| The accuracy of spatial-temporal distribution information of electric load and photovoltaic(PV)power source is significant to the decision-making of power grid planning and energy management departments.In recent years,the society,economy,science and technology have obtained continuously development at home,meanwhile,the industrial structure has been upgrading in many regions of China.As a result,power consumers' load types,electricity consumption characteristics,upstream and downstream relationships,etc.,have been taking on diversified developing trends,respectively: 1)With the vigorous development of electric vehicles(EVs)and large-scale PV intergration in the past few years,the constitution of power and load in distribution networks has changed greatly;2)The development of PV power presents uncertainties to some extent,which is similar to the uncertainties in electric load distribution and growth;3)The factors associated with power or load growth are becoming increasingly complex,and the time-lag effects of them are not clear.All of these problems pose challenges to traditional methods of eliminating uncertainties in power supply and load sides of distribution networks.With the construction and development of smart distribution grid,big data in smart distribution grid gradually formed in electric utilities,including production data,marketing data,and related socioeconomic data,etc.The formation of big data in smart distribution grid has provided a good data base for data-driven spatial-temporal forecasting of electric load and PV power source.In 2016,the national energy administration of China released “Instructions on promoting the development of internet plus smart energy”,which clearly supports the deep integration of energy and information,and the development of energy big data applications in power grid,EVs,and PV generation.However,due to that big data in smart grid has the features of variety,volume,high-dimension,etc,it is difficult for traditional spatial-temporal load or PV power source forecasting methods to mine massive information from big data,and accurately grasp the correlated factors and spatial-temporal distribution laws of load and PV power source.Hence,under the circumstance of big data in smart distribution grid,how to construct spatial-temporal forecasting models for conventional loads with increasingly complex correlated factors,EV charging load,and PV power,it is a question worthy of further study and reflection.Based on analyzing features,research paradigms,and theoretical framework of big data in smart distribution grid,this article has proposed a multi-source spatial-temporal information correlated model for load and PV power forecasting.In addition,the spatial-temporal forecasting methods for consumers' conventional load,electric taxi charging load,and regional PV installed capacity have been studied.Main research works of this article are as following.1.The present situation,characteristics,and research paradigms of big data in smart distribution grid are systematically combed or expounded,and a theorical structure is constructed for big data in smart distribution grid.Based on the multi-level analysis for main application technologies of big data in smart distribution grid,the big data application roadmap is proposed for spatial-temporal load and PV power forecasting.Based on the logical association and spatial-temporal association between multiple data sources,a hierarchical correlation model of multi-source spatial-temporal information is constructed for load and PV power forecasting.2.Based on the hierarchical correlation model of multi-source spatial-temporal information,the data inside and outside smart distribution grid,such as electricity consumption data,consumer profile data,GIS(Geographic Information System)data,weather information,and socio-econonic data,etc.,are used to generate multiple correlated data layers.On this basis,the spatial-temporal distribution of consumers' electric load are forecasted considering aggeragation features of electricity cells and uncertainties in data sources: The concepts of cell and cell features are proposed,and a hierarchical spatial clustering method based on cellular features is proposed.Based on the complex network model,the electricity consumption relationship and "upstream and downstream" relationship between different industries are analyzed.Combining the results of cell clustering and electricity consumption correlation analysis between different industries,a sparse least squares support vector regression networks algorithm is proposed to realize the class-oriented cellular load prediction.Use sampled blind number to characterize the uncertainties in correlated factors of cellular load variation.By analyzing the transfer characteristics of uncertainty in prediction models,the interval prediction of cellular load under various confidence conditions is realized.3.Based on the hierarchical correlation model of multi-source spatio-temporal information,the datasets belonging to big data in smart distribution grid,such as power grid data,road network data,electricity consumption data,large-scale taxi GPS data,and GIS data,etc.,are used to forecast the spatial-temporal distribution of electric taxis' charging load: Based on road network,power grid and large-scale taxi trajectory data,the data fusion model of road network and power grid is built,besides,the spatial-temporal trajectory model for large-scale taxis is also established.The travel demands of passengers are simulated based on the spatial-temporal trajectory model,and then a taxi operation simulation model considering information interaction between multiple agents and pricing game between public charging stations is established.Considering the mutual influences between electric taxis' charging behaviors and public charging station configurations,the spatial-temporal charging load forecasting for electric taxis under multi-source data fusion condition is realized.4.Based on the hierarchical correlation model of multi-source spatial-temporal information,the time-series data inside and outside the study region,including economic data,population data,weather information,residents' living standard data,energy and environment data,city construction data,global PV installed capacity data,and PV module price data,etc.,are used to forecast regional cumulative PV installed capacity: According to the data of global PV installed capacity and PV module cost,the levelized costs of regional PV generation is predicted in the time dimension.The time series stationary of potential associated factors for regional PV development are tested by cointegration analysis.In addition,Granger causality test is used to identify the associated factors and analyze their time lags.On the basis of associated factors reduction and principal component regression,the time series forecasting of regional PV installed capacity is realized considering the time lags of associated factors and PV generation costs.5.Based on the hierarchical correlation model of multi-source spatial-temporal information,the datasets belonging to big data in smart distribution grid,such as installed capcacity,time and location of distributed PV systems,power grid topology data,electricity consumption data,type and contour data of buildings,road information,and government subsidy policies,etc.,are used to analyze and foreacast the diffusion tendency of distributed PV systems in a data-driven approach: Based on spatial clustering and eigenvector analysis,the distribution and movement laws of PV clusters are investigated in spatial dimension,and the factors associated with the development of distributed PV systems are analyzed in time dimension.Then,the spatial-temporal forecasting of distributed PV systems are realized with a data-driven approach,which consists of four steps,namely time series prediction of cumulative distributed PV installed capacity,development state estimation of PV cells considering multiple associated factors,probability density approximation for the statistical distribution of newly installed PV capacity per cell,and spatial allocation of PV installed capacity based on entropy-weight TOPSIS method.Take some regions in East China as examples,and the hierarchical correlation model of multi-source spatial-temporal information is constructed for these regions.Based on the hierarchical correlation model,the spatial-temporal electric load and PV power source forecasting methods proposed in this article are implemented and verified. |
PDF Full Text Request