The thesis focuses on analysing the behaviour of systems by connecting log analysis with event processing, and subsequently making conclusions in real time about the future state of the system. The analysis is based on typical patterns contained in the time series of logs. Based on this analysis it is possible to predict for example critical situations.
The thesis presents the process of trying to find typical patterns in the data series of logs with the method of searching for shapelets, and classifying data series in real time based on the shapelets found. Pattern searching is possible even in multidimensional data series.
The document is organized as follows: after introducing the background it describes the algorithm of computation and some optimization methods of time-consuming pattern search process. The following chapters of the thesis present the features of offline and online diagnostics and the use of semantic information in the definition of parameters. Finally, the paper demonstrates various possibilities of application through case studies.