Mobile QR Code QR CODE : Journal of the Korean Institute of Illuminating and Electrical Installation Engineers

Journal of the Korean Institute of Illuminating and Electrical Installation Engineers

ISO Journal TitleJ Korean Inst. IIIum. Electr. Install. Eng.

  1. (Researcher, Gridwiz Inc.)
  2. (Associate Research Engineer, Gridwiz Inc.)



Clustering, Energy Storage System, Load Pattern, Pattern Classification Labeling, Peak Shaving, Scheduling

1. ์„œ ๋ก 

์ตœ๊ทผ ์ „๋ ฅ์‚ฐ์—…์—์„œ ํƒˆ์›์ „ ๋ฐ ๋ฏธ์„ธ๋จผ์ง€, ์ด์ƒ๊ธฐํ›„ ๋“ฑ์˜ ์ด์Šˆ๋“ค์ด ๋Œ€๋‘๋˜๋ฉด์„œ DR(Demand Response), ์‹ ์žฌ์ƒ์—๋„ˆ์ง€, ESS(Energy Storage System), EMS (Energy Management System)์™€ ๊ฐ™์€ ์ฒญ์ •์—๋„ˆ์ง€ ๋ฐ ์—๋„ˆ์ง€ ํšจ์œจํ™”์˜ ์ค‘์š”์„ฑ์ด ๋‚˜๋‚ ์ด ๋†’์•„์ง€๊ณ  ์žˆ๋‹ค[1-2]. DR์˜ ์ ์ •์šฉ๋Ÿ‰ ์‚ฐ์ •, ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜ ์ตœ์  ์šด์˜ ๊ณ„ํš, ์‹ ์žฌ์ƒ์—๋„ˆ์ง€ ๋ฐœ์ „๋Ÿ‰ ์˜ˆ์ธก๊ณผ ๊ฐ™์€ ์—๋„ˆ์ง€ ์†”๋ฃจ์…˜ ๊ด€๋ จ ์—ฐ๊ตฌ๊ฐ€ ํ™œ๋ฐœํžˆ ์ง„ํ–‰๋˜๊ณ  ์žˆ์œผ๋ฉฐ[3-5], ์ด ์ค‘์—์„œ ๋ถ€ํ•˜๋Ÿ‰ ์˜ˆ์ธก์€ ์—๋„ˆ์ง€ ํšจ์œจํ™”์˜ ์ค‘์š”ํ•œ ๋ถ€๋ถ„์ด๋‹ค. ํŠนํžˆ ๊ณต์žฅ ๋ถ€ํ•˜์šฉ ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ๊ฒฝ์šฐ ๋ถ€ํ•˜ ํŒจํ„ด์˜ ์˜ˆ์ธก ๊ฒฐ๊ณผ์— ๋”ฐ๋ผ ์ถฉ/๋ฐฉ์ „ ์‹œ์ ์ด ๋ณ€๋™ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์˜ˆ์ธก์˜ ์ •ํ™•๋„๋Š” ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜ ์ตœ์  ์šด์˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์ตœ์ ํ™”์™€ ์ง์ ‘์ ์ธ ๊ด€๋ จ์ด ์žˆ๋‹ค. ๋ถ€ํ•˜ ํŒจํ„ด์˜ ์ผ๋ฐ˜์ ์ธ ์˜ˆ์ธก ๊ธฐ๋ฒ•์€ ์„ ํ˜•ํšŒ๊ท€ ๋ถ„์„๋ฒ•, ์ง€์ˆ˜ํ‰ํ™œํ™”๋ฒ•, ์‹œ๊ณ„์—ด ๋ถ„์„๋ฒ• ๋“ฑ์„ ์ฃผ๋กœ ์‚ฌ์šฉํ•œ๋‹ค[6-7]. ์ตœ๊ทผ์—๋Š” ๋””์ง€ํ„ธ ๋ฐ์ดํ„ฐ(ํ•™์Šต ๋ฐ์ดํ„ฐ)์˜ ์ฆ๊ฐ€์™€ ๊ธฐ๊ณ„ํ•™์Šต ๊ธฐ์ˆ ์ด ๋ฐœ์ „ํ•จ์— ๋”ฐ๋ผ ๋จธ์‹ ๋Ÿฌ๋‹, ์ธ๊ณต์‹ ๊ฒฝ๋ง ๋“ฑ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ๋ถ€ํ•˜ ํŒจํ„ด ์˜ˆ์ธก ๊ธฐ๋ฒ•์ด ํ™œ๋ฐœํžˆ ์—ฐ๊ตฌ๋˜๊ณ  ์žˆ๋‹ค[8-9].

๊ธฐ๊ณ„ ํ•™์Šต์„ ํ†ตํ•ด ๋ถ€ํ•˜ ํŒจํ„ด์˜ ์˜ˆ์ธก ์‹ ๋ขฐ๋„์™€ ์ •ํ™•๋„๋ฅผ ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•ด์„œ ํ•™์Šต ์ „ ๋ถ€ํ•˜ ํŒจํ„ด์„ ๋ถ„๋ฅ˜ํ•˜์—ฌ ๋ผ๋ฒจ๋งํ•˜๋Š” ๊ณผ์ •์ด ํ•„์š”ํ•˜๋‹ค. ํŠนํžˆ ๋ณต์žกํ•˜๊ณ  ๋ถˆ๊ทœ์น™ํ•œ ๋ถ€ํ•˜ ํŒจํ„ด์˜ ๊ฒฝ์šฐ ์ผ์ • ๊ธฐ์ค€์œผ๋กœ ๋ผ๋ฒจ๋งํ•˜์—ฌ ํ•™์Šตํ•  ๊ฒฝ์šฐ ๊ฒฐ๊ณผ๊ฐ’์ด ํ–ฅ์ƒ๋œ๋‹ค. ์ฆ‰ ํ•™์Šต ์ „ ํŒจํ„ด ๋ถ„๋ฅ˜ ๋ผ๋ฒจ๋ง์€ ์ธ๊ณต์ง€๋Šฅ(AI)์ด ํ•™์Šต ๋ฐ์ดํ„ฐ์— ๋„ˆ๋ฌด ์ง€๋‚˜์น˜๊ฒŒ ๋งž์ถ”์–ด ์ผ๋ฐ˜ํ™” ์„ฑ๋Šฅ์ด ๋–จ์–ด์ง€๋Š” ๊ณผ์ ํ•ฉ(Over Fitting)๊ณผ ์ ์ • ์ˆ˜์ค€์˜ ํ•™์Šต์„ ํ•˜์ง€ ๋ชปํ•˜์—ฌ ์‹ค์ œ ์˜ˆ์ธก ์„ฑ๋Šฅ์ด ๋–จ์–ด์ง€๋Š” ๊ณผ์†Œ์ ํ•ฉ(Under Fitting) ์ฆ์ƒ์„ ๋ฐฉ์ง€ํ•˜๋ฉฐ, ๋ถ€ํ•˜ ํŒจํ„ด ์˜ˆ์ธก ๊ฒฐ๊ณผ์— ๋Œ€ํ•œ ์˜ค์ฐจ์œจ์„ ๊ฐœ์„ ํ•˜์—ฌ ์˜ˆ์ธก์— ๋Œ€ํ•œ ์ •ํ™•๋„๋ฅผ ํ–ฅ์ƒ์‹œํ‚ฌ ์ˆ˜ ์žˆ๋‹ค.

๋ฐ์ดํ„ฐ ๊ตฐ์ง‘ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์ค‘ ๋Œ€ํ‘œ์ ์ธ ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ์žˆ๋‹ค[10]. K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ด์šฉํ•œ ๋ฐ์ดํ„ฐ ๊ตฐ์ง‘์€ ๋Œ€์šฉ๋Ÿ‰ ๋ฐ์ดํ„ฐ์˜ ์‚ฌ์ „์ •๋ณด ์—†์ด ์˜๋ฏธ ์žˆ๋Š” ๊ตฌ์กฐ๋ฅผ ์ฐพ์•„๋‚ด๊ธฐ ์ ํ•ฉํ•˜๋ฉฐ, ๋ฐ์ดํ„ฐ๋ฅผ ํ˜•ํƒœ์— ๋งž๊ฒŒ๋งŒ ๋ณ€ํ™˜ํ•œ๋‹ค๋ฉด ๋‹ค์–‘ํ•œ ํ˜•ํƒœ์˜ ๋ฐ์ดํ„ฐ์— ์ ์šฉ ๊ฐ€๋Šฅํ•˜๋‹ค๋Š” ์žฅ์ ์ด ์žˆ๋‹ค. ํ•˜์ง€๋งŒ ๋‹จ์ ์œผ๋กœ ์ดˆ๊ธฐ ๊ตฐ์ง‘ ์ˆ˜์— ๋Œ€ํ•ด ์‚ฌ์šฉ์ž๊ฐ€ ๋ณ€์ˆ˜๋ฅผ ์„ค์ •ํ•˜์—ฌ์•ผ ํ•˜๋ฉฐ, ์„ค์ • ๊ฐ’์ด ์ ํ•ฉํ•˜์ง€ ์•Š์„ ๊ฒฝ์šฐ ์ •ํ™•ํ•œ ๋ถ„๋ฅ˜๋ฅผ ํ•˜์ง€ ๋ชปํ•˜๊ณ , ๋…ธ์ด์ฆˆ๋‚˜ ์•„์›ƒ๋ผ์ด์–ด๊ฐ€ ์กด์žฌํ•˜์—ฌ ๋ถ„์„ ๋ฐ์ดํ„ฐ์˜ ์œ ์‚ฌ์„ฑ์ด ์ƒ์ดํ•œ ๊ฒฝ์šฐ ํ•ด๋‹น ๊ฐ’์— ๋ฏผ๊ฐํ•˜๊ฒŒ ๋ฐ˜์‘ํ•˜์—ฌ ๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ์— ์˜ํ–ฅ์„ ์ฃผ๊ฒŒ ๋œ๋‹ค. ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ์šด์ „ ๊ณ„ํš์„ ์‚ฐ์ •ํ•˜๊ธฐ ์œ„ํ•œ ๋ชฉ์ ์œผ๋กœ ๋จธ์‹ ๋Ÿฌ๋‹ ๋ถ€ํ•˜ ์˜ˆ์ธก์„ ์‚ฌ์šฉํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ์ถฉ/๋ฐฉ์ „ ์‹œ์  ๊ฒฐ์ •์ด ๊ฐ€์žฅ ์ค‘์š”ํ•˜๋‹ค. K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฒฝ์šฐ ๋ถ€ํ•˜๋Ÿ‰์˜ ์ฆ๊ฐ€๊ฐ์†Œ ํŒจํ„ด์ด ์•„๋‹Œ ๋ถ€ํ•˜๋Ÿ‰์˜ ๋ถ„ํฌ๋กœ ๋ถ„๋ฅ˜ํ•˜์—ฌ ํ•ด๋‹น ๊ฒฐ๊ณผ๋กœ ๋งŒ๋“ค์–ด์ง€๋Š” ํ•™์Šต ๋ฐ์ดํ„ฐ์— ํฐ ์˜ค์ฐจ๊ฐ€ ๋ฐ˜์˜๋  ์ˆ˜ ์žˆ๋‹ค.

๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๋ถ€ํ•˜ ์˜ˆ์ธก AI๋ฅผ ํ•™์Šต์‹œํ‚ค๊ธฐ ์œ„ํ•ด ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜ ์ถฉ/๋ฐฉ์ „ ๊ณ„ํš ์‚ฐ์ •์— ์ ํ•ฉํ•˜๋„๋ก ๊ฐœ์„ ๋œ ํŒจํ„ด ๋ถ„๋ฅ˜ ๋ผ๋ฒจ๋ง ์•Œ๊ณ ๋ฆฌ์ฆ˜(PCL; Pattern Classification Labeling Algorithm)์„ ์ œ์•ˆํ•œ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ ์ œ์•ˆํ•œ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋ถ€ํ•˜์˜ ๋ณ€์œ„๋ฅผ ๊ธฐ์ค€์œผ๋กœ ๋ถ„๋ฅ˜ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜ ์ถฉ/๋ฐฉ์ „ ์‹œ์ ์„ ๊ฒฐ์ •ํ•˜๋Š” AI์˜ ํ•™์Šต ๋ฐ์ดํ„ฐ๋กœ ์ ํ•ฉํ•˜๋‹ค.

2์žฅ์—์„œ๋Š” ์ œ์•ˆํ•œ ๋ชจ์ง‘๋‹จ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋Œ€ํ•ด ์„ค๋ช…ํ•˜๊ณ , 3์žฅ์—์„œ๋Š” k-means ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ ์ œ์•ˆํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ(์‚ฌ๋ก€ ์—ฐ๊ตฌ)๋ฅผ ๋น„๊ตํ•œ๋‹ค.

2. ํŒจํ„ด ๋ถ„๋ฅ˜ ๋ผ๋ฒจ๋ง(PCL) ์•Œ๊ณ ๋ฆฌ์ฆ˜

2.1 PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ตฌ์„ฑ

Fig. 1. The composition of the PCL algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig1.png

๋ณธ ์—ฐ๊ตฌ์—์„œ ์ œ์‹œํ•˜๋Š” PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ๋ถ€ํ•˜ ํŒจํ„ด์„ ๋‹จ์ˆœํ™”ํ•˜์—ฌ ๋ถ„๋ฅ˜ํ•œ๋‹ค. ๊ทธ๋ฆผ 1๊ณผ ๊ฐ™์ด PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋ฐ์ดํ„ฐ ๋ถ„๋ฅ˜, ๋ฐ์ดํ„ฐ ํ†ต๊ณ„, ํŒจํ„ด ์ถœ๋ ฅ์œผ๋กœ ๊ตฌ์„ฑ๋œ๋‹ค. ๋ฐ์ดํ„ฐ ๋ถ„๋ฅ˜ ๊ณผ์ •์—์„œ๋Š” ๊ธฐ๋ณธ ๋ฐ์ดํ„ฐ์ธ ๋ถ€ํ•˜ ๋ฐ์ดํ„ฐ๋ฅผ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ๋กœ ๋ณ€ํ™˜ํ•˜๊ณ , ๋ณ€ํ™˜๋œ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ์˜ ํŒจํ„ด์„ ๋‹จ์ˆœํ™”ํ•œ๋‹ค. ๋ฐ์ดํ„ฐ ํ†ต๊ณ„ ๊ณผ์ •์—์„œ๋Š” ์ผ์ž๋ณ„๋กœ ๋‹จ์ˆœํ™”ํ•œ ํŒจํ„ด์„ ๋™์ผํ•œ ํŒจํ„ด ๋ณ„๋กœ ๋ถ„๋ฅ˜ํ•˜๊ณ , ๋ถ„๋ฅ˜๋œ ๊ฐ ๋ฐ์ดํ„ฐ ์…‹(Class) ๋ณ„ ์˜ค์ฐจ์œจ์„ ๊ณ„์‚ฐํ•œ๋‹ค. ์˜ค์ฐจ์œจ์ด ๋ชฉํ‘œ์น˜์— ๋งŒ์กฑํ•˜์ง€ ๋ชปํ•œ ๊ฒฝ์šฐ, ํŒจํ„ด ๋‹จ์ˆœํ™” ๊ธฐ์ค€๊ฐ’์„ ์žฌ์‚ฐ์ •ํ•˜์—ฌ ๋ฐ์ดํ„ฐ ๋ถ„๋ฅ˜ ๊ณผ์ •์˜ ๋‹จ์ˆœํ™”๋ถ€ํ„ฐ ๋‹ค์‹œ ์ˆ˜ํ–‰ํ•œ๋‹ค. ๊ฐ Class์˜ ์˜ค์ฐจ์œจ์ด ๋ชฉํ‘œ์น˜๋ฅผ ๋งŒ์กฑํ•  ๊ฒฝ์šฐ, ํ•ด๋‹น Class์˜ ์‚ฐ์ˆ  ํ‰๊ท ๊ฐ’์„ ๋Œ€ํ‘œ ํŒจํ„ด์œผ๋กœ ํŒ๋‹จํ•˜๊ณ  ์ถœ๋ ฅํ•œ๋‹ค.

2.2 PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ฐ์ดํ„ฐ ๋ถ„๋ฅ˜ ๊ณผ์ •

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ฐ์ดํ„ฐ ๋ถ„๋ฅ˜ ๊ณผ์ •์„ ์ง„ํ–‰ํ•˜๊ธฐ ์œ„ํ•ด ์‚ฌ์—…์žฅ์˜ ๋ถ€ํ•˜ ๋ฐ์ดํ„ฐ๋ฅผ ์‹œ๊ฐ„๋Œ€๋ณ„๋กœ ๋‚˜๋ˆ„์–ด ๋ถ„๋ฅ˜ํ•˜๋Š” ์ž‘์—…์ด ์„ ํ–‰๋˜์–ด์•ผ ํ•˜๋ฉฐ, ์ถ”๊ฐ€์ ์œผ๋กœ ์‹œ๊ฐ„๋Œ€๋ณ„ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์„ ์‚ฐ์ •ํ•ด์•ผํ•œ๋‹ค. ์ดํ›„ ์ผ๋ณ„ ๋ถ€ํ•˜ ๋ฐ์ดํ„ฐ์˜ ํŒจํ„ด๋ณ„ ๊ตฌ๋ถ„์„ ์œ„ํ•ด ์›์‹œ ๋ถ€ํ•˜ ๋ฐ์ดํ„ฐ๋ฅผ $-2\alpha\sim 2\alpha$์˜ ์ •์ˆ˜๋กœ ๋‹จ์ˆœํ™”ํ•œ๋‹ค. ์—ฌ๊ธฐ์„œ $\alpha$๋Š” ํŒจํ„ด ๋‹จ์ˆœํ™” ๊ธฐ์ค€๊ฐ’์ธ $Step_{\alpha}$์˜ ๋‹จ๊ณ„์ด๋‹ค. ํ•ด๋‹น 24์‹œ๊ฐ„ ๋ถ€ํ•˜๋ฅผ ์‹œ๊ฐ„๋ณ„ ์ •์ˆ˜๋กœ ๋‹จ์ˆœํ™”ํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ๊ทธ๋ฆผ 2์™€ ๊ฐ™์ด ์›์‹œ ๋ฐ์ดํ„ฐ๋ฅผ ์ฆ๊ฐ์— ๋Œ€ํ•œ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ ($\Delta Load_{t,\:n}$)๋กœ ๋ณ€ํ™˜ํ•œ ๋’ค ๋ณ€ํ™”๋Ÿ‰์— ๋”ฐ๋ฅธ ์ •์ˆ˜๊ฐ’ ($Code_{t,\:n}$)์œผ๋กœ ์น˜ํ™˜ํ•œ๋‹ค. ์—ฌ๊ธฐ์„œ $t$๋Š” ์‹œ๊ฐ„, $n$์€ ๋‚ ์งœ๋ฅผ ์˜๋ฏธํ•œ๋‹ค.

Fig. 2. Daily load(top) and load variation(bottom)
../../Resources/kiiee/JIEIE.2019.33.12.021/fig2.png

$\Delta Load_{t,\:n}$๋กœ๋ถ€ํ„ฐ ๋‹จ์ˆœํ™”๋œ ์ •์ˆ˜๊ฐ’ $Code_{t,\:n}$๋ฅผ ์‚ฐ์ถœํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ๋ณ€ํ™”๋Ÿ‰์— ๋Œ€ํ•ด $2\alpha$๊ฐœ์˜ ๋“ฑ๋ถ„์œผ๋กœ ๋‚˜๋ˆ„์–ด ์ฃผ์–ด์•ผ ํ•œ๋‹ค. ์ด ๋•Œ ๋‹จ์ˆœํ™” ๊ธฐ์ค€๊ฐ’ $Step_{\alpha}$์€ ์‹ (1)๊ณผ ๊ฐ™์ด ๊ตฌํ•  ์ˆ˜ ์žˆ๋‹ค. $Step_{\alpha}$์€ $2\alpha$๊ฐœ ๋“ฑ๋ถ„ํ•  ๋•Œ ๋“ฑ๋ถ„๊ฐ’์˜ ํฌ๊ธฐ๋ฅผ ์˜๋ฏธํ•œ๋‹ค.

(1)
$\begin{align*} Step_{\alpha}\\ \\ =\dfrac{\max(\triangle Load_{t,\:n})-\min(\triangle Load_{t,\:n})}{2\alpha}\forall t,\:n \end{align*}$

์ •์ˆ˜๋กœ ๋‹จ์ˆœํ™”๋œ ํŒจํ„ด $Code_{t,\:n}$๋Š” ์‹ (2)์™€ ๊ฐ™์ด ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์ด $Step_{\alpha}$๋ฅผ ์ดˆ๊ณผํ•  ๋•Œ ๊ฐ’์ด ์ƒˆ๋กœ ์‚ฐ์ •๋œ๋‹ค. $C_{t,\:n}$์€ ์‹ (3)๊ณผ ๊ฐ™์ด ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰๊ณผ $Step_{\alpha}$์„ ๋‚˜๋ˆ„์–ด ๋‚ด๋ฆผํ•œ ๊ฐ’์ด๋‹ค. ๋ชจ๋“  ๋‚ ์งœ $n$์— ๋Œ€ํ•ด $Code_{t,\:n}$๋ฅผ ๊ตฌํ•œ ํ›„ ๊ฐ™์€ 24์‹œ๊ฐ„์— ๋Œ€ํ•œ $Code_{t,\:n}$๋ฅผ ๊ฐ€์ง„ ํŒจํ„ด์„ ์ทจํ•ฉํ•˜๋ฉด ๊ทธ๋ฆผ 3๊ณผ ๊ฐ™์ด ๋ถ„๋ฅ˜ํ•  ์ˆ˜ ์žˆ๋‹ค.

Fig. 3. Classified daily loads(top) and simplified load pattern(bottom)
../../Resources/kiiee/JIEIE.2019.33.12.021/fig3_1.png../../Resources/kiiee/JIEIE.2019.33.12.021/fig3_2.png

(2)
$$Code_{t,\:n}=Code_{t-1,\:n}+C_{t,\:n}$$

(3)
$$C_{t,\:n}=Round Down(\Delta Load_{t,\:n}/Step_{\alpha})$$

2.3 PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ฐ์ดํ„ฐ ํ†ต๊ณ„ ๊ณผ์ •

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ฐ์ดํ„ฐ ํ†ต๊ณ„ ๊ณผ์ •์—์„œ๋Š” ๋‹จ์ˆœํ™” ๋ถ„๋ฅ˜๋œ ํŒจํ„ด์„ ๊ฒ€์ฆํ•˜์—ฌ, ๋Œ€ํ‘œ ํŒจํ„ด์„ ์‚ฐ์ถœํ•œ๋‹ค. ๋ถ„๋ฅ˜ํ•œ ํŒจํ„ด์ด ์–ผ๋งˆ๋‚˜ ์œ ์‚ฌํ•œ ํŒจํ„ด๋“ค๋กœ ๊ตฐ์ง‘ํ•˜์˜€๋Š”์ง€ ์œ ์‚ฌ๋„๋ฅผ ํŒ๋‹จํ•˜๊ธฐ ์œ„ํ•ด, ๊ตฐ์ง‘ํ•œ ํŒจํ„ด์˜ ํ‰๊ท ๊ฐ’๊ณผ ๊ฐœ๋ณ„ ํŒจํ„ด์˜ ์˜ค์ฐจ์œจ์„ ๊ณ„์‚ฐํ•œ๋‹ค. ์˜ค์ฐจ์œจ์€ ์‹ (4)์™€ ๊ฐ™์ด ํ‰๊ท ์ ˆ๋Œ€๋น„์œจ์˜ค์ฐจ(MAPE; Mean Absolute Percentage Error) ๊ณ„์‚ฐ ๋ฐฉ์‹์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ณ„์‚ฐํ•œ๋‹ค.

(4)
$\begin{align*} MAPE_{error}\\ =\dfrac{1}{24}\sum_{t=1}^{24}\dfrac{\left |\Delta Load_{t,\:n}\right | -X_{t}}{X_{t}}\times 100(\%) \end{align*}$

(5)
$$X_{t}=\dfrac{1}{N}\sum_{n=1}^{N}\left |\Delta Load_{t,\:n}\right |$$

์‹ (5)์—์„œ $X_{t}$๋Š” $\Delta Load_{t,\:n}$์˜ ์‹œ๊ฐ„๋Œ€๋ณ„ ํ‰๊ท ๊ฐ’์„ ์˜๋ฏธํ•˜๊ณ  $N$์€ ๋ฐ์ดํ„ฐ ์ˆ˜(๋‚ ์งœ ์ˆ˜)๋ฅผ ์˜๋ฏธํ•œ๋‹ค. ๊ฒ€์ฆ ๊ธฐ์ค€๊ฐ’์€ 50%๋กœ ๊ฐ€์ •ํ•˜์˜€์œผ๋ฉฐ, $MAPE_{error}$๊ฐ€ 50%๋ณด๋‹ค ํฌ๋ฉด ๋™์ผํ•œ ํŒจํ„ด์œผ๋กœ ๊ตฐ์ง‘์ด ๋˜์ง€ ์•Š์€ ๊ฒƒ์œผ๋กœ ํŒ๋‹จํ•˜๊ณ , ๋ฐ์ดํ„ฐ ๋ถ„๋ฅ˜ ๊ณผ์ •์˜ $Step_{\alpha}$์˜ $\alpha$์„ 1์”ฉ ์ฆ๊ฐ€์‹œ์ผœ ํŒจํ„ด ๋ถ„๋ฅ˜ ๊ณผ์ •์„ ๋ฐ˜๋ณตํ•œ๋‹ค. ๋ฐ˜๋Œ€์˜ ๊ฒฝ์šฐ, ๋ถ„๋ฅ˜๊ฐ€ ์ž˜๋œ ๊ฒƒ์œผ๋กœ ํŒ๋‹จํ•˜์—ฌ ๊ฐ ๋ถ„๋ฅ˜๋ณ„ ํ‰๊ท ๊ฐ’์„ ๋Œ€ํ‘œ ํŒจํ„ด์œผ๋กœ ์ถœ๋ ฅํ•œ๋‹ค.

3. ์‚ฌ๋ก€ ์—ฐ๊ตฌ

3.1 ๊ณต์žฅ ๋ถ€ํ•˜ ํŒจํ„ด ๋ถ„๋ฅ˜

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋Œ€ํ•œ ๊ฒฐ๊ณผ๋ฅผ ๋ถ„์„ํ•˜๊ธฐ ์œ„ํ•ด ๋Œ€ํ‘œ์ ์ธ ๋ฐ์ดํ„ฐ ๊ตฐ์ง‘ ๋ฐฉ๋ฒ• ์ค‘ ํ•˜๋‚˜๋กœ ๋งŽ์ด ์‚ฌ์šฉ๋˜๋Š” K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ ๋น„๊ตํ•˜์˜€๋‹ค. ์›์‹œ ๋ฐ์ดํ„ฐ๋Š” A์‚ฌ ๊ฐ€์Šค๊ณต์žฅ์˜ 2018๋…„ 6์›”์—์„œ 9์›”๊นŒ์ง€ 4๊ฐœ์›” ๊ฐ„ ๋ถ€ํ•˜๋Ÿ‰ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ง„ํ–‰ํ•˜์˜€์œผ๋ฉฐ, ํ•ด๋‹น ๊ณต์žฅ์˜ ๋ถ€ํ•˜๋Ÿ‰ ๋ฐ์ดํ„ฐ๋Š” ๊ทธ๋ฆผ 4์™€ ํ‘œ 1๊ณผ ๊ฐ™๋‹ค.

Fig. 4. Daily load of the plant
../../Resources/kiiee/JIEIE.2019.33.12.021/fig4.png

Table 1. Load data of the plant

๊ตฌ๋ถ„

๋ถ€ํ•˜๋Ÿ‰(kWh)

์‹œ๊ฐ„๋‹น ์ตœ๋Œ€ ๋ถ€ํ•˜๋Ÿ‰

36,404.98

์‹œ๊ฐ„๋‹น ์ตœ์†Œ ๋ถ€ํ•˜๋Ÿ‰

17,612.32

์ผํ‰๊ท  ๋ถ€ํ•˜๋Ÿ‰

734,391.94

์‹œ๊ฐ„๋‹น ์ตœ๋Œ€ ๋ถ€ํ•˜์ฆ๊ฐ€๋Ÿ‰

4,185.70

์‹œ๊ฐ„๋‹น ์ตœ๋Œ€ ๋ถ€ํ•˜๊ฐ์†Œ๋Ÿ‰

4,064.19

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ํŒจํ„ด ๋ถ„๋ฅ˜ํ•˜๊ธฐ ์œ„ํ•ด ๊ฐ€์Šค๊ณต์žฅ์˜ ๋ถ€ํ•˜๋Ÿ‰ ๋ฐ์ดํ„ฐ๋ฅผ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ๋กœ ๋ณ€ํ™˜ํ•œ ๊ทธ๋ž˜ํ”„๋Š” ๊ทธ๋ฆผ 5์™€ ๊ฐ™๋‹ค. ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ๋กœ๋ถ€ํ„ฐ ๋ถ€ํ•˜์˜ ์ฆ๊ฐ€๊ฐ์†Œ๋ฅผ ํŒ๋‹จํ•˜๊ณ  $Code_{t,\:n}$์„ ๊ณ„์‚ฐํ•˜๊ธฐ ์œ„ํ•ด $Step_{n}$์„ ์‚ฐ์ •ํ•ด์•ผ ํ•œ๋‹ค. ๋‚ฎ์€ $Step$์œผ๋กœ ํŒจํ„ด ๋ถ„๋ฅ˜ํ•  ๊ฒฝ์šฐ ํŒจํ„ด์˜ ์ •ํ™•๋„๊ฐ€ ๋‚ฎ์•„์ง€๋ฉฐ, ๋†’์€ $Step$์œผ๋กœ ํŒจํ„ด ๋ถ„๋ฅ˜ํ•  ๊ฒฝ์šฐ ๊ฐ™์€ ํŒจํ„ด์œผ๋กœ ๋ถ„๋ฅ˜๋˜๋Š” ๋ฐ์ดํ„ฐ๊ฐ€ ์ ์–ด์ง€๋ฏ€๋กœ ์ ์ ˆํ•œ $Step$ ์ ์šฉ์ด ํ•„์š”ํ•˜๋‹ค.

$Step_{n}$์˜ $n$์„ 1โˆผ3์œผ๋กœ ์ฆ๊ฐ€์‹œํ‚ค๋ฉด์„œ ๋“ฑ๋ถ„ ๊ฐ’์˜ ํฌ๊ธฐ๋ฅผ ๊ณ„์‚ฐํ•œ ๊ฒฐ๊ณผ๋Š” ํ‘œ 2์™€ ๊ฐ™์œผ๋ฉฐ, ๊ทธ๋ฆผ 6~8์€ ๊ณต์žฅ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰๊ณผ $Step_{n}$์„ ๋น„๊ตํ•œ ๊ทธ๋ž˜ํ”„์ด๋‹ค.

Table 2. Judgment standard by step

๊ตฌ๋ถ„

ํŒ๋‹จ ๊ธฐ์ค€ ๋ถ€ํ•˜๋ณ€ํ™”๋Ÿ‰(kwh)

$Step_{1}$

4,124.95

$Step_{2}$

2,062.47

$Step_{3}$

1,031.24

Fig. 5. Daily load variation of the plant
../../Resources/kiiee/JIEIE.2019.33.12.021/fig5.png

Fig. 6. Comparison of load variation with $Step_{1}$
../../Resources/kiiee/JIEIE.2019.33.12.021/fig6.png

Fig. 7. Comparison of load variation with $Step_{2}$
../../Resources/kiiee/JIEIE.2019.33.12.021/fig7.png

Fig. 8. Comparison of load variation with $Step_{3}$
../../Resources/kiiee/JIEIE.2019.33.12.021/fig8.png

๊ทธ๋ฆผ 6์˜ Min/Max์„ ์€ ๊ณต์žฅ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ์˜ ์‹œ๊ฐ„๋‹น ์ตœ๋Œ€ ๋ถ€ํ•˜ ์ฆ๊ฐ€๋Ÿ‰๊ณผ ์ตœ๋Œ€ ๋ถ€ํ•˜ ๊ฐ์†Œ๋Ÿ‰์„ ๋‚˜ํƒ€๋‚ธ ์„ ์ด๋‹ค. ๊ทธ๋ž˜ํ”„์˜ Min/Max์„  ์‚ฌ์ด๋ฅผ 2๋“ฑ๋ถ„ํ•˜์—ฌ ํŒ๋‹จํ•˜๊ณ ์žํ•˜๋Š” ์‹œ๊ฐ„๋Œ€์˜ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์ด ๊ทธ๋ž˜ํ”„์˜ C=1์˜ ํฌ๊ธฐ๋ณด๋‹ค ํฌ๊ฒŒ ์ฆ๊ฐ€ํ•  ๊ฒฝ์šฐ ํ•ด๋‹น ์‹œ๊ฐ„๋Œ€์˜ $C$ ๊ฐ’์€ 1๊ณผ 2 ์‚ฌ์ด์—์„œ ๊ฒฐ์ •๋˜๋ฉฐ, ๋ฐ˜๋Œ€๋กœ ๋‘ ์„ ์˜ ๊ฐ„๊ฒฉ๋ณด๋‹ค ํฌ๊ฒŒ ๊ฐ์†Œํ•  ๊ฒฝ์šฐ ํ•ด๋‹น ์‹œ๊ฐ„๋Œ€ $C$ ๊ฐ’์€ โ€“1๊ณผ โ€“2์‚ฌ์ด์—์„œ ๊ฒฐ์ •๋œ๋‹ค. ์ด๋ฅผ ์ œ์™ธํ•œ ๊ฒฝ์šฐ์˜ $C$ ๊ฐ’์€ 0์ด๋‹ค. $Step_{2}$๋Š” ๊ทธ๋ฆผ 7๊ณผ ๊ฐ™์ด Min/Max์„  ์‚ฌ์ด๋ฅผ 4๋“ฑ๋ถ„ํ•˜์—ฌ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์— ๋”ฐ๋ผ $C$๊ฐ€ -4์™€ 4์‚ฌ์ด์—์„œ ์ •ํ•ด์ง„๋‹ค. ์ด๋ ‡๊ฒŒ ๋ถ„๋ฅ˜๋œ $C$์˜ ๊ฐ’์— ๋”ฐ๋ผ $Code$๊ฐ€ ๊ฒฐ์ •๋˜๋ฉฐ, ์ด๋ฅผ ํ†ตํ•ด ์ผ๋ณ„ ํŒจํ„ด์„ ๋‹จ์ˆœํ™”ํ•˜์—ฌ ๋ถ„๋ฅ˜ํ•œ๋‹ค.

3.2 PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ถ€ํ•˜ ์ฆ๊ฐ€๊ฐ์†Œ๋ฅผ ํŒ๋‹จํ•˜๋Š” ๊ธฐ์ค€์„ $Step_{2}$๋กœ ๊ฐ€์ •ํ•˜์—ฌ ํŒจํ„ด์„ ๋ถ„๋ฅ˜ํ•œ ๊ฒฐ๊ณผ๋Š” ๊ทธ๋ฆผ 9~12์™€ ๊ฐ™์ด 4๊ฐ€์ง€ ๋ถ€ํ•˜ ํŒจํ„ด๊ณผ ๊ฐ™๋‹ค.

k-means ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฐ€์žฅ ํฐ ์ฐจ์ด์ ์€ ๊ทธ๋ฆผ 9์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค. PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฒฝ์šฐ ์ „๊ตฌ๊ฐ„์˜ $Code_{t,\:n}$๊ฐ’์ด ๋ชจ๋‘ 0์œผ๋กœ ํ‘œํ˜„๋˜์–ด ํ•˜๋‚˜๋กœ ๋ถ„๋ฅ˜ํ•˜์˜€์œผ๋ฉฐ K-means๋Š” ์ค‘์‹ฌ๊ฐ’์„ ๊ธฐ์ค€์œผ๋กœ ํ•˜๋ฏ€๋กœ ์„œ๋กœ ๋‹ค๋ฅด๊ฒŒ ๋ถ„๋ฅ˜ํ•œ๋‹ค. ๋ถ€ํ•˜ ๊ตฐ์ง‘ ๊ฒฐ๊ณผ๋กœ ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜ ์ถฉ๋ฐฉ์ „ ๊ณ„ํš์„ ์ˆ˜ํ–‰ํ•œ๋‹ค๋Š” ๊ด€์ ์„ ๋ณด์•˜์„ ๋•Œ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ๊ฐ€ ๋” ํƒ€๋‹นํ•œ ๋ถ„๋ฅ˜๋ผ๋Š” ๊ฒƒ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค. ๊ทธ๋ฆผ 10~12๋„ ๋ชจ๋‘ ๋น„์Šทํ•œ ํŒจํ„ด์ด ๋ถ„๋ฅ˜๋œ ๊ฒƒ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค.

Fig. 9. Pattern 1 classified by PCL algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig9.png

Fig. 10. Pattern 2 classified by PCL algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig10.png

Fig. 11. Pattern 3 classified by PCL algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig11.png

Fig. 12. Pattern 4 classified by PCL algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig12.png

๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์— ๋Œ€ํ•œ MAPE ์˜ค์ฐจ์œจ๋„ ํ‘œ 3๊ณผ ๊ฐ™์ด ์•ฝ 32.2%โˆผ44.2%๋กœ ๊ฐ ํŒจํ„ด์˜ ์˜ค์ฐจ์œจ ์ฐจ์ด ๋˜ํ•œ ์ผ์ •ํ•˜๊ฒŒ ๋‚˜ํƒ€๋‚ฌ๋‹ค. PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฒฝ์šฐ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์„ ๊ธฐ์ค€์œผ๋กœ ๊ตฐ์ง‘์„ ๋ถ„๋ฅ˜ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๊ทธ๋ฆผ 9๊ณผ ๊ฐ™์ด ๋ถ€ํ•˜๋Ÿ‰์˜ ํฌ๊ธฐ๊ฐ€ ๋‹ค๋ฅด๋”๋ผ๋„ ๋™์ผํ•œ ํŒจํ„ด์œผ๋กœ ๋ถ„๋ฅ˜ํ•˜์˜€์œผ๋ฉฐ, ์˜ค์ฐจ์œจ์ด ํฌ๊ฒŒ ์ฐจ์ด ๋‚˜์ง€ ์•Š๋Š” ์ ์€ ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ์šด์ „ ๊ฒฐ์ • ๊ณ„ํš์— ํฐ ๊ฐ•์ ์„ ๊ฐ€์ง„๋‹ค.

Table 3. Error rate of patterns classified by PCL algorithm

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜ ํŒจํ„ด

์˜ค์ฐจ์œจ(%)

1

32.2

2

35.0

3

33.5

4

44.2

3.3 K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ

K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ k๊ฐ’์„ ์ง€์ •ํ•ด์ฃผ์–ด์•ผ ํ•œ๋‹ค๋Š” ํŠน์ง•์ด ์žˆ์œผ๋ฉฐ, ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ์˜ ๋น„๊ต๋ฅผ ์œ„ํ•ด k ๊ฐ’์„ 4๋กœ ์ง€์ •ํ•˜์—ฌ ๊ทธ๋ฆผ 13~16๊ณผ ๊ฐ™์ด ๋ถ„๋ฅ˜ํ•˜์˜€๋‹ค. K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ถ„๋ฅ˜ ๊ฒฐ๊ณผ๋Š” ๊ทธ๋ฆผ 14์—์„œ ๋ถ€ํ•˜๋Ÿ‰์ด ๊ตฐ์ง‘์—์„œ ์ผ์ • ์ˆ˜์ค€์ด์ƒ ๋ฒ—์–ด๋‚˜๋Š” ๋ฐ์ดํ„ฐ๋“ค์ด ์ƒ๋‹น์ˆ˜ ์กด์žฌํ•œ๋‹ค. ์ฆ‰ K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฒฝ์šฐ ๋ถ€ํ•˜๋Ÿ‰์˜ ํฌ๊ธฐ๊ฐ€ ๋น„์Šทํ•˜์ง€๋งŒ ๋™์ผํ•˜์ง€ ์•Š๋Š” ํŒจํ„ด์„ ํฌํ•จํ•˜์—ฌ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋น„ํ•ด ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜ ์Šค์ผ€์ค„ ์‚ฐ์ •์— ์˜ค์ฐจ ๋ฐœ์ƒ ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค. K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ MAPE ์˜ค์ฐจ์œจ์€ ํ‘œ 4์™€ ๊ฐ™๋‹ค.

Fig. 13. Pattern 1 classified by K-means algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig13.png

Fig. 14. Pattern 2 classified by K-means algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig14.png

Fig. 15. Pattern 3 classified by K-means algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig15.png

Fig. 16. Pattern 4 classified by K-means algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig16.png

Table 4. Error rate of patterns classified by K-means algorithm

K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜ ํŒจํ„ด

์˜ค์ฐจ์œจ(%)

1

32.1

2

56.0

3

22.8

4

56.1

3.4 ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋น„๊ต

๋ถ€ํ•˜๋Ÿ‰์˜ ์ฆ๊ฐ€๊ฐ์†Œ ํŒจํ„ด์€ ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ์ตœ์ ์˜ ์Šค์ผ€์ค„ ์‚ฐ์ •์— ์žˆ์–ด ๋งค์šฐ ์ค‘์š”ํ•œ ์š”์†Œ์ด๋‹ค. ๋ณธ ์ ˆ์—์„œ๋Š” PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ํŒจํ„ด๋ณ„ ์‹œ๊ฐ„๋Œ€ ๋ถ€ํ•˜๋Ÿ‰ ๋ณ€์œ„๊ฒฐ๊ณผ์— ๋Œ€ํ•ด ๋ถ„์„ํ–ˆ๋‹ค.

๊ทธ๋ฆผ 17์€ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ๋ถ„๋ฅ˜๋œ ํŒจํ„ด๋ณ„ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๊ทธ๋ž˜ํ”„์ด๋‹ค. ์—ฌ๊ธฐ์„œ x์ถ• 1์นธ์€ 1์‹œ๊ฐ„์„ ์˜๋ฏธํ•œ๋‹ค. ๋”ฐ๋ผ์„œ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํŒจํ„ด์„ ๋ถ„๋ฅ˜ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๋ถ„๋ฅ˜๋œ ํŒจํ„ด์˜ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์ด ํŒจํ„ด๋ณ„๋กœ ํ•ด๋‹น์‹œ๊ฐ„๋Œ€์—์„œ ์ผ์ •ํ•œ ๊ฐ’์„ ๊ฐ€์ง„๋‹ค. PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์ผ์ •ํ•œ ๋ณ€ํ™”๋Ÿ‰์„ ๊ฐ€์ง„ ํŒจํ„ด์œผ๋กœ ๋ถ„๋ฅ˜ํ•˜์—ฌ ๋ผ๋ฒจ๋ง์„ ํ•˜๊ณ  ๋ถ„๋ฅ˜ ์˜ค์ฐจ ์—ญ์‹œ ํŽธ์ฐจ๊ฐ€ ์ ์œผ๋ฏ€๋กœ ๋จธ์‹  ๋Ÿฌ๋‹ ์˜ˆ์ธก ์‹œ ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ์ถฉ๋ฐฉ์ „ ๊ณ„ํš์„ ๊ฒฐ์ •ํ•จ์— ์žˆ์–ด ์ข€ ๋” ์œ ์šฉํ•œ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์ผ ์ˆ˜ ์žˆ๋‹ค.

Fig. 17. Load variation of patterns classified by PCL algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig17.png

๊ทธ๋ฆผ 18์€ K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ๋ถ„๋ฅ˜๋œ ํŒจํ„ด๋ณ„ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๊ทธ๋ž˜ํ”„์ด๋‹ค. K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์‹œ๊ฐ„๋Œ€๋ณ„ ๋ถ€ํ•˜๋Ÿ‰ ๋ฐ์ดํ„ฐ ๊ฐ„์˜ ๊ฑฐ๋ฆฌ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ตฐ์ง‘์„ ๋ถ„๋ฅ˜ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๊ทธ๋ฆผ 18 (b)์™€ ๊ฐ™์ด ๊ฐ™์€ ์‹œ๊ฐ„๋Œ€์— ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์ด ํฌ๊ฒŒ ์ฐจ์ด๋‚˜๋Š” ๋ฐ์ดํ„ฐ๋“ค์ด ํ•˜๋‚˜๋กœ ๋ถ„๋ฅ˜๋˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋ฐœ์ƒํ•œ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฐ์ดํ„ฐ ๋ถ„๋ฅ˜๋ฅผ ํ†ตํ•ด ๋ถ€ํ•˜๋Ÿ‰์„ ์˜ˆ์ธกํ•˜๊ฒŒ ๋  ๊ฒฝ์šฐ ํฌ๊ฒŒ ์ฐจ์ด๋‚˜๋Š” ๋ช‡๋ช‡ ๋ฐ์ดํ„ฐ๋“ค์€ ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ์ตœ์  ์ถฉ๋ฐฉ์ „ ๊ณ„ํš์„ ๊ฒฐ์ •ํ•จ์— ์žˆ์–ด ์˜ค์ฐจ๋ฅผ ๋ฐœ์ƒ์‹œํ‚ฌ ์—ฌ์ง€๊ฐ€ ํฌ๋‹ค.

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฒฐ๊ณผ์— ๋Œ€ํ•œ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ์˜ค์ฐจ์œจ ๋น„๊ต ์‹œ ํ‘œ 5์™€ ๊ฐ™์ด ์ตœ์†Œ ์˜ค์ฐจ์œจ์„ ์ œ์™ธํ•˜๊ณ  ๋ชจ๋‘ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์˜ค์ฐจ์œจ์ด ๋‚ฎ๊ฒŒ ๋‚˜ํƒ€๋‚ฌ๋‹ค. ์ „์ฒด ๋Œ€๋น„ ๊ฐ ํŒจํ„ด ๋ฐ์ดํ„ฐ์˜ ๋น„์œจ์„ ๊ณ ๋ คํ•˜์—ฌ ์˜ค์ฐจ์œจ์˜ ํ‰๊ท ์„ ์‚ฐ์ •ํ•œ ๊ฐ€์ค‘ ํ‰๊ท  ์˜ค์ฐจ์œจ์˜ ๊ฒฝ์šฐ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฒฐ๊ณผ๋ณด๋‹ค 12.2% ๋‚ฎ์•˜๋‹ค.

Fig. 18. Load variation of patterns classified by K-means algorithm
../../Resources/kiiee/JIEIE.2019.33.12.021/fig18.png

Table 5. Error rate of load variation by algorithm

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜

K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜

๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰

์ตœ์†Œ ์˜ค์ฐจ์œจ(%)

32.2

20.5

๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰

์ตœ๋Œ€ ์˜ค์ฐจ์œจ(%)

44.2

54.0

๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰

ํ‰๊ท  ์˜ค์ฐจ์œจ(%)

36.2

37.7

๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰

๊ฐ€์ค‘ํ‰๊ท  ์˜ค์ฐจ์œจ(%)

34.3

46.5

์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ํ”ผํฌ์ €๊ฐ์„ ์œ„ํ•œ ์Šค์ผ€์ค„ ์‚ฐ์ •์€ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์— ๋ฏผ๊ฐํ•˜๊ฒŒ ๋ฐ˜์‘ํ•œ๋‹ค. ์ฆ‰ ๋ถ€ํ•˜๋ณ€ํ™”๋Ÿ‰ ์˜ค์ฐจ์œจ์ด ๋†’์œผ๋ฉด ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜ ์ตœ์ ์˜ ์Šค์ผ€์ค„ ์‚ฐ์ •์— ์žˆ์–ด ์˜ค์ฐจ ๋ฐœ์ƒ ๊ฐ€๋Šฅ์„ฑ์ด ๋†’์•„์ง„๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰์˜ ์˜ค์ฐจ์œจ์„ ๋น„๊ตํ•จ์œผ๋กœ์„œ PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋น„ํ•ด ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜ ์ตœ์ ์˜ ์Šค์ผ€์ค„ ์‚ฐ์ •์— ์žˆ์–ด ์˜ค์ฐจ ๋ฐœ์ƒ ๊ฐ€๋Šฅ์„ฑ์ด ๋‚ฎ์€ ํŒจํ„ด ๋ถ„๋ฅ˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ž„์„ ํ™•์ธํ•˜์˜€๋‹ค.

4. ๊ฒฐ ๋ก 

์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜๋Š” ๋ถ€ํ•˜๋Ÿ‰ ์˜ˆ์ธก์— ๊ธฐ๋ฐ˜์„ ๋‘์–ด ์Šค์ผ€์ค„์ด ์‚ฐ์ •๋œ๋‹ค. ๊ทธ๋Ÿฌ๋ฏ€๋กœ ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ์ตœ์  ์šด์˜์— ์žˆ์–ด ์ •ํ™•ํ•œ ๋ถ€ํ•˜๋Ÿ‰ ์˜ˆ์ธก์€ ๋งค์šฐ ์ค‘์š”ํ•œ ์‚ฌํ•ญ์ด๋‹ค. ๋˜ํ•œ ์—๋„ˆ์ง€์ €์žฅ์žฅ์น˜์˜ ์Šค์ผ€์ค„ ์‚ฐ์ •ํ•จ์— ์žˆ์–ด ๋ถ€ํ•˜์˜ ์ฆ๊ฐ€๊ฐ์†Œ ํŒจํ„ด์ด ์ค‘์š”ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ํ•ด๋‹น ์‚ฌํ•ญ์„ ๊ณ ๋ คํ•œ ํŒจํ„ด ๋ถ„๋ฅ˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ํ•„์š”ํ•˜๋‹ค.

๋ณธ ์—ฐ๊ตฌ๋Š” AI ํ•™์Šต ๊ธฐ๋ฐ˜์˜ ๋ถ€ํ•˜๋Ÿ‰ ์˜ˆ์ธก ๊ธฐ๋ฒ•์„ ์‚ฌ์šฉํ•˜๊ธฐ์— ์•ž์„œ, ๋ถ€ํ•˜์˜ ์ฆ๊ฐ€๊ฐ์†Œ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํŒจํ„ด์„ ๊ตฐ์ง‘ํ•˜๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋Œ€ํ•ด ์ œ์•ˆํ–ˆ๋‹ค. ์ œ์•ˆํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋Œ€ํ‘œ์ ์ธ ๊ตฐ์ง‘ ๋ฐฉ๋ฒ•์ธ K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ ๋น„๊ตํ•˜์˜€์œผ๋ฉฐ, PCL์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๊ฐ€์ค‘ํ‰๊ท  ์˜ค์ฐจ์œจ์ด K-means ์•Œ๊ณ ๋ฆฌ์ฆ˜๋ณด๋‹ค 12.2% ๋‚ฎ์€ ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ๋‹ค.

PCL ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๋ฐ์ดํ„ฐ ๋ถ„๋ฅ˜ ๊ณผ์ •์€ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฐ์ถœํ•˜์—ฌ ์ฝ”๋“œํ™” ๊ณผ์ •์ด ์ง„ํ–‰๋˜๋ฉฐ, ์ฝ”๋“œํ™” ๊ณผ์ •์—์„œ ๋ถ€ํ•˜ ๋ณ€ํ™”๋Ÿ‰ ๋ฐ์ดํ„ฐ์˜ ๋‹จ์ˆœํ™” ๊ธฐ์ค€๊ฐ’์„ ์„ค์ •ํ•˜๊ฒŒ ๋œ๋‹ค. ํ•ด๋‹น ๊ธฐ์ค€๊ฐ’์— ๋”ฐ๋ผ ํŒจํ„ด์˜ ๋ถ„๋ฅ˜๊ฐ€ ์ƒ์ดํ•˜๊ฒŒ ๋„์ถœ๋  ์ˆ˜ ์žˆ์–ด ํ–ฅํ›„ ์—ฐ๊ตฌ์—์„œ ๋ณด๋‹ค ์ ์ ˆํ•œ ๊ธฐ์ค€๊ฐ’ ์‚ฐ์ •์— ๋Œ€ํ•ด ๊ณ ๋ คํ•˜์—ฌ ๋ฐ˜์˜ํ•œ๋‹ค๋ฉด ๋ณด๋‹ค ๊ฐœ์„ ๋œ ๋ถ„๋ฅ˜๊ฒฐ๊ณผ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์„ ๊ฒƒ์œผ๋กœ ๊ธฐ๋Œ€๋œ๋‹ค.

Acknowledgements

๋ณธ ์—ฐ๊ตฌ๋Š” ๊ตญํ† ๊ตํ†ต๋ถ€/๊ตญํ† ๊ตํ†ต๊ณผํ•™๊ธฐ์ˆ ์ง„ํฅ์›์˜ ์ง€์›์œผ๋กœ ์ˆ˜ํ–‰๋˜์—ˆ์Œ(๊ณผ์ œ๋ฒˆํ˜ธ 19AUDP-B119346-04).

References

1 
Min D. K., Ryu J. H., Choi D. G., 2019, Evaluating the Effects of Koreaโ€™s Nuclear & Coal Phase-out and Renewable Energy Shift on the Power System Reliability, Korea Energy Economic Review, Vol. 17, No. 1, pp. 1-35Google Search
2 
Bae Y. H., 2018, Experiences and Prospects of New and Renewable Energy in Korea, KSCE magazine, Vol. 66, No. 3, pp. 18-21Google Search
3 
Lee K. J., Park J. M., Park J. A., 2019, Implementation of an Adaptive EMS System Based on Solar Power Generation and Power Demand Pattern, JKIIT, Vol. 17, No. 1, pp. 53-62Google Search
4 
Han S. K., Kim E. J., 2019, A Study on the Impact Analysis of Power Demand for ESS and SmartGrid, Trans. Korean. Inst. Elect. Eng., Vol. 68p, No. 2, pp. 100-105DOI
5 
Yu S. J., Lee A. G., Beak J. S., Cha D. J., Kim J. C., 2018, "Study on ESS Operation Plan for Energy Reduction in Industrial Customer, Trans. Korean. Inst. Elect. Eng, Vol. 67, No. 9, pp. 1152-1158DOI
6 
Oh B. C., Kim S. Y., 2019, "Development of SVR based Short-term Load Forecasting Algorithm, Trans. Korean. Inst. Elect. Eng, Vol. 68p, No. 2, pp. 95-99DOI
7 
Lee J. H., Oh S. J., Yoon Y. C., Ahn Y. H., Kim J. S., et al. , 2019, Analysis of Time Series to Support Decision Making on V2G Using Energy Consumption Data, JKDISS, Vol. 30, No. 2, pp. 401-414DOI
8 
Lee D. G., Kim S. H., Jung H. C., Sun Y. G., Sim I. S., et al. , 2019, Power Consumption Prediction Scheme Based on Deep Learning for Powerline Communication Systems, Journal of IKEEE, Vol. 22, No. 3, pp. 822-828DOI
9 
Ahn J. Y., Park S. M., Kim C. B., 2019, A Study on Neural Network Model for Winter Electric Power Demand Prediction, JKIIT, Vol. 15, No. 9, pp. 1-9Google Search
10 
Yoon T. S., Shim K. S., 2012, K-means Clustering for Handling High-dimensional Large Data, KIISE Trans. Computing Practices, Vol. 18, No. 1, pp. 55-59Google Search

Biography

Young-Il Kim
../../Resources/kiiee/JIEIE.2019.33.12.021/au1.png

Young-Il kim received B.S., M.S degrees in Energy IT from Gachon university, Seongnam, Korea, in 2016, 2018. he is working a researcher in Gridwiz Inc.

His research interests are electric power economy, Renewable Energy and ESS.

Sung-Man Choi
../../Resources/kiiee/JIEIE.2019.33.12.021/au2.png

Sung-Man Choi received B.S, M.S. degrees in Electrical Engineering from Changwon National University, Changwon, Korea in 2013, 2015, respectively.

At present, he is working as a direction in Gridwiz Inc.

His interests are power conditioning system and microgrids.

Min-Kyu Baek
../../Resources/kiiee/JIEIE.2019.33.12.021/au3.png

Min-Kyu Baek received the B.S., M.S., and Ph.D. degrees from the Konkuk University, Seoul, South Korea all in electrical engineering.

He is currently a associate research engineer in Gridwiz Inc.

His current research interests include operating algorithm of the battery energy storage system and artificial intelligence.

Bok-Deok Shin
../../Resources/kiiee/JIEIE.2019.33.12.021/au4.png

Bok-Deok Shin received B.S., M.S., and Ph.D. degrees in Computer Engineering from Kyungnam University, Changwon, Korea in 1997, 2001, and 2005, respectively.

At present, he is working as a direction in Gridwiz Inc.

His interests are power system and microgrids.