Research On Time Series Data Compression And Recovery Technology For Power Edge Computing

With the development of information technology,time series data as a common type of data widely exists in the fields of power grid,medical and meteorology,and its related research has become a popular topic for global research scholars to explore.After entering the 21 st century,the volume of data has been climbing,which is due to the 5G communication technology and the update of terminal devices that make data collection and transmission more efficient.In addition,the regularity of the devices also makes the generated data highly redundant,which makes the data face greater pressure in terms of storage and how to effectively carry out subsequent analysis,and data compression can solve the above problems.The goal of data compression algorithms is to reduce the size and cost of the data while keeping the information characteristics as complete as possible.Due to the diversity of data volume and structure,traditional data compression algorithms face problems such as inefficient computation and waste of resources caused by storage in cloud servers.Facing the above problems,this thesis proposes a data compression and recovery scheme for time-series data in the context of "research on security protection technology of edge computing in power system".The scheme eliminates redundancy by calculating the similarity between data,and also models and classifies data according to the characteristics of time series data to reduce data redundancy and ensure optimal data recovery.This thesis focuses on data compression and recovery in edge-oriented power system scenarios,with the following main work.1.To address the redundancy of data and the privacy leakage of data caused by the transmission process or stored in the cloud,this thesis designs a data de-duplication scheme based on dynamic time regularization.The scheme collects data from grid terminals and transmits them to edge nodes,analyzes the data at the edge,segments the time series data using sliding windows,calculates the similarity between subsequences by dynamic time regularization algorithm,and de-duplicates the redundant data according to the similarity.The experimental results show that this scheme saves storage space while compressing the data.2.In order to compress the power time series data more effectively,for the idea of dynamic time regularization algorithm and the limitations such as low efficiency of data segmentation,this thesis designs a data compression and recovery scheme based on improved differential exponential golomb coding algorithm combined with neural network model CBN-VAE.This thesis uses a sliding window function to classify the data into stable and fluctuating data.The improved differential exponential Columbus coding algorithm is used to process the smooth data and the CBN-VAE model to process the fluctuating data to achieve better compression and recovery results.The feasibility of the power timing data de-duplication technique is experimentally demonstrated,and the effectiveness of the data compression and recovery scheme based on an improved differential exponential Columbian coding algorithm combined with CBN-VAE is demonstrated.In this thesis,we use real power timing datasets for experimental validation,and verify the effectiveness and feasibility of the timing data compression and recovery scheme for power edge computing by the above two schemes.