Research On Real - Time Fault Diagnosis Method Of Synchronous Multidimensional Data Stream Based On Improved Principal Component Analysis

With the rapid development of the computer, the communication technology and the network technology, which led to explosive growth in the amount of information, and there has emerged in massive high-speed, multi-dimensional dynamic data stream in many application areas. It has become the focus of attention that analyzing the data security by abnormality diagnosing for the data stream. Currently, the research of the data stream focused on mining the data elements inside a single data stream. The abnormality diagnosis and the prediction alarm of the practical application engineering were obtained by mining the inherent laws of the data stream. However, the study of multi-dimensional data streams are more concentrated in the data after dimensionality reduction, and combining with other data algorithms.In this paper, according to the characteristics of synchronous multi-dimensional data stream, an improved principal component analysis method for the abnormal information diagnostic of the data stream is presented. In this method, the problem of the original data stream variation tendency is mapped to the eigenvector space, and the steady-state eigenvector is solved, then the abnormal changes of the synchronous multi-dimensional data stream can be diagnosed by the relationship between the instantaneous eigenvector and the steady-state eigenvector. The main contents of this paper includes three parts:Firstly, according to the deep study of the synchronous multi-dimensional data stream mining technology, the feasibility of using the principal component analysis in the dynamic processes abnormal diagnosis and the multi-dimensional data stream compression is systematically summarized. Secondly, according to the key issues of the large dimension and the unobtrusive feature information for the multi-dimensional data stream, an improved principal component analysis method is proposed, it combines the principal component analysis technology with sliding window technology. In a sliding window, the relationship between the original data stream and the eigenvector obtained by using the principal component analysis is constructed. The improved principal component analysis model and the anomaly detection algorithm process for the synchronous multi-dimensional data stream are analyzed. Finally, the feasibility of the theoretical approach is verified by the actual industrial production multidimensional data stream.The study can realize the dynamic analysis of the synchronous multi-dimensional data stream abnormal information, it also can be applied to non-periodic sampling data analysis, and it expands the application fields of principal component analysis method. This method also provides an effective method for the safety warning and fault detection of the actual industrial process data.