Leveraging Neural Networks to Locate Short-Lived Anomalies in Gas Consumption

The core objective of this research revolves around the creation of a robust system designed to identify outliers with a novel approach. This approach combines the strengths of both Artificial Neural Networks (ANNs) and Auto-regressive Integrated Moving Averages (ARIMA) models. While ARIMA excels in linear prediction, ANNs demonstrate their prowess in tackling nonlinear prediction challenges. Through their symbiotic integration, they effectively encapsulate the intricate and intricate nonlinear relationships that underlie the interaction between meteorological variables and gas consumption patterns. The system excels at labeling and pinpointing outliers, allowing these anomalies to be promptly communicated to building managers for further scrutiny and potential rectification measures within HVAC systems. This intervention, in turn, contributes to the reduction of energy wastage. The system functions within a dual-phase framework: first, it forecasts short-term (hourly) gas consumption by tapping into historical time-series data; second, it employs a deviation-based strategy to identify anomalies from projected or expected values. Notably, the identification of these anomalies is accomplished without the requirement for pre-annotated instances. This is attributed to the remarkable precision of the predictions, characterized by a Root Mean Square Error (RMSE) that spans from 8 cubic meters to 2.5 cubic meters.