30-07-2012, 10:56 AM
Design of adaptive load shedding by artificial neural networks
Design of adaptive load shedding by artificial neural.pdf (Size: 239.8 KB / Downloads: 27)
Abstract:
The design of an adaptive load-shedding strategy by executing an artificial neural
network (ANN) and transient stability analysis for an electric utility system is presented. To
prepare the training data set for an ANN, the transient stability analysis of an actual power system
has been performed to solve for minimum load shedding with various operation scenarios without
causing the tripping problem of generators. The Levenberg–Marquardt algorithm has been
adopted and incorporated into the back-propagation learning algorithm for training feedforward
neural networks. By selecting the total power generation, total load demand and frequency decay
rate as the input neurons of the ANN, the minimum amount of load shedding is determined to
maintain the stability of power systems. To demonstrate the effectiveness of the proposed ANN
minimum load-shedding scheme, a utility power system has been selected for computer simulation
and the amount of load shedding is verified by stability analysis.
1 Introduction
Due to the load growth in Taiwan the power shortage
during the summer peak period has been a very serious
problem for the Taiwan Power Company (Taipower) in
recent years. With the longitudinal power system, a large
amount of power has to be transmitted over 345kV EHV
lines from south to north due to the deficiency of area
generation in the northern region. System disturbances
caused by the outage of large generation units or tripping of
EHV transmission lines may result in severe system
instability. Without taking the proper remedial action in
time, the dynamic response of the power system will
deteriorate. For instance, the EHV transmission tower
collapsing in 1999 initiated the tripping of tie lines between
southern and northern-central areas and resulted in the
total system blackout [1].
To enhance the system reliability of the Taipower system,
an effective load-shedding scheme has to be derived and
implemented to maintain power system stability. Insufficient
load shedding will cause serious system frequency
decay and a stability problem. On the other hand, excessive
load shedding will trip the load too much to cause an
unnecessary power outage problem. It is a common practice
for utility companies to perform load shedding by using
under-frequency relays to trip the predetermined load with
many shedding steps when the frequency drops below set
values [2, 3]. A load-shedding method which considers the
frequency decay rate is also applied for utilities in [4]. To
determine the minimum load shedding required for
maintaining system stability in fault contingencies an
adaptive ANN model is proposed. The amount of load to
be tripped can be determined very efficiently according to
the input parameters so that the control signals to
disconnect the less critical loads can be issued at the instant
of tie line tripping.
Up to now the ANN has been successfully used to deal
with the data by imitating the human neural network. It has
been applied in the areas of function approximation and
pattern recognition. It is also suitable for multivariable
applications because of the ability to easily identify the
interaction between the inputs and outputs. In recent years
the ANN has been applied in the design of power system
controllers [5, 6], load forecasting [7], harmonic prediction
[8, 9] and fault protection [10]. Because the variation of
system frequency is highly dependent on the initial
operation condition, the seriousness of fault contingency,
the response of governor systems, and so on, it becomes
more difficult to determine the minimum amount of load
shedding by the traditional methods for a large power
system. By performing the stability analysis for various fault
scenarios the training data set of ANN model can be
created. With the ANN model derived, the optimal load
shedding at the instant of tie lines tripping is determined
according to the input neurons of the neural network.