF算法对特征变量进行选择,在此基础上利用高斯混合模型(Gaussian mixture model,GMM)聚类方法将天气分为晴天、晴转多云和阴雨天3种类型,将处理好的数据输入到CNN-GRU模型中,并利用向量加权平均(weighted mean of vectors algorithm,INFO)优化算法对模型超参数进行调参,将分位数回归模型(quantile regression,QR)与INFO-CNN-GRU模型相结合得到光伏功率条件分布,结合核密度估计法从条件分布中获得概率密度函数,完成概率预测.以实际光伏电站数据作为基础,将提出的INFO优化算法与其他几种传统的优化算法进行对比,结果表明INFO的优化效果更好,在此基础上进行概率预测,得到的概率预测结果相较于点预测能提供更多有效信息,更具有应用价值.
Short-term Probabilistic Prediction of Rural Distributed Photovoltaic Power Generation Based on Improved INFO-CNN-QRGRU Model
With the advancement of the"dual carbon"goal,the proportion of clean energy has increased significantly,and distributed photovoltaic power generation has developed rapidly in rural areas of China,but its randomness and intermittent characteristics have brought great challenges to the consumption of new energy and the stability of the power grid.Photovoltaic power generation prediction can to improve the new energy consumption and reduce the impact of its instability on the power grid.Therefore,in order to improve the accuracy of photovoltaic power generation power prediction,this paper proposes a photovoltaic power generation probability prediction method based on the improved vector weighted average algorithm to optimize the CNN-QRGRU network.First,the ReliefF algorithm is used to select the feature variables,based on which the Gaussian mixture model(GMM)clustering method is used to classify the weather into three types:sunny,sunny to cloudy and rainy,the processed data are input into the CNN-GRU model,and the INFO optimization algorithm is used to tune the model hyperparameters,and the quantile regression(QR)model is combined with the INFO-CNN-GRU model to obtain the PV power conditional distribution,and the kernel density estimation method is combined to obtain the probability density function from the conditional distribution to complete the probability prediction.Using the actual PV plant data as the research basis,the optimization effect of the newly proposed INFO optimization algorithm is compared with several other traditional optimization algorithms,and the results show that the optimization effect of INFO is better,and the probability prediction results obtained on this basis can provide more effective information compared with the point prediction,which is more valuable for application.
photovoltaic power outputGaussian mixture model clusteringgated recurrent unitweighted mean of vectors algorithmquantile regressionprobabilistic forecast
国家科技期刊平台
NSTL
万方数据
