λ°μ’
μ
(Jong-young Park)
β iD
μ΄νλ―Ό
(Hanmin Lee)
1
μ‘°κ·μ
(Gyu-Jung Cho)
1
μ νΈμ±
(Hosung Jung)
1
νλ¬Έμ
(Moonseob Han)
1
-
(Electrification System Research Division, Korea Railroad Research Institute, Korea)
Copyright Β© The Korean Institute of Electrical Engineers(KIEE)
Key words
High-resistance ground fault, Deep learning, CNN, Distribution system, Protection system
1. μ λ‘
λ°°μ κ³ν΅ λ΄ κ³ μ ν μ¬κ³ (High Resistance Fault, HRF)λ λ°°μ μ μ΄ μ΄λ ν μμΈ λλ¬Έμ μκ°, λͺ¨λ, μλͺ© λ± μ νμ΄ ν° λ¬Όμ§μ
μ μ΄νμ¬ λ°μνλ μ¬κ³ μ΄λ©°, λ¨μ μ§λ½μ¬κ³ κ° λ°°μ κ³ν΅ μ¬κ³ μ 70% μ΄μμ μ°¨μ§νλ€. κ³ν΅μ μ μμ μ΄μμ μνμ¬ κ³ μ ν μ¬κ³ λ λ€λ₯Έ μ¬κ³ μ²λΌ μ λ’°μ±
μλ κ²μΆ λ° μ κ±°κ° νμνλ, κ³ μ ν μ¬κ³ λ μ¬κ³ μ λ₯ μ¦κ°λμ΄ λΆν μ λ₯μ λΉν΄ ν¬μ§ μμΌλ©°, μν¬λ‘ λΆνμ μ μ¬ν νΉμ±μ 보μ¬μ μ¬κ³ νμ μ΄ μ΄λ ΅λ€(1). μ΄μ κ°μ μ΄μ λ‘ μΌλ°μ μΈ κ³Όμ λ₯ κ³μ κΈ° λ±μΌλ‘ μ¬κ³ κ²μΆνλ κ²μ΄ μ΄λ €μ°λ©°, κ²μΆμ μν΄ κ³ μ ν μ¬κ³ μ λΉμ ν, μλ³μ μΈ νΉμ±μ νμ©ν κΈ°λ²λ€μ΄
μ°κ΅¬λμ΄μλ€. νΉν λ°°μ λ§μ λΆμ°μ μμ μ μ©μ΄ λμ΄λκ³ μλλ°, λΆμ°μ μμ΄ μλ λ°°μ κ³ν΅μμ κ³ μ₯ λ°μμ΄ λ°μνλ©΄ κ³ μ₯ μμΉμ λΆμ°μ μ μμΉμ κ΄κ³μ
λ°λΌ μ λ₯ λ°©ν₯ λ± κ³ μ₯μ λ₯μ μμμ΄ λ¬λΌμ§λ€(2).
λ°°μ κ³ν΅ νΌλμ κ³ μ₯μ μ ννκ³ μ μνκ² κ°μ§λκ³ μ κ±°λμ΄μΌ κ³ μ₯μ νμ°μ λ°©μ§ν μ μλ€(3,4). λ¨μ μ§λ½μ¬κ³ μ λν μ¬λ¬ κ°μ§ 보νΈκΈ°λ²λ€μ΄ μ μλμ΄ μμΌλ©°(4), ν¬κ² μΈ κ°μ§λ‘ λΆλ₯ν μ μλλ° 1) μ μμν μ νΈ κΈ°λ° λΆμ, 2) κ³Όλμν μ νΈ κΈ°λ° λΆμ, 3) ITμ΅ν© κΈ°μ κΈ°λ° λΆμμ΄ μλ€. μ΄ μ€
μμ λ κ°μ§ λ°©λ²μ κ³ν΅ ꡬ쑰λ μ¬κ³ κ° λ³΅μ‘ν μν©μ λν΄ νκ³κ° μλ€. ITκΈ°μ μ΄ λ°μ ν¨μ λ°λΌ λ΄λ΄ λ€νΈμν¬(5), μν¬νΈ λ²‘ν° λ¨Έμ (6), μ μ μκ³ λ¦¬μ¦, μ λ¬Έκ° μμ€ν
, νΌμ§ μ΄λ‘ (7) λ±μ΄ μ μ©λμ΄ μλ€. Farshadμ Sadeh(8)λ k-μ΅κ·Όμ μκ³ λ¦¬μ¦μ μ΄μ©ν νκ·λΆμμ μ μ©νμ¬ λ¨μ μ§λ½μ¬κ³ μ μμΉλ₯Ό νμ νμλ€. λ
Όλ¬Έ (9)μμλ μ¨μ΄λΈλ¦Ώ λ³νκ³Ό λ² μ΄μμ κΈ°λ²μ μ΄μ©νμ¬ κ³ μ₯ νΌλλ₯Ό νλ³νμκ³ , (10)μμλ μΈκ³΅μ κ²½λ§(Artificial Neural Network, ANN)μ μ΄μ©νμ¬ μ¬κ³ νΌλμ 건μ νΌλλ₯Ό νλ³νμλ€. νμ§λ§ μ΄λ¬ν λ°©λ²μ κ³ μ₯μΌλ‘
λ°μνλ μ νΈμ λν μ μ ν μ νΈμ²λ¦¬ λ°©λ²μ΄ νμνλ€. μ΅κ·Όμλ μ¨μ΄λΈλ¦Ώ λ³νκ³Ό λ¨Έμ λ¬λμ ν΅ν ν¨ν΄ λΆλ₯λ₯Ό κ²°ν©ν λ°©λ²μ΄ μ μλκ³ μλ€. (11)μμλ μ΄μ° μ¨μ΄λΈλ¦Ώ λ³νκ³Ό λΉμμ νλ ¬ λΆν΄λ₯Ό μμ± λΆμμ μ μ©νμκ³ , (12)μμλ μ§ν₯μ± μ¨μ΄λΈλ¦Ώ λ³νμ μ΄μ©ν μ¬μΈ΅ CNN(Convolutional Neural Network, ν©μ±κ³± μ κ²½λ§)μ κΈ°λ° μ μ λ Xμ μ»΄ν¨ν°
λ¨μΈ΅μ΄¬μ(CT)μ μ μνμλ€. μ΄λ¬ν μ¨μ΄λΈλ¦Ώ λ³νκ³Ό CNNμ μ μ©νμ¬ λ°°μ λ§μμ κ³ μ₯ νΌλ νλ³μ ν΅ν΄ κ³ μ₯ μμΉλ₯Ό νλ³νκ±°λ(13), νλ² λ₯΄νΈ-ν© λ³ν(Hilbert-Huang Transform)κ³Ό CNNμ μ μ©νμ¬ λ°°μ λ§ μ¬κ³ μ’
λ₯λ₯Ό νλ³νλ λ°©λ²μ΄ μ μλμλ€(14).
λ³Έ λ
Όλ¬Έμμλ λΆμ°μ μμ΄ μ€μΉλ λ°°μ κ³ν΅μμ κ³ μ νλ¨μ μ§λ½μ¬κ³ κ° λ°μνμμ λ κ·Έκ²μ νλ³νλ λ°©λ²μ μ μνκ³ μ νλ€. μ μ¬μλ°μ λ± λΆμ°μ μ μ¦κ°λ
μ λ ₯λ³νμ₯μΉ λμ
λ±μΌλ‘ μΈν΄ λ°°μ κ³ν΅μ΄ λ 볡μ‘ν΄μ Έ μ¬κ³ λ°μμ κ³ μ₯μ λ₯μ μμμ΄ λ³΅μ‘ν΄μ§κ³ μμΌλ©°, μ΄μ λ°λΌ κ³ μ₯μ νΈ μ²λ¦¬μ μ μ ν λ°©λ²μ μ°ΎκΈ°κ°
μ΄λ €μμ§κ³ μλ€. λ°°μ κ³ν΅μμ μμ£Ό λ°μνλ κ³ μ ν λ¨μ μ§λ½μ¬κ³ λ κ·Έ νΉμ± λλ¬Έμ κ·Έ νλ³μ΄ λμ± μ½μ§ μλ€. μ΄μ λ°λΌ λ€λ₯Έ λΆμΌμμλ λ§μ΄ μ μ©λκ³
μλ μ¨μ΄λΈλ¦Ώ λ³νκ³Ό CNNμ νμ©νμ¬ κ³ μ ν μ§λ½μ¬κ³ λ₯Ό νλ³νλ λ°©λ²μ μ μνκ³ μ νλ€. μ΄λ₯Ό ν΅νμ¬ λ€μν λ°°μ κ³ν΅μ ꡬ쑰λ μ¬κ³ μν©μ λ°λΌ
λ¬λΌμ§λ κ³ μ ν μ§λ½μ¬κ³ λ₯Ό νλ³ν μ μλ€.
2. κ³ μ₯μν© λ°μ΄ν° μμ±
κ³ μ ν μ¬κ³ κ²μΆ μκ³ λ¦¬μ¦μ κ°λ°νκΈ° μν΄μ κ³ν΅μμ μ€μ λ‘ λ°μν κ³ μ₯μ΄λ μ€νμ ν΅νμ¬ κ΅¬ν κ³ μ ν μ¬κ³ μ μ μκ³Ό μ λ₯λ₯Ό λΆμν΄μλ€. μ΄λ¬ν λ°μ΄ν°λ₯Ό
λΆμν λ λμΌν μ μ΄λ¬Όμ§μ μν κ³ μ ν μ¬κ³ λΌ νλλΌλ μ¬κ³ μμ μ λΆνμ©λμ΄λ κΈ°ν κ³ν΅μν©μ μν΄ κ³μ°μ μμμ μ μ, μ λ₯κ° λ¬λΌμ§λ€(1). λ³Έ λ
Όλ¬Έμμλ μ λ ₯κ³ν΅ λͺ¨μλ₯Ό ν΅νμ¬ μ¬κ³ μ λ°μ΄ν°μ μΌλ° μ΄μ μμ λ°μ΄ν°λ₯Ό μμ±νμΌλ―λ‘ κ³ μ ν μ¬κ³ μ μ μ, μ λ₯ νΉμ±μ μ λνλΌ μ μλ
μ λ ₯κ³ν΅ λͺ¨λΈλ§μ΄ νμνλ€. λ°°μ κ³ν΅μ μ€κ³ν΅μ κΈ°λ°ν λͺ¨λΈμ λΆμ°μ μμ΄ μ€μΉλμλ€κ³ κ°μ νμ¬ κ³ν΅μ λͺ¨λΈλ§νμλ€. κ³ μ ν μ¬κ³ μ νΉμ±μΌλ‘λ μ¦κ°νμ,
λ©μΆ€νμ, λΉμ νμ±, λΉλμΉμ± λ±μ΄ μλ€. (15), (16) 보νΈκ³μ κΈ°μμλ νλ μΈμ΄ν΄ λ΄μ μ¬κ³ λ₯Ό κ²μΆν΄μΌ νλ―λ‘ μ§§μ μκ°μ λνλλ νΉμ±μΈ λΉμ νμ±, λΉλμΉμ±μ λ°μν λͺ¨λΈμ ꡬμ±νμλ€.
2.1 λ°°μ κ³ν΅ λͺ¨λΈλ§
κ·Έλ¦Ό 1 λΆμ°μ μμ΄ μλ λ°°μ κ³ν΅ λͺ¨λΈλ§
Fig. 1 Distribution system model with a distribution generation
λ°°μ κ³ν΅μμ λ°μνλ κ³ μ ν μ¬κ³ μ λ°μ΄ν°λ₯Ό λ§λ€κΈ° μν΄μ λͺ¨λΈ μ λ ₯κ³ν΅μ ꡬμ±νκ³ μ¬κ³ μν©μ ꡬμ±νμλ€. λμ μ λ ₯κ³ν΅μ λΆμ°μ μμ΄ μλ λ°°μ κ³ν΅μμ
κ³ μ ν μ¬κ³ κ° λ°μνλ μν©μ μμ νμλ€. κ·Έλ¦¬κ³ λΉκ΅λ₯Ό μνμ¬ λΆνλμ΄ μ¦κ°νμ¬ μ λ₯κ° μ»€μ§λ μν©μ μμ νμλ€.
κ·Έλ¦Ό 1μ λ°μ΄ν° μμ±μ μν΄ μ¬μ©ν λͺ¨λΈ λ°°μ κ³ν΅μ ꡬμ±λμ΄λ€. λ³μ μμμ λκ°λ νΌλ ν κ°λ₯Ό λͺ¨λΈλ§νμμΌλ©°, νΌλ λ΄ μ΄ λΆνλμ 10MW μ΄λ©° μλ₯ μ
95%λ‘ νμλ€. κ·Έλ¦¬κ³ λΆμ°μ μμ΄ μλ λ°°μ κ³ν΅μ΄λ©°, μ μ¬μ λΆμ°μ μμ μ΅λ λ°μ μ©λμ 8MWλ‘ κ°μ νμλ€. λ°μ΄ν° μμ± μ μ¬λ¬ κ°μ§ κ²½μ°λ₯Ό μμ νκΈ°
μν΄, λΆμ°μ μμ λ°μ λμ΄ 8, 4.5, 1MWμΈ κ°κ°μ κ²½μ°μ λν΄ μ¬κ³ κ° λ°μν μν©μ λͺ¨μνμλ€.
μ λ‘ λͺ¨λΈμ μ€κ³ν΅ λ°μ΄ν°μ μ λ‘κΈΈμ΄, μ μ’
μ κ³ λ €νμ¬ λͺ¨λΈλ§νμμΌλ©°, PI λ±κ°λͺ¨λΈλ‘ ꡬννμμΌλ©°, λΆνλ μ μ λ ₯κ³Ό μ μνΌλμ€ λΆνλ‘ κ΅¬μ±νμλ€.
μ μ νΌλμΈ‘μ μ μ§λ Y-Ygλ‘ νμμΌλ©°, λΆμ°μ μ μΈ‘μ Yg-β³λ‘ νμλ€.
κ·Έλ¦Ό 2 μ μ νΌλμΈ‘ μ μ§ λͺ¨λΈλ§
Fig. 2 Grounding modeling of generation side
κ·Έλ¦Ό 3 λΆμ°μ μμΈ‘ μ μ§ λͺ¨λΈλ§
Fig. 3 Grouding modeling of distributed generation side
2.2 κ³ μ ν μ¬κ³ λͺ¨λΈλ§
λΆμ°μ μμ΄ μλ λ°°μ κ³ν΅μμ λ°μνλ κ³ μ ν μ§λ½μ¬κ³ λ₯Ό λΆμνκΈ° μνμ¬ μ¬κ³ λ°μ΄ν° λ° κ·Έκ²κ³Ό λΉκ΅νκΈ° μν μ μ λ°μ΄ν°λ₯Ό μμ±νμλ€. κ³ μ ν μ§λ½μ¬κ³ λ₯Ό
λΆμμ μν λ°μ΄ν°λ₯Ό ꡬμΆνλ λ° μ€μν κ²μ κ·Έ νΉμ±μ κ°λ₯ν μ ννκ² λ°μνλ κ²μ΄λ€. κ·Έλ¦Ό 4λ κ° μκ°μμ κ³ μ ν μ¬κ³ μ€νμ ν΅ν΄ μ»μ μ λ₯ ννμ 보μ¬μ£Όκ³ μμΌλ©°, μ¦κ° νμκ³Ό λ©μΆ€ νμμ λ³Ό μ μμΌλ©°, λ€λ₯Έ μμΈμ μν κ³ μ ν μ§λ½μ¬κ³ λ
λΉμ·ν νΉμ±μ νλλΈλ€. κ·Έλ¦Ό 5λ μ¬κ³ λ°μ ν 20μ£ΌκΈ°μ 40μ£ΌκΈ°μμ μ λ₯ ννμ λΉκ΅ν΄μ 보μ¬μ£Όκ³ μμΌλ©°, λΉμ νμ±κ³Ό λΉλμΉμ±μ΄ μ¬κ³ λ°μ ν λͺ¨λ μκ° μμμμ λνλκ³ μλ€(1).
κ³ μ‘°ν κ²μΆμ ν΅νμ¬ κ³ μ ν μ§λ½μ¬κ³ λ₯Ό νλ³νκΈ° μν΄ μ€μν νΉμ±μ μ΄λ¬ν λΉμ νμ±μ΄λ―λ‘ μ΄λ¬ν νΉμ±μ λͺ¨λΈλ§μ ν΅νμ¬ κ΅¬ννμμΌλ©°, μ΄λ κ² μμ±λ
λ°μ΄ν°μ νΉμ±μ λΆμνμ¬ κ³ μ ν μ§λ½μ¬κ³ λ₯Ό νλ³ν μ μμ κ²μ΄λ€. κ³ μ ν μ§λ½μ¬κ³ μ νΉμ±μ λ°μνκΈ° μν΄ μ΄λ¬ν λΉμ ν νΉμ±μ κ°μ§λ λͺ¨λΈμ μμ±νμ¬
λͺ¨μνμλ€.
κ·Έλ¦Ό 4 κ° μκ°μμμ κ³ μ ν μ¬κ³ μ λ₯ (1)
Fig. 4 HRF current on robust pebbles (1)
κ·Έλ¦Ό 5 μ¬κ³ λ°μ ν μκ°μ λ°λ₯Έ μ λ₯ νν (1)
Fig. 5 Current with the time after a HRF (1)
κ³ μ ν μ¬κ³ λ₯Ό λͺ¨μνκΈ° μνμ¬ μλ³μ νμ λͺ¨λΈμ μ¬μ©νμμΌλ©°, κ·Έλ¦Ό 6κ³Ό κ°μ΄ κ°λ³μ νμ μ΄μ©νμ¬ κ³ μ₯μ νμ λͺ¨λΈλ§νμλ€. κ·Έλ¦Ό 7(a)μμλ μ΄ λͺ¨λΈμ ν΅νμ¬ μ»μ κ³ μ₯μ λ₯ ννμ΄ κ·Έλ¦Ό 5μ μ μ¬ν ννμμ λ³Ό μ μκ³ , μ΄λ¬ν κ³ μ₯μ λ₯ μ
λ ₯μ΄ μμ λ κ³ν΅μμ νλ₯΄λ κ³ μ₯μ λ₯λ₯Ό κ·Έλ¦Ό 7(b)μμ λ³Ό μ μλ€. μ΄λ¬ν κ³ μ₯μ λ₯ ννμμ λͺ¨λΈμ μν κ³ μ₯μ λ₯μ λΉμ νμ±κ³Ό λΉλμΉμ±μ νμΈν μ μλ€.
κ·Έλ¦Ό 6 κ°λ³μ νμ μ΄μ©ν κ³ μ₯μ ν λͺ¨λΈ
Fig. 6 Fault resistance model using variable resistors
κ·Έλ¦Ό 7 λ°°μ κ³ν΅ λͺ¨λΈμμ μμ±λ κ³ μ₯μ λ₯ νν
Fig. 7 Fault current by simulation with the distribution system model
2.3 λΆνκΈμ¦ λͺ¨λΈλ§
κ³ μ₯μ΄ μλ λΆνμ μ μμ μΈ μ¦κ° μμλ κ³μ κΈ°κ° μΈ‘μ νλ μ λ₯λμ΄ μ»€μ§ μ μλ€. κ³ μ ν κ³ μ₯μ΄ λ°μνλ κ³ μ₯μ λ₯μ μμ΄ ν¬μ§ μκΈ° λλ¬Έμ λ¨μν
μ λ₯ λ³νλ μΈ‘μ μ ν΅ν΄μλ μ μμ μΈ λΆν μ¦κ°μ λ°λΌ λμ΄λλ μ λ₯μ ꡬλ³νκΈ° μ½μ§ μμΌλ©°, μ΄ λ
Όλ¬Έμμλ μ¨μ΄λΈλ¦Ώ λ³νκ³Ό CNN κΈ°λ²μ μ μ©νμ¬
κ³ μ ν κ³ μ₯μ νλ³νκ³ μ νλ€. μ΄λ₯Ό μν λ°μ΄ν° μμ±μ μνμ¬ κ³ μ₯ μν©λΏ μλλΌ λΆν μ¦κ°μ λν λ°μ΄ν°κ° νμνλ©°, μ΄λ₯Ό μνμ¬ λ€μκ³Ό κ°μ΄
λͺ¨λΈμ ꡬμ±νμ¬ λͺ¨μνμλ€. μκ°μ μΌλ‘ μ¦κ°νλ λΆνλ₯Ό λͺ¨μνκΈ° μν΄ κ° μλ³λ‘ λΆνλμ΄ κΈμ¦νλ μν©μ λͺ¨μνμμΌλ©°, νΌλμ κ°κΉμ΄ μ§μ , μ€κ°
μ§μ , λΆμ°μ μμ κ°κΉμ΄ μ§μ μμ λΆνκ° κΈμ¦νλλ‘ κ·Έλ¦Ό 8κ³Ό κ°μ΄ λͺ¨λΈλ§νμλ€. κ·Έλ¦Ό 9λ μ΄ λͺ¨λΈμ λ°λΌ λΆνκ° μ¦κ°ν λ λͺ¨μ λ³ μ μμ λ³νλ₯Ό λνλ΄κ³ μλ€.
κ·Έλ¦Ό 8 μκ° λΆνκΈμ¦ λͺ¨λΈλ§
Fig. 8 Modeling for rapid increase in load
κ·Έλ¦Ό 9 μκ° λΆνκΈμ¦μ κ° λͺ¨μ λ³ μ μκ°ν
Fig. 9 Voltage drop by rapid increase in load
μ΄μκΉμ§ λ₯λ¬λμ μν λͺ¨μλ°μ΄ν° μμ± μ‘°κ±΄μ λν΄ μ 리νλ©΄ ν 1κ³Ό κ°λ€. μ΄λ¬ν 쑰건μ λν΄μ λͺ¨μλ₯Ό μννκ³ , κ·Έ κ²°κ³Όλ₯Ό ν
μ€νΈ νμμ λ°μ΄ν° νμΌλ‘ μ μ₯νμλ€. κ° μ‘°κ±΄μ λν΄ λͺ¨μλ₯Ό μννμμΌλ©°, μνλ§
νμμ μμ© λμ§νΈ κ³μ κΈ°μ κ°κ² 50ΞΌsecλ‘ νμλ€.
ν 1 λͺ¨μλ°μ΄ν° μμ± μ‘°κ±΄
Table 1 Conditions for generating simulation data
λΆμ°μ μ μ©λ [MW]
|
8.0, 4.5, 1.0, 0.0
|
κ³ μ₯ / λΆν κΈμ¦ μμΉ
|
2-3 λͺ¨μ μ¬μ΄ / 7-8 λͺ¨μ μ¬μ΄ / 13-14 λͺ¨μ μ¬μ΄
|
κ³ μ₯ μ’
λ₯
|
κ³ μ ν λ¨μ μ§λ½ κ³ μ₯: κ³ μ₯μ ν 100 / 500 / 1,000 Ξ©
|
3. λ₯λ¬λμ ν΅ν κ³ μ₯ νλ³
3.1 κ³ μ₯ λ°μ΄ν°μ μκ°ν
λͺ¨μλ₯Ό ν΅νμ¬ μ»μ κ³ μ₯λ°μ΄ν° λ° μ μμνμ λ°μ΄ν°λ μ μκ³Ό μ λ₯μ μκ³μ΄ λ°μ΄ν°μ΄λ€. μ΄ λ°μ΄ν°λ€μ CNNμ ν΅νμ¬ κ΅¬λΆνκΈ° μνμ¬ μκ³μ΄ λ°μ΄ν°λ₯Ό
μκ°ννμλ€. κ³ μ₯μ νλ³νλ μΌλ°μ μΈ λ³΄νΈκ³μ κΈ°μ κ°μ΄ μ λ₯ λΆμμ ν΅νμ¬ κ³ μ₯ νλ³μ ν μ μλλ‘ μ λ₯μ μκ³μ΄ λ°μ΄ν°λ₯Ό λμμΌλ‘ λΆμνμλ€.
λ³Έ λ
Όλ¬Έμμλ κ³ μ ν μ¬κ³ κ° μλ μΌλ°μ μΈ μ¬κ³ λ κΈ°μ‘΄ κ³μ κΈ° μκ³ λ¦¬μ¦μΌλ‘ νλ³ν μ μλ€κ³ κ°μ νκ³ , κ³μ κΈ°μ κΈ°λ³Έ κΈ°λ₯μΌλ‘ νλ³μ΄ μ΄λ €μ΄ κ³ μ ν
μ¬κ³ μ λΆνλ μ¦κ°μ κ΅¬λ³ λ°©λ²μ μ μνκ³ μ νλ€. λ°λΌμ λμ λ°μ΄ν°λ κ·Έ λ κ²½μ°μ λ°μ΄ν°λ₯Ό λ€λ£¨μλ€.
μ λ ₯κ³ν΅μμ κ³ μ₯μ΄ λ°μνμμ λ κ³ν΅μ 보νΈνκΈ° μν΄μλ λΉ λ₯΄κ² κ³ μ₯ μ¬λΆλ₯Ό νλ¨ν μ μμ΄μΌ νλ€. μΌλ° κ³μ κΈ°λ λ μ£ΌκΈ° λ΄μ νλ¨ν μ μμ΄μΌ
νλ©°, λ³Έ λ
Όλ¬Έμμλ ν μ£ΌκΈ° λ΄μ νλ³ν μ μλλ‘ ν μ£ΌκΈ°μ μ λ₯ λ°μ΄ν°λ₯Ό μ¬μ©νμ¬ νλ³ν μ μλλ‘ νμλ€.
μ€μ λͺ¨λ₯Όλ μ¨μ΄λΈλ¦Ώμ λͺ¨ν¨μλ λ€μκ³Ό κ°λ€.
λ³ν κ²°κ³Όμ κ³μλ₯Ό ꡬνκΈ° μν΄μ pywt λͺ¨λμ pywt.cwt() ν¨μλ₯Ό μ΄μ©νμκ³ , matplotlib.pyplotμ μ¬μ©νμ¬ κ³μλ€μ κ·Έλν½μΌλ‘
λνλ΄μλ€. μμ λ₯μ ν μ£ΌκΈ° λ°μ΄ν°μ λͺ¨λ₯Όλ μ¨μ΄λΈλ¦Ώ λ³ν(Morlet wavelet transform)μ μ μ©νκ³ , κ·Έ κ³μμ λνμ¬ κ·Έλν½ λ°μ΄ν°λ₯Ό
μμ±νμλ€. κ·Έλ¦Ό 10μ μμ±λ μ λ₯ λ°μ΄ν°λ₯Ό κ·Έλν½ λ°μ΄ν°λ‘ λ³νν μμ΄λ€. λ³ννλ μ λ₯ μκ³μ΄ λ°μ΄ν°μ κ°μλ ν μ£ΌκΈ°μ ν΄λΉνλ 32κ°μ μκ³μ΄ κ°μ΄λ©°, λͺ¨λ₯Όλ
μ¨μ΄λΈλ¦Ώ λ³νμ κ³μλ₯Ό μμΌλ‘ νννμ¬ 32Γ48 ν½μ
μ μκ°-μ£Όνμ κ·Έλν½ λ°μ΄ν°λ‘ ꡬμ±νμλ€.
κ·Έλ¦Ό 10 μ λ₯ λ°μ΄ν°μ μκ°-μ£Όνμ κ·Έλν½ λ°μ΄ν° λ³ν
Fig. 10 Time-frequency energy picture of current data
λͺ¨μλ₯Ό ν΅ν΄ μ»μ κ³ μ νμ§λ½μ¬κ³ μν©κ³Ό λΆνλ μ¦κ°μ μν μ λ₯ λ³νμν©μ λν λ°μ΄ν°μ λνμ¬ μ¨μ΄λΈλ¦Ώ λ³νμ ν΅νμ¬ κ·Έλν½ λ°μ΄ν°λ‘ λ³ννμκ³ ,
λ³νλ κ·Έλν½λ°μ΄ν°μ λνμ¬ κ³ μ ν μ¬κ³ μ λΆνμ¦κ°λ‘ λΌλ²¨λ§νμλ€. ν 2μ λΆμ λμ λ°μ΄ν°μ λν λ΄μ©μ μ 리νμλ€.
ν 2 CNN νμ΅μ μ¬μ©ν λ°μ΄ν°
Table 2 Amount of data for CNN
λ°μ΄ν° μ’
λ₯
|
μ΄ κ°μ
|
νλ ¨
(Train)
|
κ²μ¦
(Validation)
|
ν
μ€νΈ
(Test)
|
κ³ μ νμ¬κ³
|
7,200
|
3,600
|
2,160
|
1,440
|
λΆνμ¦κ°
|
2,400
|
1,200
|
720
|
480
|
ν©κ³
|
9,600
|
4,800
|
2,880
|
1,920
|
3.2 CNNλͺ¨λΈ κ΅¬μ± λ° νμ΅ κ²°κ³Ό
κ·Έλ¦Ό 11 CNN νμ΅ μν κ³Όμ
Fig. 11 Structure of CNN algorithm
κ·Έλ¦Ό 12 CNN νμ΅ κ³‘μ
Fig. 12 Learning curve of CNN
μμ±λ κ·Έλν½ λ°μ΄ν°μ λν΄ CNNμ μ μ©νμμΌλ©°, λ¬Έμ λ μ΄μ§ λΆλ₯λ¬Έμ λ‘ κ΅¬μ±νμλ€. (κ³ μ ν μ§λ½κ³ μ₯ / μ μ μ΄μ ) κ·Έλ¦Ό 11μ λ₯λ¬λ μ΄μ©ν κ³ μ ν μ§λ½ κ³ μ₯μ λΆμνλ μμλμ μ μ©ν CNNμ κ³μΈ΅ ꡬμ±μ΄λ€. νμ΄μ¬(Python) νκ²½μμ tensorflow, keras
λͺ¨λμ νμ©ν΄ CNN νμ΅μ μννμμΌλ€. κ·Έ νμ΅κ³Όμ μ κ·Έλ¦Ό 12μ κ°μΌλ©°, κ°κ° νμ΅μ λ°λ₯Έ μ νλ(accuracy)μ μμ€(loss)μ λ³νλ₯Ό λνλ΄κ³ μλ€. νμ΅μ ν΅ν΄ μμ±λ λͺ¨λΈμ ν
μ€νΈ λ°μ΄ν°μ μ μ©ν
κ²°κ³Ό, μ νλ 98.29%λ‘ κ³ μ νμ§λ½κ³ μ₯μ νλ³νμλ€. μ΄λ¬ν μ νλλ κ³μ κΈ°μ μ μ© κ°λ₯ν μμ€μΌλ‘ νλ¨λλ€.
μ΄λ κ² μμ±λ κ³ μ₯νλ¨ λͺ¨λΈμ κ³μ κΈ°μ νμ¬ν¨μΌλ‘μ¨ κ³ μ ν μ¬κ³ λ₯Ό νλ³νκ² λλ€. μ΄λ κ² νμ±λ κ³ μ ν νλ³ μκ³ λ¦¬μ¦μ ν¬ν¨ν λ°°μ κ³ν΅μμμ κ³ μ₯νλ³
κ³Όμ μ μ 리νλ©΄ κ·Έλ¦Ό 13κ³Ό κ°λ€. 보νΈκ³μ κΈ°μ μ΄λ¬ν μκ³ λ¦¬μ¦μ μ μ©ν¨μΌλ‘μ¨ κ³ μ ν μ¬κ³ κ° μλ κ³ μ₯μλ κΈ°μ‘΄μ μΌλ°μ μΈ κ³ μ₯κ³μ κΈ° μκ³ λ¦¬μ¦μ ν΅νμ¬ λμνκ³ , κ³ μ ν μ¬κ³ μ
λν΄μλ μ΄ λ
Όλ¬Έμμ μ μν μκ³ λ¦¬μ¦μΌλ‘ λμν μ μλ€.
κ·Έλ¦Ό 13 κ³ μ ν κ³ μ₯νλ³μ ν¬ν¨ν λ°°μ κ³ν΅ 보νΈκ³μ κΈ° λμ
Fig. 13 Protction relay operation including with HRF detection
4. κ²° λ‘
λ³Έ λ
Όλ¬Έμμλ λΆμ°μ μ μ€μΉ λ±μΌλ‘ ꡬμ±μ΄ 볡μ‘ν΄μ§ λ°°μ κ³ν΅μμ κ³ μ νμ§λ½μ¬κ³ κ° λ°μνμμ λ μ΄λ₯Ό νλ³νκΈ° μν λ°©λ²μ μ μνμλ€. κ³ μ νμ§λ½μ¬κ³ λ
μ¬κ³ μ λ₯μ ν¬κΈ°κ° ν¬μ§ μμ μΌλ°μ μΈ λ³΄νΈκ³μ κΈ°μ μκ³ λ¦¬μ¦μΌλ‘ νλ³νκΈ° μ½μ§ μμμ κ³ μ νμ§λ½μ¬κ³ μ λ°μνλ κ³ μ₯μ λ₯μ νΉμ±μ νμ©ν μ£ΌνμλΆμ
λ±μ κΈ°λ²μ΄ μ μ©λμ΄ μλ€. λΆμ°μ μ λμ
λ±μΌλ‘ κ³ μ₯μ λ₯μ μμμ΄ λ³΅μ‘ν΄μ§λ©΄ κ³ μ νμ¬κ³ μ νλ³μ΄ λμ± μ΄λ €μμ§κ² λλ©°, μ΄λμλ λ³νμ§ μλ κ³ μ₯νΉμ±μ
νμ©ν μ μλ κΈ°λ²μΌλ‘μ κ³ μ₯μ λ°μνλ μ λ₯λ₯Ό μκ°λ°μ΄ν°λ‘ λ³ννκ³ , μ΄μ λν΄ CNN κΈ°λ²μ μ μ©νμ¬ κ³ μ νμ§λ½μ¬κ³ λ₯Ό νλ³νλ λ°©λ²μ μ μνμλ€.
CNN κΈ°λ²μ νμ΅νκΈ° μν λ°μ΄ν°λ λͺ¨λΈκ³ν΅μ λͺ¨μλ₯Ό ν΅νμ¬ μμ±νμλ€. λ°μ΄ν° μμ±μ μν΄ λͺ¨λΈκ³ν΅μμ κ³ μ νμ§λ½μ¬κ³ κ° λ°μν κ²½μ°μ λΆνκ° μ¦κ°ν
κ²½μ°μ λν΄μ κ³ μ₯μ νμ ν¬κΈ°, λ°μ μμΉ, λΆμ°μ μμ λ°μ λ λ±μ λ³ννλ©΄μ λͺ¨μλ₯Ό μννμλ€. μ΄λ κ² μμ±λ λ°μ΄ν°μ λͺ¨λ₯Όλ μ¨μ΄λΈλ¦Ώ λ³νμ μ μ©νμ¬
κ·Έλν½ λ°μ΄ν°λ‘ λ³νν λ€μ CNNμ μ μ©νμ¬ νμ΅μ μννμλ€. νμ΅ κ²°κ³Ό 98.29%μ μ νλλ‘ κ³ μ νμ§λ½μ¬κ³ λ₯Ό νλ³νμμΌλ©°, μ΄λ₯Ό κΈ°μ‘΄ 보νΈκ³μ κΈ°
μκ³ λ¦¬μ¦κ³Ό μ‘°ν©ν¨μΌλ‘μ¨ λ°°μ κ³ν΅μμ λ°μν κ³ μ νμ§λ½μ¬κ³ μ λμν μ μλ 보νΈκ³μ μκ³ λ¦¬μ¦μ μ μνμλ€. μΆν μ€μ κ³ν΅μμ λ°μν λ°μ΄ν°λ₯Ό ν΅νμ¬
μ μ λ°©λ²μ κ²μ¦κ³Ό κ°μ μ λν μ°κ΅¬λ₯Ό μ μνλ€.
Acknowledgements
This work was supported by KETEP (Korea Institute of Energy Technology Evaluation
and Planning) grant funded by the Korea government (MOTIE) (No. 20191210301890)
References
Soon-Ryul Nam, Yong-Chul Kang, Jong-Keun Park, Nov 1999, Modeling of a high impedance
fault in a distribution system using time-varying resistance, Proc. KIEE Fall Conf.
Korea, pp. 113-115
Won-Wook Jung, Hak-Ju Lee, Seong-Chul Kwon, Woo-Kyu Chae, Sep 2010, Application of
directional over current protection schemes considering the fault characteristics
in the distribution system with dispersed generation, Journ. KIIEE, Vol. vol 24, No.
9, pp. 97-107
Z. Zhixia, L. Xiao, P. Zailin, Feb 2014, Fault line detection in neutral point ineffectively
grounding power system based on phase-locked loop, IET Generat. Transmiss., Distrib.,
Vol. 8, No. 2, pp. 273-280
Y. Wang, J. Zhou, Z. Li, Z. Dong, Y. Xu, Aug 2015, Discriminant-analysisbased single-phase
earth fault protection using improved PCA in distribution systems, IEEE Trans. Power
Del., Vol. 30, No. 4, pp. 1974-1982
T. Ji, Q. Pang, X. Liu, Aug 2006, Study on fault line detection based on genetic artificial
neural network in compensated distribution system, Proc. IEEE Int. Conf. Inf. Acquisition,
Vol. weihai, No. china, pp. 1427-1431
X. Deng, R. Yuan, Z. Xiao, K. L. L. Wang, Feb 2015, Fault location in loop distribution
network using SVM technology, Int. J. Electr. Power Energy Syst., Vol. 65, pp. 254-261
Y. Y. Wang, X. J. Zeng, Z. Y. Dong, Y. Xu, J. Yuan, Y. Huang, Dec 2013, Stator single-phase-to-ground
fault protection for bus-connected powerformers based on hierarchical clustering algorithm,
IEEE Trans. Energy Convers., Vol. 28, No. 4, pp. 991-998
M. Farshad, J. Sadeh, Oct 2012, Accurate single-phase fault-location method for transmission
feeders based on K-nearest neighbor algorithm using one-end voltage, IEEE Trans. Power
Del., Vol. 27, No. 4, pp. 2360-2367
N. I. Elkalashy, A. M. Elhaffar, T. A. Kawady, N. G. Tarhuni, M. Lehtonen, Oct 2010,
Bayesian selectivity technique for earth fault protection in medium-voltage networks,
IEEE Trans. Power Del., Vol. 25, No. 4, pp. 2234-2245
S. Zheng, L. C. Wang, H. Zhang, Oct 2016, A fault line selection method for small
current grounding system based on big data, in Proc. IEEE PES Asia.Pacific Power Energy
Eng. Conf. (APPEEC), Vol. xiβan, No. china, pp. 2470-2474
S.-S. Wang, A. Chern, Y. Tsao, J. Hung, X. Lu, Y.-H. Lai, B. Su, Aug 2016, Wavelet
speech enhancement based on nonnegative matrix factorization, IEEE Signal Process.
Lett., Vol. 23, No. 8, pp. 1101-1105
E. Kang, J. Min, J. C. Ye, Oct 2016, A deep convolutional neural network using directional
wavelets for low-dose X-ray CT reconstruction, Comput. Vis. Pattern Recognit., Vol.
44, No. 10, pp. e360-e375
M. F. Guo, X. D. Zeng, D. Y. Chen, N. C. Yang, Feb 2018, Deep-learning-based earth
fault detection using continuous wavelet transform and convolutional neural network
in resonant grounding distribution systems, IEEE Sensors Journ., Vol. 18, No. 3, pp.
1291-1300
M. F. Guo, N. C. Yang, W. F. Chen, April 2019, Deep-learning- based fault classification
using Hilbert-Huang transform and convolutional neural network in power distribution
systems, IEEE Sensors Journ., Vol. 19, No. 16, pp. 6905-6913
D. I. Jeerings, J. R. Linders, April 1989, Ground resistance- revisited, IEEE Trans.
Power Deliv., Vol. 4, No. issue 2, pp. 949-956
A. Aljohani, I. Habiballah, High-Impedance Fault Diagnosis: A Review, Energies 2020,
13, 6447
μ μμκ°
λ°μ’
μ (Jong-young Park)
Jong-young Park received the B.S., M.S., and Ph.D. degrees from Seoul National University,
Seoul, Korea, in 1999, 2001, and 2007, respectively.
He was a Senior Researcher at LS Electric Co., Ltd., Korea from 2009 to 2013.
Currently, he is a Senior Researcher at Korea Railroad Research Institute (KRRI) since
2013.
His recent research interests include the optimal operation of power systems in railway
with the smart grid technology.
Hanmin Lee received the M.S. and Ph.D. degrees from Korea University, Seoul, Korea,
in 2006.
Currently, he is a chief Researcher at Korea Railroad Research Institute (KRRI) since
2000.
His research interests include power quality and energy storage systems.
Gyu-Jung Cho (Sβ14) was born in South Korea, in 1986.
He received the B.S., M.S. and Ph.D. degrees, in 2012, 2014 and 2019, respectively,
from the College of Electrical and Computer Engineering, Sungkyunkwan University,
Suwon, South Korea.
He is currently a Senior Researcher with the Smart Electrical & Signaling Division,
Korea Railroad Research Institute, Uiwang, South Korea.
His research interests include power system dynamics, electric railway system operation
and protection, integration of renewable energy resources, and distribution system
planning.
He received a B.S and M.S. degree in Electrical engineering from Sungkyunkwan University,
Republic of Korea, in 1995 and 1998, respectively.
He received a Ph.D. degree from the Electrical Electronic and Computer Engineering
from Sungkyunkwan University in 2002.
He is currently a chief Researcher with the Smart Electrical & Signaling Division,
Korea Railroad Research Institute, Uiwang, South Korea.
His research interests are railway electrification, energy system and power protection
system.
He received a B.S and M.S. degree in Electrical engineering from Inha University,
Republic of Korea, in 1987 and 1989, respectively.
He is currently a Researcher with the Smart Electrical & Signaling Division, Korea
Railroad Research Institute, Uiwang, South Korea.