分布式光伏集群功率短期预测的空间互补特性初探

Preliminary Study on Spatial Complementarity Characteristics of Short-term Power Prediction for Distributed Photovoltaic Clusters

随着分布式光伏装机容量的快速增长,分布式光伏集群功率预测对电网调度的影响日益增强。分布式光伏集群功率预测存在“先累加后预测”“先预测后累加”以及“先聚类再预测”3种技术路线。通过在含600多个分布式光伏站点实测功率数据集上的实验发现,分布式光伏集群功率预测存在“空间互补”现象,即3类集群预测路线得到的预测误差相比于单个场站的平均预测误差都有所降低。为了探究空间互补现象的产生机理及影响因素,首先从机理分析的角度将其归纳为“出力曲线互补”和“预测误差互补”两类空间互补特性。然后,提出了“空间互补系数”这一概念,对互补效果进行量化评估。最后,探究了集群规模、分布范围、天气类型和聚类簇数4种因素对空间互补特性的影响。结果表明,两类空间互补特性对分布式光伏集群短期功率预测精度提升效果显著,且预测误差互补在精度上优于出力曲线互补。研究结果可为分布式光伏集群划分提供依据,有助于实现更加高效且准确的功率预测。

With the rapid growth of distributed photovoltaic(PV)installed capacity,the impact of power prediction for distributed PV clusters on grid dispatch is becoming increasingly significant.There are three technical routes for power prediction of distributed PV clusters,i.e.,accumulation before prediction,prediction before accumulation and clustering before prediction.Through experiments on a dataset of over 600 distributed PV sites,the spatial complementarity is found in the power prediction for distributed PV clusters,whereby the prediction errors of the three cluster prediction routes are lower than the average prediction error of a single site.To explore the generation mechanism and influencing factors of the spatial complementarity,this paper first categorizes it into two types of spatial complementarity characteristics,namely power curve complementarity and prediction error complementarity,based on the mechanism analysis.Secondly,the concept of spatial complementarity coefficient is proposed to quantitatively evaluate the complementary effect.Finally,the effects of cluster scale,distribution range,weather type and number of clusters on the spatial complementarity characteristics are explored.The results show that the two types of spatial complementarity characteristics have significant effects on improving the short-term power prediction accuracy of distributed PV clusters,with prediction error complementarity superior to power curve complementarity.The research results can provide a basis for the division of distributed PV clusters and contribute to achieving more efficient and accurate power prediction.