A Systematic Energy Information Collection Methodology for Improved Energy Analytics
Final Report Abstract
The growing number of installed smart electricity meters makes it possible to capture power consumption data in an unprecedented spatial and temporal resolution. A multitude of data processing approaches have been proposed in scientific literature, seeking to maximize the information content that can be extracted from meter readings and targeting to enable services like forecasts of future demand or the detection of consumption anomalies. Most scientific contributions in this domain aim towards the creation and improvement of data processing algorithms, however, while no specific consideration is usually given to the fundamental question of whether the used input data actually permit the extraction of the desired features. Substantial variations and anomalies exist in such data sets (including, but not limited to, different sampling rates, outliers in the data, different measured quantities, extended periods of appliance inactivity) and have a non-negligible impact on the data processing. As such, an in-depth understanding of the required data properties for repeatable and reliable energy data analytics is strongly needed. Despite the inherent reliance on electrical consumption data, a widely adopted methodology defining how to instrument the environment with sensors in order to attain data of appropriate quality has not emerged to date. The key objective of the SECoM project was therefore to investigate a systematic energy information collection methodology, and thus enable researchers and developers to get a better understanding of the challenges and opportunities that lie within the data collection setup. A primary field of investigation was how the resolution of electrical consumption data impacts the data processing step, for which we were able to demonstrate that data reporting rates in the range from 10–30 seconds are often sufficient for macroscopic load analysis tasks, yet higher resolutions can provide further increments in many cases. Data reporting rates of one measurement per 15-minute interval (used by most state-of-the-art smart meters) are, in contrast, often too low to extract sufficient meaningful information. A secondary observation was that different types of energy data analytics methods not only have different requirements to the temporal and spatial resolution of their input data, but they also need to be suitably parameterized to achieve optimal results. During our research, it became apparent to us that many widely used energy data analytics implementations rely on parameters that are not adapted to the available data properties, but set as fixed constants. In combination with the heterogeneity of the publicly available energy data sets frequently used by researchers, this leads to incomparable results and non-generalizable conclusions as a result thereof. Lastly, besides having identified dependencies and recommendations for parameter configurations of contemporary energy analytics methods, our investigation into the collection of data from supplementary (non-electrical) sensing modalities has also shown the potential positive impact on analytics methods. The holistic approach we have followed in SECoM has allowed us to understand interdependencies between different components of the entire energy data processing sequence and assess their impacts on practical energy data processing. Throughout the project, we have ensured a high practical relevance of our research results by operating on data sets collected in real-world scenarios, either within the project itself or by other researchers, or simulation models derived from such data. The possibility to transfer the project results to other practical use cases and application scenarios is hence inherently given.
Publications
- “A Study on the Impact of Data Sampling Rates on Load Signature Event Detection.” In: Energy Informatics 2, Supplement 1.24 (2019)
Jana Huchtkoetter and Andreas Reinhardt
(See online at https://doi.org/10.1186/s42162-019-0096-9) - “Electricity Consumption Data Sets: Pitfalls and Opportunities.” In: Proceedings of the 6th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys). 2019, pp. 159–162
Christoph Klemenjak, Andreas Reinhardt, Lucas Pereira, Mario Berges, Stephen Makonin, and Wilfried Elmenreich
(See online at https://doi.org/10.1145/3360322.3360867) - “Reliable Streaming and Synchronization of Smart Meter Data over Intermittent Data Connections.” In: Proceedings of the IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm). 2019, pp. 366–371
Chenfeng Zhu and Andreas Reinhardt
(See online at https://doi.org/10.1109/SmartGridComm.2019.8909705) - “How does Load Disaggregation Performance Depend on Data Characteristics? Insights from a Benchmarking Study.” In: Proceedings of the 11th ACM International Conference on Future Energy Systems (e-Energy). 2020, pp. 167–177
Andreas Reinhardt and Christoph Klemenjak
(See online at https://doi.org/10.1145/3396851.3397691) - “On the Impact of the Sequence Length on Sequenceto-Sequence and Sequence-to-Point Learning for NILM.” In: Proceedings of the 5th International Workshop on Non-Intrusive Load Monitoring (NILM). 2020, pp. 75–78
Andreas Reinhardt and Mazen Bouchur
(See online at https://doi.org/10.1145/3427771.3427857) - “Watt’s up at Home? Smart Meter Data Analytics from a Consumer-Centric Perspective.” In: MDPI Energies 14.3 (2021)
Benjamin Völker, Andreas Reinhardt, Anthony Faustine, and Lucas Pereira
(See online at https://doi.org/10.3390/en14030719)
