Anomaly Detection Algorithm Using a Hybrid Modelling Approach for Energy Consumption Time Series

EasyChair Preprint 6888

Abstract

Many energy time series captured by real-time systems contain errors or anomalies that prevent accurate forecasts of time series evolution. However, accurate forecasting of load time series and fluctuating renewable energy feed-in as well as subsequent optimization of the dispatch of controllable generators, storage and loads is crucial to ensure a cost-effective, sustainable and reliable energy supply. Therefore, we investigate methods and approaches for a system solution that automatically detect and replace anomalies in time series to enable accurate forecasts.

Here, we present an anomaly detection method for the energy sector inhibiting a large automization potential due to the ability to handle previously unknown data sets, which classical approaches like regression models can hardly provide.

At first, we define anomalies as data, which do not correspond to the normal characteristics of a respective time series. In contrast, errors are parts of time series, which are know to be erroneous due to external information, e.~g., information of a fallen power pole. To classify anomalies appearing in consumption time series, we distinguish outliers, incomplete data, change points and anomalous part of time series based on mathematical considerations. We are not aware of any literature presenting a similar formal classification.

To investigate our detection methods, we manipulated real, highly accumulated energy consumption time series, that were manually verified and corrected externally. A classical approach to detect anomalies is to calculate the difference between estimation and observation. e.~g., by using regression. Unfortunately, these approaches are only applicable if a sufficient amount of data for the respective time series are available to calculate an appropriate estimation. Alternative approaches like Shannon entropy and classification are able to detect anomalies in previously unknown time series, yet yielding improvable results.

Keyphrases: Time Series Processing, anomaly detection, energy consumption

@booklet{EasyChair:6888,
  author    = {Florian Rippstein and Steve Lenk and Andre Kummerow and Lucas Richter and Stefan Klaiber and Peter Bretschneider},
  title     = {Anomaly Detection Algorithm Using a Hybrid Modelling Approach for Energy Consumption Time Series},
  howpublished = {EasyChair Preprint 6888},
  year      = {EasyChair, 2021}}