Академический Документы
Профессиональный Документы
Культура Документы
The variables that need to optimize from (4) and (5) are δ and
Data preparation, data structure
|V| so the iterations (2) can be written as follows: (load data, Admittance data)
δ i( k +1) = δ i( k ) + ∆δ i( k )
(6)
( k +1)
= Vi (k )
+ ∆ Vi (k ) Construct
Vi Ybus
(7)
The eq. (3) can be written as (8-10)
−1 Initialize
∆δ J1 J 2 ∆P Voltage and Q according to
∆ | V | J 3 J 4 ∆Q
(8)
Construct Jacobian matrix
∆Pi( k ) = Pi sch − Pi( k ) according to the kind of
buses (eq.11-18)
(9)
∆Qi( k ) = Qisch − Qi( k ) Iteration k=1
(10)
The J1, J2, J3, and J4 actually are related to kind of buses, for Calculated P and Q
according to the kind of
PV-buses the variable need to optimize are δ and Q, and for buses using eq.(4) and (5)
PQ-buses the variable need to optimize are δ and |V|, so J1, J2,
No For PV-bus
J3, and J4 can be written as follows: Check
Q > Qmax
∂Pi
become PQ bus
Yes
= ∑ Vi V j Yij sin(θ ij − δ i + δ j )
∂δ i j ≠1
Set Q=Qmin
And change bus status Construct Jacobian matrix
become PQ bus according to the kind of
(11) buses (eq.11-18)
∂Pi
= − Vi V j Yij sin(θ ij − δ i + δ j ) j≠i
∂δ j
Yes
(12)
Any bus status changed
(13)
∂Pi
( )
Yes
(14)
∂Qi
= ∑ Vi V j Yij cos(θ ij − δ i + δ j )
∂δ i j ≠1
(15)
Fig.1: Newton Raphson power flow
The checking limit methods was used in Fig.1 is rectangular This makes the computation extremely difficult. To simplify
limit (Pmin-Pmax /Qmin-Qmax ). The dashed line is procedure to the computation process, we convert all the (P, Q) pairs into
limit P and Q, many power flow program applied rectangular the polar coordinates, (R,θ) pairs as shown in Fig. 3. Once θ is
limit [5, 19-20] [21] as a checking limit. Some commercial chosen, we only need to compute the length, R. Therefore, θ
software [22] used the several line to construct GCC, this will be our input for the training process and R will be the
method better than just rectangular. Actually limit for output of the NN. The proper weighting and number of
operating point of generator is GCC, but it is to complicate neurons in the hidden layer are then determined to construct
applying in power flow. The proposed method is introducing the complete GCC curves. The one of results Comparison
the easy algorithm to account GCC limit in power flow based between GCC data sheet and GCC based on NN can be seen
on NN to replace dashed line procedure, the detail process will in Fig.4.
be described in the next subsection.
PAITON CAPABILITY CURVE
B. Develop neural network (NN) models for generator GCC data sheet
600
capability curves(GCC) limits GCC based on NN
400
The proposed NN model for a generator capability curve
Reactive Power(MVar)
is very straightforward as it only has one input, one output and 200
one hidden layer, as shown in Fig. 2. The number of neuron in
hidden layer is constructed automaticaly by using constructive 0
backpropagation method [18].
-200
-400
-600
Q
θ = tan −1
R = P2 + Q2 Convert P and Q
become θ and R
Fig. 6. Relationship between P,Q, θ, R and Rref
Q
θ = tan −1
R = P2 + Q2
P
III. SIMULATION RESULT AND ANALYSIS
A. Data Simulation
The simulation data used are the IEEE data test 30 bus. To
show the advantages of using GCC limit based on NN, the
load data each buses is increased by 1.3 times.
Because of the IEEE data test 30 buses don’t have the GCC
data, it only provide (Pmin-Pmax/Qmin-Qmax) data. To verify the
proposed method the GCC data from [16] is adopted to
represent model of GCC in IEEE data test 30 bus , the scaling
is needed in matching the generator capacity.
B. GCC based on NN
The NN model used to model GCC has one input and one
output. The input is θ and the output is R. the range of θ are
between +90o and -90o, and R is magnitude of complex power.
The NN model only has 1 hidden layer with numbers of
neurons are 17, the activation function used is sigmoid. The
sigmoid choose because the output is R that always positive. Fig. 7. Newton Raphson power flow with limit GCC based on
After training process success with 1e-6 error tolerance, the NN
plotting between R data and R resulted by NN can be seen in
Fig.8. The number of data used in training process are 200
sampling point.
Mag. (p.u.)
0.8
1
0.7 0.5
R(pu)
0.6 0
0 5 10 15 20 25 30
Bus Number
0.5
0
0.4
Angle. (rad)
-0.5
0.3
R by data
R by NN NR rectangular
0.2 NR GCC-NN
0 20 40 60 80 100 120 140 160 180 200 -1
Number of data (pu) 0 5 10 15 20 25 30
Bus Number
Fig. 8. Comparison between R data and R resulted by NN
Fig. 10. Voltage magnitude and angle each bus.
2
Fig. 8 shows that the value of R data and R resulted by NN are NR rectangular
exactly coincide both of them. The R values then convert to P 1.5 NR GCC-NN
Mag. (p.u.)
and Q according to their θ resulted the GCC as shown in 1
Fig.9.
0.5
Comparison between GCC data and GCC based on NN
1
GCC data 0
0 5 10 15 20 25 30
GCC NN Bus Number
0.8
0
0.6 -0.2
Angle. (rad)
-0.4
0.4 -0.6
Q(pu)
-0.8 NR rectangular
0.2 -1 NR GCC-NN
0 5 10 15 20 25 30
0 Bus Number
0.6
The result of simulation using IEEE data test 30 buses is
Reactive Power (pu)
[14] Abdel-Akher, M., K.M. Nor, and A.H.A. Rashid, Improved Three-Phase
Power-Flow Methods Using Sequence Components. Power Systems,
It is clear, that using GCC limit can decrease the losses and
IEEE Transactions on, 2005. 20(3): p. 1389-1397.
also can maintain the performance of system better than just [15] Mayordomo, J.G., et al., Decoupled Newton Algorithms in the Harmonic
rectangular limit. Because of the capability generator are Domain for the Harmonic Interaction of Line Commutated Converters
utilized until the exact limit, all. With AC Systems. Power Delivery, IEEE Transactions on, 2010. 25(3):
p. 1721-1733.
[16] Sutherland, P.E. Generator capability study for offshore oil platform. in
IV. CONCLUSION Industrial & Commercial Power Systems Technical Conference -
The proposed method is success in replacing rectangular limit Conference Record 2009 IEEE. 2009.
[17] Sharma, S.K., Constructive Neural Network Review. International
using GCC based on NN as a limit in Newton Raphson power Journal of Engineering Science and Technology 2010. 2(12): p. 7847-
flow. Using GCC limit the power flow still can keep the 7855.
voltages each bus at the set point when the load is increased [18] Gunaseeli, N. and N. Karthikeyan. A Constructive Approach of Modified
Standard Backpropagation Algorithm with Optimum Initialization for
by 30%. At this load condition, Using GCC limit also can
Feedforward Neural Networks. in Conference on Computational
reduce the losses by 11.401% for active power and -0.22451 Intelligence and Multimedia Applications, 2007. International
for reactive power. The Algorithm of security checking limit Conference on. 2007.
of proposed method is very easy because only use simple NN [19] Saadat, H., Power System Analysis. 2004: McGraw-Hill Education.
model that only has one layer with 17 neurons. The checking [20] John J. Grainger, W.D.S., Power System analysis: McGraw-Hill.
[21] Chowdhury, B.H. and S. Rahman, A review of recent advances in
process is only doing in one variable, i.e. R resulted by root economic dispatch. Power Systems, IEEE Transactions on, 1990. 5(4):
square calculation and R resulted by NN. p. 1248-1259.
[22] ETAP Operation Technology, Inc. Power and Energy Magazine, IEEE,
2006. 4(5): p. 81-81.
[23] Mat Syai'in. Adi Soeprijanto. Takashi Hiyama., Generator Capability
ACKNOWLEDGMENT Curve Constraint for PSO Based Optimal Power Flow. International
Thank you to Power system simulation Laboratory, Journal of Electrical and Electronics Engineering 2010. 4(6): p. 371-376.
REFERENCES
[1] Dunstan, L.A., Digital Load Flow Studies. Power Apparatus and
Systems, Part III. Transactions of the American Institute of Electrical
Engineers, 1954. 73(1): p. 825-832.
[2] Ward, J.B. and H.W. Hale, Digital Computer Solution of Power-Flow
Problems. Power Apparatus and Systems, Part III. Transactions of the
American Institute of Electrical Engineers, 1956. 75(3): p. 398-404.
[3] Tinney, W.F. and J.W. Walker, Direct solutions of sparse network
equations by optimally ordered triangular factorization. Proceedings of
the IEEE, 1967. 55(11): p. 1801-1809.
[4] Stott, B., Review of load-flow calculation methods. Proceedings of the
IEEE, 1974. 62(7): p. 916-929.
[5] Stott, B., Decoupled Newton Load Flow. Power Apparatus and Systems,
IEEE Transactions on, 1972. PAS-91(5): p. 1955-1959.
[6] Stott, B. and O. Alsac, Fast Decoupled Load Flow. Power Apparatus
and Systems, IEEE Transactions on, 1974. PAS-93(3): p. 859-869.
[7] AlHajri, M.F. and M.E. El-Hawary, Exploiting the Radial Distribution
Structure in Developing a Fast and Flexible Radial Power Flow for
Unbalanced Three-Phase Networks. Power Delivery, IEEE Transactions
on, 2010. 25(1): p. 378-389.
[8] Chen, T.H., et al., Distribution system power flow analysis-a rigid
approach. Power Delivery, IEEE Transactions on, 1991. 6(3): p. 1146-
1152.
[9] Chen, T.H. and N.C. Yang, Three-phase power-flow by direct ZBR
method for unbalanced radial distribution systems. Generation,
Transmission & Distribution, IET, 2009. 3(10): p. 903-910.
[10] Cheng, C.S. and D. Shirmohammadi, A three-phase power flow method
for real-time distribution system analysis. Power Systems, IEEE
Transactions on, 1995. 10(2): p. 671-679.
[11] Chih-Wen, L., et al., Toward a CPFLOW-based algorithm to compute
all the type-1 load-flow solutions in electric power systems. Circuits and
Systems I: Regular Papers, IEEE Transactions on, 2005. 52(3): p. 625-
630.
[12] Tsai-Hsiang Chen, N.-C.y., Loop frame of based three-phase power flow
for unbalanced radial distribution systems. Electric Power Systems
Research, 2010. 80: p. 799-806.
[13] Yi-Shan, Z. and C. Hsiao-Dong, Fast Newton-FGMRES Solver for
Large-Scale Power Flow Study. Power Systems, IEEE Transactions on,
2010. 25(2): p. 769-776.