基于XTimesNet的光伏并网低压台区电能质量预测研究

打开文本图片集

中图分类号： TP274+ .3 文献标志码：A 文章编号：1007-2683（2025）06-0119-09

Abstract：Aiming attheproblemoflow powerqualityprediction accuracyinPVgrid-conected low-voltage stations，this paper proposesan XTimesNet prediction modelforpowerqualitysteady-state index.Byimproving theconvolution modulein TimesBlockand adoptingteXception-basedmethod，thepredictionaccuracyofthemodelissignificantlyimproved.Dataafectingthesteadystateof powerquality，suchasvoltagedeviation，harmonicdistortionrateandfrequencydeviation，arecolectedusingthegridmonitoring systemincombinationwithenvironmentalfactorsafectingpowergeneration，andseveralpredictionmodelsarecomparedthrough experiments.Theresultsshowthattheproposedmethodhasthebestperformanceinthepredictionofthreetypicalfactorsffecting powerqualitysteadystate.Inthepredictionofvoltagedeviation，theaccuracyof thepredictionresultsoftheproposedmethodhasben improved by 53% and 35% relative to the BP neural network model and the traditional GRU prediction model，so that the proposed methodhasahigheraccuracyinthepredictionofpowerqualtof thelow-voltagedistributionareaswithdistributedPVgrid-cocted areas.

Keywords：distributedphotovoltaicgridconnection；low-voltagedistributionarea；powerqualityprediction;TimesNet；Xception; time series model depthwise separable convolution

0 引言

全球环境污染导致气候变化不断加剧，“绿色低碳”已经成为未来能源发展的主题。（剩余14399字）