Detection of Change Point in Process Signals by Cascade Classification
Sergey Gavrin, Damir A. Murzagulov, Alexander Zamyatin · 2018 International Russian Automation Conference (RusAutoCon) · 2018
Nowadays, industrial companies are more than ever forced to dynamically adapt their business process executions to currently existing business situations in order to keep up with increasing market demands in global competition. Companies that are able to analyze the current state of their processes, forecast its most optimal progress and proactively control them basing on reliable predictions will stay significantly ahead their competitors. This result can be achieved by embedding predictive analytics systems, such systems allow to prevent emergencies and optimize existing processes on the basis of historical process data. The analysis and processing of process signals are at the heart of these systems. These tasks are reduced to a number of other subtasks, one of the most important of them is the detection of a change point in the process signals. Change points are abrupt variations in process signals. Such abrupt changes may represent transitions that occur between states of equipment or process stages. The paper proposes the application of the cascade classification of time series approach for detecting the region of a changepoint in process signals. It is assumed that this approach to the refinement of the moment of a changepoint is more suitable for signals characterizing the different flow stages of the processes. The rationale for using this approach in the field of automated control systems is presented, the results of experiments performed on model data with different levels of noise are demonstrated.