Uticaj generisanih sintetičkih vremenskih serija na tačnost predikcije u sistemima daljinskog grejanja
##plugins.themes.bootstrap3.article.sidebar##
##plugins.themes.bootstrap3.article.main##
http://orcid.org/0009-0001-9160-6379
Ivan Ćirić
http://orcid.org/0000-0003-0430-8937
Ljiljana Radović
http://orcid.org/0000-0002-4099-1289
Vladan Jovanović
http://orcid.org/0009-0002-8564-9900
Marko Ignjatović
http://orcid.org/0000-0001-7205-7987
Miloš Simonović
http://orcid.org/0000-0003-1364-7746
Apstrakt
SCADA sistemi u sistemima daljinskog grejanja (DHS) često beleže podatke u neregularnim vremenskim intervalima, što je posledica različitih tehničkih, infrastrukturnih i operativnih faktora. Ovakva neregularnost otežava primenu standardnih modela mašinskog učenja, koji obično zahtevaju vremenske serije sa ujednačenim uzorkovanjem. Pored toga, neregularne vremenske serije negativno utiču na rad brojnih specijalizovanih softverskih alata u energetici, uključujući one koji se koriste za simulaciju i optimizaciju energetskih sistema, upravljanje potrošnjom, obračun troškova i izveštajne module koji zahtevaju precizne vremenske agregacije. Kako bi se prevazišao ovaj izazov, u ovom radu se predlaže metodologija za generisanje sintetičkih vremenskih serija koje realistično oponašaju operativno ponašanje sistema i dopunjuju neregularne ili nepotpune zapise podataka. Poseban akcenat stavlja se na upotrebu modela zasnovanih na generativnim neuronskim mrežama, kao i na statističke pristupe za rekonstrukciju podataka. U eksperimentalnom delu rada biće sprovedena uporedna validacija kvaliteta rekonstrukcije neregularnih serija dobijenih: (1) proširenjem realnih podataka sintetičkim generacijama i (2) interpolacijom. Rezultati će pokazati u kojoj meri sintetički podaci i metode popunjavanja poboljšavaju tačnost, robusnost i pouzdanost daljih analiza i softverskih alata za energetske sisteme.
##plugins.themes.bootstrap3.article.details##
Reference
[2] Z. Xiao et al., “FLEXTSF: A Universal Forecasting Model for Time Series with Variable Regularities,” arXiv preprint, arXiv:2306.08892, 2023.
[3] X. Zhang et al., “Irregular Multivariate Time Series Forecasting: A Transformable Patching Graph Neural Networks Approach,” arXiv preprint, arXiv:2306.12543, 2023.
[4] J. Klötergens et al., “Functional Latent Dynamics for Irregularly Sampled Time Series Forecasting,” arXiv preprint, arXiv:2306.00594, 2023.
[5] S. Grosswindhager, M. Kozek, A. Voigt, and L. Haffner, “Fuzzy predictive control of district heating network,” International Journal of Modelling, Identification and Control, vol. 19, no. 2, pp. 161–170, 2013.
[6] C. Tian, Y. Ye, Y. Lou, W. Zuo, G. Zhang, Chengdong Li, et al., “Daily power demand prediction for buildings at a large scale using a hybrid of physics-based model and generative adversarial network,” Building Simulation, vol. 15, pp. 1685-1701, 2022. DOI: 10.1007/s12273-022-0887-y.
[7] G. Chaudhary, H. Johra, L. Georges, and B. Austbø, “Synconn_build: A python based synthetic dataset generator for testing and validating control-oriented neural networks for building dynamics prediction,” Energy and Buildings, vol. 293, p. 113247, 2023. DOI: 10.1016/j.enbuild.2023.113247.
[8] X. Wang, J. Harrison, R. Teigland, and A. Hollberg, “Machine Learning (ML) as a Surrogate Model for Early-stage Heating Demand Optimization,” in Proc. Eleventh National Conference of IBPSA-USA (SimBuild 2024), Denver, CO, USA, May 21–23, 2024.
[9] Z. Wang and T. Hong, “Generating realistic building electrical load profiles through the Generative Adversarial Network (GAN),” Lawrence Berkeley National Laboratory, Building Technology and Urban Systems Division, Berkeley, CA, USA, 2022.
[10] *** National Renewable Energy Laboratory (NREL), “ARPA-E GRID DATA and Smart-DS: Large-scale synthetic transmission and distribution datasets for power system planning and optimization,” NREL Technical Report, 2021.
[11] L. Richter, T. Bender, S. Lenk, and P. Bretschneider, “Generating Synthetic Electricity Load Time Series at District Scale Using Probabilistic Forecasts,” Energies, vol. 17, no. 7, p. 1634, 2024. DOI: 10.3390/en17071634
[12] H. Li, Z. Wang, and T. Hong, “A synthetic building operation dataset,” Scientific Data, vol. 8, no. 1, Art. 213, 2021. DOI: 10.1038/s41597-021-00989-6.
[13] D. Patel et al., “EMIT: Event-Based Masked Auto Encoding for Irregular Time Series,” arXiv preprint, arXiv:2307.03696, 2023.
[14] J. Oh et al., “DualDynamics: Synergizing Implicit and Explicit Methods for Robust Irregular Time Series Analysis,” arXiv preprint, arXiv:2307.00117, 2023.
[15] V. Yalavarthi et al., “ProFITi: Probabilistic Forecasting of Irregularly Sampled Time Series with Missing Values,” arXiv
[16] J. Chen, J. Yuan, W. Chen, A. Zeb, M. Suzauddola, and Y. A. Nanehkaran, “Research on Interpolation Method for Missing Electricity Consumption Data,” Computers, Materials & Continua, vol. 78, no. 2, pp. 2576–2591, Feb. 2024, doi:10.32604/cmc.2024.050414.
[17] M. Schaffer, T. Tvedebrink, and A. Marszal-Pomianowska, “Three years of hourly data from 3021 smart heat meters installed in Danish residential buildings,” Scientific Data, vol. 9, p. 420, 2022, doi: 10.1038/s41597-022-01502-3.
[18] S. Chatterjee, P. K. Jana, and A. Bandyopadhyay, “Prediction of household-level heat consumption using PSO-kSVR with accumulated smart meter data,” arXiv preprint, arXiv:2112.01908, Dec. 2021.