• ๋Œ€ํ•œ์ „๊ธฐํ•™ํšŒ
Mobile QR Code QR CODE : The Transactions of the Korean Institute of Electrical Engineers
  • COPE
  • kcse
  • ํ•œ๊ตญ๊ณผํ•™๊ธฐ์ˆ ๋‹จ์ฒด์ด์—ฐํ•ฉํšŒ
  • ํ•œ๊ตญํ•™์ˆ ์ง€์ธ์šฉ์ƒ‰์ธ
  • Scopus
  • crossref
  • orcid

  1. (Digital Solution Laboratory, KEPRI, Korea)
  2. (Safety & Security Department, Korea Electric Power Corporation(KEPCO), Korea)
  3. (Department of Energy Systems, Ajou University, Korea)



Natural Disaster, Typhoon, Power System Facilities, Damage Prediction, Artificial Neural Network, AI

1. ์„œ ๋ก 

์ตœ๊ทผ ์ „ ์„ธ๊ณ„์ ์œผ๋กœ ํญ์—ผ, ๊ฐ€๋ญ„, ์ง‘์ค‘ํ˜ธ์šฐ์™€ ๊ฐ™์€ ๊ธฐํ›„ ๋ณ€ํ™”๋กœ ์ธํ•œ ์ด์ƒ๊ธฐ์ƒ ํ˜„์ƒ์ด ์ž์ฃผ ๋ฐœ์ƒํ•˜๊ณ  ์žˆ๋‹ค. ๊ตญ๋‚ด์—์„œ๋„ ๊ธฐ์ƒ์ฒญ์˜ ๊ธฐํ›„๋ณด๊ณ ์„œ์— ๋”ฐ๋ฅด๋ฉด ์šฐ๋ฆฌ๋‚˜๋ผ 6๋Œ€ ๋„์‹œ์˜ ํ‰๊ท ๊ธฐ์˜จ์ด 100๋…„๊ฐ„ 1.7โ„ƒ๊ฐ€ ์ƒ์Šนํ•˜์˜€๋‹ค. ์ด๋Š” ์ „ ์ง€๊ตฌ ํ‰๊ท  ์ƒ์Šนํญ์ธ 0.75โ„ƒ์˜ 2๋ฐฐ๊ฐ€ ๋„˜๋Š”๋‹ค. ๊ฐ•์ˆ˜๋Ÿ‰์€ 19% ์ฆ๊ฐ€ํ•˜์˜€์œผ๋ฉฐ, ํ•ด์ˆ˜๋ฉด์€ 8cm ์ด์ƒ ์ƒ์Šนํ•˜์˜€๋‹ค. ์ด๋Ÿฌํ•œ ๊ธฐํ›„๋ณ€ํ™”์˜ ๊ฒฐ๊ณผ๋“ค์€ ์žฌ๋‚œ์˜ ๊ฐ•๋„๋ฅผ ์ฆ๊ฐ€์‹œํ‚จ๋‹ค. ์ „๋ ฅ์„ค๋น„ ๊ด€๋ฆฌ ์ธก๋ฉด์—์„œ๋„ ํƒœํ’, ํ˜ธ์šฐ, ํ•œํŒŒ, ํญ์—ผ ๊ฐ™์€ ์ž์—ฐ์žฌ๋‚œ์— ์˜ํ•œ ์ „๋ ฅ์„ค๋น„ ํ”ผํ•ด๊ฐ€ ์ ์  ์ฆ๊ฐ€ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ[1], ์ž์—ฐ์žฌ๋‚œ์— ๋Œ€ํ•œ ๋น ๋ฅธ ๋Œ€์‘๊ณผ ํ”ผํ•ด๋ฅผ ์ตœ์†Œํ™”ํ•˜๊ธฐ ์œ„ํ•ด ์žฌ๋‚œ์— ๋Œ€ํ•œ ์„ค๋น„ํ”ผํ•ด ์˜ˆ์ธก ๊ธฐ์ˆ ์˜ ๊ฐœ๋ฐœ์ด ์š”๊ตฌ๋œ๋‹ค.

๊ตญ์™ธ์—์„œ๋„ ์žฌ๋‚œ ํ˜„์ƒ์œผ๋กœ ์ธํ•œ ์ „๋ ฅ์‹œ์Šคํ…œ ํ”ผํ•ด์˜ ์‹ฌ๊ฐ์„ฑ์„ ์ธ์ง€ํ•˜๊ณ  ์žฌ๋‚œ ํ˜„์ƒ ๋ฐœ์ƒ์— ๋”ฐ๋ฅธ ์ „๋ ฅ์‹œ์Šคํ…œ ๋Œ€์‘ ๋ฐฉ์•ˆ์— ๋Œ€ํ•œ ์—ฐ๊ตฌ๋ฅผ ์ง„ํ–‰ํ•˜๊ณ  ์žˆ๋‹ค. ํŠนํžˆ, ํ—ˆ๋ฆฌ์ผ€์ธ Sandy๋‚˜ Katrina์™€ ๊ฐ™์€ ์น˜๋ช…์ ์ธ ์ƒํ™ฉ์€ ์•ˆ์ •์ ์ธ ์ „๋ ฅ๊ณต๊ธ‰์— ํฐ ์˜ํ–ฅ์„ ๋ฏธ์ณ ์‚ฌํšŒ ๊ฒฝ์ œ์  ์‹œ์Šคํ…œ์— ๋ง‰๋Œ€ํ•œ ์†์‹ค์„ ์ดˆ๋ž˜ํ•˜์˜€์œผ๋ฉฐ, ๋ฏธ๊ตญ ์ „๊ธฐํšŒ์‚ฌ๋Š” ์žฌ๋‚œ์— ๊ด€ํ•œ ์—ฌ๋Ÿฌ ํ”„๋กœ์ ํŠธ๋ฅผ ์ง„ํ–‰ ์ค‘์— ์žˆ๋‹ค. North American Electric Reliability Corporation(NREL)์—์„œ๋Š” ํ—ˆ๋ฆฌ์ผ€์ธ Sandy์— ๊ด€ํ•œ ๋ถ„์„ ๋ณด๊ณ ์„œ๋ฅผ ๋ฐœํ‘œํ•˜์˜€๋‹ค[2][3].

์œ ๋Ÿฝ์—ฐํ•ฉ์€ 2014๋…„๋ถ€ํ„ฐ 2017๋…„๊นŒ์ง€ 15๊ฐœ์˜ ๊ธฐ๊ด€์ด ์ฐธ์—ฌํ•œ RAIN(Risk Analysis of Infrastructure Networks in Response to Extreme Weather) ํ”„๋กœ์ ํŠธ๋ฅผ ํ†ตํ•ด ์žฌ๋‚œ ๋ชจ์˜์‹คํ—˜ ํ”„๋กœ๊ทธ๋žจ์„ ๊ฐœ๋ฐœํ•˜์—ฌ ์žฌ๋‚œ ๋ฐœ์ƒ ์‹œ ํ”ผํ•ด์ƒํ™ฉ์„ ์˜ˆ์ธกํ•˜๊ณ  ์‹ ์†ํ•œ ๋ณต๊ตฌ๋ฅผ ์œ„ํ•ด ์ „๋ ฅ์‹œ์Šคํ…œ์— ํ™œ์šฉํ•˜๋Š” ์—ฐ๊ตฌ๋ฅผ ํ•˜์˜€๋‹ค. RAIN ํ”„๋กœ์ ํŠธ๋Š” 1) ์‹œ์Šคํ…œ ์ทจ์•ฝ์  & ์œ„ํ—˜ ์‹๋ณ„ 2) ์œก์ƒ ๋ฐ ๊ตํ†ต ์ธํ”„๋ผ์— ๋Œ€ํ•œ ์ทจ์•ฝ์  3) ์—๋„ˆ์ง€ ๋ฐ ํ†ต์‹  ์ธํ”„๋ผ์— ๋Œ€ํ•œ ์ทจ์•ฝ์ ์˜ 3๊ฐ€์ง€ ๋ฒ”์œ„์—์„œ ์ˆ˜ํ–‰๋˜์—ˆ๋‹ค[4].

IBM T.J Watson Research Center์—์„œ๋Š” ๋น…๋ฐ์ดํ„ฐ๋ฅผ ๋ถ„์„ํ•˜๊ธฐ ์œ„ํ•ด ํŠน์ • ์‹œ๊ณต๊ฐ„ ํ†ต๊ณ„ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ํŠนํžˆ 2003๋…„ ์œ ๋Ÿฝ์—์„œ ๋ฐœ์ƒํ•œ ํญ์—ผ๊ณผ ๊ฐ™์€ ์žฌ๋‚œ์„ ๋ชจ๋ธ๋งํ•˜๊ธฐ ์œ„ํ•ด ๊ณต๊ฐ„-์‹œ๊ฐ„ ๋ฐ์ดํ„ฐ๋กœ๋ถ€ํ„ฐ ์ธ๊ณผ ๊ด€๊ณ„๋ฅผ ์ถ”๋ก ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๊ฐœ๋ฐœํ•˜์˜€๋‹ค. ๋˜ํ•œ ๋ฐฐ์ „์‹œ์Šคํ…œ์— ํญํ’๊ณผ ๊ฐ™์€ ์žฌ๋‚œ์ด ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์„ ์กฐ์‚ฌํ•˜๊ณ  ๊ณ„์ธต์  ํฌ์•„์†ก ํšŒ๊ท€ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ์˜ˆ์ธกํ•˜์˜€๋‹ค[5].

๋ฏธ๊ตญ์˜ The Connecticut Light and Power company(CL&P)๋Š” ํญํ’ ๋ฐœ์ƒ ์‹œ ๊ณ ์žฅ ๋ฐœ์ƒ๊ณผ ๋ฒ”์œ„๋ฅผ ํ™•๋ฅ ๋ก ์ ์œผ๋กœ ๊ฒฐ์ •ํ•˜๊ธฐ ์œ„ํ•ด ๋‘ ๊ฐœ์˜ ์„ ํ˜• ๋ชจ๋ธ์„ ๊ฒฐํ•ฉ(Logistic ๊ณผ Gamma ํšŒ๊ท€ ๋ชจ๋ธ)ํ•˜์—ฌ ์žฌ๋‚œํ”ผํ•ด๋ฅผ ์˜ˆ์ธกํ•˜๋Š” ์—ฐ๊ตฌ๋ฅผ ํ•˜์˜€๋‹ค[6].

๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” 6๊ฐ€์ง€ ์žฌ๋‚œ ์œ ํ˜•์ธ ํƒœํ’, ๊ฐ•ํ’, ํ˜ธ์šฐ, ํญ์„ค, ํ•œํŒŒ, ํญ์—ผ์— ๋Œ€ํ•˜์—ฌ ์ „๋ ฅ์„ค๋น„๋ณ„ ๊ณ ์žฅ๊ฑด์ˆ˜์™€ ์„ค๋น„๊ณ ์žฅ์œผ๋กœ ์ธํ•œ ์ •์ „๊ฐ€๊ตฌ์ˆ˜๋ฅผ ์˜ˆ์ธกํ•˜๊ธฐ ์œ„ํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœ ๊ณผ์ •์„ ๊ธฐ์ˆ ํ•˜๊ณ  ์žˆ๋‹ค. 2์žฅ์—์„œ๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœ์— ํ•„์š”ํ•œ ๊ธฐ์ƒ ๋ฐ์ดํ„ฐ์™€ ์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ ๋ฐ์ดํ„ฐ๋ฅผ ๋ถ„์„ํ•˜๊ณ , ํ”ผ์–ด์Šจ ์ƒ๊ด€๊ด€๊ณ„ ๋ถ„์„ ๊ธฐ๋ฒ•์„ ํ™œ์šฉํ•˜์—ฌ ์ž…๋ ฅ ๋ฐ์ดํ„ฐ ์„ ์ •์— ๋Œ€ํ•œ ๊ณผ์ •์„ ๊ธฐ์ˆ ํ•œ๋‹ค. 3์žฅ์—์„œ๋Š” ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก์„ ์œ„ํ•œ ์ธ๊ณต์‹ ๊ฒฝ๋ง(Artificial Neural Network, ANN) ๋ชจ๋ธ์„ ๊ฐœ๋ฐœํ•˜๋Š” ๊ณผ์ •์„ ๊ธฐ์ˆ ํ•˜๋ฉฐ, 4์žฅ์—์„œ๋Š” ๊ฐœ๋ฐœํ•œ ANN ๋ชจ๋ธ๊ณผ ์„ ํ˜•ํšŒ๊ท€๋ถ„์„(Linear Regression Analysis, REG) ๋ชจ๋ธ์˜ ์˜ˆ์ธก๊ฒฐ๊ณผ๋ฅผ ๋น„๊ตํ•˜์—ฌ ํ‰๊ฐ€ํ•˜๊ณ , ์‹ค์ œ ํƒœํ’ โ€˜์†”๋ฆญโ€™์— ๋Œ€ํ•œ ์˜ˆ์ธก ๊ฒฐ๊ณผ๋ฅผ ๊ธฐ์ˆ ํ•˜๊ณ  ์žˆ๋‹ค. ๋งˆ์ง€๋ง‰์œผ๋กœ 5์žฅ์—์„œ๋Š” ํ–ฅํ›„ ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์ •ํ™•๋„ ํ–ฅ์ƒ์„ ์œ„ํ•œ ํ›„์† ์—ฐ๊ตฌ๋ฐฉํ–ฅ์„ ์ œ์‹œํ•˜๊ณ  ์žˆ๋‹ค.

2. ๋ฐ์ดํ„ฐ ์ˆ˜์ง‘ ๋ฐ ์ƒ๊ด€๊ด€๊ณ„ ๋ถ„์„

์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ๋ชจ๋ธ์€ ๊ธฐ๊ณ„ํ•™์Šต(Machine Learning, ML)๋ฒ•์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ๋ฐœํ•œ๋‹ค. ์˜ˆ์ธก ๋ชจ๋ธ์ด ํ•™์Šตํ•˜๊ธฐ ์œ„ํ•ด ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ๊ฐ€ ํ•„์š”ํ•˜๋ฉฐ, ์ด๋Š” ๊ธฐ์ƒํŠน๋ณด๊ฐ€ ๋ฐœ์ƒํ–ˆ๋˜ ๊ธฐ๊ฐ„์˜ {๊ณผ๊ฑฐ๊ธฐ์ƒ๋ฐ์ดํ„ฐ(ํ’์†, ํ˜ธ์šฐ, ์˜จ๋„ ๋“ฑ), ๊ณ ์žฅ๋ฐ์ดํ„ฐ(๊ณ ์žฅ๊ฑด์ˆ˜, ์ •์ „๊ณ ๊ฐ์ˆ˜)}๋กœ ๊ตฌ์„ฑ๋œ๋‹ค. ๋˜ํ•œ ํ•™์Šต๋œ ๋ชจ๋ธ์„ ์ด์šฉํ•˜์—ฌ ์˜ˆ์ธกํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ์˜ˆ์ธก์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ๊ฐ€ ํ•„์š”ํ•˜๋ฉฐ, ์ด๋Š” ๊ธฐ์ƒํŠน๋ณด๊ฐ€ ๋ฐœ๋ น๋œ ์ง€์—ญ์˜ {๋ฏธ๋ž˜๊ธฐ์ƒ๋ฐ์ดํ„ฐ(ํ’์†, ํ˜ธ์šฐ, ์˜จ๋„ ๋“ฑ)}์œผ๋กœ ๊ตฌ์„ฑ๋œ๋‹ค. ๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” ์ด์™€ ๊ฐ™์€ ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ์™€ ์˜ˆ์ธก ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ๊ตฌ์ถ•ํ•˜๊ณ  ๋ถ„์„ํ•˜๊ธฐ ์œ„ํ•ด 2007๋…„ ๋ถ€ํ„ฐ 2018๋…„ ๊นŒ์ง€์˜ ๊ตญ๋ฏผ์•ˆ์ „์ฒ˜์—์„œ ์ œ๊ณตํ•˜๋Š” โ€œ๊ธฐ์ƒํŠน๋ณด ๋ฐ์ดํ„ฐโ€, ๊ธฐ์ƒ์ฒญ์—์„œ ์ œ๊ณตํ•˜๋Š” โ€œ๊ธฐ์ƒ๊ด€์ธก ๋ฐ์ดํ„ฐโ€, โ€œ๋™๋„ค์˜ˆ๋ณด ๋ฐ์ดํ„ฐโ€, ๊ทธ๋ฆฌ๊ณ  ์ „๋ ฅํšŒ์‚ฌ์˜ โ€œ์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ๋ฐ์ดํ„ฐโ€๋ฅผ ์ˆ˜์ง‘ํ•˜์˜€์œผ๋ฉฐ, ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์ž…๋ ฅ์œผ๋กœ ์œ ํšจํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์„ ์ •ํ•˜๊ธฐ ์œ„ํ•ด ์„ค๋น„๊ณ ์žฅ์— ์˜ํ–ฅ๋ ฅ์ด ์žˆ๋Š” ๊ธฐ์ƒ์š”์ธ(์˜จ๋„, ๊ฐ•์šฐ๋Ÿ‰, ํ’์† ๋“ฑ)์— ๋Œ€ํ•œ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๋ถ„์„ํ•˜์˜€๋‹ค.

2.1 ์ž์—ฐ์žฌ๋‚œ์— ๋”ฐ๋ฅธ ์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ ์˜ํ–ฅ ์ธ์ž ๋ถ„์„ ๋ฐ ๋ฐ์ดํ„ฐ ์ˆ˜์ง‘

2.1.1 ๊ตญ๋ฏผ์•ˆ์ „์ฒ˜ ๊ธฐ์ƒํŠน๋ณด ๋ฐ์ดํ„ฐ

๊ตญ๋ฏผ์•ˆ์ „์ฒ˜๋Š” ํƒœํ’, ๊ฐ•ํ’, ํ˜ธ์šฐ, ํญ์„ค, ํ•œํŒŒ, ํญ์—ผ์— ๋Œ€ํ•˜์—ฌ ๊ฒฝ๋ณด/์ฃผ์˜๋ณด์˜ ํ˜•ํƒœ๋กœ ๋ฐœ๋ น ๊ธฐ๊ฐ„๊ณผ ๋Œ€์ƒ์ง€์—ญ์ด ํฌํ•จํ•˜๋Š” ๊ธฐ์ƒํŠน๋ณด๋ฅผ ๋ฐœํ‘œํ•œ๋‹ค. ํ”ผํ•ด์ง€์—ญ๊ณผ ํ”ผํ•ด๊ทœ๋ชจ ๋ฐ ํ”ผํ•ด์‹œ๊ฐ„์— ๋Œ€ํ•œ ํ•™์Šต๊ณผ ์˜ˆ์ธก์„ ์œ„ํ•ด โ€œ๊ธฐ์ƒํŠน๋ณด ๋ฐ์ดํ„ฐโ€๋ฅผ ํ™œ์šฉํ•˜์˜€๋‹ค.

2.1.2 ๊ธฐ์ƒ์ฒญ ๊ธฐ์ƒ๊ด€์ธก ๋ฐ์ดํ„ฐ

ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ๊ตฌ์„ฑํ•˜๋Š” ๊ณผ๊ฑฐ ๊ธฐ์ƒ๋ฐ์ดํ„ฐ๋Š” ๊ธฐ์ƒ์ฒญ์˜ ๊ธฐ์ƒ๊ด€์ธก์‹œ์Šคํ…œ(Automatic Weather System, AWS)์—์„œ ์ˆ˜์ง‘ํ•˜์˜€๋‹ค. ๊ธฐ์ƒ๊ด€์ธก ๋ฐ์ดํ„ฐ๋Š” ๊ด€์ธก์ผ์‹œ, ์ง€์ (๊ฒฝ์œ„๋„ ๋ฐ ๊ณ ๋„), ๋‚ ์”จ, ๊ธฐ์˜จ, ํ’ํ–ฅ, ํ’์†, ๊ฐ•์šฐ๋Ÿ‰, ์ ์„ค๋Ÿ‰, ์Šต๋„ ๋“ฑ์„ ํฌํ•จํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๊ธฐ์ƒ์ฒญ์˜ 1์‹œ๊ฐ„ ๊ธฐ์ƒ๊ด€์ธก ๋ฐ์ดํ„ฐ๋ฅผ ํ™œ์šฉํ•˜์˜€๋‹ค.

2.1.3 ๊ธฐ์ƒ์ฒญ ๋™๋„ค์˜ˆ๋ณด ๋ฐ์ดํ„ฐ

์˜ˆ์ธก์— ์‚ฌ์šฉ๋˜๋Š” ๋ฏธ๋ž˜ ๊ธฐ์ƒ๋ฐ์ดํ„ฐ๋Š” ๊ธฐ์ƒ์ฒญ์˜ ๋™๋„ค์˜ˆ๋ณด ๋ฐ์ดํ„ฐ๋ฅผ ํ™œ์šฉํ•˜์—ฌ ์ˆ˜์ง‘ํ•œ๋‹ค. ๋™๋„ค์˜ˆ๋ณด ๋ฐ์ดํ„ฐ๋Š” ์˜ˆ์ธก์ผ์‹œ, ์ง€์ (๋™/์/๋ฉด), ๋‚ ์”จ, ๊ธฐ์˜จ, ํ’ํ–ฅ, ํ’์†, ๊ฐ•์šฐ๋Ÿ‰, ์ ์„ค๋Ÿ‰, ์Šต๋„ ๋“ฑ์„ ํฌํ•จํ•˜๊ณ  ์žˆ๋‹ค. ๋™๋„ค์˜ˆ๋ณด ๋ฐ์ดํ„ฐ๋Š” ๊ทธ๋ฆผ 1๊ณผ ๊ฐ™์ด 3์‹œ๊ฐ„ ๋‹จ์œ„๋กœ ์˜ˆ๋ณด๋œ๋‹ค.

๊ทธ๋ฆผ. 1. ๊ธฐ์ƒ์ฒญ ๋™๋„ค์˜ˆ๋ณด ์‚ฌ๋ก€(May 28, 2018, ๋น›๊ฐ€๋žŒ๋™)

Fig. 1. A case of the KMA's (May 28, 2018, Bitgaram-dong)

../../Resources/kiee/KIEE.2019.68.9.1085/fig1.png

๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๊ธฐ์ƒ์ฒญ์˜ 3์‹œ๊ฐ„ ๋‹จ์œ„ ์˜ˆ๋ณด๋ฅผ ์˜ˆ์ธก ์ž…๋ ฅ๋ฐ์ดํ„ฐ๋กœ ์‚ฌ์šฉํ•˜๊ธฐ ์œ„ํ•˜์—ฌ, 1์‹œ๊ฐ„ ๊ธฐ์ƒ๊ด€์ธก ๋ฐ์ดํ„ฐ๋ฅผ ๊ทธ์— ์ƒ์‘ํ•˜๋Š” 3์‹œ๊ฐ„ ๋‹จ์œ„๋กœ ๋ณ€๊ฒฝํ•˜๋Š” ์ „์ฒ˜๋ฆฌ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜์˜€๋‹ค. ์ „์ฒ˜๋ฆฌ ์ž‘์—…์€ 3์‹œ๊ฐ„ ๋‹จ์œ„์˜ ์‹œ๊ฐ„๊ฐ„๊ฒฉ ์œˆ๋„์šฐ๋ฅผ 1์‹œ๊ฐ„ ๋‹จ์œ„๋กœ ์Šฌ๋ผ์ด๋”ฉ ์ด๋™ํ•˜๋Š” ๋ฐฉ์‹(0์‹œ-3์‹œ, 1์‹œ-4์‹œ, ... ,23์‹œ-์ต์ผ2์‹œ)์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ํ‘œ 1๊ณผ ๊ฐ™์€ 1์‹œ๊ฐ„ ๋‹จ์œ„์˜ ํ’์†๋ฐ์ดํ„ฐ๋ฅผ ์ด์šฉํ•˜์—ฌ ํ‘œ 2์™€ ๊ฐ™์€ 3์‹œ๊ฐ„ ์œˆ๋„์šฐ๋ฅผ ํ™œ์šฉํ•˜์—ฌ 3์‹œ๊ฐ„ ๋‹จ์œ„์˜ ํ‰๊ท ํ’์†์„ ๊ตฌํ•˜์˜€๊ณ , ์ด๋ฅผ ํ†ตํ•ด ์˜ˆ์ธก ์ž…๋ ฅ๋ฐ์ดํ„ฐ์™€ ๊ฐ™์€ 3์‹œ๊ฐ„ ๋‹จ์œ„์˜ ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ๋ฅผ ๊ตฌ์ถ•ํ•˜์˜€๋‹ค.

ํ‘œ 1. 0์‹œ~24์‹œ AWS ํ’์† ๋ฐ์ดํ„ฐ ์ธก์ • ์‚ฌ๋ก€

Table 1. Measurement of AWS wind speed data between 0 and 24 hours

์‹œ๊ฐ„๋Œ€

(h)

1์‹œ๊ฐ„

ํ‰๊ท ํ’์†(m/s)

์‹œ๊ฐ„๋Œ€

(h)

1์‹œ๊ฐ„

ํ‰๊ท ํ’์†(m/s)

0-1

5

12-13

5

1-2

5

13-14

3

2-3

10

14-15

3

3-4

15

15-16

2

4-5

15

16-17

1

5-6

15

17-18

1

6-7

15

18-19

1

7-8

10

19-20

1

8-9

10

20-21

1

9-10

5

21-22

0

10-11

5

22-23

0

11-12

5

23-00

0

ํ‘œ 2. 3์‹œ๊ฐ„ ๋‹จ์œ„ ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ ์ • : ํ•˜๋ฃจ(24์‹œ๊ฐ„)๋ฅผ 3์‹œ๊ฐ„ ๋‹จ์œ„ 24๊ฐœ ์œˆ๋„์šฐ๋กœ ๊ตฌ๋ถ„ํ•จ

Table 2. Selection of 3 hour learning input data: 24 hours divided into 24 windows per 3 hours

์‹œ๊ฐ„๋Œ€(h)

3์‹œ๊ฐ„

ํ‰๊ท ํ’์†

(m/s)

ํŠน๋ณด์ˆ˜์ค€

๋ฐ์ดํ„ฐ์„ ์ •

0-3

6.7

-

1-4

10

-

2-5

13.3

์ฃผ์˜๋ณด

์„ ์ •(์ฃผ์˜๋ณด)

3-6

15

๊ฒฝ๋ณด

์„ ์ •(๊ฒฝ๋ณด)

4-7

15

๊ฒฝ๋ณด

์„ ์ •(๊ฒฝ๋ณด)

5-8

13.3

์ฃผ์˜๋ณด

์„ ์ •(์ฃผ์˜๋ณด)

6-9

11.7

์ฃผ์˜๋ณด

์„ ์ •(์ฃผ์˜๋ณด)

7-10

8.3

-

๏ฝž

11-14

4.3

-

2.2 ์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ ๋ฐ์ดํ„ฐ ๋ถ„์„

ํ‘œ 3. ์ž์—ฐ์žฌ๋‚œ์œผ๋กœ ์ธํ•œ ์ „๋ ฅ์„ค๋น„์˜ ๊ณ ์žฅ ๋ฐœ์ƒ(2007~2018) ํ˜„ํ™ฉ

Table 3. Status of power facility failures due to natural disasters(2007-2018)

์ „๋ ฅ์„ค๋น„

์ž์—ฐ์žฌ๋‚œ(๊ฑด)

์ „์ฒด๊ณ ์žฅ

(๊ฑด)

ํƒœํ’

๊ฐ•ํ’

ํ˜ธ์šฐ

๊ณ„

๋ณ€์••๊ธฐ

166

39

87

292

18,746

์ „์„ 

2,790

1,080

809

4,679

20,472

๊ฐœํ๊ธฐ

80

28

37

145

7,693

์• ์ž

149

51

39

239

7,628

ํ”ผ๋ขฐ๊ธฐ

177

78

54

309

16,973

์ „์ฃผ

254

116

260

630

2,065

COS

342

191

98

631

6,120

๊ณ„

3,958

1,583

1,384

6,925

79,697

์œ„์˜ ์ „์ฒ˜๋ฆฌ ๊ณผ์ •์„ ํ†ตํ•ด ๊ธฐ์ƒ์žฌํ•ด ์ข…๋ฅ˜ ๋ฐ ๋ฐœ์ƒ์‹œ๊ฐ์„ ํŒŒ์•…ํ•˜๊ณ  ์„ค๋น„๊ณ ์žฅ ์‹œ๊ฐ๊ณผ ๋น„๊ตํ•จ์œผ๋กœ์จ ๊ทธ ๊ณ ์žฅ์˜ ์›์ธ์„ ๋งค์นญํ•˜์˜€๋‹ค. ๊ทธ ๊ฒฐ๊ณผ๋Š” ํ‘œ 3๊ณผ ๊ฐ™๋‹ค.

2.3. ๊ธฐ์ƒ๊ณผ ์„ค๋น„๊ณ ์žฅ ๋ฐ์ดํ„ฐ ์ƒ๊ด€๊ด€๊ณ„ ๋ถ„์„

2.3.1 ์ƒ๊ด€๊ด€๊ณ„ ๋ถ„์„ ๊ฐœ์š”

๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์˜ˆ์ธก๋ชจ๋ธ ์ถœ๋ ฅ(์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ๊ฑด์ˆ˜, ์ •์ „๊ณ ๊ฐ์ˆ˜)์— ์˜ํ–ฅ์„ ๋ฏธ์น˜๋Š” ์ฃผ์š” ์ž…๋ ฅ๋ณ€์ˆ˜(๊ธฐ์ƒ ์š”์†Œ) ์„ ์ •์„ ์œ„ํ•ด ํ”ผ์–ด์Šจ ์ƒ๊ด€๊ด€๊ณ„ ๋ถ„์„๋ฒ•์„ ์‚ฌ์šฉํ•˜์˜€๊ณ , ๊ธฐ์ƒ ์š”์†Œ๋Š” ํ‰๊ท  ๊ธฐ์˜จ, ์ตœ์†Œ ์˜จ๋„, ์ตœ๋Œ€ ์˜จ๋„, ํ‰๊ท  ๊ฐ•์ˆ˜๋Ÿ‰, ์ตœ์†Œ ๊ฐ•์ˆ˜๋Ÿ‰, ์ตœ๋Œ€ ๊ฐ•์ˆ˜๋Ÿ‰, ํ‰๊ท  ํ’์†, ์ตœ์†Œ ํ’์†, ์ตœ๋Œ€ ํ’์†, ํ‰๊ท  ํ’ํ–ฅ, ์ตœ์†Œ ํ’ํ–ฅ, ์ตœ๋Œ€ ํ’ํ–ฅ์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ํ”ผ์–ด์Šจ ์ƒ๊ด€๊ณ„์ˆ˜๋Š” ๋‹ค์Œ ๋ฐฉ์ •์‹์œผ๋กœ ๊ณ„์‚ฐํ•  ์ˆ˜ ์žˆ๋‹ค.

(1)
$R_{m}=\dfrac{n\times(\sum X_{n}Y)-(\sum X_{m)}\times(\sum Y)}{\sqrt{(n\times\sum X_{m}^{2})\times(\sum X_{m})^{2}\times(n\times\sum Y^{2})-(\sum Y)^{2}}}$

์ˆ˜์‹(1)์—์„œ $R_{m}$์€ ๊ธฐ์ƒ ๋ฐ์ดํ„ฐ m์— ๋Œ€ํ•œ ํ”ผ์–ด์Šจ ์ƒ๊ด€๊ณ„์ˆ˜๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค. $n$์€ ๋ฐ์ดํ„ฐ ์ˆ˜์ด๋ฉฐ, $X_{m}$์€ ๊ธฐ์ƒ ๋ฐ์ดํ„ฐ์ด๋‹ค. $Y$๋Š” ์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ ์ˆ˜๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค.

ํ”ผ์–ด์Šจ ์ƒ๊ด€๊ณ„์ˆ˜($R_{m}$)์€ $-1$๊ณผ $1$์‚ฌ์ด์˜ ๊ฐ’์„ ๊ฐ–๋Š”๋‹ค. ์ž…๋ ฅ๋ณ€์ˆ˜๋ฅผ ์„ ์ •ํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ์ž„๊ณ„๊ฐ’์ด ํ•„์š”ํ•˜๋ฉฐ ์ด๋Š” $0$๊ณผ $1$ ์‚ฌ์ด์˜ ๋ฒ”์œ„์—ฌ์•ผ ํ•œ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๊ธฐ๋ณธ ์ž„๊ณ„๊ฐ’์„ $0.2$๋กœ ์ ์šฉํ•˜์˜€๋‹ค. ์ „๋ ฅ์„ค๋น„๋Š” ๊ธฐ์ƒ ์•…ํ™” ์ƒํ™ฉ์—์„œ๋„ ์ •์ƒ์œผ๋กœ ์šด์˜๋  ์ˆ˜ ์žˆ๋„๋ก ๊ฐ•ํ™”๋œ ์„ค๊ณ„ ๊ธฐ์ค€์— ๋”ฐ๋ผ ๊ด€๋ฆฌ๋˜๋Š” ๊ด€๊ณ„๋กœ ์ž์—ฐ์žฌ๋‚œ ์ƒํ™ฉ์—์„œ๋„ ๊ณ ์žฅ์ด ๋“œ๋ฌผ๊ฒŒ ๋ฐœ์ƒํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์ผ๋ฐ˜์ ์œผ๋กœ $0.5$๋ณด๋‹ค ๋‚ฎ์€ $0.2$๋ฅผ ์ ์šฉํ•  ์ˆ˜ ์žˆ๋‹ค.

2.3.2 ์ƒ๊ด€๊ด€๊ณ„ ๋ถ„์„ ๊ฒฐ๊ณผ

ํ‘œ 4. ํƒœํ’์žฌ๋‚œ์— ๋Œ€ํ•œ ๊ณ ์••๋ฐฐ์ „์„ค๋น„(๊ณ ์žฅ, ์ •์ „๊ณ ๊ฐ์ˆ˜)์™€ ๊ธฐ์ƒ์š”์†Œ ์‚ฌ์ด์˜ ์ƒ๊ด€๊ด€๊ณ„ ๋ถ„์„ ๊ฒฐ๊ณผ(์„œ๊ท€ํฌ์ง€์—ญ)

Table 4. Correlation analysis results between distribution facilities failure and meteorological factors for typhoon(Seogwipo area)

๊ตฌ๋ถ„

ํ‰๊ท 

์˜จ๋„

์ตœ์†Œ

์˜จ๋„

์ตœ๋Œ€

์˜จ๋„

ํ‰๊ท 

ํ’์†

์ตœ์†Œ

ํ’์†

์ตœ๋Œ€

ํ’์†

ํ‰๊ท 

ํ’ํ–ฅ

์ตœ์†Œ

ํ’ํ–ฅ

์ตœ๋Œ€

ํ’ํ–ฅ

ํ‰๊ท 

๊ฐ•์ˆ˜๋Ÿ‰

์ตœ์†Œ

๊ฐ•์ˆ˜๋Ÿ‰

์ „์ฃผ

๊ณ ์žฅ๊ฑด์ˆ˜

-0.060

-0.115

-0.037

0.275

0.170

0.274

0.012

-0.037

-0.013

0.303

0.098

์ •์ „๊ณ ๊ฐ์ˆ˜

-0.058

-0.058

-0.034

0.244

0.196

0.245

0.028

-0.030

0.005

0.321

0.092

์ „์„ 

๊ณ ์žฅ๊ฑด์ˆ˜

-0.037

-0.156

-0.045

0.561

0.332

0.542

-0.108

0.015

-0.196

0.447

0.371

์ •์ „๊ณ ๊ฐ์ˆ˜

-0.057

-0.158

-0.061

0.547

0.340

0.525

-0.086

0.008

-0.162

0.480

0.467

์• ์ž

๊ณ ์žฅ๊ฑด์ˆ˜

-0.027

-0.058

-0.034

0.244

0.194

0.245

0.028

-0.030

0.005

0.321

0.092

์ •์ „๊ณ ๊ฐ์ˆ˜

-0.040

-0.002

-0.028

0.044

-0.00

0.074

0.028

0.089

0.011

-0.027

-0.020

ํ”ผ๋ขฐ๊ธฐ

๊ณ ์žฅ๊ฑด์ˆ˜

-0.026

-0.241

-0.021

0.391

0.130

0.365

-0.023

-0.016

-0.131

0.183

0.221

์ •์ „๊ณ ๊ฐ์ˆ˜

0.012

-0.214

-0.006

0.356

0.077

0.330

-0.077

-0.024

-0.156

0.158

0.234

CoS

๊ณ ์žฅ๊ฑด์ˆ˜

-0.120

-0.143

-0.078

0.398

0.262

0.399

0.023

-0.009

-0.057

0.439

0.320

์ •์ „๊ณ ๊ฐ์ˆ˜

-0.016

-0.081

-0.007

0.339

0.198

0.337

-0.083

0.030

-0.162

0.305

0.209

๋ณ€์••๊ธฐ

๊ณ ์žฅ๊ฑด์ˆ˜

-0.006

-0.134

-0.006

0.371

0.181

0.352

-0.059

0.014

-0.162

0.188

0.182

์ •์ „๊ณ ๊ฐ์ˆ˜

-0.005

-0.075

0.002

0.363

0.234

0.342

-0.028

0.045

-0.137

0.154

0.135

ํ‘œ 4๋Š” ํƒœํ’ ๋ฐœ์ƒ ์‹œ ์ œ์ฃผ์ง€์—ญ ์„œ๊ท€ํฌ์ง€์‚ฌ์˜ ๊ธฐ์ƒ์š”์†Œ์™€ ๋ฐฐ์ „์„ค๋น„ ๊ณ ์žฅ์— ๋Œ€ํ•œ ์ƒ๊ด€๋ถ„์„ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ์ƒ๊ด€๊ณ„์ˆ˜์˜ ์ ˆ๋Œ€ ๊ฐ’์ด 0.2 ์ด์ƒ์ธ ๊ฒฝ์šฐ๋Š” ํ‘œ 4์— ์Œ์˜์œผ๋กœ ํ‘œ์‹œํ•˜์˜€๋‹ค. ์ƒ๊ด€๊ณ„์ˆ˜๊ฐ€ 0.2 ์ด์ƒ์ธ ๊ฒฝ์šฐ ๋‘ ๋ณ€์ˆ˜๋Š” ์•ฝํ•œ ์„ ํ˜•๊ด€๊ณ„ ์ด์ƒ์ž„์„ ๋‚˜ํƒ€๋‚ธ๋‹ค. ํ•ด๋‹น ํ‘œ์—์„œ ์ „์ฃผ ๊ณ ์žฅ์— ๋Œ€ํ•ด์„œ๋Š” ํ‰๊ท ๊ฐ•์ˆ˜๋Ÿ‰์ด 0.303์œผ๋กœ ๊ฐ€์žฅ ํฐ ์ƒ๊ด€๊ณ„์ˆ˜๋ฅผ ๊ฐ€์ง€๋Š” ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ๋‹ค. ์ฆ‰, ์ „์ฃผ ๊ณ ์žฅ์€ ํ‰๊ท ๊ฐ•์ˆ˜๋Ÿ‰์ด ๊ฐ€์žฅ ํฐ ์˜ํ–ฅ์„ ๋ฏธ์น  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒ๋œ๋‹ค. ์ „์ฃผ ๊ณ ์žฅ์œผ๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜์˜ ๊ฒฝ์šฐ๋„ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ํ‰๊ท ๊ฐ•์ˆ˜๋Ÿ‰์ด ๊ฐ€์žฅ ํฐ ์ƒ๊ด€๊ณ„์ˆ˜๋ฅผ ๊ฐ€์ง„๋‹ค.

์ƒ๊ด€๊ณ„์ˆ˜๊ฐ€ ๊ธฐ์ค€์น˜๋ฅผ ์ดˆ๊ณผํ•˜๋Š” ๊ธฐ์ƒ์š”์†Œ๋“ค์€ ํ•ด๋‹น ์ง€์—ญ์˜ ์ธ๊ณต์‹ ๊ฒฝ๋ง ๊ธฐ๊ณ„ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ํ•™์Šต๊ณผ ์˜ˆ์ธก๊ณผ์ •์—์„œ ์ž…๋ ฅ ์š”์†Œ๋กœ ์‚ฌ์šฉ๋œ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด ์ „์„ ์˜ ๊ณ ์žฅ์„ ํ•™์Šตยท์˜ˆ์ธกํ•˜๋Š” ๊ฒฝ์šฐ์—๋Š” ๊ธฐ์ƒ์š”์†Œ์—์„œ {ํ‰๊ท ํ’์†, ์ตœ์†Œํ’์†, ์ตœ๋Œ€ํ’์†, ํ‰๊ท ๊ฐ•์ˆ˜๋Ÿ‰, ์ตœ์†Œ๊ฐ•์ˆ˜๋Ÿ‰, ์ตœ๋Œ€๊ฐ•์ˆ˜๋Ÿ‰}์„ ์‚ฌ์šฉํ•˜๋ฉฐ, CoS์˜ ๊ณ ์žฅ๊ฑด์ˆ˜์™€ ์ •์ „๊ณ ๊ฐ์ˆ˜๋ฅผ ํ•™์Šตยท์˜ˆ์ธกํ•˜๋Š” ๊ฒฝ์šฐ์—๋Š” ์ „์„ ๊ณผ ๋‹ค๋ฅด๊ฒŒ {์ตœ์†Œํ’์†}์ด ์ œ์™ธ๋œ {ํ‰๊ท ํ’์†, ์ตœ๋Œ€ํ’์†, ํ‰๊ท ๊ฐ•์ˆ˜๋Ÿ‰, ์ตœ์†Œ๊ฐ•์ˆ˜๋Ÿ‰, ์ตœ๋Œ€๊ฐ•์ˆ˜๋Ÿ‰}์„ ๊ธฐ์ƒ์š”์†Œ์˜ ์ž…๋ ฅ์œผ๋กœ ์‚ฌ์šฉํ•œ๋‹ค.

3. ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœ ๋ฐ ํ‰๊ฐ€

3.1 ๊ฐœ๋ฐœ ๋ชจ๋ธ ๊ฐœ์š”

3.1.1 ์ธ๊ณต์‹ ๊ฒฝ๋ง(Artificial Neural Network, ANN) ๋ชจ๋ธ

ANN์€ ์ธ๊ฐ„ ๋˜๋Š” ๋™๋ฌผ์˜ ๋‘๋‡Œ๋ฅผ ๋งŒ๋“œ๋Š” ์ƒ๋ฌผํ•™์  ์‹ ๊ฒฝ ๋„คํŠธ์›Œํฌ์—์„œ ์˜๊ฐ์„ ์–ป์€ ์ปดํ“จํ„ฐ ์‹œ์Šคํ…œ์ด๋‹ค. ANN์€ ๋ฐ์ดํ„ฐ์™€ ๊ทธ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ํ•™์Šตํ•˜์—ฌ ๊ฒฐ๊ณผ๋ฅผ ์ถ”์ •ํ•œ๋‹ค. ๋น„์„ ํ˜•์ ์ธ ์ข…์†์„ฑ์„ ํ•ด๊ฒฐํ•˜๊ณ  ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ํฌํ•จ๋œ ๋…ธ์ด์ฆˆ๋ฅผ ์ œ๊ฑฐํ•˜๊ธฐ ์œ„ํ•ด์„œ ๋ฐ์ดํ„ฐ๋ฅผ ํ•™์Šตํ•œ๋‹ค. ANN์€ ๋‰ด๋Ÿฐ์ด๋ผ ๋ถˆ๋ฆฌ๋Š” ์—ฐ๊ฒฐ ๋‹จ์œ„ ๋ชจ์Œ์œผ๋กœ ๊ตฌ์„ฑ๋œ๋‹ค. ํ•˜๋‚˜์˜ ๊ธฐ๋ณธ ๋‰ด๋Ÿฐ์€ ์ตœ์†Œํ•œ ์ž…๋ ฅ, ์ถœ๋ ฅ ๋ฐ ๊ฐ€์ค‘ ์ž…๋ ฅ์„ ์ถœ๋ ฅ์— ๋งคํ•‘ํ•˜๋Š” ํ™œ์„ฑํ™” ํ•จ์ˆ˜๊ฐ€ ์žˆ์–ด์•ผํ•œ๋‹ค. ์ด ๊ธฐ๋Šฅ์„ ํ™œ์„ฑํ™” ๊ธฐ๋Šฅ ๋˜๋Š” ์ „๋‹ฌ ๊ธฐ๋Šฅ์ด๋ผ๊ณ  ํ•œ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ํ™œ์„ฑํ™”ํ•จ์ˆ˜๋กœ softplus ํ•จ์ˆ˜๋ฅผ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ๋˜ํ•œ, ์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ ์˜ˆ์ธก๊ณผ ๊ฐ™์€ ์„ ํ˜• ์˜ˆ์ธก์„ ์œ„ํ•ด ANN์€ ๋‹ค์ค‘ ์ž…๋ ฅ ๋ฐ ํ™œ์„ฑํ™” ํ•จ์ˆ˜ ๋ฐ ๋‹จ์ผ ์ถœ๋ ฅ์œผ๋กœ ๊ตฌ์„ฑํ•  ์ˆ˜ ์žˆ๋‹ค. ๊ทธ๋ฆผ 2๋Š” ANN์˜ ๊ตฌ์กฐ๋ฅผ ๋ณด์—ฌ์ค€๋‹ค.

๊ทธ๋ฆผ. 2. ANN ๊ตฌ์กฐ

Fig. 2. ANN Structure

../../Resources/kiee/KIEE.2019.68.9.1085/fig2.png

๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ๊ฐ€์žฅ ๋งŽ์€ ๋ถ„์•ผ์—์„œ ์‚ฌ์šฉ๋˜๋Š” Gradient descent๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋น„์šฉ ํ•จ์ˆ˜๋ฅผ ์ตœ์†Œํ™”ํ•˜์˜€๋‹ค.

ANN ๋ชจ๋ธ ๊ฐœ๋ฐœ ๊ณผ์ •์—์„œ ๊ฒฐ์ •ํ•ด์•ผํ•˜๋Š” ์„ธ ๊ฐ€์ง€ ์„ค์ •์ด ์žˆ๋‹ค. ์ด ์„ค์ •์€ ํžˆ๋“  ๋ ˆ์ด์–ด์˜ ๋…ธ๋“œ ์ˆ˜, ์ •๊ทœํ™” ํŒŒ๋ผ๋ฏธํ„ฐ($\lambda$) ๋ฐ ์ž„์˜ ์ดˆ๊ธฐํ™” ํŒŒ๋ผ๋ฏธํ„ฐ($\epsilon$)์ด๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ ์–ป์–ด์ง„ ANN ์„ค์ • ์ง‘ํ•ฉ ์ค‘ ํžˆ๋“  ๋ ˆ์ด์–ด๋Š” 3๊ฐœ์˜ ๋…ธ๋“œ๋กœ ์„ค์ •ํ•˜์˜€๋‹ค. ํžˆ๋“  ๋ ˆ์ด์–ด๋ฅผ 4๊ฐœ ์ด์ƒ์œผ๋กœ ์ฆ๊ฐ€์‹œํ‚ค๋Š” ๊ฒฝ์šฐ ๊ธฐ๊ณ„ ํ•™์Šต ์‹œ๊ฐ„๊ณผ ์˜ˆ์ธก ์‹œ๊ฐ„์ด ์ฆ๊ฐ€๋˜์—ˆ์œผ๋ฉฐ, ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์„ ๋Œ€์ƒ์œผ๋กœ Test Error์˜ ์ •ํ™•๋„๊ฐ€ ๋‚ฎ์•„์ง€๋Š” ํ˜„์ƒ์ด ๋ฐœ์ƒํ•˜์˜€๋‹ค. ์ดˆ๊ธฐํ™” ํŒŒ๋ผ๋ฏธํ„ฐ($\epsilon$), ์ •๊ทœํ™” ํŒŒ๋ผ๋ฏธํ„ฐ($\lambda$)๋Š” ์ง€์ˆ˜์ ์œผ๋กœ ์ฆ๊ฐ€ํ•˜๋Š” ๊ฐ’์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฒฐ์ •ํ•œ๋‹ค. ์ง€์ˆ˜์ ์œผ๋กœ ์ฆ๊ฐ€ํ•˜๋Š” ๊ฐ’์€ ์ˆ˜์‹(2)์™€ ์ˆ˜์‹(3)์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ณ„์‚ฐํ•œ๋‹ค. ANN ๋ชจ๋ธ ๊ธฐ๊ณ„ํ•™์Šต ๊ณผ์ •์—์„œ ์ดˆ๊ธฐํ™” ํŒŒ๋ผ๋ฏธํ„ฐ($\epsilon$), ์ •๊ทœํ™” ํŒŒ๋ผ๋ฏธํ„ฐ($\lambda$)๋Š” ์ง€์—ญ๋ณ„, ์„ค๋น„๋ณ„ ํ•™์Šต ๊ณผ์ •์—์„œ ์•„๋ž˜์˜ ๊ฐ’๋“ค ์ค‘์—์„œ ์ตœ์ ์˜ ๊ฐ’์œผ๋กœ ์ž๋™์œผ๋กœ ์„ ์ •ํ•˜์—ฌ ๊ด€๋ฆฌํ•œ๋‹ค.

(2)
\begin{align*} \epsilon & =\left\{0.01\times\left\{2^{0},\: 2^{1},\: 2^{2},\: 2^{3},\: 2^{4},\: 2^{5},\: 2^{6},\: 2^{7}\right\}\right\}\\ & =\{0.01,\: 0.02,\: 0.04,\: 0.08,\: 0.16,\: 0.64,\: 1.28\} \end{align*}

(3)
\begin{align*} \lambda & =\left(0.01\times\left(e^{0},\: e^{1},\: e^{2},\: e^{3},\: e^{4},\: e^{5},\: e^{6},\: e^{7}\right\}\right\}\\ & =\{0.01,\: 0.03,\: 0.07,\: 0.2,\: 0.5,\: 1.5,\: 4,\: 11\} \end{align*}

3.1.2 ๋น„๊ต ํ‰๊ฐ€๋ฅผ ์œ„ํ•œ ์„ ํ˜•ํšŒ๊ท€๋ถ„์„(Linear Regression Analysis, REG) ๋ชจ๋ธ

๋น„๊ต๋ฅผ ์œ„ํ•œ REG ๋ชจ๋ธ์€ ์ฃผ์–ด์ง„ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ {$y_{i},\: x_{i},\: ...,\: x_{ip}$}์— ๋Œ€ํ•ด, ์ข…์† ๋ณ€์ˆ˜ $y_{i}$์™€ p ๊ฐœ์˜ ๋…๋ฆฝ๋ณ€์ˆ˜ $x_{i}$ ์‚ฌ์ด์˜ ์„ ํ˜•๊ด€๊ณ„๋ฅผ ๋ชจ๋ธ๋ง ํ•œ๋‹ค. REG ๋ชจ๋ธ์€ ์ˆ˜์‹ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ํ˜•ํƒœ๋ฅผ ๊ฐ–๋Š”๋‹ค.

(4)
\begin{align*} y_{i}=\beta_{1}x_{i1}+\cdots +\beta_{p}x_{ip}+e_{i}=X_{i}^{T}\beta +e_{i},\:\\ \\ i=1,\:...,\:n \end{align*}

์ˆ˜์‹(4)์—์„œ $\beta_{i}$๋Š” ๊ฐ ๋…๋ฆฝ๋ณ€์ˆ˜์˜ ๊ณ„์ˆ˜์ด๋ฉฐ, $p$๋Š” ์„ ํ˜•ํšŒ๊ท€๋กœ ์ถ”์ •๋˜๋Š” ๋ชจ์ˆ˜์˜ ๊ฐœ์ˆ˜์ด๋‹ค. T๋Š” ์ „์น˜๋ฅผ ์˜๋ฏธํ•˜๋ฉฐ $X_{i}^{T}\beta$๋Š” $X_{i}$์™€ $\beta$์˜ ๋‚ด์ ์„ ์˜๋ฏธํ•œ๋‹ค. $e_{i}$๋Š” ์˜ค์ฐจํ•ญ์œผ๋กœ ๊ด€์ฐฐ๋˜์ง€ ์•Š์€ ํ™•๋ฅ ๋ณ€์ˆ˜์ด๋ฉฐ ์ข…์†๋ณ€์ˆ˜์™€ ๋…๋ฆฝ๋ณ€์ˆ˜์‚ฌ์ด์— ์˜ค์ฐจ๋ฅผ ์˜๋ฏธํ•œ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ ๋…๋ฆฝ๋ณ€์ˆ˜๋Š” ํ•ด๋‹น ์žฌ๋‚œ ๋ฐœ์ƒ ์‹œ์˜ ๋‚ ์”จ๋ณ€์ˆ˜๋“ค์„ ์‚ฌ์šฉํ•˜์˜€๊ณ  ์˜ˆ์ธก๋ณ€์ˆ˜๋Š” ๊ฐ ์„ค๋น„์— ๋Œ€ํ•œ ๊ณ ์žฅ๊ฑด์ˆ˜ ๋ฐ ์ •์ „๊ณ ๊ฐ์ˆ˜๋ฅผ ์ ์šฉํ•˜์˜€๋‹ค. ์ตœ์†Œ์ œ๊ณฑ๋ฒ•์€ ๊ฐ€์žฅ ๋‹จ์ˆœํ•˜๊ณ  ๋งŽ์ด ์“ฐ์ด๋Š” ์ถ”์ • ๋ฐฉ๋ฒ•์ด๋‹ค. ์ด๋Š” ๊ฐœ๋…์ ์œผ๋กœ ๋‹จ์ˆœํ•˜๊ณ , ๊ณ„์‚ฐ์ด ๊ฐ„๋‹จํ•˜๋‹ค. ์ตœ์†Œ์ œ๊ณฑ๋ฒ• ์ถ”์ •์€ ์ผ๋ฐ˜์ ์œผ๋กœ ์‹คํ—˜์ด๋‚˜ ๊ด€์ธก์น˜์— ์ ์šฉํ•˜๊ณ ์ž ํ•  ๋•Œ ์‚ฌ์šฉํ•œ๋‹ค. ์ตœ์†Œ์ œ๊ณฑ๋ฒ•์€ ์˜ค์ฐจ์˜ ์ œ๊ณฑ์˜ ํ•ฉ์„ ์ตœ์†Œํ™”ํ•˜๋Š” ๊ธฐ๋ฒ•์œผ๋กœ, ์ถ”์ •ํ•˜๊ณ ์ž ํ•˜๋Š” ํŒŒ๋ผ๋ฏธํ„ฐ ฮฒ์— ๋Œ€ํ•œ ํ‘œํ˜„์‹์„ ์ˆ˜์‹(5)์™€ ๊ฐ™์ด ๊ตฌํ•  ์ˆ˜ ์žˆ๋‹ค.

(5)
$\hat\beta =(X^{T}X)^{-1}X^{T}y$

3.2 ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ๋ชจ๋ธ ๊ฐœ๋ฐœ

๊ธฐ์ƒ ์ •๋ณด๋Š” ์—ฌ๋Ÿฌ ๊ธฐ์ƒ๊ด€์ธก์„ผํ„ฐ์—์„œ ์ˆ˜์ง‘๋˜๋Š” ๋ฐ˜๋ฉด ์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ์ •๋ณด๋Š” ํ•œ์ „์ง€์‚ฌ์—์„œ ๊ด€๋ฆฌํ•˜๊ณ  ์žˆ๋‹ค. ํ•˜๋‚˜์˜ ์ง€์‚ฌ ์˜์—ญ ๋‚ด์— ์—ฌ๋Ÿฌ ๊ธฐ์ƒ๊ด€์ธก์„ผํ„ฐ๊ฐ€ ์กด์žฌํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๋ชจ๋“  AWS์—์„œ ์ˆ˜์ง‘ํ•œ ๊ธฐ์ƒ ์ •๋ณด๋ฅผ ๋ฐ˜์˜ํ•  ํ•„์š”๊ฐ€ ์žˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ์ž…๋ ฅ๋ณ€์ˆ˜๋กœ ๊ธฐ์ƒ ๋ณ€์ˆ˜์˜ ํ‰๊ท ๊ฐ’, ์ตœ์†Ÿ๊ฐ’, ์ตœ๋Œ“๊ฐ’์„ ์‚ฌ์šฉํ•˜์˜€์œผ๋ฉฐ, ๊ธฐ์ƒ ๋ณ€์ˆ˜ ์‚ฌ์šฉ์— ๋”ฐ๋ผ ํ‰๊ท ๊ฐ’๋งŒ์„ ์‚ฌ์šฉํ•œ ํ‰๊ท ๋ชจ๋ธ, ์ตœ์†Œ ๋ฐ ์ตœ๋Œ“๊ฐ’๋งŒ์„ ์‚ฌ์šฉํ•œ ๊ทนํ•œ๋ชจ๋ธ, ๋ชจ๋“  ๊ธฐ์ƒ ๋ณ€์ˆ˜๋ฅผ ์‚ฌ์šฉํ•œ ํ†ตํ•ฉ๋ชจ๋ธ์„ ๊ฐ๊ฐ ๊ฐœ๋ฐœํ•˜๊ณ  ๊ฒ€์ฆํ•˜์˜€๋‹ค. ์‹คํ—˜ ๊ฒฐ๊ณผ ํ‰๊ท ๋ชจ๋ธ, ๊ทนํ•œ๋ชจ๋ธ, ํ†ตํ•ฉ๋ชจ๋ธ ์ค‘์—์„œ ๊ธฐ์ƒ ๋ณ€์ˆ˜์˜ ํ‰๊ท ๊ฐ’์„ ์ ์šฉํ•œ ํ‰๊ท ๋ชจ๋ธ์˜ ์ •ํ™•๋„๊ฐ€ ๊ฐ€์žฅ ์šฐ์ˆ˜ํ•˜์˜€๋‹ค.

๊ทธ๋ฆผ 3์€ ๊ธฐ์ƒ๊ด€์ธก์„ผํ„ฐ์™€ ์ง€์‚ฌ๊ด€๋ฆฌ ์˜์—ญ์˜ ์˜ˆ์‹œ๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ํƒ€์›์€ ๊ธฐ์ƒ์ฒญ์˜ ๊ธฐ์ƒ๊ด€์ธก์„ผํ„ฐ(AWS)๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ ๊ฒ€์ •์ƒ‰ ์ ์„ ์€ ์‚ฌ์—…์†Œ์˜์—ญ์„ ๋‚˜ํƒ€๋‚ธ๋‹ค. ํŒŒ๋ž€์ƒ‰ ์‹ค์„ ์€ ๊ธฐ์ƒ์ฒญ ๋™๋„ค์˜ˆ๋ณด ๊ตฌ์—ญ์„ ์˜๋ฏธํ•œ๋‹ค.

๊ทธ๋ฆผ. 3. ๊ธฐ์ƒ๊ด€์ธก์„ผํ„ฐ ๋ฐ ์‚ฌ์—…์†Œ ์˜์—ญ ๊ตฌ์„ฑ ์‚ฌ๋ก€

Fig. 3. A case of AWS and branch office composition

../../Resources/kiee/KIEE.2019.68.9.1085/fig3.png

๋ณธ ๋…ผ๋ฌธ์˜ ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์ง€์—ญ(์ง€์‚ฌ) ๋‹จ์œ„๋กœ ํ•™์Šต๊ณผ ์˜ˆ์ธก์„ ์ˆ˜ํ–‰ํ•œ๋‹ค. ๊ทธ๋ฆผ 3๊ณผ ๊ฐ™์ด ์„œ๋Œ€์ „์ง€์—ญ์˜ ํ•™์Šต ๋‹จ๊ณ„์—์„œ๋Š” AWS 101์—์„œ 104๊นŒ์ง€ 4๊ฐœ์˜ AWS ๋ฐ์ดํ„ฐ์™€ ์„œ๋Œ€์ „์ง€์‚ฌ์—์„œ ๋ฐœ์ƒํ•œ ์ „๋ ฅ์„ค๋น„ ํ”ผํ•ด ์ •๋ณด๋ฅผ ์ด์šฉํ•œ ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ํ™œ์šฉํ•œ๋‹ค. ์˜ˆ์ธก ๋‹จ๊ณ„์—์„œ๋Š” ๋™๋„ค์˜ˆ๋ณด ๋ฐ์ดํ„ฐ์ธ ๋„๋งˆ๋™, ๊ดด์ •๋™, ๋ณ€๋™, ๋Œ€ํฅ๋™, ๋ฌธํ™”๋™์˜ 5๊ฐœ ์˜ˆ๋ณด๋ฐ์ดํ„ฐ๋ฅผ ์ด์šฉํ•˜์—ฌ ์˜ˆ์ธก ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ํ™œ์šฉํ•œ๋‹ค. ๊ฐ™์€ ๋ฐฉ์‹์œผ๋กœ ๋Œ€๋•์œ ์„ฑ์ง€์—ญ์˜ ํ•™์Šต์—๋Š” AWS 201์—์„œ 203๊นŒ์ง€ 3๊ฐœ์˜ ๊ธฐ์ƒ๊ด€์ธก๋ฐ์ดํ„ฐ์™€ ๋Œ€๋•์œ ์„ฑ์ง€์‚ฌ์—์„œ ๋ฐœ์ƒํ•œ ์ „๋ ฅ์„ค๋น„ ํ”ผํ•ด ์ •๋ณด๋ฅผ ์ด์šฉํ•œ ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ํ™œ์šฉํ•˜๋ฉฐ, ์˜ˆ์ธก ๋‹จ๊ณ„์—์„œ๋Š” ๋™๋„ค์˜ˆ๋ณด ๋ฐ์ดํ„ฐ์ธ ์ „๋ฏผ๋™, ๋ฌธ์ง€๋™, ๊ด€ํ‰๋™์˜ 3๊ฐœ ์˜ˆ๋ณด๋ฐ์ดํ„ฐ๋ฅผ ์ด์šฉํ•˜์—ฌ ์˜ˆ์ธก ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ํ™œ์šฉํ•œ๋‹ค.

์—ฌ๊ธฐ์„œ, ์ง€์—ญ(์ง€์‚ฌ)๋‹จ์œ„๋ณด๋‹ค ํฐ ์ง€์—ญ(์ „๊ตญ ๋˜๋Š” ํŠน๋ณ„์‹œ, ๋„ ์˜์—ญ)์˜ ๋„“์€ ์ง€์—ญ์„ ๋Œ€์ƒ์œผ๋กœ ํ•˜๋Š” ๊ด‘๋ฒ”์œ„ํ•œ ์˜์—ญ์— ๋Œ€ํ•œ ํ•™์Šต๊ณผ ์˜ˆ์ธก์€ ์ง€์—ญ๋ณ„ ํŠน์„ฑ์„ ์ •ํ™•ํžˆ ๋ฐ˜์˜ํ•˜๋Š”๋ฐ ํ•œ๊ณ„๊ฐ€ ์žˆ๊ธฐ ๋•Œ๋ฌธ์— ๋‚ฎ์€ ์˜ˆ์ธก ์ •ํ™•๋„๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค. ๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” ์ด๋ฅผ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด ๋„“์€ ์ง€์—ญ์„ ์ข์€ ์ง€์—ญ์œผ๋กœ ํ•œ์ •ํ•˜๋Š” ์ง€์—ญ ๋‹จ์œ„์˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๊ฐœ๋ฐœํ•˜์˜€์œผ๋ฉฐ, ์ œ์ฃผ์ง€์—ญ ์„œ๊ท€ํฌ์ง€์‚ฌ๋ฅผ ๋Œ€์ƒ์œผ๋กœ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœยท๊ฒ€์ฆยท์ ์šฉ ๊ฒฐ๊ณผ๋ฅผ ๊ธฐ์ˆ ํ•œ๋‹ค.

์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๊ธฐ๋ณธ์ ์œผ๋กœ ANN ๋ชจ๋ธ๋กœ ๊ฐœ๋ฐœํ•˜์˜€์œผ๋ฉฐ REG ๋ชจ๋ธ๊ณผ ๋น„๊ต ๊ฒ€์ฆํ•˜์˜€๋‹ค. ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœ์„ ์œ„ํ•œ ์ฃผ์š” ๋ฐ์ดํ„ฐ๋Š” ๊ทธ๋ฆผ 4์™€ ๊ฐ™์€ ํ˜•ํƒœ๋กœ ๊ตฌ์„ฑ๋œ ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์ด๋‹ค. ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์žฌ๋‚œํŠน๋ณด ์ง€์—ญ์˜ ๋™๋„ค์˜ˆ๋ณด๋ฅผ ์ž…๋ ฅ๋ฐ์ดํ„ฐ๋กœ ๊ฐ ์ „๋ ฅ ์„ค๋น„์˜ ๊ณ ์žฅ ์ˆ˜๋ฅผ ์˜ˆ์ธกํ•œ๋‹ค.

๊ทธ๋ฆผ. 4. ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค ๊ตฌ์„ฑ๋„

Fig. 4. Unified Database Configuration

../../Resources/kiee/KIEE.2019.68.9.1085/fig4.png

์œ„ ๊ทธ๋ฆผ 4์—์„œ โ€œXโ€๋กœ ํ‘œ์‹œํ•œ ์ž…๋ ฅ๋ณ€์ˆ˜๋Š” ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์˜ ๊ธฐ์ƒ ๋ณ€์ˆ˜์ด๋‹ค. ์ด ๊ธฐ์ƒ ๋ณ€์ˆ˜๋Š” ํ‰๊ท  ์˜จ๋„, ์ตœ์†Œ ์˜จ๋„, ์ตœ๋Œ€ ์˜จ๋„, ์ตœ๋Œ€ ํ’์†, ํ‰๊ท  ํ’์† ๋ฐ ์ตœ๋Œ€ ํ’ํ–ฅ, ํ‰๊ท  ๊ฐ•์ˆ˜๋Ÿ‰, ์ตœ์†Œ ๊ฐ•์ˆ˜๋Ÿ‰, ์ตœ๋Œ€ ๊ฐ•์ˆ˜๋Ÿ‰, ํ‰๊ท  ์ ์„ค๋Ÿ‰, ์ตœ์†Œ ์ ์„ค๋Ÿ‰, ์ตœ๋Œ€ ์ ์„ค๋Ÿ‰์ด๋‹ค. ๋ฐ˜๋ฉด, โ€œYโ€๋กœ ํ‘œ์‹œํ•œ ์ถœ๋ ฅ ๋ฐ์ดํ„ฐ๋Š” ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์˜ ์ „๋ ฅ ์„ค๋น„ ๊ณ ์žฅ ์ˆ˜ ๋ฐ ์ •์ „๊ณ ๊ฐ์ˆ˜์ด๋‹ค. ๋ ˆ์ฝ”๋“œ๋Š” ์žฌ๋‚œ ํ˜„์ƒ ๋ฐœ์ƒ ์ผ์„ ๋‚˜ํƒ€๋‚ธ๋‹ค.

ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋Š” ๊ธฐ์ƒ ๋ฐ์ดํ„ฐ (๊ธฐ์ƒ์ฒญ ์ œ๊ณต), ์ „๋ ฅ ์„ค๋น„ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค ๋ฐ ๊ณ ์žฅ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋กœ ๊ตฌ์„ฑํ•œ๋‹ค. ์žฌ๋‚œ ํ˜„์ƒ์˜ ๊ฐ ์œ ํ˜•์€ ํƒœํ’, ํญ์—ผ, ํ•œํŒŒ, ํ˜ธ์šฐ, ๊ฐ•ํ’, ํญ์„ค๋กœ ๊ฐ๊ฐ์˜ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๊ฐ€ ์กด์žฌํ•œ๋‹ค. ๋”ฐ๋ผ์„œ ๊ฐ ์œ ํ˜•์˜ ์žฌ๋‚œ ํ˜„์ƒ์— ๋Œ€ํ•ด ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๊ฐœ๋ฐœํ•˜์—ฌ ๊ฐ ์ „๋ ฅ ์„ค๋น„์˜ ๊ณ ์žฅ ์ˆ˜๋ฅผ ์˜ˆ์ธกํ•œ๋‹ค. ๊ทธ๋ฆผ 5๋Š” ์ „์ œ ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ตฌ์„ฑ ๋ฐ ๋™์ž‘ ํ๋ฆ„์„ ๋ณด์—ฌ์ค€๋‹ค.

๊ทธ๋ฆผ. 5. ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋™์ž‘ ํ๋ฆ„๋„

Fig. 5. Operational flow chart of the algorithms for prediction

../../Resources/kiee/KIEE.2019.68.9.1085/fig5.png

์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก์„ ์œ„ํ•ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์—์„œ ๋ฐ์ดํ„ฐ ๊ฒ€์ƒ‰์„ ์‹œ์ž‘ํ•œ๋‹ค. ์ด๋•Œ ์ƒˆ๋กœ์šด ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ๊ฐ€ ์žˆ๋Š” ๊ฒฝ์šฐ ์ƒˆ๋กœ์šด ํ•™์Šต์„ ํ†ตํ•ด ์‹ ๊ทœ ๋ชจ๋ธ์„ ๊ฐœ๋ฐœํ•˜๊ฑฐ๋‚˜, ๊ฐœ๋ฐœ๋œ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜๋Š” ๋‘ ๋ฐฉ์‹์ค‘ ํ•˜๋‚˜๋ฅผ ์„ ํƒํ•˜์—ฌ ์ˆ˜ํ–‰ํ•œ๋‹ค. ํ•™์Šต๋œ ๋ชจ๋ธ์€ ANN ํŒŒ๋ผ๋ฏธํ„ฐ{$\Theta^{(1)}$(๊ธฐ์ƒ์š”์†Œ), $\Theta^{(2)}$(์ •์ „๊ณ ์žฅ์ˆ˜, ์ •์ „๊ณ ๊ฐ์ˆ˜)}, ํ”ผ์ณ ์Šค์ผ€์ผ๋ง ํŒŒ๋ผ๋ฏธํ„ฐ{p1(๊ธฐ์ƒ์š”์†Œ ๋ณด์ •), p2(์ •์ „ํ”ผํ•ด ๋ณด์ •)}, ๊ทธ๋ฆฌ๊ณ  ์ƒˆ๋กœ์šด ANN ์„ค์ •(๋…ธ๋“œ ์ˆ˜, ์ž„์˜ ์ดˆ๊ธฐํ™” ํŒŒ๋ผ๋ฏธํ„ฐ($\epsilon$), ์ •๊ทœํ™” ํŒŒ๋ผ๋ฏธํ„ฐ($\lambda$)๋กœ ๊ตฌ์„ฑ๋œ๋‹ค. ์„ ํ˜•ํšŒ๊ท€๋ถ„์„ ๋ชจ๋ธ๋„ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ๊ฐ๊ฐ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ์™€ ์„ค์ • ๊ฐ’์„ ๋ชจ๋ธ์— ์ €์žฅํ•œ๋‹ค.

4. ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ๋ฐ ํ‰๊ฐ€

4.1 ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ๋ฐ ์„ฑ๋Šฅ ์ตœ์ ํ™”

๊ฐœ๋ฐœํ•œ ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์„ฑ๋Šฅ ์ตœ์ ํ™”๋ฅผ ์œ„ํ•ด ์‹ค์ œ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์„ ์ˆ˜ํ–‰ํ•˜์˜€๋‹ค. ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์„ค๋น„ ๊ณ ์žฅ ์ˆ˜, ์„ค๋น„ ๊ณ ์žฅ์— ๋”ฐ๋ฅธ ์ •์ „๊ณ ๊ฐ์ˆ˜๋ฅผ ์˜ˆ์ธกํ•˜๊ธฐ ์œ„ํ•ด ์‚ฌ์šฉํ•œ๋‹ค. ์‹œ๋ฎฌ๋ ˆ์ด์…˜์„ ์œ„ํ•ด ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ, ๊ต์ฐจ๊ฒ€์ฆ ๋ฐ์ดํ„ฐ, ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ ์„ธ ๊ฐ€์ง€ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์„ ๊ตฌ์„ฑํ•œ๋‹ค. ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ(๊ต์ฐจ๊ฒ€์ฆ ๋ฐ์ดํ„ฐ ํฌํ•จ)๋Š” ์ „์ฒด ๋ฐ์ดํ„ฐ์˜ 80%, ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ๋Š” ๋‚˜๋จธ์ง€ 20%๋ฅผ ์‚ฌ์šฉํ•œ๋‹ค. ๊ฐ ๋ชจ๋ธ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋Š” ๊ฐ ์žฌ๋‚œ ์œ ํ˜• ๋ฐ ๊ฐ ์ „๋ ฅ ์„ค๋น„ ์œ ํ˜•์— ๋”ฐ๋ผ ์ตœ์ ํ™”๋œ ๊ฐ’์„ ์‚ฌ์šฉํ•œ๋‹ค. ๋˜ํ•œ, ์˜ˆ์ธก ์ •ํ™•๋„ ํ‰๊ฐ€๋ฅผ ์œ„ํ•ด ํ‰๊ท  ์ ˆ๋Œ€ ์˜ค์ฐจ(Mean Absolute Error, MAE)๋ฅผ ์‚ฌ์šฉํ•œ๋‹ค. MAE ๋ฐฉ์ •์‹์€ ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

(6)
$MAE=\dfrac{\sum_{i=1}^{m}\left | Y_{i}-H_{i}\right |}{m}$

์ˆ˜์‹(6)์—์„œ, MAE๋Š” ํ‰๊ท  ์ ˆ๋Œ€ ์˜ค์ฐจ, m์€ ๋ฐ์ดํ„ฐ ์ˆ˜, $Y_{i}$๋Š” ์‹ค์ œ ๋ฐ์ดํ„ฐ, $H_{i}$ ์˜ˆ์ธกํ•œ ์ถœ๋ ฅ์„ ๋‚˜ํƒ€๋‚ธ๋‹ค.

4.1.1 ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ๋ฐ์ดํ„ฐ

๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” 6๊ฐ€์ง€ ์žฌ๋‚œ ์ค‘ ํƒœํ’์— ๋Œ€ํ•ด์„œ ์ˆ˜ํ–‰ํ•œ ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœ ๊ฒฐ๊ณผ๋ฅผ ๊ธฐ์ˆ ํ•˜์˜€๋‹ค.

ํƒœํ’ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์€ ํƒœํ’์ด ์ผ์–ด๋‚ฌ์„ ๋•Œ์˜ ์ƒํ™ฉ์œผ๋กœ ์ •์˜ํ•œ๋‹ค. ํƒœํ’์˜ ๊ธฐ๊ฐ„์€ ์ˆ˜ ์‹œ๊ฐ„์—์„œ ์ˆ˜ ์ผ ๋™์•ˆ ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ ํƒœํ’ ์‹œ ๊ฐ€์žฅ ์ค‘์š”ํ•œ ์š”์†Œ๋Š” ํ’์†๊ณผ ๊ฐ•์ˆ˜๋Ÿ‰์ด๋‹ค. ๋‚ ์”จ๋ณ€์ˆ˜์— ๋Œ€ํ•ด์„œ๋Š” ๊ธฐ์ƒ ๋ณ€์ˆ˜์˜ ํ‰๊ท ๊ฐ’๋งŒ ์‚ฌ์šฉํ•˜๋Š” ํ‰๊ท ๋ชจ๋ธ์„ ์ ์šฉํ•˜์˜€๋‹ค.

ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์€ ํƒœํ’์ด ๋ฐœ์ƒํ–ˆ๋˜ ๊ธฐ๊ฐ„์˜ ๊ณ ์žฅ๋ฐ์ดํ„ฐ์™€ ๋‚ ์”จ๋ฐ์ดํ„ฐ๋ฅผ ํ†ตํ•ฉ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์—์„œ ๋ฐ›์•„ ์‚ฌ์šฉํ•œ๋‹ค. ๊ฐ๊ฐ์˜ ์ „๋ ฅ์ง€์‚ฌ์— ๋”ฐ๋ผ ๋ชจ๋ธ์„ ๊ฐœ๋ฐœํ•˜๋ฉฐ ํ•ด๋‹น ์ง€์‚ฌ๋Š” ๊ณ ์œ ์˜ ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋ฅผ ๊ฐ–๋Š”๋‹ค. ์˜ˆ์ธก์— ์‚ฌ์šฉํ•œ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์€ ๋‘ ๊ฐ€์ง€๋กœ ๋ถ„๋ฅ˜๋œ๋‹ค.

์ฒซ ๋ฒˆ์งธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์€ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ๋กœ ๋ชจ๋ธ ๊ฐœ๋ฐœ์— ์‚ฌ์šฉ๋œ๋‹ค. ๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” 2007๋…„๋ถ€ํ„ฐ 2017๋…„๊นŒ์ง€ ์šฐ๋ฆฌ๋‚˜๋ผ์— ์˜ํ–ฅ์„ ๋ฏธ์นœ 29๊ฐœ ํƒœํ’์ค‘์— 2007๋…„๋ถ€ํ„ฐ 2016๋…„ ์ค‘๋ฐ˜๊นŒ์ง€์˜ 28๊ฐœ๋ฅผ ๋ฐ์ดํ„ฐ๋ฅผ ํ•™์Šต์— ์‚ฌ์šฉํ•˜์˜€๋‹ค. ๋‘ ๋ฒˆ์งธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์€ ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์œผ๋กœ ๋ชจ๋ธ์„ ๊ฒ€์ฆํ•˜๊ธฐ ์œ„ํ•ด ์‚ฌ์šฉํ•œ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์ œ์ฃผ์‹œ ์„œ๊ท€ํฌ์ง€์‚ฌ๋ฅผ ๋Œ€์ƒ์œผ๋กœ 2016๋…„ 10์›”์˜ ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™ ๋ฐ์ดํ„ฐ๋ฅผ ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์œผ๋กœ ํ™œ์šฉํ•˜์˜€๋‹ค.

ํ‘œ 5์™€ ๊ฐ™์ด ๋ณธ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์—์„œ ๋ชจ๋ธ ๊ฐœ๋ฐœ์— ์‚ฌ์šฉํ•˜๋Š” ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ๋Š” 255๊ฐœ์ด๋‹ค. ๊ต์ฐจ๊ฒ€์ฆ(Cross Validation) ๋ฐ์ดํ„ฐ๋Š” 2015-07-11 15:00 ~ 2017-07-04 15:00 ์ด๋ฉฐ 63๊ฐœ์ด๋‹ค. ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ๋Š” ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™(2016-10-04 15:00 ~ 2016-10-05 15:00)์— ๋Œ€ํ•œ 9๊ฐœ๋กœ์„œ ์˜ˆ์ธก ์ •ํ™•๋„ ํ‰๊ฐ€์— ์‚ฌ์šฉํ•˜์˜€๋‹ค. ๋ชจ๋ธ์€ ๋ฐฐ์ „์‹œ์Šคํ…œ์˜ ๊ฒฝ์šฐ ์ „์ฃผ, ์• ์ž, ์ „์„ , ํ”ผ๋ขฐ๊ธฐ, CoS, ๋ณ€์••๊ธฐ์˜ ์ด 6๊ฐ€์ง€ ์„ค๋น„์— ๋Œ€ํ•ด ๊ฐœ๋ฐœํ•˜์˜€๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์‹ค์ œ๋กœ ํƒœํ’ ๊ธฐ๊ฐ„์— ๋ชจ๋“  ์„ค๋น„๊ฐ€ ๊ณ ์žฅ๋‚˜๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ฉฐ ๊ณ ์žฅ ๋ฐœ์ƒ์ด ๊ฑฐ์˜ ์—†๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์•„ ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™ ๋ฐœ์ƒ ๊ธฐ๊ฐ„ ๋™์•ˆ์— ๊ณ ์žฅ์ด ๋ฐœ์ƒํ•œ ์„ค๋น„์ธ ์ „์„ ๊ณผ CoS์˜ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ๊ฒฐ๊ณผ๋งŒ์„ ์ฒจ๋ถ€ํ•˜์˜€๋‹ค.

ํ‘œ 5. ์„œ๊ท€ํฌ ์ง€์‚ฌ ๊ธฐ๊ณ„ํ•™์Šต ๋ฐ์ดํ„ฐ ์ˆ˜

Table 5. Number of machine learning data for Seogwipo branch offics

์„œ๊ท€ํฌ์ง€์‚ฌ

๋ฐ์ดํ„ฐ ์ˆ˜

ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ(Train Data)

255

๊ต์ฐจ๊ฒ€์ฆ ๋ฐ์ดํ„ฐ(CV Data)

63

ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ(Test Data)

9

4.1.2 ANN ๋ชจ๋ธ ๊ฒ€์ฆ๊ฒฐ๊ณผ

๊ทธ๋ฆผ 6์€ ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™๋กœ ์ธํ•ด ๋ฐœ์ƒํ•œ ์ „์„  ๊ณ ์žฅ๊ฑด์ˆ˜์™€ ์ „์„  ๊ณ ์žฅ์œผ๋กœ ์ธํ•ด ๋ฐœ์ƒํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜์— ๋Œ€ํ•œ ANN๊ณผ REG์„ ์ ์šฉํ•œ ๋ชจ๋ธ ๊ฒ€์ฆ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ๊ทธ๋ฆผ 6์˜ ์™ผ์ชฝ ๊ทธ๋ž˜ํ”„๋Š” ์ „์„  ๊ณ ์žฅ๊ฑด์ˆ˜์— ๋Œ€ํ•œ ์‹ค์ œ๊ณ ์žฅ(์‹ค์„ ), ANN ์˜ˆ์ธก๊ฒฐ๊ณผ(๋Œ€์‰ฌ ์ ์„ ), ๊ทธ๋ฆฌ๊ณ  REG ์˜ˆ์ธก๊ฒฐ๊ณผ(์ ์„ )๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ์˜ค๋ฅธ์ชฝ ๊ทธ๋ž˜ํ”„๋Š” ์ „์„  ๊ณ ์žฅ์œผ๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜์— ๋Œ€ํ•œ ์˜ˆ์ธก ๊ฒฐ๊ณผ๋ฅผ ์‹ค์ œ ์ •์ „๊ณ ๊ฐ์ˆ˜, ANN, REG ๋ชจ๋ธ๋กœ ์˜ˆ์ธกํ•œ ๊ฒฐ๊ณผ๋ฅผ ๋™์‹œ์— ๋ณด์—ฌ์ค€๋‹ค.

๊ทธ๋ฆผ. 6-1. ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™๋กœ ์ธํ•œ ์ „์„ ํ”ผํ•ด ์˜ˆ์ธก ๊ฒ€์ฆ๊ฒฐ๊ณผ

Fig. 6-1. The analysis of line damage caused by Typhoon Chaba

../../Resources/kiee/KIEE.2019.68.9.1085/fig6_1.png

๊ทธ๋ฆผ. 6-2. ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™๋กœ ์ธํ•œ ์ „์„ ๊ณ ์žฅ ์ •์ „๊ณ ๊ฐ์ˆ˜ ์˜ˆ์ธก ๊ฒ€์ฆ๊ฒฐ๊ณผ

Fig. 6-2. The analysis of the # of households(caused by line damage)

../../Resources/kiee/KIEE.2019.68.9.1085/fig6_2.png

โ€˜์ฐจ๋ฐ”โ€™๋กœ ์ธํ•ด ์‹ค์ œ ๋ฐœ์ƒํ•œ ์ „์„  ๊ณ ์žฅ๊ณผ ๊ทธ๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜์— ๋Œ€ํ•˜์—ฌ ANN, REG ๋ชจ๋ธ ๋ชจ๋‘ ์‹ค์ œ์™€ ์œ ์‚ฌํ•œ ์˜ˆ์ธก์„ ์ˆ˜ํ–‰ํ•˜๋Š” ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ๋‹ค. ANN ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ ์˜ˆ์ธกํ•œ ๊ฒฐ๊ณผ์น˜๊ฐ€ ์‹ค์ œ์™€ ๋น„์Šทํ•œ ํŒจํ„ด์„ ๊ทธ๋ฆฌ๋Š” ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ๋‹ค. ๊ฐ ๊ทธ๋ž˜ํ”„์—์„œ โ€˜xโ€™์ถ•์€ ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™๊ฐ€ ์˜ํ–ฅ์„ ๋ฏธ์น˜๋Š” ์‹œ์ž‘์‹œ๊ฐ„(2016-10-04 15:00) ๋ถ€ํ„ฐ ์ข…๋ฃŒ์‹œ๊ฐ„(2016-10-05 15:00) ๊นŒ์ง€๋ฅผ 3์‹œ๊ฐ„ ๊ฐ„๊ฒฉ์œผ๋กœ ๋‚˜ํƒ€๋‚ด๊ณ  ์žˆ์œผ๋ฉฐ, ์ขŒใƒป์šฐ์ธก ๊ทธ๋ž˜ํ”„์˜ โ€˜yโ€™์ถ•์€ ๊ฐ๊ฐ ์ „์„  ๊ณ ์žฅ๊ฑด์ˆ˜(๊ฑด)์™€ ์ „์„ ๊ณ ์žฅ์œผ๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค.

๊ทธ๋ฆผ. 7-1. ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™๋กœ ์ธํ•œ CoS ํ”ผํ•ด ์˜ˆ์ธก ๊ฒ€์ฆ๊ฒฐ๊ณผ

Fig. 7-1. The analysis of CoS damage caused by Typhoon Chaba

../../Resources/kiee/KIEE.2019.68.9.1085/fig7_1.png

๊ทธ๋ฆผ. 7-2. ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™๋กœ ์ธํ•œ CoS ๊ณ ์žฅ ์ •์ „๊ณ ๊ฐ์ˆ˜ ์˜ˆ์ธก ๊ฒ€์ฆ๊ฒฐ๊ณผ

Fig. 7-2. The analysis of the # of households(caused by CoS damage)

../../Resources/kiee/KIEE.2019.68.9.1085/fig7_2.png

๊ทธ๋ฆผ 7์€ ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™๋กœ ๋ฐœ์ƒํ•œ CoS ๊ณ ์žฅ๊ฑด์ˆ˜์™€ CoS ๊ณ ์žฅ์œผ๋กœ ์ธํ•ด ๋ฐœ์ƒํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜์— ๋Œ€ํ•œ ANN๊ณผ REG์„ ์ ์šฉํ•œ ๋ชจ๋ธ ๊ฒ€์ฆ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ๊ทธ๋ฆผ 7์˜ ์™ผ์ชฝ ๊ทธ๋ž˜ํ”„๋Š” CoS ๊ณ ์žฅ๊ฑด์ˆ˜์— ๋Œ€ํ•œ ์‹ค์ œ๊ณ ์žฅ(์‹ค์„ ), ANN ์˜ˆ์ธก๊ฒฐ๊ณผ(๋Œ€์‰ฌ ์ ์„ ), ๊ทธ๋ฆฌ๊ณ  REG ์˜ˆ์ธก๊ฒฐ๊ณผ(์ ์„ )๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ์˜ค๋ฅธ์ชฝ ๊ทธ๋ž˜ํ”„๋Š” CoS ๊ณ ์žฅ์œผ๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜์— ๋Œ€ํ•œ ์˜ˆ์ธก ๊ฒฐ๊ณผ๋ฅผ ์‹ค์ œ ์ •์ „๊ณ ๊ฐ์ˆ˜, ANN, REG ๋ชจ๋ธ๋กœ ์˜ˆ์ธกํ•œ ๊ฒฐ๊ณผ๋ฅผ ๋™์‹œ์— ๋ณด์—ฌ์ค€๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ •์ „๊ณ ๊ฐ์˜ ๊ฒฝ์šฐ์—๋Š” ๋ฐœ์ƒ ์‹œ๊ฐ„์„ ์ •ํ™•ํžˆ ์˜ˆ์ธกํ•˜์ง€๋Š” ๋ชปํ•˜์˜€๋‹ค. ์ •์ „๊ณ ๊ฐ์ˆ˜ ๋ฐœ์ƒ ํŒจํ„ด์„ ์‹œ๊ฐ„์ถ•์„ ๊ธฐ์ค€์œผ๋กœ ์ด๋™ํ•˜๋ฉด ์˜ˆ์ธก์น˜๊ฐ€ ์‹ค์ œ ๊ฐ’์„ ๋”ฐ๋ผ๊ฐ€๋Š” ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ์œผ๋ฉฐ ์˜ˆ์ธกํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜๊ฐ€ ์‹ค์ œ์™€ ๋น„์Šทํ•œ ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ๋‹ค. ์ด๋Ÿฌํ•œ ์›์ธ์€ ์ „๋ ฅ์„ค๋น„ ์ค‘ CoS ๊ณ ์žฅ๋ฐœ์ƒ์˜ ๋นˆ๋„๊ฐ€ ์ „์„ ์˜ ๊ณ ์žฅ๋ฐœ์ƒ ๋นˆ๋„์— ๋น„ํ•ด ๋‚ฎ๊ฒŒ ๋‚˜ํƒ€๋‚˜๊ณ , ์ƒ๋Œ€์ ์œผ๋กœ ์ ์€ ์ˆ˜์˜ ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ๋กœ ๊ธฐ๊ณ„ํ•™์Šต์„ ์ˆ˜ํ–‰ํ•˜์˜€๊ธฐ ๋•Œ๋ฌธ์œผ๋กœ ๋ถ„์„ํ•˜์˜€๋‹ค.

4.1.3 MAE ํ™œ์šฉ ์˜ˆ์ธก ์ •ํ™•๋„ ํ‰๊ฐ€

ํ‘œ 6์€ ANN ๋ชจ๋ธ๊ณผ REG ๋ชจ๋ธ์„ ์ ์šฉํ•œ ์ „์„ , COS์— ๊ด€ํ•œ MAE ๊ฐ’์„ ์š”์•ฝํ•œ ๊ฒƒ์ด๋‹ค. MAE๋Š” ๊ทธ ๊ฐ’์ด 0์— ๊ฐ€๊นŒ์šธ ์ˆ˜๋ก ๋ชจ๋ธ์˜ ์ •ํ™•๋„๊ฐ€ ์ข‹์Œ์„ ๋‚˜ํƒ€๋‚ธ๋‹ค. ์ด๋Š” ๋ชจ๋ธ์ด ๊ณ ์žฅ์ธ ๊ฒฝ์šฐ์™€ ๊ทธ๋ ‡์ง€ ์•Š์€ ๊ฒฝ์šฐ๋ฅผ ์ž˜ ๊ตฌ๋ถ„ํ•จ์„ ๋ณด์—ฌ์ค€๋‹ค. ์„œ๊ท€ํฌ์ง€์‚ฌ์˜ 2016๋…„ ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™์— ๋Œ€ํ•ด์„œ ํ‰๊ฐ€ํ•œ ๊ฒฐ๊ณผ๋กœ๋Š” ์ „์„  ํ”ผํ•ด์™€ ์ „์„  ๊ณ ์žฅ์œผ๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜ ๋ถ„์•ผ๋Š” ANN ๋ชจ๋ธ์ด REG ๋ชจ๋ธ์— ๋น„ํ•ด ๋‹ค์†Œ ์˜ค์ฐจ์œจ์ด ์ ์—ˆ์œผ๋ฉฐ, CoS ํ”ผํ•ด๊ฑด์ˆ˜๋Š” REG ๋ชจ๋ธ์ด ANN ๋ชจ๋ธ์— ๋น„ํ•ด ์šฐ์ˆ˜ํ•œ ๊ฒƒ์œผ๋กœ ํ‰๊ฐ€๋˜์—ˆ์œผ๋‚˜, CoS๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜์— ๋Œ€ํ•œ ์˜ˆ์ธก์ •ํ™•๋„๋Š” ANN์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ๋ณด๋‹ค ์šฐ์ˆ˜ํ•œ ๊ฒƒ์œผ๋กœ ํ‰๊ฐ€๋˜์—ˆ๋‹ค. ๋‹ค๋งŒ, ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ์™€ ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ ๋ชจ๋‘ ์ „๋ฐ˜์ ์œผ๋กœ CoS์˜ ํ”ผํ•ด๊ฑด์ˆ˜๊ฐ€ ์ „์„ ์˜ ํ”ผํ•ด๊ฑด์ˆ˜๋ณด๋‹ค ์ ์—ˆ๊ธฐ ๋•Œ๋ฌธ์— ํ•™์Šต ์ž…๋ ฅ๋ฐ์ดํ„ฐ๊ฐ€ ๋งŽ์€ ์ „์„ ์— ๋Œ€ํ•œ ํ‰๊ฐ€ ๊ฒฐ๊ณผ๊ฐ€ CoS์— ๋Œ€ํ•œ ํ‰๊ฐ€ ๊ฒฐ๊ณผ๋ณด๋‹ค ์ •ํ™•๋„๊ฐ€ ์ข€ ๋” ์šฐ์ˆ˜ํ•˜๋‹ค๊ณ  ๋ถ„์„ํ•˜์˜€๋‹ค.

ํ‘œ 6. MAE ํ™œ์šฉ ๋ชจ๋ธ ์ •ํ™•๋„ ํ‰๊ฐ€ ๊ฒฐ๊ณผ

Table 6. Results of accuracy evaluation using MAE

๊ตฌ๋ถ„

MAE

์ „์„ ๊ณ ์žฅ

์ •์ „๊ณ ๊ฐ์ˆ˜ (์ „์„ ๊ณ ์žฅ)

CoS๊ณ ์žฅ

์ •์ „๊ณ ๊ฐ์ˆ˜ (CoS๊ณ ์žฅ)

REG

0.190

631.2222

0.0476

49.2857

ANN

0.127

429.0476

0.0476

28.4286

4.2 ํƒœํ’ ์†”๋ฆญ์— ๋Œ€ํ•œ ํ”ผํ•ด ์˜ˆ์ธก ๊ฒฐ๊ณผ

2018๋…„ ํƒœํ’ โ€˜์†”๋ฆญโ€™์ด ์„œ๊ท€ํฌ ์ง€์‚ฌ์— ๋‚ด์Šตํ•œ ๊ธฐ๊ฐ„์€ 2018-08-22 15:00 ~ 2018-08-24 12:00์ด๋ฉฐ, ๋‚ด์Šต ๊ธฐ๊ฐ„๋™์•ˆ ๋™๋„ค ์˜ˆ๋ณด ๋ฐœํšจ ๊ฑด์ˆ˜๋Š” 3์‹œ๊ฐ„ ๋‹จ์œ„์˜ 16๊ฐœ์ด๋ฉฐ, ์ด 16๊ฐœ์˜ ๋™๋„ค ์˜ˆ๋ณด๋ฅผ ์ž…๋ ฅ์œผ๋กœ ์‹ค์ œ ํƒœํ’ ๋‚ด์Šต ์ด์ „์— ์ˆ˜ํ–‰๋œ ์˜ˆ์ธก๊ณผ ์‹ค์ œ ๊ณ ์žฅ๊ฒฐ๊ณผ๋ฅผ ๋น„๊ตํ•˜์˜€๋‹ค. โ€˜์†”๋ฆญโ€™์œผ๋กœ ์ธํ•ด ์„œ๊ท€ํฌ์ง€์‚ฌ์—์„œ๋Š” ์ด 16๊ฐœ์˜ ์˜ˆ์ธก ๊ธฐ๊ฐ„ ๋™์•ˆ 9๊ฐœ์˜ ๊ธฐ๊ฐ„์—์„œ ๊ณ ์žฅ์ด ๋ฐœ์ƒํ•˜์˜€์œผ๋ฉฐ, ์ „์„  ์ด์™ธ์˜ ๋‹ค๋ฅธ ์„ค๋น„ํ”ผํ•ด๋Š” ๋ฐœ์ƒํ•˜์ง€ ์•Š์•˜๋‹ค.

๊ทธ๋ฆผ. 8. ์ „์„ ์„ ๋Œ€์ƒ์œผ๋กœ ํ•œ ํ”ผํ•ด ์˜ˆ์ธก ๊ฒฐ๊ณผ(2018๋…„ ํƒœํ’โ€˜์†”๋ฆญโ€™)

Fig. 8. Prediction Result for the line damage (2018 Typhoon 'SOULIC')

../../Resources/kiee/KIEE.2019.68.9.1085/fig8.png

ANN๊ณผ REG ๋ชจ๋ธ ์ ์šฉ ๊ฒฐ๊ณผ๋Š” ๊ทธ๋ฆผ 8๊ณผ ๊ฐ™๋‹ค. ๊ฐ ๊ทธ๋ž˜ํ”„์—์„œ โ€˜xโ€™์ถ•์€ ํƒœํ’ โ€˜์†”๋ฆญโ€™์ด ์˜ํ–ฅ์„ ๋ฏธ์น˜๋Š” ์‹œ์ž‘์‹œ๊ฐ„(2018-08-22 15:00) ๋ถ€ํ„ฐ ์ข…๋ฃŒ์‹œ๊ฐ„(2018-08-24 12:00) ๊นŒ์ง€๋ฅผ 3์‹œ๊ฐ„ ๊ฐ„๊ฒฉ์œผ๋กœ ๋‚˜ํƒ€๋‚ด๊ณ  ์žˆ์œผ๋ฉฐ, โ€˜yโ€™์ถ•์€ ์ „์„  ๊ณ ์žฅ๊ฑด์ˆ˜(๊ฑด)์™€ ์ „์„ ๊ณ ์žฅ์œผ๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค. ์˜ˆ์ธก์—์„œ ๊ณ ์žฅ์ด ๋ฐœ์ƒ(๋ฏธ๋ฐœ์ƒ)ํ•œ๋‹ค๊ณ  ํŒ์ •ํ•œ ๊ฒฝ์šฐ์™€ ์‹ค์ œ ๊ณ ์žฅ์ด ๋ฐœ์ƒ(๋ฏธ๋ฐœ์ƒ)ํ•œ ๊ฒฝ์šฐ ๊ทธ ๊ฒฐ๊ณผ๋ฅผ ์„ฑ๊ณต์œผ๋กœ ํŒ์ •ํ•˜์˜€๋‹ค. ํ‘œ 7๊ณผ ๊ฐ™์ด ANN ๋ชจ๋ธ์€ 16๊ฐœ ์‹œ๊ฐ„๋Œ€์— ๋Œ€ํ•˜์—ฌ 9๊ฐœ ์‹œ๊ฐ„๋Œ€์˜ ๊ณ ์žฅ ๋ฐœ์ƒ ์—ฌ๋ถ€๋ฅผ ํŒ๋ณ„ํ•˜๋Š”๋ฐ ์„ฑ๊ณตํ•˜์˜€๋‹ค. MAE๋ฅผ ์ ์šฉํ•œ ์ •ํ™•๋„๋Š” 1.25๋กœ ์ธก์ •๋˜์—ˆ๋‹ค. REG ๋ชจ๋ธ๋„ ANN ๋ชจ๋ธ๊ณผ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ 16๊ฐœ ์‹œ๊ฐ„๋Œ€์— ๋Œ€ํ•˜์—ฌ 9๊ฐœ ์‹œ๊ฐ„๋Œ€์˜ ๊ณ ์žฅ ๋ฐœ์ƒ ์—ฌ๋ถ€๋ฅผ ํŒ๋ณ„ํ•˜๋Š”๋ฐ ์„ฑ๊ณตํ•˜์˜€์œผ๋ฉฐ, MAE๋ฅผ ์ ์šฉํ•œ ์˜ˆ์ธก ์ •ํ™•๋„๋Š” 1.75๋กœ ์ธก์ •๋˜์—ˆ๋‹ค.

ํ‘œ 7. ์‹œ๊ฐ„๋Œ€๋ณ„ ์‹ค์ œ ๊ณ ์žฅ ๋ฐ ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ณ ์žฅ ์˜ˆ์ธก ๊ฒฐ๊ณผ(2018๋…„ ํƒœํ’โ€˜์†”๋ฆญโ€™)

Table 7. Results of Comparision actual failure with prediction (2018 Typhoon 'SOULIC')

์‹œ๊ฐ„๋Œ€

์‹ค์ œ

REG

ANN

์˜ˆ์ธก

๊ฒฐ๊ณผ

์˜ˆ์ธก

๊ฒฐ๊ณผ

2018-08-22 15:00

0

3

0

2

0

2018-08-22 18:00

1

3

์„ฑ๊ณต

2

์„ฑ๊ณต

2018-08-22 21:00

2

6

์„ฑ๊ณต

3

์„ฑ๊ณต

2018-08-23 0:00

0

6

0

3

0

2018-08-23 3:00

5

7

์„ฑ๊ณต

3

์„ฑ๊ณต

2018-08-23 6:00

6

5

์„ฑ๊ณต

2

์„ฑ๊ณต

2018-08-23 9:00

2

4

์„ฑ๊ณต

2

์„ฑ๊ณต

2018-08-23 12:00

3

5

์„ฑ๊ณต

2

์„ฑ๊ณต

2018-08-23 15:00

2

2

์„ฑ๊ณต

1

์„ฑ๊ณต

2018-08-23 18:00

2

2

์„ฑ๊ณต

1

์„ฑ๊ณต

2018-08-23 21:00

0

2

0

1

0

2018-08-24 0:00

0

1

0

1

0

2018-08-24 3:00

0

1

0

1

0

2018-08-24 6:00

0

1

0

1

0

2018-08-24 9:00

0

1

0

0

์„ฑ๊ณต

2018-08-24 12:00

0

0

์„ฑ๊ณต

0

์„ฑ๊ณต

ํ‘œ 8. ํ”ผํ•ด ์˜ˆ์ธก ์ •ํ™•๋„ ๋น„๊ต(2018๋…„ ํƒœํ’โ€˜์†”๋ฆญโ€™)

Table 8. Comparison of the predicted damage MAE(2018 Typhoon 'SOULIC')

๊ตฌ๋ถ„

์„ฑ๊ณต์ˆ˜(๊ฑด)/์ „์ฒด์ˆ˜(๊ฑด)

MAE(๊ณ ์žฅ๊ฑด์ˆ˜)

REG

9/16

1.75

ANN

9/16

1.25

๋‘ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋ชจ๋‘ ๋™์ผํ•œ ์ž…๋ ฅ์ธ ์„œ๊ท€ํฌ์ง€์‚ฌ์˜ ๋™๋„ค์˜ˆ๋ณด ๊ธฐ์ƒ ์š”์†Œ(๊ฐ•์šฐ๋Ÿ‰, ํ’์†)๋ฅผ ์ž…๋ ฅ์œผ๋กœ ์˜ˆ์ธก์„ ์ˆ˜ํ–‰ํ•˜์˜€๋‹ค. ๋‘ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋ชจ๋‘ ๊ณ ์žฅ๋ฐœ์ƒ์— ๋Œ€ํ•œ ํŒ๋ณ„์€ 9๊ฑด์”ฉ์œผ๋กœ ๊ฐ™์€ ๊ฒฐ๊ณผ๋ฅผ ๋‚˜ํƒ€๋ƒˆ์œผ๋ฉฐ, ์˜ˆ์ธก ์ •ํ™•๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋Š” MAE ๊ฐ’์— ์žˆ์–ด์„œ๋Š” ANN ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด REG ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋น„ํ•ด ์šฐ์ˆ˜ํ•จ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์—ˆ๋‹ค.

5. ๊ฒฐ ๋ก 

๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” ANN์„ ์ด์šฉํ•˜์—ฌ ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ๋ชจ๋ธ์„ ๊ฐœ๋ฐœํ•˜๊ณ  REG ๊ธฐ๋ฒ•๊ณผ ๋น„๊ตํ•˜์—ฌ ์ œ์•ˆ๋œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์ •ํ™•๋„๋ฅผ ํ‰๊ฐ€ํ•˜์˜€๋‹ค. ํ”ผ์–ด์Šจ ์ƒ๊ด€๊ด€๊ณ„ ๋ถ„์„ ๊ธฐ๋ฒ•์œผ๋กœ ์˜ˆ์ธก ๋ชจ๋ธ ์•Œ๊ณ ๋ฆฌ์ฆ˜์—์„œ ์‚ฌ์šฉ๋  ์ž…๋ ฅ ๋ณ€์ˆ˜๋ฅผ ์„ ์ •ํ•˜์˜€์œผ๋ฉฐ, ๊ต์ฐจ ๊ฒ€์ฆ ๋ฐฉ๋ฒ•์„ ํ™œ์šฉํ•˜์—ฌ ํžˆ๋“  ๋ ˆ์ด์–ด ์ˆ˜, ์ดˆ๊ธฐํ™” ํŒŒ๋ผ๋ฏธํ„ฐ ๋ฐ ์ •๊ทœํ™” ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ์ตœ์ ํ™” ํ•˜์˜€์œผ๋ฉฐ, ์ด๋•Œ ์ตœ์  ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ์ฐพ๋Š” ๋ฐฉ๋ฒ•์œผ๋กœ๋Š” gradient descent ๋ฐฉ๋ฒ•์„ ์ ์šฉํ•˜์˜€๋‹ค.

์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฒ€์ฆ์— ์‚ฌ์šฉํ•˜๋Š” ๋ฐ์ดํ„ฐ๋Š” ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋ฅผ ๊ตฌ์ถ•ํ•˜์—ฌ ์‚ฌ์šฉํ•œ๋‹ค. ์ด ํ†ตํ•ฉ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋Š” ์žฌ๋‚œ ํŠน๋ณด์— ํ•ด๋‹นํ•˜๋Š” ๊ธฐ์ƒ์กฐ๊ฑด์„ ์ถ”์ถœํ•˜์—ฌ ๊ตฌ์ถ•ํ•œ๋‹ค. ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœ ๋ฐ ๊ฒ€์ฆ์„ ์œ„ํ•˜์—ฌ ๋ฐ์ดํ„ฐ๋ฅผ ๋‘ ๊ฐ€์ง€ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์œผ๋กœ ๊ตฌ๋ถ„ํ•œ๋‹ค. ์ฒซ ๋ฒˆ์งธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์€ ์žฌ๋‚œํ”ผํ•ด ์˜ˆ์ธก ๋ชจ๋ธ์„ ๊ฐœ๋ฐœํ•˜๋Š”๋ฐ ์‚ฌ์šฉํ•˜๋Š” ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ๊ณผ ๊ต์ฐจ๊ฒ€์ฆ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์ด๋ฉฐ, ๋‘ ๋ฒˆ์งธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์€ ์˜ˆ์ธก๊ฒฐ๊ณผ๋ฅผ ๊ฒ€์ฆํ•˜๊ธฐ ์œ„ํ•œ ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์ด๋‹ค. ์ •ํ™•๋„ ๊ฒ€์ฆ์€ ๊ฐœ๋ฐœ ๋ชจ๋ธ์„ ์ ์šฉํ•˜์—ฌ ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์˜ ์„ค๋น„๊ณ ์žฅ ๊ฑด์ˆ˜์™€ ๋น„๊ต๋ฅผ ํ†ตํ•˜์—ฌ ๊ฒ€์ฆํ•˜์˜€์œผ๋ฉฐ, MAE๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์˜ˆ์ธก ์ •ํ™•๋„๋ฅผ ํ‰๊ฐ€ํ•˜์˜€๋‹ค. ๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” 6๊ฐœ ์žฌ๋‚œ ์œ ํ˜•์— ๋Œ€ํ•œ ์˜ˆ์ธก ๋ชจ๋ธ์„ ์„ค๊ณ„ํ•˜๋Š” ๊ณผ์ •์„ ๊ธฐ์ˆ ํ•˜์˜€์œผ๋ฉฐ, ์ œ์ฃผ ์„œ๊ท€ํฌ์ง€์—ญ์˜ ํƒœํ’ ์žฌ๋‚œ์„ ๋Œ€์ƒ์œผ๋กœ ๊ฐœ๋ฐœ๋œ ANN ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ REG ๋ฐฉ์‹๊ณผ ๋น„๊ต ๋ถ„์„ํ•˜์˜€๋‹ค. 2016๋…„ ํƒœํ’ โ€˜์ฐจ๋ฐ”โ€™์— ๋Œ€ํ•œ ์ „์„  ๊ณ ์žฅ๋ฐœ์ƒ ์˜ˆ์ธก์˜ ๊ฒฝ์šฐ ANN์˜ MAE๊ฐ€ 0.127๋กœ REG์˜ MAE์ธ 0.190 ๋ณด๋‹ค ๋‚ฎ์€ ๊ฐ’์„ ๊ฐ€์ง€๋ฏ€๋กœ ANN์ด ๋ณด๋‹ค ์ •ํ™•ํ•˜๊ฒŒ ์˜ˆ์ธกํ•จ์„ ํ™•์ธํ•˜์˜€์œผ๋ฉฐ, CoS ๊ณ ์žฅ์œผ๋กœ ์ธํ•œ ์ •์ „๊ณ ๊ฐ์ˆ˜์˜ ๊ฒฝ์šฐ์—๋„ ์—ญ์‹œ ANN์˜ MAE๊ฐ€ ๋‚ฎ์€ ๊ฐ’์„ ๋ณด์˜€๋‹ค. ๋˜ํ•œ, 2018๋…„ ํƒœํ’ โ€˜์†”๋ฆญโ€™์„ ๋Œ€์ƒ์œผ๋กœ ์ˆ˜ํ–‰ํ•œ ์‹ค์ œ ์˜ˆ์ธก์ˆ˜ํ–‰ ๊ฒฐ๊ณผ๋„ ANN์ด ๊ณ ์žฅ๊ฑด์ˆ˜์˜ MAE๊ฐ€ 1.25๋กœ REG 1.75 ๋ณด๋‹ค ๋‚ฎ์Œ์„ ํ™•์ธํ•˜์˜€๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ANN์˜ ์˜ˆ์ธก ์ •ํ™•๋„๊ฐ€ REG ๋ฐฉ์‹๋ณด๋‹ค ์šฐ์ˆ˜ํ•จ์„ ํ™•์ธํ•˜์˜€๋‹ค.

ํ–ฅํ›„ ์ „๋ ฅ์„ค๋น„ ๊ณ ์žฅ ์œ ๋ฐœ ์š”์ธ์ธ ์„ค๋น„์ œ์›, ๋…ธํ›„๋„ ๋“ฑ ์„ค๋น„์˜ ์ถ”๊ฐ€์  ํŠน์„ฑ์„ ๊ณ ๋ คํ•˜๊ณ , ๋‚ ๋กœ ๋ฐœ์ „ํ•˜๋Š” ์ธ๊ณต์ง€๋Šฅ ๊ธฐ์ˆ ์„ ์ถ”๊ฐ€์ ์œผ๋กœ ์ ‘๋ชฉํ•จ์œผ๋กœ์จ ์˜ˆ์ธก ์ •ํ™•๋„๊ฐ€ ๋ณด๋‹ค ์šฐ์ˆ˜ํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœ์„ ์œ„ํ•œ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•˜๋‹ค.

References

1 
Executive Office of the President, 2013, Council of Economic Advisers, Economic Benefits of Increasing Electric Grid Resilience to Weather Outages, The CouncilGoogle Search
2 
D. T. Ton, W.-T. P. Wang, April 2015, A more resilient grid: The US department of energy joins with stakeholders in an R&D plan, IEEE Power and Energy Magazine, Vol. 13, No. 3, pp. 26-34DOI
3 
NERC, Jan. 2014, Hurricane Sandy Event Analysis ReportGoogle Search
4 
http://rain-project.eu/, RAIN Project Consortium, European UnionGoogle Search
5 
Hongfei Li, Sep. 2010, http://researcher.watson.ibm.com/files/us-egaioni/HongfeiLi-IBMworkshop.pptGoogle Search
6 
D. W. Wanik, 2012, Weather-Based Damage Prediction Models for Electric Distribution Networks, Masterโ€™s Thesis, University of ConnecticutGoogle Search

์ €์ž์†Œ๊ฐœ

์ตœ๋ฏผํฌ(Min-Hee Choi)
../../Resources/kiee/KIEE.2019.68.9.1085/au1.png

He received his M.S. Degree from University of Seoul.

He has worked in Korea Electric Power Research Institute(KEPRI) as a researcher for 10 years.

His research interests include SW Engineering for power system stability and power quality.

์ •๋‚จ์ค€(Nam-Joon Jung)
../../Resources/kiee/KIEE.2019.68.9.1085/au2.png

He received his Ph.D. from Hanbat University.

He has worked in Korea Electric Power Research Institute (KEPRI) as a researcher for 24 years.

He is presently a Chief Researcher of Software Platform Laboratory.

His research interests include Smart Grid and SW Engineering.

์ด๊ทœ์ฒ (Kyu-Chul Lee)
../../Resources/kiee/KIEE.2019.68.9.1085/au3.png

He received B.S. Degree from Korea University.

He has worked in Korea Electric Power Corporation (KEPCO) for 36 years as a General Manager of Safety and Security Department.

His research interests include Disaster Management and HSSE.

์ •์žฌ์„ฑ(Jae-Sung Jeong)
../../Resources/kiee/KIEE.2019.68.9.1085/au4.png

He received B.S. degree in electrical engineering from Chungnam National University, Korea; M.S. degree in electrical engineering from North Carolina State University, Raleigh, NC; Ph.D. degree in electrical engineering from Virginia Tech, Blacksburg, VA.

He is currently a faculty member in the Department of Energy Systems Research at Ajou University, Korea.

His research interests include the development and deployment of renewable and sustainable energy technologies.

์„œ์ธ์šฉ(In-Yong Seo)
../../Resources/kiee/KIEE.2019.68.9.1085/au5.png

He received the B.S. and M.S. degree in Electrical Engineering from Sungkyunkwan University and Busan National University in 1984 and 1989, respectively.

He received his Ph.D. in Electrical Engineering from Brown University, United States, in 2003.

He is a vice president of Korea Electric Power Research Institute(KEPRI).

His current research interests are digital solutions in energy new industry and modeling of Complex System.