Abstract
Frequency control is essential to ensure reliability and quality of power systems. North American Electric Reliability Corporation’s (NERC) Control Performance Standard 1 (CPS1) is widely adopted by many operating authorities to examine the quality of the frequency control. The operating authority would have a strong interest in knowing how the frequency-sensitive features affect the CPS1 score and finding out more effective unit-dispatch schedules for reaching the CPS1 goal. As frequency-sensitive features usually possess multi-variable and high-correlated characteristics, this paper employed an ensemble learning technique (the Gradient Boosting Decision Tree algorithm, GBDT) to construct Frequency Response Model (FRM) of the Taipower system in Taiwan to evaluate by CPS1 score. The proposed CPS1 model was then integrated with Unit Commitment (UC) program to determine the unit-dispatch that achieves the targeted CPS1 score. The feasibility and effectiveness of the proposed CPS1-UC platform were validated and compared with the other benchmark model-based UC methods by two operating cases. The proposed model shows promising results: the system frequency could be maintained well, especially in the periods of the early morning or the high renewable penetration.
Highlights
The problem of determining the optimal unit commitment (UC) in power generation systems is solved by minimizing the total system cost while maintaining the load generation balance
We focus on the daytime periodowwheerngtehneerPaVtiopno.wInerthgiesncearsaet,iwone wfoacsusacotnivteh.eTdaabyletim3 sehpoewrisodth(efrcoomst7coam.m-. ptoar3ispo.mn .b)ywthheendtihffeePreVntptohwreeer gmeentehroadtison(i.we.a, sRaOctLivFeC.-TUaCblevs3. sChPoSw1sUtChevcso.sLt ScTomMp-UarCis)o. nFibgyuthree 1d1iffsehroewntstthhraetetmheetthhroedesm(i.oed.,eRl-ObaLsFeCd-UUCC avlsl.aCchPiSe1vUedCtvhse. tLaSrTgeMt -CUPCS)1. sFcigourer.eH1o1wsheovwers, tthheatvtahleuethsroefeinmeordtieal,rbaamsepdrUatCe aanlldacAhGieCverdestehrevtea(rAgeGtCCPuSp1asncdorAe.GHCodwoewvenr), athree dviaflfuereesnotf
The proposed Frequency Response Model (FRM) was constructed by an ensemble learning technique, the gradient boost decision tree (GBDT) algorithm, to deal with feature identification and model training
Summary
The problem of determining the optimal unit commitment (UC) in power generation systems is solved by minimizing the total system cost while maintaining the load generation balance. Several statistical-based methods have been proposed to evaluate the CPS score by constructing a reference based on the load frequency control (LFC) model or UC model. This paper attempts to introduce the supervised machine learning algorithm to construct the frequency response model (FRM), and integrate the model into the unit commitment program. (4) Comparative results from different methodologies have shown that the proposed FRM model is superior in portraying system frequency response over the simplified model that did not take the multivariate and nonlinear characteristics into account.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
