GREEND: An energy consumption dataset of households in Austria and Italy
3 likes2,922 views
The document presents an energy consumption dataset from households in Italy and Austria, highlighting the need for diverse datasets to support energy management and research. It details the features of the GREEND dataset, its methodology for data collection, and its potential applications in non-intrusive load monitoring and occupancy detection. The authors also discuss future work, including enhancements to the dataset and further analyses to support energy conservation strategies.
GREEND: An energy consumption dataset of households in Austria and Italy
- 1. An energy consumption dataset of households in Italy and Austria Andrea Monacchi, Dominik Egarter, Wilfried Elmenreich, Salvatore D’Alessandro, Andrea M. Tonello
- 2. Motivation ● Recent years ○ Massive installation of RE & electrical loads (e.g. vehicles) ○ Growth of research on energy management & sustainability ● Demand response & HEMS ○ Monitoring and rationalization of local energy ○ User in the loop VS autonomous operation ● Datasets necessary to validate research
- 3. Energy consumption datasets ■ Few households, very short campaign, very high resolution ➢ Interesting for load disaggregation community ■ Few households, long-term campaign, very low resolution ➢ necessary for energy consumption modeling & forecast ■ Many households, medium length, low resolution ➢ Necessary for statistical significance ■ Many devices, no household association, low duration & resolution ➢ Source of device power profiles to extract signatures
- 4. ACS-F1 Switzerland 1 hour session N/A 100 (10 types) I, V, Q, f,Φ 10 secs Tracebase Germany N/A N/A 158 (43 types) P 1 - 10 secs Dataset Location Duration Houses Sensors Features Resolution BLUED Pennsylvania 8 days 1 Agg. I, V, switch events 12 KHz REDD Boston 3-19 days 6 9-24 V + Q (Agg), P (Sub) 15 KHz UK-DALE UK 499 days 4 5 - 53 (house 1) P (Agg), P (Sub) 16 KHz, 6 KHz AMPds Vancouver 1 year 1 19 I, V, pf, f, P, Q, S 1 min iHEPCDS France 4 years 1 3 circuits I, V, P, Q 1 min HES UK 1 month 251 13-51 P 2 min OCTES FI, IS, Scot. 4-13 months 33 Agg. P, energy price 7 secs GREEND Austria, Italy 1 year 8 9 P 1 Hz iAWE India 73 days 1 33 (10 dev. level) I, V, f, P, S, E, Φ 1 Hz Sample Texas 7 days 10 12 S 1 min Smart* Massachusetts 3 months 1 sub + 2 (sub & agg) 25 circuits, 29 device monitors P + S (circuits), P (submetered) 1 Hz
- 5. The GREEND dataset ● Features: ○ minimal requirements given by NILM ○ active power (P) @ 1Hz, 1 year ● Scenario selection: ○ greedy common devices (survey study in CAR & FVG) ○ aim at diversity in dwelling and residents characteristics A. Monacchi, W. Elmenreich, S. D’Alessandro, A. M. Tonello. Strategies for Domestic Energy Conservation in Carinthia and Friuli-Venezia Giulia. In Proceedings of the 39th Annual Conference of the IEEE Industrial Electronics Society (IECON), Vienna, Austria, 2013. Zeifman, M. Disaggregation of home energy display data using probabilistic approach. IEEE Transactions on Consumer Electronics, vol.58, no.1, pp.23-31, February 2012. GREEND = GREEND Electrical ENergy Dataset
- 6. Deployments House Residents Place Deployment 0 retired couple (2 p) Spittal / Drau (AT) Coffee machine, washing machine, radio, water kettle, fridge w/ freezer, dishwasher, kitchen lamp, TV, vacuum cleaner 1 young couple (2 p) Klagenfurt (AT) Fridge, dishwasher, microwave, water kettle, washing machine, radio w/ amplifier, drier, kitchenware (mixer and fruit juicer), bedside light 2 mature couple with adult son (3 p) Spittal / Drau (AT) TV, NAS, washing machine, drier, dishwasher, notebook, kitchenware, coffee machine, bread machine 3 mature couple with 2 young kids (4 p) Klagenfurt (AT) Entrance outlet, Dishwasher, water kettle, fridge w/o freezer, washing machine, hair drier, computer, coffee machine, TV 4 young couple (2 p) Udine (IT) Total outlets, total lights, kitchen TV, living room TV, fridge w/ freezer, electric oven, computer w/ scanner and printer, washing machine, hood 5 mature couple with adult son (3 p) Colloredo di Prato (IT) Plasma TV, lamp, toaster, stove, iron, computer w/ scanner and printer, LCD TV, washing machine, fridge w/ freezer 6 mature couple with 2 young kids (4 p) Udine (IT) Total ground and first floor (including lights and outlets, with whitegoods, air conditioner and TV), total garden and shelter, total third floor. 7 retired couple (2 p) Basiliano (IT) TV w/ decoder, electric oven, dishwasher, hood, fridge w/ freezer, kitchen TV, ADSL modem, freezer, laptop w/ scanner and printer
- 7. Handling the dataset ● Dataset modeled using NILM metadata ○ building, appliances, residents information alongside data ● Importer for NILM toolkit ○ benchmarking nilm approaches ○ general tool for consumption data processing ○ GREEND Data/metadata directly accessible for analysis J. Kelly and W. Knottenbelt. Metadata for Energy Disaggregation. In The 2nd IEEE International Workshop on Consumer Devices and Systems (CDS 2014), Västerås, Sweden, Jul. 2014. N. Batra, J. Kelly, O. Parson, H. Dutta, W. Knottenbelt, A. Rogers, A. Singh, and M. Srivastava. Nilmtk: An open source toolkit for non-intrusive load monitoring. In Proceedings of the 5th International Conference on Future Energy Systems, ser. e-Energy ’14. ACM, 2014, pp. 265-276. http://nilmtk.github.io
- 8. The measurement platform 1. Plugwise smart plugs 2. ARM-based gateway 3. Gateway dæmon ○ Periodic remote update ○ Best-effort round-robin collection ○ 4 storage modalities Codebase freely available at: http://sourceforge.net/projects/monergy/
- 9. Use case 1: NILM D. Egarter, V. P. Bhuvana, and W. Elmenreich. PALDi: Online load disaggregation via particle filtering. IEEE Transactions on Instrumentation and Measurement. 2014. no.99, pp.1,1. Factorial HMM
- 10. Use case 1: NILM House #0 House #1 Type Accuracy Type Accuracy TV 0.91 hair dryer 0.99 coffee machine 0.99 light 0.97 dishwasher 0.96 dishwasher 0.46 fridge 0.93 fridge 0.90 vacuum cleaner 0.99 water kettle 0.99 water kettle 0.99 washing machine 0.80 Accuracy = (TP + TN) / N ● House #0 and #1 for 7 days ● 1000 particles (empirically chosen) D. Egarter, V. P. Bhuvana, and W. Elmenreich. PALDi: Online load disaggregation via particle filtering. IEEE Transactions on Instrumentation and Measurement. 2014. no.99, pp.1,1.
- 11. Use case 2: occupancy detection ● Inferring presence in environments ○ Normally through specific sensors (motion, doors, acoustic, cameras) ○ Few recent works based on energy use: cheap, non-intrusive ● NIOM (Non-intrusive occupancy monitoring) ○ does not require further HW and works with low frequency data ○ baseline threshold values computed during inactivity periods (night) ■ mostly non-user-driven devices (fridge, HVAC, etc.) ■ inaccurate as night unusage does not imply unoccupancy D. Chen, S. Barker, A. Subbaswamy, D. Irwin, and P. Shenoy. Non-intrusive occupancy monitoring using smart meters. in Proc. of the 5th ACM Workshop on Embedded Systems For Energy-Efficient Buildings (BuildSys’13), 2013.
- 12. Use case 2: occupancy detection D. Chen, S. Barker, A. Subbaswamy, D. Irwin, and P. Shenoy. Non-intrusive occupancy monitoring using smart meters. in Proc. of the 5th ACM Workshop on Embedded Systems For Energy-Efficient Buildings (BuildSys’13), 2013. Weekdays Weekends NIOM on House #4 Actuation Example occupancy over 3 weekdays Credit: NEST thermostat
- 13. Use case 3: appliance usage mining ● Forecasting device operation ○ Usage events easily extracted from consumption dataset ○ Previous work with ANNs, BN, HMMs, EGHMMs ● Usage model as a Bayesian network ○ The BN is simple to use and provides a conditional probability distribution, useful to model device starting L. Hawarah, S. Ploix, and M. Jacomino. User behavior prediction in energy consumption in housing using bayesian networks. in Artificial Intelligence and Soft Computing, Lecture Notes in Computer Science. Springer Berlin Heidelberg, 2010, vol. 6113, pp. 372--379 C. Chang, P.-A. Verhaegen, J. R. Duflou, M. M. Drugan, and A. Nowe. Finding Days-of-week Representation for Intelligent Machine Usage Profiling. Journal of Industrial and Intelligent Information, vol. 1, no. 3, pp. 148-154, 2010.
- 14. Use case 3: appliance usage mining ● Bayesian network: ○ DAG G = (V, E) ■ V = {X1 , … ,Xn } random variables ■ ∀ e ϵ E, e = Xi → Xj , e quantified as P(Xj |Xi ) ○ ∀ Xi ϵ V, its CPD given as P(X1 , … ,Xn ) = P(Xi | pa(Xi )) ● CPT learning: ○ expectation maximization (EM) ● Inference: ○ exact (Junction tree alg.) ○ approximated (MCMC algs.) Example starting probability for house #0 Coffee
- 15. Conclusions ● GREEND: Yet another dataset! ○ Dataset and metadata openly released ○ Measurement platform codebase freely released ● Showed possible use of the data in current research topics
- 16. Future work ● Dataset ○ Aggregated consumption at higher frequency ○ Production from PV in residential environment ○ Commercial and office buildings ● Further data analysis ○ Extraction of appliance usage models for simulation tools ○ Management/Control strategies based on given models
- 17. Questions? Andrea Monacchi Smart Grid group Institute of Networked and Embedded Systems Alpen-Adria Universität Klagenfurt E: [email protected] W: http://wwwu.aau.at/amonacch