Deep autoencoders for anomaly detection in electricity consumption of rotor spinning system

The anomaly detection of electricity quantity in the rotor spinning system is crucial for achieving energy savings in yarn production. The characteristics of varied working condition and special process of the rotor spinning system, make previous data-driven methods hard to deal with, since the time series data of electricity consumption is influenced by various factors, the dynamics of which are often intertwined in many modalities. Motivated by the deep autoencoders that can be adapted to work with a wide variety of input modalities, this paper constructs four types of autoencoder-based models for detecting anomalies in electricity consumption of rotor spinning system, which are composed of convolutional, recurrent, and attention layers. The autoencoder-based model with best performance is selected and benchmarked against the state-of-the-art methods in experiments. Experimental results on real data sets show that the proposed LSTM VAE method has more advantages than several existing mainstream methods, and offers an innovative approach to model selection for identifying energy consumption anomalies in the rotor rotation system.