Виявлення аномальних даних у процесі калібрування вимірювальної техніки за допомогою автоенкодерів

Abstract

The article considers the problem of automated detection of anomalous values in data arrays generated during the calibration of measuring instruments. A method based on neural networks of the “autoencoder” type is proposed for effective filtering of outliers in measurement data. The theoretical foundations of the functioning of autoencoders, their architectural features, and training algorithms are considered. The nature of the occurrence of anomalies in the calibration process is analyzed in detail, their types are classified, and statistical methods of preliminary data processing are described.