Anomaly Detection on Compressed-Decompressed Time Series Data: A Deep Learning Approach
Industry 4.0, Anomaly Detection, Symbolic Aggregate Approximation, Swinging Door Trending, Multi-layer Perceptron, Long Short-Term Memory
The development of technologies such as Cloud Computing and the Internet of Things promoted extensive data generation in Industry 4.0. This growth allowed to enhance process variables monitoring and controlling. It increased the energy consumed in data warehouses, CO2 emission, and the computational cost of applying data mining algorithms like anomaly detection. To overcome this, in this project, we propose the usage of compression algorithms such as Symbolic Aggregate Approximation and Swinging Door Trending in multi-variable industrial time series to reduce the load on databases, while also evaluating the performance of Multi-layer Perceptron and Long Short-Term Memory Neural Networks for detecting anomalies on compressed-decompressed and the original data of the Tennessee Eastman Process benchmark simulation. We will use binary classification metrics of accuracy, precision, recall, and F1-score for anomaly detection models and the compression metrics of compression rate and compression error to validate the algorithms used to compress time series to validate our approach.