基于多源数据融合的分布式光伏聚合超短期预测方法

Ultra-short-term Prediction Method of Distributed Photovoltaic Aggregation Based on Multi-source Data Fusion

分布式光伏聚合发电的超短期预测是支撑其功率快速调节的前提保障,由于规模化接入的分布式光伏容量小、分布广,其发电时序特性差异性大、非平稳性强,导致其超短期预测精度难以保证。为此,文章提出基于多源数据融合的分布式光伏聚合超短期预测方法。该方法基于变分模态分解法,充分挖掘分布式光伏聚合发电非平稳性特性,并采用核主成分分析法对引发光伏发电非平稳性的影响因素即温度、湿度、光照、云量等多源数据进行量化解析,同时结合改进的长短期记忆神经网络,创建了多源数据融合方法,实现了分布式光伏聚合发电超短期预测。仿真结果表明,该方法有效提升了模型的预测精度。与传统方法相比,提出的预测方法对随机性波动严重的光伏超短期预测具有显著优势。

Ultra-short-term prediction of distributed photovoltaic aggregated power generation is a prerequisite guarantee to support its rapid power regulation.However,due to the small capacity and wide distribution of distributed photovoltaic with large-scale access,the timing characteristics of its power generation have great differences and strong non-stationarity,causing difficulty in ensuring the accuracy of its ultra-short-term prediction.Therefore,this paper proposes an ultra-short-term prediction method for distributed photovoltaic aggregated power generation based on multi-source data fusion.Based on the variational mode decomposition,the method fully explores the non-stationarity characteristics of distributed photovoltaic power generation,and uses the kernel principal component analysis method to quantitatively analyze the influencing factors of photovoltaic power generation non-stationarity,that is,temperature,humidity,light,cloud cover and other multi-source data.At the same time,combined with the improved long short-term memory neural network,the distributed photovoltaic power generation ultra-short term prediction is realized.Simulation results show that this method can effectively improve the prediction accuracy of the model.Compared with the traditional method,the prediction method proposed in this paper has a significant advantage for the ultra-short-term prediction of PV with severe random fluctuations.