Hybridization of Ontologies and Neural Networks in the Problems of Detecting Anomalies of Time Series

The article describes the results of the development of an algorithm for detecting anomalies of time series taking into account the features of the subject area. The algorithm involves finding a forecast of time series using recurrent neural networks, detecting anomalies according to the obtained forecast, filtering the detected anomalies in accordance with possible deviations of the time series values from the trend reflected in ontology, and logical output of search results using a set of rules. The effectiveness of the proposed approach is confirmed by a number of experiments conducted on the benchmark of data on the operation of oil rigs.