Research On Key Technologies Of Big Data Processing And Analysis In Distribution And Utilization System

With the widespread use of smart sensor devices in distribution and utilization system,power companies have accumulated unprecedented amounts of data.Fully exploiting the big data value of distribution and utilization system can improve the level of power service and promote the digitization,informationization and intellectualization of distribution and utilization system.Therefore,it is of great significance to effectively collect,process,analyze and apply big data of distribution and utilization system.This thesis focuses on the big data acquisition,processing and analysis in distribution and utilization system,involving fine-grained power data acquisition,data cleaning and compression,data value mining and power privacy information protection,the main work is summarized as follows:1)To overcome the shortage of fine-grained user power data,a non-invasive load identification algorithm based on improved VI trajectory characteristics and deep forest is proposed,which implements the indirect collection of device-level power data.By adding discrete color coding background,the VI trajectory and power information are fused,and the improved features are classified by deep forest.The defects that VI trajectory cannot reflect power information are overcome,and the load identification effect is improved.2)For big data storage and data quality problems of power distribution and utilization system,a method of data compression for smart meters based on convolution auto-encoder and a method of rebuilding measurement missing data based on improved generation of anti-network are presented.The former learns power curve characteristics through unsupervised training of convolution auto-encoder,and realizes power curve coding and decoding through forward operation,which reduces data compression error and improves calculation efficiency.The latter achieves the exact repair of missing values by generating potential rules against the historical data of network learning measurements and by optimizing the reconstruction loss and context similarity loss.3)Based on the big data of smart meters,an ultra-short-term aggregated load prediction method based on gated recurrent unit network and an aggregated load dayahead load prediction method based on two-terminal sparse coding and deep neural network fusion are proposed.The prediction accuracy is improved by means of cluster analysis,dynamic time modeling,data dimension reduction,model fusion,etc.This thesis presents a line loss analysis method based on MSApriori for low voltage station area and a general analysis method for power saving based on power consumption behavior analysis.It can fully consider the complex factors that are not easy to quantify in the traditional model,analyze the line loss and the key influencing factors of power saving for users,so as to guide the work of power saving and loss reduction.4)In order to protect the privacy of power consumption information,a digital watermarking based and asymmetric encryption method for smart grid user privacy information protection is proposed.The former is used to hide sensitive data without changing the original data form and increasing bandwidth,while the latter is used to achieve end-to-end confidentiality.Considering the privacy risk brought by the joint meter reading in the multi-energy system,a new privacy protection system based on dictionary learning is proposed.The data use needs of energy users and suppliers are extracted and analyzed,and a multi-level privacy protection scheme is designed and proposed to meet the diverse data use needs of users and achieve the trade-off between privacy protection and convenient energy services.