STEAMCODER: Spatial and Temporal Adaptive Dynamic Convolution Autoencoder for Anomaly Detection

doi:10.1016/j.knosys.2023.110929

STEAMCODER: Spatial and Temporal Adaptive Dynamic Convolution Autoencoder for Anomaly Detection

2023-11-01

3

PRE

AI

DOI：10.1016/j.knosys.2023.110929

Original

Original request for help

Save

Abstract

En 中文

The anomaly detection algorithm greatly improves the reliability of equipment operation. Traditional anomaly detection algorithms are mostly designed for large data sets, making it difficult to detect anomalies when where is not enough accumulated equipment data. Therefore, detecting anomalies during the cold-start stage is also challenging. In industrial production, the working conditions of the equipment are constantly adjusted, and the amount or characteristics of data for training also change frequently. Hence, an adaptive algorithm is required to address these problems. We design a novel algorithm called Spatial-TEmporal Adaptive dynaMic Convolution autoencODER for Anomaly Detection (STEAMCODER), specifically. This algorithm first converts the data into a spatial-temporal anomaly feature matrix and then utilizes a dynamic convolution autoencoder to analyze the matrix and detect anomalies. Finally, we conduct extensive experiments to validate the performance of STEAMCODER and the results demonstrated its superiority compared to state-of-the-art algorithms in adaptive anomaly detection, including F1 score and other indicators. Furthermore, STEAMCODER is capable of filtering false positives caused by glitch data, enabling the early detection of equipment anomalies. (c) 2023 Elsevier B.V. All rights reserved.

Keywords:

Anomaly detection

Dynamic CNN

Autoencoder

STEAMCODER: Spatial and Temporal Adaptive Dynamic Convolution Autoencoder for Anomaly Detection

Abstract

Journal

Organization