光伏发电系统运行数据时间序列补齐与降维方法研究

Research on time series completing and dimensionality reduction methods for photovoltaic power generation system operation data

分布式光伏发电系统(photovoltaic generation system,PGS)往往存在运行数据采集缺失及维数过多问题,这给平稳电力供应的PGS发电功率准确预测建模带来困难。为此,该文提出一种应用插补+主成分分析(principal component analysis,PCA)的PGS数据时间序列补齐与降维方法,以采集时间戳的环境、光伏发电数据完整为目标,采用线性插值、多重插补、生成对抗插补网络等3种典型插补算法对PGS数据时间序列补齐构成完整数据集,并对数据集所包含13种输入变量、3种输出变量进行PCA处理,有效地减少数据维度及减低建模难度。实验选取环境温度、光伏发电功率缺失数据,时间序列补齐数据表明多重插补相对于线性插值、生成对抗插补网络效果更佳;对环境温度、光伏发电功率补齐数据PCA处理,5主成分累计贡献率>95%,较好地完成时间序列补齐与降维处理,有助于数据后续建模、预测及其他挖掘工作,其研究工作具有重要的推广应用价值。

Photovoltaic Generation System(PGS) often suffers from missing operational data collection and too many dimensions,which makes it difficult to accurately predict and model the power generation of PGS for smooth power supply.To this end,this paper proposes a method of complementary and dimensionality reduction of PGS data time series by applying interpolation and Principal Component Analysis(PCA),with the goal of collecting the integrity of time-stamped environmental and photovoltaic generation data.Interpolation algorithms such as linear interpolation,multiple imputation and generative adversarial imputation network were used to complement the time series of PGS data to form a complete data set,and PCA processing was performed on 13 input variables and 3 output variables contained in the data set,effectively reducing the data dimension and modeling difficulty.The experiment selects the missing data of environmental temperature and photovoltaic power,and the time series complement data show that multiple imputation is better than linear interpolation and the generated adversarial imputation network.For PCA processing of environmental temperature and photovoltaic power complement data,the cumulative contribution rate of 5 principal components is greater than 95%,and the time series complementary and dimensionality reduction processing are well completed,which is helpful for subsequent data modeling,prediction and other data mining work,and the research work has important promotion and application value.