Time Series Data Mining In FY-3MERSI Research

The Meteorological administration researchers use digital number (DN) of Earth View (EVDN) observation and Space View (SVDN) observation of medium resolution spectral imager (MERSI) which is a sensor carried on the FY-3Meteorological satellite to do radiometric calibration. Through the observation of SVDN value, the researchers found the SVDN value was influenced by some factors, so the inaccurate SVDN value will lead to inaccurate radiation calibration result. The purpose of this paper is by means of research and improvement of relevant time series data mining algorithms, and then applied these algorithms to the analysis of SVDN value of FY-3MERSI. In this paper, we use the nearly three year’s data to find out which factors influence the SVDN value. The researchers of Meteorological administration can use the result of this time series data mining to guide they work.Based on the in-depth study of time series data mining algorithms and combination with the background of the research, this paper proposes a method of piecewise linear representation of time series based on angle between lines (PLR_AG) and a time series clustering method based on principal component analysis, and implements the FY-3MERSI time series mining platform. In this paper, the main content and innovation points are as follows:1.Due to the characteristic of high time dimension of time series which will be researched in this paper, through study the exiting time series piecewise linear representation method and found those methods cannot extract the time series which has a small range of fluctuation but a large range of slope change. In order to solve this problem, this paper proposes a method of piecewise linear representation of time series based on angle between lines (PLR_AG). The method can select the cutting-points of the time series by judging the size of angle which is formed by connecting the adjacent points in time series and the vertical distance between adjacent vertexes of angle. Using a new time series that composed by these piecewise points can fit the original time series. Compared with several before piecewise linear representation methods, It can be easily calculated and has a low degree of fitting error.2. According to the high dimensions of feature that will lead to slow speed and wrong time series clustering result. Through study the method of feature dimension reduction and time series clustering, this paper proposes the idea of time series clustering analysis method based on principal component analysis. Firstly, applying principal component analysis to time series dataset, by way of dimension reduction, obtained the corresponding coefficient matrix and eigenvalues. Secondly, using clustering method based on Euclidean distance on the calculated coefficient matrix, the clustering result of coefficient matrix is consistent with time series dataset. This algorithm can guarantee that under the premise of the same or higher clustering accuracy, the time complexity of time series clustering method proposed in this paper is significantly better than the algorithm based on Euclidean distance.3. On the basis of the above two algorithm, a FY-3MERSI time series mining platform was be designed and implemented. The main two functions of this platform are:1. this platform can display and compare the existing method and the proposed method in this paper;2. by using this platform can finally find out the factors which are influence the different channel of FY-3MERSI SVDN value.