05-05-2011, 11:40 AM
reactive power.pdf (Size: 123.2 KB / Downloads: 54)
Abstract
This paper presents an artiÞcial neural network-based approach (ANN) for reactive power
(VAR) optimization of interconnected power systems. The reactive power resources are scheduled
to minimize the total transmission losses of the network. The proposed ANN for this study
is a three-layer feed-forward network with a sigmoidal transfer function. Di¤erent loading
conditions of reactive power are used as input pattern to train the ANN. The desired output is
the optimal voltages at the VAR-controlled buses. Since the resulting state from the ANN may
not be feasible and some voltage limits are exceeded, a rule-based approach is used for control
variable adjustments.
Simplicity, high processing speed and ability to model non-linear functions using ANN make
the proposed approach a viable option for VAR optimization. The proposed approach is
applied on a real power system and the presented test results demonstrate its applicability for
real-time VAR optimization. ( 1998 Elsevier Science B.V. All rights reserved.
Keywords: Reactive power optimization; Interconnected power systems
1. Introduction
Recently, reactive power control has received an ever-increasing interest from
electric utilities especially due to limited transmission capabilities of high-voltage
network to accommodate additional electric loads. Any changes in system demand
may result in lower-voltage proÞles. In order to maintain the desired voltage proÞle
and reactive power ßow along the transmission lines under various operating conditions,
power system operators can select number of control tools such as switching
VAR compensators, changing generator voltages and adjusting transformer tap
settings. By an optimal adjustment of these control devices, the scheduling of the
reactive power could minimize the transmission losses of the network
The solution of optimal rescheduling of reactive power involves the minimization of
the non-linear network losses subjected to non-linear operating constraints [3,9,10].
Extensive computational experience indicates that non-linear programming techniques
is very demanding task, especially for large-scale power systems. Linear
programming approaches employ linearized version of system constraints and transmission
losses [2,6]. These approaches are iterative and linear programming is run
successively until the real losses are optimized. Moreover, it is only applicable to small
system deviations from the desired optimal voltage set points.
The complexity of VAR rescheduling problem and the demand for ßexibility in
applying the proposed approach on real-time loss minimization have challenged the
researcher to investigate the applicability of ANN algorithm to solve this problem
[11Ð13]. The learning capability of ANN o¤ers some unique opportunities to speed
up the real-time solution of such control problems. Most of the required computing
time is spent in the o¤-line training from the input/output data patterns. Once the
ANN is completely trained, the on-line operation would involve a chain of simple
arithmetic operations for which the processing time should be very short compared to
the analytical programming technique. If a forecasted reactive-power proÞle is received,
the ANN will match it with the closest pattern available in the training set, and
the corresponding control variable to minimize the network losses will be retrieved as
a new solution. The output pattern from ANN may lead to a few system states, which
violate some operating constraints. The subsequent FDLF solution examines the
system state corresponding to the ANN output pattern. A rule-based approach is
employed to reÞne the solution with minimum adjustment of control variables to
alleviate limit voilations.