考虑计及测量不完备性的区域分布式光伏功率预测研究

Research on Regional Distributed Photovoltaic Power Prediction Considering Measurement Incompleteness

中文关键词:测量不完备性  分布式光伏  功率预测模型  气象数据  相似日  灰色关联度

keywords:measurement incompleteness  distributed photovoltaic  power prediction model  meteorological data  similar days  grey correlation degree

中文摘要:针对光伏功率预测的准确性在信息缺失条件下会受影响的问题，设计一种计及测量不完备性的区域分布式光伏功率预测模型。首先，基于历史功率数据，利用K-means聚类算法建立电站间的关联关系，实现数据共享与互补；其次，选取具有完备气象数据的电站作为基准站，通过灰色关联分析来构建相似日数据集；再次，采用粒子群算法优化核极限学习机，建立基准电站功率预测模型，并计算其与目标电站的关联度；最后，将平均关联度输入一维卷积神经网络，实现目标电站在信息缺失条件下的功率预测。实验结果表明，所提模型在不同天气条件下均能实现有效预测，在晴天时预测功率与实际功率之间的误差较小，其最大值为-6.49%，阴天时误差稍大，其最大值为8.25%；对各分布式光伏输出功率的预测误差均低于4%。

Abstract:The accuracy of photovoltaic （PV） power prediction is compromised under conditions of information deficiency. To address this issue，a regional distributed PV power prediction model accounting for measurement incompleteness is proposed. Firstly，based on historical power data，the K-means clustering algorithm is employed to establish correlations among power stations，enabling data sharing and complementarity. Secondly，power stations with complete meteorological data are selected as reference stations，and a similar-day dataset is constructed via grey relational analysis （GRA）. Subsequently，the particle swarm optimization （PSO） algorithm is utilized to optimize the Kernel Extreme Learning Machine （KELM），with which a power prediction model for reference stations is established. The correlation degree between reference stations and the target power station is then calculated. Finally，the average correlation degree is fed into a one-dimensional convolutional neural network （1DCNN） to realize power prediction for the target power station under conditions of information incompleteness. Experimental results demonstrate that the proposed model achieves effective prediction under various weather conditions. On sunny days，the error between the predicted and actual powers is smaller，with a maximum value of -6.49%. On cloudy days，the error is slightly larger，reaching a maximum of 8.25%. Notably，the prediction error for the output power of all distributed PV systems is below 4%.