Open Radar Time Series Analysis And Prediction Platform

As a product of modern technology,radar has been widely used in many fields.When the radar works in a complex environment for a long time,we need to strengthen the monitoring and the health condition of radar to ensure the stable operation of the radar system.When the radar works,it not only receives the target data but also generates a large amount of health information data.These data will put tremendous pressure on data storage and archiving.Therefore,how to store these data reasonably and predict the health status of the radar have become the research focus.This thesis researches on the radar health data and implements an open platform for radar time series analysis and prediction.The platform mainly uses Hadoop distributed clusters,including based on HDFS distributed data storage platforms and based on spark distributed data processing frameworks,and takes advantage of the performance of computer clusters to achieve efficient and secure storage and management of radar data.Radar health data storage is based on the platform,it provides data query and analysis mining by building a Hive-based data warehouse.The platform stores the radar data from different sources in a data warehouse for unified management through message decoding and data preprocessing.Radar health data prediction is mainly divided into single variable time series prediction and multivariate variable time series prediction according to the number of variables.For single time series prediction,this thesis uses the traditional Auto Regressive and Moving Average model(ARMA).It is often used for fitting stationary time series.For non-stationary series,we need to use differential signal conversion.Owing to many unknown parameters in ARMA modeling process,and the subjectivity of the estimation of parameter,they will affect accuracy.Therefore,taking advantage of features of the convex optimization model of Support Vector Regression(SVR),SVR model uses the residual data after using ARMA model.Comparing to the single model used to predicting,the method of the combination of ARMA and SVR model show higher precision and better stability.For multivariate time series prediction,in order to reduce the impact of redundant variables,this thesis uses the causality method to perform feature extraction,while retaining the lag value of the series,and then uses support vector regression to predict.It reduces the amount ofcalculation and improves the accuracy of prediction.This thesis completes the development of the data layer,business layer,and interaction layer through the requirements analysis,outline design,detailed design,and the final page design.Platform functions such as radar data message decoding,data import and export,data cleaning,data warehouse establishment,data feature extraction and predictive analysis are implemented.