Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Scanned with CamScanner
Swami Keshvanand Institute of Technology, Management & Gramothan,
Ramnagaria, Jagatpura, Jaipur-302017, INDIA
Approved by AICTE, Ministry of HRD, Government of India
Recognized by UGC under Section 2(f) of the UGC Act, 1956
Tel. : +91-0141- 5160400 Fax: +91-0141-2759555
E-mail: info@[Link] Web: [Link]
Computational Issues in Large Scale Power Systems
In this section, GS and NR methods are compared when both use Ybus as the network model. It is experienced
that the GS method works well when programmed using rectangular coordinates, whereas NR requires more
memory when rectangular coordinates are used. Hence, polar coordinates are preferred for the NR method.
The GS method requires the fewest no. of arithmetic operations to complete an iteration. This is
because of the sparsity of the network matrix and the simplicity of the solution techniques. Consequently,
this method requires less time per iteration, With the NR method, the elements of the Jacobian are to be
computed in each iteration, so the time is considerably longer. For typical large systems, the time per
iteration in the NR method is roughly equivalent to 7 times that of the GS method. The time per iteration in
both these methods increases almost directly as the number of buses of the network.
The rate of convergence of the GS method is slow (linear convergence characteristic), requiring a
considerably greater number of iterations to obtain a solution than the NR method which has quadratic
convergence characteristics and is the best among all methods from the stand point of convergence. In
addition, the number of iterations for the GS method increases directly as the number of buses of the
network, whereas the number of iterations for the NR method remains practically constant, independent of
system size. The NR method needs 3 to 5 iterations to reach an acceptable solution for a large system. In
the GS method and other methods, convergence is affected by the choice of slack bus and the presence of
series capacitor, but the sensitivity of the NR method is minimal to these factors which cause poor
convergence.
Therefore, for large systems the NR method is faster, more accurate and more reliable than the GS
method or any other known method. In fact, it works for any size and kind of problem and is able to solve
a wider variety of ill conditioned problems. Its programming logic is considerably more complex and it has
the disadvantage of requiring a large computer memory even when a compact storage scheme is used for
the Jacobian and admittance matrices. In fact, it can be made even faster by adopting the scheme of optimally
renumbered buses. The method is probably best suited for optimal load flow studies because of its high
accuracy which is restricted only by round-off errors.
The chief advantage of the GS method is the ease of programming and most efficient utilization of
core memory. It is, however, restricted in use of small size system because of its doubtful convergence and
longer time needed for solution of large power networks. Thus the NR method is decidedly more suitable
than the GS method for all but very small systems.