๊ด€๋ฆฌ ๋ฉ”๋‰ด

Done is Better Than Perfect

[๋…ผ๋ฌธ ๋ฆฌ๋ทฐ] ๐Ÿค–๐Ÿ”Œ ๊ฐ•ํ™”ํ•™์Šต์„ ์‚ฌ์šฉํ•œ ์—ดํ™” ์ตœ์†Œํ™” ๊ธ‰์† ์ถฉ์ „ ํ”„๋กœํŒŒ์ผ ๋ณธ๋ฌธ

๐Ÿ”‹ ์ด์ฐจ ์ „์ง€/๋…ผ๋ฌธ ๋ฆฌ๋ทฐ

[๋…ผ๋ฌธ ๋ฆฌ๋ทฐ] ๐Ÿค–๐Ÿ”Œ ๊ฐ•ํ™”ํ•™์Šต์„ ์‚ฌ์šฉํ•œ ์—ดํ™” ์ตœ์†Œํ™” ๊ธ‰์† ์ถฉ์ „ ํ”„๋กœํŒŒ์ผ

jimingee 2025. 2. 23. 17:43

๐Ÿ”—  MSCC-DRL: Multi-Stage constant current based on deep reinforcement learning for fast charging of lithium ion battery

  • Journal of Energy Storage, 2024

 

์ „๊ธฐ์ฐจ(EV) ์‹œ์žฅ์˜ ๊ธ‰์„ฑ์žฅ๊ณผ ํ•จ๊ป˜ ๊ทนํ•œ ๊ธ‰์† ์ถฉ์ „(Extreme Fast Charging, XFC)์— ๋Œ€ํ•œ ํ•„์š”์„ฑ์ด ์ ์  ๋” ์ปค์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ, ๋ฐฐํ„ฐ๋ฆฌ๋ฅผ 10~15๋ถ„ ๋‚ด์— 80% SOC(State of Charge)๊นŒ์ง€ ์ถฉ์ „ํ•˜๋Š” ๊ฒƒ์ด ์—…๊ณ„์˜ ์ฃผ์š” ๋ชฉํ‘œ๋กœ ์„ค์ •๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

 

ํ•˜์ง€๋งŒ ๊ธ‰์† ์ถฉ์ „์„ ์ง„ํ–‰ํ•  ๊ฒฝ์šฐ, ๋ฐฐํ„ฐ๋ฆฌ ๋‚ด๋ถ€ ์ƒํƒœ๊ฐ€ ๊ธ‰๊ฒฉํ•˜๊ฒŒ ๋ณ€๋™ํ•˜๋ฉด์„œ ๋ฐฐํ„ฐ๋ฆฌ์˜ ์—ดํ™”๋„ ๋น ๋ฅด๊ฒŒ ์ง„ํ–‰๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ, ๋ฐฐํ„ฐ๋ฆฌ์˜ ์—ดํ™”๋ฅผ ์ตœ์†Œํ™”ํ•˜๋ฉด์„œ๋„ ๋น ๋ฅด๊ฒŒ ์ถฉ์ „ํ•  ์ˆ˜ ์žˆ๋Š” ์ถฉ์ „ ํ”„๋กœํŒŒ์ผ์„ ์ฐพ์•„๋‚ด๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ๊ฐœ๋ฐœ์ด ์ค‘์š”ํ•œ ๊ณผ์ œ๋กœ ๋– ์˜ค๋ฅด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

 

โœก๏ธ Problem Statement

์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ

๋ฆฌํŠฌ ์ด์˜จ ๋ฐฐํ„ฐ๋ฆฌ๋Š” ์ „๊ธฐ์ฐจ(EV) ๋ฐ ๋‹ค์–‘ํ•œ ์—๋„ˆ์ง€ ์ €์žฅ ์‹œ์Šคํ…œ์—์„œ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋น ๋ฅธ ์ถฉ์ „ ๊ณผ์ •์—์„œ ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐฐํ„ฐ๋ฆฌ ์—ดํ™” ๋ฐ ์•ˆ์ „ ๋ฌธ์ œ๋Š” ์—ฌ์ „ํžˆ ํ•ด๊ฒฐํ•ด์•ผ ํ•  ์ค‘์š”ํ•œ ๊ณผ์ œ์ž…๋‹ˆ๋‹ค. ๋ณธ ์—ฐ๊ตฌ๋Š” ๊ฐ•ํ™” ํ•™์Šต(Deep Reinforcement Learning, DRL) ๊ธฐ๋ฐ˜์˜ ๋‹ค๋‹จ๊ณ„ ์ •์ „๋ฅ˜ ์ถฉ์ „(Multi-Stage Constant Current, MSCC) ์ตœ์ ํ™” ๊ธฐ๋ฒ•์„ ์ œ์•ˆํ•˜์—ฌ, ์ถฉ์ „ ์†๋„๋ฅผ ํ–ฅ์ƒ์‹œํ‚ค๋ฉด์„œ๋„ ๋ฐฐํ„ฐ๋ฆฌ์˜ ์•ˆ์ „์„ฑ์„ ์œ ์ง€ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์—ฐ๊ตฌํ•ฉ๋‹ˆ๋‹ค.

์—ฐ๊ตฌ ๋ชฉํ‘œ

๋ณธ ์—ฐ๊ตฌ๋Š” ๊ฐ•ํ™” ํ•™์Šต์„ ํ™œ์šฉํ•˜์—ฌ ์ถฉ์ „ ์‹œ๊ฐ„์„ ๋‹จ์ถ•ํ•˜๋ฉด์„œ๋„ ๋ฐฐํ„ฐ๋ฆฌ์˜ ์—ดํ™” ๋ฐ ์•ˆ์ „์„ฑ์„ ์œ ์ง€ํ•  ์ˆ˜ ์žˆ๋Š” ์ตœ์  ์ถฉ์ „ ํ”„๋กœํŒŒ์ผ์„ ํ•™์Šตํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ํŠนํžˆ, ๊ธฐ์กด์˜ CC-CV ๋ฐฉ์‹๊ณผ ๋น„๊ตํ•˜์—ฌ ๋‹ค๋‹จ๊ณ„ ์ •์ „๋ฅ˜ ์ถฉ์ „(MSCC) ๋ฐฉ์‹์ด ์–ด๋–ป๊ฒŒ ๊ฐ•ํ™” ํ•™์Šต์„ ํ†ตํ•ด ์ตœ์ ํ™”๋  ์ˆ˜ ์žˆ๋Š”์ง€ ๋ถ„์„ํ•ฉ๋‹ˆ๋‹ค.

๊ธฐ์กด ์ถฉ์ „ ๋ฐฉ์‹

1. CC-CV (Constant Current-Constant Voltage) ๋ฐฉ์‹

: CC-CV ๋ฐฉ์‹์€ ์ดˆ๊ธฐ ์ •์ „๋ฅ˜(CC) ์ถฉ์ „ ํ›„, ํŠน์ • ์ „์••์— ๋„๋‹ฌํ•˜๋ฉด ์ •์ „์••(CV) ๋ชจ๋“œ๋กœ ์ „ํ™˜๋˜๋Š” ์ „ํ†ต์ ์ธ ์ถฉ์ „ ๋ฐฉ์‹์ž…๋‹ˆ๋‹ค.

๋‹จ์ 

  • ์ถฉ์ „ ์‹œ๊ฐ„ ์ฆ๊ฐ€: CV ๋‹จ๊ณ„์—์„œ ์ถฉ์ „ ์ „๋ฅ˜๊ฐ€ ๊ฐ์†Œํ•˜์—ฌ ์ „์ฒด ์ถฉ์ „ ์‹œ๊ฐ„์ด ๊ธธ์–ด์ง
  • ๋ฐฐํ„ฐ๋ฆฌ ์—ดํ™” ์œ„ํ—˜: ๋†’์€ ์ถฉ์ „์œจ์—์„œ ๋ฆฌํŠฌ ๋„๊ธˆ(Lithium Plating)์ด ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ์Œ
  • ์˜จ๋„ ์ƒ์Šน ๋ฌธ์ œ: ๋†’์€ C-rate์—์„œ ๊ณผ์—ด ๊ฐ€๋Šฅ์„ฑ ์กด์žฌ
  • ์ถฉ์ „ ์ตœ์ ํ™” ์–ด๋ ค์›€: ๋ฐฐํ„ฐ๋ฆฌ ์ƒํƒœ์— ๋”ฐ๋ฅธ ์ตœ์ ํ™”๋œ ์ถฉ์ „ ํŒจํ„ด์„ ์ ์šฉํ•˜๊ธฐ ์–ด๋ ค์›€

 

2. ๋‹ค๋‹จ๊ณ„ ์ •์ „๋ฅ˜ ์ถฉ์ „(MSCC) ๋ฐฉ์‹

: ๋‹ค๋‹จ๊ณ„ ์ •์ „๋ฅ˜ ์ถฉ์ „์€ ์—ฌ๋Ÿฌ ๋‹จ๊ณ„๋กœ ๋‚˜๋ˆ„์–ด C-rate๋ฅผ ์ ์ง„์ ์œผ๋กœ ๊ฐ์†Œ์‹œํ‚ค๋ฉฐ ์ถฉ์ „ํ•˜๋Š” ๋ฐฉ์‹์ž…๋‹ˆ๋‹ค.

์žฅ์ 

  • ์ถฉ์ „ ์‹œ๊ฐ„ ๋‹จ์ถ•: CV ๋‹จ๊ณ„ ์—†์ด ์ถฉ์ „์ด ๊ฐ€๋Šฅ
  • ๋ฐฐํ„ฐ๋ฆฌ ์—ดํ™” ๊ฐ์†Œ: ์ดˆ๊ธฐ ๋†’์€ C-rate์—์„œ ์ ์ง„์ ์œผ๋กœ ๊ฐ์†Œ → ๋ฐœ์—ด ๋ฐ ์—ดํ™”๋ฅผ ๋ฐฉ์ง€
  • ๊ฐ•ํ™” ํ•™์Šต ์ ์šฉ ๊ฐ€๋Šฅ: ๋ฐฐํ„ฐ๋ฆฌ ๊ตฌ์กฐ ํŠน์„ฑ์— ๋งž๋Š” ์ตœ์  ์ถฉ์ „ ํ”„๋กœํŒŒ์ผ์„ ํ•™์Šตํ•  ์ˆ˜ ์žˆ์Œ

 

โœก๏ธ  Methodology

๊ฐ•ํ™” ํ•™์Šต ๊ธฐ๋ฐ˜ ์ตœ์  ์ถฉ์ „ ๋ฌธ์ œ ์ •์˜

  • ๋ชฉํ‘œ: ์ถฉ์ „ ์†๋„๋ฅผ ์ตœ๋Œ€ํ™”ํ•˜๋ฉด์„œ ๋ฐฐํ„ฐ๋ฆฌ์˜ ์•ˆ์ „์„ฑ์„ ์œ ์ง€ํ•˜๋Š” ์ตœ์  ์ถฉ์ „ ์ „๋žต ํ•™์Šต
  • ์ œ์•ฝ ์กฐ๊ฑด:
    • ์ „์•• ์ดˆ๊ณผ ๊ธˆ์ง€: $V_{max} \leq 4.2V$
    • ์˜จ๋„ ์ดˆ๊ณผ ๊ธˆ์ง€: $T_{max} \leq 70^\circ C$
    • ๋ชฉํ‘œ SOC ๋„๋‹ฌ: $SOC_{target} = 80%$

๐Ÿค– ๊ฐ•ํ™” ํ•™์Šต ์š”์†Œ ์ •์˜

State SOC, ์ „์•• $V$, ์˜จ๋„ $T$
Action ์ถฉ์ „ ์ „๋ฅ˜ $C_r$ (C-rate, 1C-7C), ์ง€์† ์‹œ๊ฐ„ $\Delta t$ (1-5๋ถ„)
Reward ์ถฉ์ „ ์†๋„ ์ฆ๊ฐ€ ๋ณด์ƒ, ๋ฐฐํ„ฐ๋ฆฌ ์•ˆ์ „ ์œ ์ง€ ๋ณด์ƒ

 

 

โœก๏ธ  Experiment & Analysis

๐Ÿ”‹ ์‹คํ—˜ ํ™˜๊ฒฝ ๋ฐ ๋ฐฐํ„ฐ๋ฆฌ ๋ชจ๋ธ

  • ๋ฐฐํ„ฐ๋ฆฌ ๋ชจ๋ธ: PyBaMM์˜ SPMe(Single Particle Model with Electrolyte)
  • ๊ฐ•ํ™” ํ•™์Šต ์•Œ๊ณ ๋ฆฌ์ฆ˜: PPO (Proximal Policy Optimization)
  • ์ด ํ›ˆ๋ จ ์‹œ๊ฐ„: 336์‹œ๊ฐ„ (10,000 ์—ํ”ผ์†Œ๋“œ)

์‹คํ—˜ ๊ฒฐ๊ณผ ๋ฐ ๋ถ„์„

  • ์ถฉ์ „ ์‹œ๊ฐ„ ๋‹จ์ถ• ํšจ๊ณผ:
    • CC-CV ๋ฐฉ์‹: 25๋ถ„
    • DRL ๊ธฐ๋ฐ˜ MSCC ๋ฐฉ์‹: 14๋ถ„ (๋‘๊บผ์šด ์ „๊ทน), 6๋ถ„ (์–‡์€ ์ „๊ทน)
  • ๋ฐฐํ„ฐ๋ฆฌ ์•ˆ์ „์„ฑ: ์˜จ๋„ ๋ฐ ์ „์•• ์ดˆ๊ณผ ์—†์ด ๋ฐฐํ„ฐ๋ฆฌ ์•ˆ์ „์„ฑ ์œ ์ง€๋จ
  • ํ•™์Šต ๊ฒฐ๊ณผ: ์ถฉ์ „ ์‹œ๊ฐ„์ด ์•ˆ์ •์ ์œผ๋กœ ์ˆ˜๋ ดํ•˜๋ฉฐ ํ•™์Šต์ด ์„ฑ๊ณต์ ์œผ๋กœ ์™„๋ฃŒ๋จ

๋น„๊ต ์‹คํ—˜

  • ์ถฉ์ „ ์‹œ๊ฐ„: ๊ธฐ์กด CC-CV ๋ฐฉ์‹ ๋Œ€๋น„ ์ถฉ์ „ ์‹œ๊ฐ„์ด ์ตœ๋Œ€ 30% ๋‹จ์ถ•๋จ
  • ๋ฐฐํ„ฐ๋ฆฌ ์ˆ˜๋ช…: ์˜จ๋„ ์ƒ์Šน์ด ๋‚ฎ์•„ ๋ฐฐํ„ฐ๋ฆฌ ์ˆ˜๋ช…์ด ์—ฐ์žฅ๋  ๊ฐ€๋Šฅ์„ฑ ์žˆ์Œ
  • ํ•™์Šต๋œ ์ตœ์  ์ถฉ์ „ ํ”„๋กœํŒŒ์ผ: 6C → 4C → 3C, ๊ฐ ๋‹จ๊ณ„ 5๋ถ„ ๋ฏธ๋งŒ ์œ ์ง€

 

โœก๏ธ Conclusion

๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” DRL์„ ์ด์šฉํ•œ ๋‹ค๋‹จ๊ณ„ ์ •์ „๋ฅ˜ ์ถฉ์ „ ๋ฐฉ์‹์„ ์ œ์•ˆํ•˜์˜€์œผ๋ฉฐ, ๊ธฐ์กด ๋ฐฉ์‹ ๋Œ€๋น„ ์ถฉ์ „ ์‹œ๊ฐ„ ๋‹จ์ถ•๊ณผ ๋ฐฐํ„ฐ๋ฆฌ ์•ˆ์ „์„ฑ ์œ ์ง€์—์„œ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค.

ํ–ฅํ›„ ์—ฐ๊ตฌ ๋ฐฉํ–ฅ

  • ์‹ค์ œ ๋ฐฐํ„ฐ๋ฆฌ ํ™˜๊ฒฝ์—์„œ์˜ ์‹คํ—˜ ๊ฒ€์ฆ
  • ๋ฐฐํ„ฐ๋ฆฌ ์—ดํ™” ๋ชจ๋ธ์„ ์ถ”๊ฐ€ํ•œ ๊ฐ•ํ™” ํ•™์Šต ํ™˜๊ฒฝ ๊ตฌ์ถ•
  • ๋‹ค์–‘ํ•œ ์ตœ์ ํ™” ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ ๋น„๊ต ์—ฐ๊ตฌ ์ง„ํ–‰
  • ํ•˜๋“œ์›จ์–ด ๊ตฌํ˜„ ๋ฐ ์‹ค์‹œ๊ฐ„ ํ•™์Šต ๊ฐ€๋Šฅ์„ฑ ํ‰๊ฐ€
  • ๋‹ค์–‘ํ•œ ๋ฐฐํ„ฐ๋ฆฌ ๋ชจ๋ธ ๋ฐ ์ „๊ทน ๊ตฌ์กฐ์— ๋Œ€ํ•œ ์ผ๋ฐ˜ํ™” ์—ฐ๊ตฌ
 

 


 

OpenAI์˜ GYM ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด ๊ฐ•ํ™”ํ•™์Šต ํ™˜๊ฒฝ์„ ๊ฐ„๋‹จํ•˜๊ฒŒ ๊ตฌํ˜„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ๋ฐฐํ„ฐ๋ฆฌ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ์ธ PyBaMM์˜ ๋ฌธ์„œ๋„ ์ž˜ ์ •๋ฆฌ๋˜์–ด ์žˆ์œผ๋‹ˆ, ๊ด€์‹ฌ์ด ์žˆ๋Š” ๋ถ„๋“ค์€ ์ด๋ฅผ ์ฐธ๊ณ ํ•˜์—ฌ ๊ตฌํ˜„ํ•ด๋ณด๋Š” ๊ฒƒ๋„ ์ข‹์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค. (์ € ์—ญ์‹œ GYM๊ณผ PyBaMM์„ ํ™œ์šฉํ•ด ๊ตฌํ˜„ํ•  ๊ณ„ํš์ž…๋‹ˆ๋‹ค.๐Ÿ™‚๐Ÿ‘

๊ฐ•ํ™”ํ•™์Šต์„ ์ด์šฉํ•œ ์ตœ์  ๋ฐฐํ„ฐ๋ฆฌ ๊ณ ์†์ถฉ์ „ ํ”„๋กœํŒŒ์ผ์„ ๋‹ค๋ฃฌ ๋…ผ๋ฌธ์€ ๋งŽ์ด ์žˆ์ง€๋งŒ, ์ด MSCC-DRL ๋…ผ๋ฌธ์€ ์‹คํ—˜ ๋ถ€๋ถ„์ด ํŠนํžˆ ์ƒ์„ธํ•˜๊ฒŒ ์ •์˜๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ž˜์„œ ์ €์™€ ๊ฐ™์ด '๊ฐ•ํ™”ํ•™์Šต์„ ํ™œ์šฉํ•œ ์ตœ์  ์ถฉ์ „ ํ”„๋กœํŒŒ์ผ'์„ ๊ตฌํ˜„ํ•˜๋ ค๋Š” ๋ถ„๋“ค์—๊ฒŒ ์ด ๋…ผ๋ฌธ์„ ๋ฐ”ํƒ•์œผ๋กœ ์‹คํ—˜์„ ์žฌํ˜„ํ•ด๋ณด๋Š” ๊ฒƒ์„ ๊ฐ•๋ ฅํžˆ ์ถ”์ฒœํ•ฉ๋‹ˆ๋‹ค! ๐Ÿ˜ƒ

 

Comments