Using Generative Neural Networks to Detect Point Anomalies in Energy Consumption Data

The paper describes studies in the field of detecting point anomalies in energy consumption data using two different datasets as examples. The authors chose six methods based on generative artificial neural networks suitable for detecting point anomalies. These are methods based on autoencoders, variational autoencoders, and generative adversarial networks. Anomalies were analyzed using both energy consumption data alone and combined with outdoor air temperature data. Based on the numerical experiment results, methods were identified that performed best in detecting point anomalies. It has been established that these methods are robust to the values of their primary parameters over a wide range. The authors consider them to be the most promising for anomaly analysis of all methods based on generative artificial neural networks.