You are on page 1of 5

AN ARTIFICIAL NEURAL NETWORK BASED TRANSMISSION LOSS

ALLOCATION FOR BILATERAL CONTRACTS

Rezaul Haque N. Chowdhury


Department of Electrical Engineering Department of Electrical Engineering
University of Saskatchewan University of Saskatchewan
e-mail: shh227@mail.usask.ca e-mail: mnc960@engr.usask.ca

Abstract In recent years, some works on transmission loss


The introduction of deregulation and the subsequent open allocation have been reported in literature. Conejo et al [1]
access policy in electricity sector have opened up the door for have discussed the Pro Rata (PR) procedure, a technique used
power transactions between generators and bulk consumers in Mainland Spain for allocation of transmission loss, where
under many different market-driven contractual forms losses are globally assigned to generators and consumers, and
including bilateral contracts. Long-term bilateral contracts then a proportional allocation rule is used. The loss allocated
are attractive to many parties who want to avoid price to a generator is proportional to its level of energy generation.
volatility. With bilateral contracts it becomes necessary to PR procedure ignores the network and therefore, is not
allocate transmission loss to respective transactions. An consistent with solved power flow. Strbac et al [2] have
artificial neural network based transmission loss allocation proposed a transmission loss allocation method by tracing the
method is presented in this paper. The method is generator and load contributions to line flows. This method
computationally efficient and can provide solutions on a real- traces the contributions of each generator and of each load to
time basis. Most independent system variables can be used as the line flows instead of marginal contributions. Since the
inputs to this neural network which in turn makes the loss allocation method had been proposed on the basis of maximum
allocation process responsive to practical situations. Training flows in the lines, it does not reflect the actual load condition.
and testing of this network have been done with the help of the Conejo et al [3] proposed a loss allocation method called “Z-
IEEE 24-bus test system. A technique has been developed to bus allocation” which expresses the loss term in terms of Z bus
expedite the convergence and to improve the accuracy of the matrix and injected bus current. Power flow solution required
results. Numerical examples on loss allocations for both peak to get injected bus current and power has to be converted to
and off-peak hours have been provided and compared with current. Bialek et al [4] had proposed another method of loss
those obtained using another technique. allocation in which power flows in the lines are traced and a
Keywords: Neural network; transmission loss; loss proportional sharing principle is used. Expósito et al [5] have
allocation; bilateral contracts. proposed a method based on unbundling of branch flows. The
authors proposed four methods for splitting branch flows;
1. Introduction proportional allocation, quadratic allocation, geometric
allocation and fast geometric allocation. Bhuiya and
Many bulk power consumers enter into bilateral contracts Chowdhury [6] have proposed two methods of loss allocation
with power producers or suppliers to avoid price fluctuations of namely, Incremental Load Flow Approach (ILFA) and
energy market in a deregulated environment. In almost all Marginal Transmission Loss Approach. The former uses a
cases, the transactions are executed through independent modified load flow to assess transmission loss. In this method
market operators or independent system operators. Each at each load bus, load is increased in a discrete step while the
supplier, therefore, has to produce enough power to meet its loads at the other buses are kept constant. The resulting
contracted load and the corresponding transmission loss. In differential transmission loss is attributed to the corresponding
some jurisdictions, the independent market operators buy generator. This method is consistent with solved load flow and
power from spot market to compensate for transmission loss rewards counter flow in the system but it requires a high
and the suppliers in turn pay for the transmission loss caused computation time. The later method expresses Kron’s
by their respective bilateral transactions. In either case, the transmission loss formula in terms of loads and generations in
transmission loss caused by a bilateral transaction has to be order to allocate transmission loss to each transaction. But the
assessed. The main problem associated with this assessment is method requires many complex mathematical analyses and
the fact that transmission loss is a non-separable entity. Any operations.
attempt to separate it is further complicated by its non-linear
nature. However, in absence of a closed-form mathematical In this paper an Artificial Neural Network (ANN) based
solution, approximate solutions have been suggested and transmission loss allocation method has been proposed. The
adopted by market operators. resulting ANN provides allocation results very fast and can
give output on a real time basis. A trained ANN for loss

0-7803-8886-0/05/$20.00@2005 IEEE 2203


CCECE/CCGEI, Saskatoon, May 2005
allocation requires only operational data to calculate the loss exists between Generator B at Bus 23 and Load B at bus 19.
allocation at any instant. The corresponding ANN to simulate the loss allocation has 54
input neurons and 4 output neurons while the number of hidden
2. Neural Network Architecture layer and number of neurons in hidden layers were varied to
A neural network functions like a human brain and is well get optimum result. The ANN’s inputs and outputs are
known as a universal approximator. Due to the capability of described in Table 1.
dealing with highly nonlinear and complex problems, neural
networks had been successfully applied to a wide variety of Table:1 Description of inputs and outputs of ANN
real world problems. An artificial neural network has three Layer Neurons Description
essential segments; an input layer, a hidden layer and an output Inputs i1-i2 Real load for Contracts A & B (p.u.)
layer. Depending on the complexity of a solution process an
i3-i4 Reactive load for contracts A & B (p.u.)
ANN may require more than one hidden layers. Each layer is
made up of several neurons that transmit signal to the neurons i5-i19 Real loads on buses (p.u.)
at the next layer. Each signal path has its own weight, known i20-i34 Reactive loads on buses (p.u.)
as synaptic weight. By selecting the synaptic weights, an ANN i35-i44 Generations on buses (p.u.)
can be made to simulate a complex solution process.
Activation functions are used at the hidden and output layers to i45-i54 Bus voltages (p.u.)
transform the input signals. A fully connected feed forward Output O1-O2 Real loss for Contracts A & B (p.u.)
ANN as shown in Fig. 1 has been utilized in this project. O3-O4 Reactive loss for contracts A & B (p.u.)

3. Training
Back propagation learning, more precisely described as
steepest descent supervised learning using back propagation of
error, has been utilized to train the ANN[7]. Training data were
generated by ILFA, since it meets most of the desirable
properties of loss allocation except the requirement of a
reasonable computation time. Typical 24 hour load variation in
the system was considered for weekdays and weekend and the
Output layer O
generators were dispatched economically to minimize running
Input layer i Hidden layer j cost. The system has a peak load of 2494 MW. Fig 3 and 4
(One or more)
show the 24-hour real and reactive loads at various buses for a
Fig. 1 A feed forward neural network. weekday. Fig. 5 shows the corresponding generation at various
generation buses.

BUS 21 3.5
BUS 18 BUS 22
BUS 17 18
3.0
Gen B
BUS 23 2.5
Real loads (p.u.)

BUS 16 BUS 19
Load B BUS 20 BUS 13
14
BUS 15 Synch. 2.0
Cond.. 3 15
BUS 14
BUS 11 1.5 20 7 13
BUS 24 BUS 12

BUS 3 Load A 10
BUS 9 1.0 16 6
BUS 10
Sync.
BUS 4 cond 1
BUS 6 0.5 5 4

0.0
BUS 8
Generator
0 4 8 12 16 20 24
BUS 1 BUS 2 BUS 7
Gen A
Load Hour
Bus

Fig.2 IEEE 24 Bus reliability Test System. Fig. 3 24-hour real loads at different buses for a weekday.
The IEEE 24-Bus Test System has been utilized for this
work. It is assumed that two bilateral contracts, A and B, as
shown in Fig. 2 exist in this system. Contract A exists between The loss due to each bilateral contract was calculated by
Generator A at Bus 7 and Load A at Bus 9 and Contract B ILFA. 2600 training patterns were generated by varying all 54

2204
0.8 b) Dual activation functions:- The loss attributed to a
transaction can be positive or negative (in case of counter
0.7
flow). This aspect can be handled by the use of a hyperbolic
Reactive loads (p.u.)

0.6 tangent function. It was also observed that the reactive part of
transmission loss is 3/5 times that of the real part. Therefore,
0.5
two activation functions were used in the output layer for two
4 3
0.4 18 different types of output. Figure 6 shows the range of the
15 outputs and the activation functions used.
0.3 13
8
10
0.2
6 Y=a*tanh(b*x)
0.1 16 0.6
2 5 4
a
0.0 0.4
0 4 8 12 16 20 24
0.2
Hours Y
0
Fig. 4 24-hour reactive loads at different buses for a weekday. -4 -2 0 2 4
-0.2 Range of
real loss
6 -0.4 Range of
13 -a
reactive loss
-0.6
5
18 b*X
Generation (p.u.)

4
Fig. 6 Activation functions and output ranges.
3
1

2 23 c) Adapting threshold values:- Although threshold values are


21 normally kept constant, adapting these values with respect to
1 15 error can enhance convergence speed . Threshold values were
7 made adaptive with the following formula[8]:
16
2
0 22 ∂E (n)
0 4 8 12 16 20 24 θ k n+1 = θ k n + η kn n n −1
+ α (θ k − θ k ) (2)
Hour ∂θ k (n)
Fig. 5 24-hour generations at different buses for a weekday. where θ k is the threshold for layer k and α is the momentum
factor and n represents iteration.
inputs. Although back propagation algorithm is widely used in
the learning of ANNs, rate of convergence is very slow [7] and
can be trapped at a local minima. A momentum term can be 4. Testing
added to accelerate the convergence and to overcome the trap Simulations were carried out with a single hidden layer and
of a local minima [7]. Still it was observed that the speed of also with two hidden layers. For the single hidden layer,
convergence became very slow when higher level of accuracy number of neurons was varied from 10 to 55. It was observed
was desired. The steps to simulate the loss allocation are that increasing number of neurons beyond 29 did not improve
described below: performance. Fig. 7 shows the convergence characteristics of
a) Adapting learning rate:- Correction to synaptic weights the ANN with a single hidden layer with 29 neurons. Fig. 8
becomes very small when error tends to be very small i.e. error shows the convergence characteristics of the ANN with two
surface becomes almost flat. Adaptive learning can increase hidden layers with 21 neurons in the first and 10 in the second
error correction and enhance convergence speed. Learning rate hidden layer. It can be observed that adding another hidden
were made adaptive by delta-delta learning rule [8] using the layer did not improve the convergence.
following recursion formula: Fig. 9 shows the effect of the amplitude of the activation
∂E (n) ∂E (n − 1) function on the convergence characteristics of the ANN with a
∆η ji (n + 1) = γ (1)
single hidden layer. It can be easily noticed from the figure
∂W ji (n) ∂W ji (n − 1)
that ‘a’ plays an important role in the convergence.
where η is learning rate, γ is a positive constant which
determines step size, E is error and Wji is synaptic weight Fig. 10 shows effect of thresholds adaptation on convergence
connecting neuron i to neuron j. characteristics on single layer ANN.

2205
1.0E-05
1.8E-06
8.0E-06

Mean sqaure error


1.5E-06 Non-adaptive
Mean sqaure error

Single activation thresholds


6.0E-06 1.2E-06
function
9.0E-07
4.0E-06 Dual activation Adaptive
6.0E-07 thresholds
functions
2.0E-06 3.0E-07
0.0E+00
0.0E+00 0 1000 2000 3000 4000
0 1000 2000 3000 4000 5000
Iterations
Iteration
Fig. 7 Convergence characteristics with a single hidden layer. Fig. 10 Effect of adaptive threshold on convergence
characteristics.

2.0E-05
1.8E-05 5. Results
1.6E-05
Based on the convergence characteristics an ANN with a
Mean sqaure error

1.4E-05 Single activation single hidden layer with 29 neurons was chosen for loss
1.2E-05 function allocation. Two activation functions, one for real power and
1.0E-05 Dual activation the other for reactive power, were chosen with amplitudes of
8.0E-06 functions 0.5115 and 0.1116 in respective manner. ‘b’ was chosen as
6.0E-06 0.667 for both activation functions. A value of 3.58 for γ was
4.0E-06 chosen for the adaptive learning. A value of 0.48 for α was
2.0E-06 chosen for weight adaptation and a value of 0.27 for the same
0.0E+00 constant was chosen for threshold adaptation. The test patterns
were derived by varying the 54 inputs to simulate 24-hour load
0 1000 2000 3000 4000 5000
patterns for weekdays and weekends. Results obtained from the
Iterations trained ANN and the ILFA show that the ANN can allocate
Fig. 8 Convergence characteristics with two hidden layers losses to bilateral contracts with very good accuracy. Real and
reactive share of losses for peak and off-peak hours obtained
by both methods are shown in Figs. 11-14.

1.0E+00
0.05
1.0E-01
0.04
NN loss A
Mean sqaure error

Loss allocation (p.u.)

1.0E-02
a=1 0.03
1.0E-03 ILFA_loss A
a=0.8 0.02 NN loss B
1.0E-04
a=0.65 0.01 ILFA loss B
1.0E-05
a=0.51 0
1.0E-06
0.20
0.35
0.50
0.65
0.80
0.95
1.10
1.25
1.40
1.55
1.70

-0.01
1.0E-07
-0.02
0 1000 2000 3000 4000 5000 6000
Iteraions Real load (p.u.)

Fig. 9 Convergence characteristics for different values of “a”. Fig. 11 Real loss allocation for an off-peak hour.

2206
6. Conclusions
0.20
NN loss A A new transmission loss allocation method based on
Loss allocation (p.u.)

0.15 ILFA loss A artificial neural network that can allocate losses to bilateral
contracts with good accuracy has been presented. The
0.10
proposed ANN can be trained with little difficulty. The trained
NN loss B ANN can provide solution in a quick manner. The proposed
0.05
ILFA loss B ANN can yield negative loss allocation to reward generators or
0.00 loads that causes counter flow in the network. Although ILFA
0.06
0.11
0.15
0.20
0.24
0.29
0.33
0.38
0.42
0.47
0.51
-0.05 was utilized to generate training data, any other method of loss
allocation can be utilized for that purpose. The use of two
-0.10 activation functions with threshold adaptation improved
Reactive load (p.u.) convergence characteristics of the ANN.

Fig. 12 Reactive loss allocation for an off-peak hour.


References
1. A. J. Conejo, J. M. Arroyo, N. Alguacil, and A. L.
0.04 Guijarro, “Transmission Loss Allocation: A
NN loss B Comparison of Different Practical Algorithms”, IEEE
0.02 Transactions On Power Systems, Vol. 17, No. 3,
Loss allocation (p.u)

ILFA loss B August 2002 571


0
0.20

0.35

0.50

0.65
0.80

0.95

1.10
1.25

1.40

1.55

1.70

2. R. Allan, D. Kirschen, G.Strbac, “Contributions of


-0.02 Individual Generators to Loads and Flows”, IEEE
-0.04 Transactions on Power Systems, Vol.12, No.1,
February 1997, pp.52-60
NN loss A
-0.06 3. Antonio J. Conejo, Francisco D. Galiana, and Ivana
ILFA loss A Kockar, “Z-Bus Loss Allocation”, IEEE Transactions
-0.08
On Power Systems, Vol. 16, No. 1, February 2001
Real load (p.u.) 105
Fig. 13 Real loss allocation for a peak hour. 4. J. Bialek “Topological Generation and Load
Distribution Factors for Supplemental Charge
Allocation in Transmission Open Access”, IEEE
0.3 transaction of Power System, Vol.12, No.3, Aug
NN loss B 1997, pp.1185-1193
0.2
Loss allocation (p.u.)

ILFA loss B
0.1 5. Y. Tsukamoto , I, Iyoda “Allocation of Fixed
Transmission Cost to Wheeling Transactions by
0 Cooperative Game Theory”, IEEE Transaction on
0.06

0.11

0.15

0.20

0.24

0.29
0.33

0.38

0.42

0.47

0.51

-0.1 Power System, Vol.11, No.2, May 1996 pp.620-627


-0.2 6. A.Bhuiya, N.Chowdhury, “Allocation of Transmission
loss in Deregulated Power System Network with
-0.3 ILFA loss A
NN loss A Bilaeral Contacts”, 1999 Large Engineering Systems
-0.4 Conference on Power Engineering, Halifax, Nova
Reactive load (p.u.)
Scotia, Canada, June 20-22,1999, pp. 220-229
7. Phansalkar, V.V.; Sastry, P.S., “Analysis of the back-
Fig. 14 Reactive loss allocation for a peak hour. propagation algorithm with momentum”, Neural
Networks, IEEE Transactions on ,Volume: 5 , Issue:
3 , May 1994, Pages:505 – 506
8. Simon Haykin, Neural Network: “A Comprehensive
Foundation”, 2nd Edition, 1999

2207

You might also like