Research And Implementation Of Smart Grid Time-Series Storage And Forecasting Technique

With the rapid development of technologies in Internet of things(IOT),systems of IOT are gradually and widely used in various industries.Smart Grid is one of the typical IOT applications on which a huge number of sensors have been deployed in the whole network to gather and generate all kinds of fresh and mass time-series data incessantly.Smart Grid data in the paper are specified as data collected from smart meters deployed in buildings,which are essentially time-series.These are very important for customers and electricity companies because they can be used in lots of electricity services,such as tracking power consumption and forecasting power load.However,with the rapid lateral growth of smart grid time-series,the processing procedure is faced with great challenges,such as storage and analysis.We have to look for new solutions that satisfy applications service needs while ensuring smart grid safe and stable operation.The goal of our paper is on the processing technique of smart grid time-series,which is mainly divided into smart grid time-series storage and load forecasting technique.At first,in order to improve efficiency of mass time-series data accessing,reduce storage space and satisfy smart grid service needs,a three-dimensional data model based on HBase is proposed.According to smart grid application scenarios,there are usually two ways to access power load data: polygonal line query and section query.The former gets related measured values of a specified user smart meter over a period of time,which uses monotone increasing timestamps as the third dimension in HBase tables.The latter gets related measured values of all user smart meters at one moment across a particular area,which uses monotone increasing uids as the third dimension.The experiment results show the model increases storage capacity in a single row,and data are distributed more evenly in the HBase cluster.So data accessing speed is greatly improved.In addition,large amount of redundant information in HBase tables is reduced,which leads to less storage space.Then,in order to improve accuracy of smart grid short-term load forecasting,and reduce model complexity,a short-term load forecasting model using k-NN based on ICS-SVM framework is proposed.It takes many factors into account,such as temperature,holiday,week and so on.A hybrid approach is adopted to avoid error propagation.An improved cuckoo search algorithm is adopted to obtain appropriate SVM parameters because of its strong ability of global optimization.K-nearest neighbor instance selection strategy is adopted to reduce the complexity of SVM regressor,by extracting k training patterns closest to the test case from training set for a new thin training set.It works for univariate and multivariate cases.It also works for one-step as well as multi-step forecasting.The experiment results show the forcasting model is ingenious and fine,and regression indicators are more effective than other methods.Finally,a smart grid time-series distributed storage system is designed and implemented to overcome the shortcomings of HBase,such as native HTable API is single and thread unsafe,and storage scheme is not flexible enough.The three-dimensional data model proposed previously is used to redesign tables to support polygonal line query and section query.It extends basic function that supports users to store,retrieve,delete and update items in HBase.It not only meets the requirement of smart grid applications,but also has high fault-tolerance,flexibility and stability.