Вы находитесь на странице: 1из 6

Enhanced Merit Order and Augmented Lagrange Hopfield Network for Transmission Constrained Unit Commitment

Vo Ngoc Dieu, Weerakorn Ongsakul, and Wichit Krueasuk Energy Field of Study School of environment, Resources and Development Asian Institute of Technology, Pathumthani 12120, Thailand Email: ongsakul@ait.ac.th ABSTRACT
This paper proposes an enhanced merit order (EMO) and augmented Hopfield Lagrange neural network (ALH) for solving transmission constrained unit commitment (TUC) problem. The EMO is a merit order enhanced by a heuristic search algorithm based on the average production cost of generating units for unit scheduling, and the ALH is a continuous Hopfield network whose energy function is based on augmented Lagrangian relaxation for economic dispatch. For each hour with insufficient power due to transmission constraint, a repairing strategy based on heuristic search is used to satisfy the constraints. The proposed EMO-ALH is tested on the 24 bus IEEE reliability test system with 26-unit and 32-unit cases and compared to decomposition and co-ordination algorithms (DCA) and three-phase algorithm scheme (TAS), respectively. The proposed method can obtain less expensive cost than the others in a faster computing time. Continuously on time of unit i, in h. Tit,on t t U , Um Total input of continuous neuron corresponding to output of Vt, Vmt. t Ui Status of unit i at hour t (on = 1, off = 0). Uit,H Total input of continuous neuron corresponding to output of Vit,H. t Total input of continuous neuron correUm ,H sponding to output of Vmt,H. t t V , Vm Outputs of multiplier neurons representing Lagrangian multiplier t and mt, respectively. t Vi ,H Output of continuous neuron representing for output power Pit. Vmt,H Output of continuous neuron representing for transmission power flow Pmt. t, mt Penalty factors associated with power balance and transmission constraints, respectively. i, i, i Coefficients of start up cost function. t, mt Lagrangian multipliers associated with power balance and transmission constraints, respectively. 1.

NOMENCLATURE
ai, bi, ci Amj Dmi Dmr Iit Mt N NG NL PDt Pi,max Pi,min Pit Pm,max Pmt PRt Rt Sit T Ti,down Ti,up Tit,off Coefficients of fuel cost function of unit i. Generation shift distribution factor, representing power change on line m with respect to power change at bus j. Sensitivity coefficient for line m flow with respect to power output of unit i. Representation of power flow on line m with respect to power generation at reference bus r. Congestion index of off unit i at hour t. Number of congested lines at hour t. Total number of units. Number of buses with generation. Number of transmission lines. System load demand at hour t, in MW. Maximum output power of unit i, in MW. Minimum output power of unit i, in MW. Generation output of unit i at hour t, in MW. Maximum power flow limit of transmission line m, in MW. Power flow on transmission line m at hour t, in MW. System spinning reserve at hour t, in MW. Excessive spinning reserve at hour t, in MW. Start up cost of unit i at hour t, in $. Schedule time horizon for UC, in h. Minimum down time of unit i, in h. Minimum up time of unit i, in h. Continuously off time of unit i, in h.

INTRODUCTION

Unit commitment (UC) is used to schedule generators such that the total production cost of the system is minimized while maintaining sufficient spinning reserve and satisfying generator constraints. Many optimization methods have been proposed to solve unit commitment problem given in biography survey by Padhy [1]. Several classical methods for UC problem include priority list [2], dynamic programming (DP) [3], mixed integer programming (MIP) [4], Lagrangian relaxation (LR) [5], etc. The LR method is an appropriate coordination technique for generating feasible primal solutions while minimizing the duality gap. However, due to the non-convexity of the UC problem, optimality of the dual problem does not guarantee the feasibility of the primal UC problem. The duality gap is used to estimate the quality of the suboptimal solution. Recently, AI techniques applied to UC problem include neural network [6], simulated annealing (SA) [7], tabu search (TS) [8], genetic algorithm (GA) [9], evolutionary programming (EP) [10], etc. However, these methods require a considerable amount of computational time for large-scale problems. Since generating units of an electric utility system usually located in different areas interconnected via transmission lines, power flows are also subject to thermal limits of transmission lines in UC problem. Therefore, UC problem without transmission constraint

may cause some transmission lines to be overloaded. This may lead to rescheduling of generating units and may incur significant costs. The transmission constrained UC (TUC) problem with transmission constraint based on DC power flow has been solved by DP method [11], LR method [12], and three phase algorithm scheme (TAS) [13]. Of these methods, the TAS method can obtain better solution than the others in a reasonable amount of time. This paper proposes an enhanced merit order and augmented Lagrange Hopfield network (EMO-ALH) for solving TUC problem. The EMO is a merit order enhanced by a heuristic search algorithm based on average production cost of units, and the ALH is a continuous Hopfield network with its energy function is based on augmented Lagrangian relaxation. The proposed EMO-ALH solves the TUC problem in three stages. In the first stage, the EMO is used to solve unit scheduling problem satisfying spinning reserve, minimum up and down time constraints. In the second stage, the ALH is used to solve transmission constrained economic dispatch (TED) problem by minimizing the operation cost subject to the power balance and transmission constraints. In the last stage, for each hour with insufficient power due to transmission constraint, a repairing strategy based on heuristic search is used to satisfy the constraints and the ALH is applied to solve TED problem for final solution. To demonstrate the effectiveness of the proposed EMO-ALH, it is tested on the 24 bus IEEE reliability test system with 26-unit and 32-unit cases and results are compared to those from decomposition and co-ordination algorithms (DCA) [14] and TAS [13], respectively. The organization of this paper is follows. Section 2 addresses the TUC problem formulation. EMO-ALH for the problem is described in Section 3. Numerical results are presented in Section 4. Finally, the conclusion is given. 2. TUC PROBLEM FORMULATION The objective of the UC problem is to minimize
Min ai + bi Pi t + ci ( Pi t ) 2 + S it 1 U it 1 U it
t =1 i =1 T N

1 t U i = 0 0 or 1

if Ti ,ton < Ti ,up if Ti ,toff < Ti ,down otherwise

(5)

5) Transmission lime constraints


t Pm,max Pm Pm,max , m = 1, , NL
t where Pm = Dmi Pi t U it i =1 N

(6) (7)

6) Start up cost
1 Ti ,toff S it = i + i 1 exp i

(8)

3. 3.1.

AN EMO-ALH FOR TUC EMO FOR UNIT SCHEDULING

The unit scheduling by EMO based on average production cost was given in [15]. Moreover, in this paper, a heuristic search is developed using the EMO result as its initial base solution to reduce excessive spinning reserve and determine better solution. This happens when a load demand is decreasing. In the heuristic search for unit substitution, the peak load units with high operation cost, low start up cost and low minimum up/down time are mainly used to substitute for the intermediate units with medium operation cost, medium start up cost and medium minimum up/down time. The amount of excessive spinning reserve at each hour is determined:
t R t = Pi ,maxU it PD PRt i =1 N

(9)

The unit substitution procedure is as follows. Step 1: Set t = 1. Step 2: Calculate excessive spinning reserve Rt at this hour.

)]

(1)

subject to 1) Power balance constraints


t PD Pi tU it = 0 , t = 1, , T i =1 N

Step 3: Find the committed mediate units so that when de-committed their minimum up/down time constraints are not violated and insufficient spinning reserve can be compensated by the uncommitted peak load units. Put these units into an off candidate list. Step 4: If the list is not empty, choose a unit with the highest average production cost from the off candidate list. De-commit this unit and update Rt, Tit,on and Tit,off. Eliminate this unit from the off candidate list. Step 5: If Rt > 0, repeat Step 4 until Rt < 0.

(2)

2) Maximum capacity spinning reserve constraint


t PD + PRt Pi ,maxU it 0 , t = 1, , T i =1 N

(3)

3) Generation limit constraints


Pi ,minU it Pi t Pi ,maxU it , i = 1, , N

(4)

Step 6: Find the uncommitted peak load units and put them into an on candidate list. Step 7: Choose a unit with the lowest average production cost from the on candidate list.

4) Minimum up and down time constraints

Commit this unit and update Rt. Eliminate this unit from the on candidate list. Step 8: If Rt < 0, repeat Step 7 until Rt 0. Step 9: If t < T, t = t + 1 and return to Step 2. Step 10: Use minimum up/down time repairing procedure in [15] for committed peak load units. 3.2. SENSITIVITY COEFFICIENT CALCULATION

transmission constraints. The objective to be minimized is


Min a i + bi Pi t + ci ( Pi t ) 2 U it
t =1 i =1 T N

(12)

subject to 1) Power balance constraint


t PD Pi t = 0 i =1 N

(13)

Sensitivity methods have been used in system security and contingency analysis. In particular, generation shift distribution factor (GSDF) and outage distribution factor (ODF) are the two most important sensitivity methods used. The GSDF [16] is based on DC load flow model, and this factor is useful only when the total system generation (or load) remains unchanged. In other words, the GSDF requires the initial power flow for each different load demand. Therefore, the GSDF is not conveniently applied in transmission constrained economic dispatch problems. In this paper, a generalized generation distribution factor (GGDF) [17] is used for transmission constrained economic dispatch. One of the advantages of the GGDF is that it is not necessarily recalculated when a load demand changes. Based on the GSDF Amj, representing power change in line m with respect to power change at bus j derived from [16], the GGDF Dmj representing power on line m due to power generation at bus j is derived as follows [17]:
Dmj = Dmr + Amj , j r

2) Generation limit constraints


Pi ,minU it Pi t Pi , maxU it

(14)

3) Transmission constraint as in (6) and (7). where Uit has been determined from the EMO for the unit scheduling. 3.3.2. ALH FOR SOLVING TED

Augmented Lagrangian function is formulated for the problem:


L = a i + bi Pi t + c i ( Pi t ) 2 U it
t =1 i =1 2 N N t t 1 t t t t + PD Pi + PD Pi t =1 i =1 i =1 2 T N t t t m Pm D mi Pi T NL i =1 + 2 N t =1 m =1 t + 1 m Pm D mi Pi t t 2 i =1 T N

(15)

(10)

where r is reference bus, and


NG (0) NG (0) ( = Pm Amj PGj / PGj0 ) j =1 j =1 jr

Dmr

(11)

The energy function of the ALH is defined based on the augmented Lagrangian relaxation as follows

and power flow Pm(0) on line m is calculated corresponding to total generation PGj(0) at each bus j by DC power flow method [16]. 3.3. ALH FOR TED PROBLEM

E = a i + biVi ,tH + c i (Vi t, H ) 2 U it


t =1 i =1 2 N N t t 1 t t t t + V PD Vi , H + PD Vi , H t =1 i =1 i =1 2 T N t t t Vm Vm , H DmiVi , H T NL i =1 + 2 1 N t =1 m =1 t + m V m , H DmiVi t, H t 2 i =1 N L Vm , H N Vi , H + g 1 (V ) dV + g 1 (V )dV t =1 i =1 0 m =1 0
t t

The ALH is applied to solve the transmission constrained economic dispatch (TED) after obtaining result of unit scheduling from EMO. The ALH is a continuous Hopfield neural network whose energy function is based on augmented Lagrangian relaxation. The proposed ALH is augmented by Hopfield terms [18] and solved by gradient descent method. Moreover, the sigmoid function used for continuous neurons of the ALH is from [19], and proof of convergence is given in [20]. 3.3.1. FORMULATION OF TED PROBLEM

(16)

The objective of the TED problem is to minimize the operation cost neglecting start up costs while satisfying power balance constraints, generator operating limits and

where the sums of integral terms are Hopfield terms in which their global effect is a displacement of solutions toward the interior of the state space [18]. Dynamics of neurons are defined as follows

dU it, H dt

E = Vi ,tH

N t t t t t bi + 2c iVi , H V + PD Vk , H k =1 = NL N t t t Vm + m V m , H Dmk V kt, H Dmi + U it, H k =1 m =1 (17)

If there are M t congested lines at hour t, the congestion index Iit of uncommitted unit i is calculated as follows [21]:
I it = Pi ,max (1 U it ) Mi

D
m =1

Mt

mi

t Pm

(27)

where
t t Pm ,max Pm if Pm > Pm ,max t t t Pm = Pm ,max Pm if Pm < Pm ,max 0 otherwise

t dU m , H

dt

E t Vm , H
N

(28)

(18)

t t t = Vtm + m V m , H Dmk Vkt, H + U m , H k =1


t N dU E t =+ = PD Vi ,tH t dt V i =1

(19)

dU tm dt

=+

N E t = V m , H DmiVi ,tH t Vm i =1

(20)

Updating for inputs of neurons at iteration n is defined as follows


n n U it,(H ) = U it,(H 1) i

E Vi ,tH E t Vm , H E Vt E Vtm

(21)

Negative Dmi and positive Pmt or vice versa imply that committing unit i will help to decrease the power flow on line m, otherwise committing unit i will increase the power flow on line m. Nonzero Pmt means that the overloaded line m at hour t must be relieved. Therefore, the positive term DmiPmt implies that committing unit i may help alleviating overload on line m at hour t. This term is enhanced by Pi,max(1-Uit)/Mi to represent capacity of unit i to relieve congested line m in the least cost manner. Thus, equation (27) shows a capacity of unit i to alleviate all congested lines at hour m, in which the higher value of Iit, the better effects on alleviating congestion. At each congested hour, after the unit with highest Iit committed to relieve overloaded lines, the TED is solved by quadratic programming to check transmission violation. If no feasible solution found, the next unit in Iit will be committed until all congested lines are relieved. When all congested hours have been relieved, a repairing procedure is conducted to satisfy minimum up and down time constraints. 3.5. OVERALL PROCEDURE

t t U m(,nH) = U m(,nH1) m

(22)

t t U ( n ) = U ( n 1) +

(23)

( ( U tmn ) = U tmn 1) + m

(24)

Outputs of continuous neurons are determined as follows [19]:


Vi ,tH = g (U it, H ) Pi ,tmax Pi ,tmin = 2
t t Vm , H = g (U m , H )

Overall procedure for the EMO-ALH solving transmission constrained unit commitment problem is as follows: Step 1: Apply EMO to solve unit scheduling as in Section 3.1. Step 2: Solve economic dispatch problem without transmission constraint by the ALH as in Section 3.3.

1 + tanh U it, H + Pi ,tmin

)]

(25)

Pm ,max Pm , min = 2

t 1 + tanh U m , H + Pm ,min

)]

(26)

Step 3: Check for transmission constraint. If transmission constraint is satisfied, go to Step 10. Step 4: Solve TED problem by the ALH as in Section 3.3. Step 5: If a feasible solution is found, go to Step 10. Step 6: At each hour having insufficient power, commit one unit based on unit congestion index in Section 3.4 and eliminate this unit from the congestion index. Step 7: Check power balance and transmission constraint by quadratic programming. Step 8: If the transmission constraint is still violated,

where Pm,min = -Pm,max. Outputs of multiplier neurons are equally defined to their total inputs. Selection of parameters, initialization, termination criteria and procedure of the ALH was given in [15]. 3.4. CONGESTION MANAGEMENT

Rescheduling is performed when a solution of the TED problem is not feasible by committing units from congestion index based on evaluating their capacity to relieve all congested lines at congested hours.

return to Step 6. Step 9: Solve final TED problem by the ALH again to find feasible solution. Step 10: Calculate total production cost and stop. The main difference between the EMO-ALH here and the one in [15] is that the latter does not consider transmission constraint. The EMO used here introduces unit substitution to reduce excessive spinning reserve, thereby leading to cost savings. 4. NUMERICAL RESULTS

The generating unit data, load demand and transmission limits are given in [25]. The total production cost and computational time comparison for the test system between the proposed method and TAS methods in [13] are given in Table 2. The proposed method obtains less expensive cost and faster computing time than the TAS. Note the computer used for the TAS was HP 700 workstation. 5.

CONCLUSIONS

The proposed EMO-ALH is tested on the 24-bus IEEE reliability test system with 26-unit and 32-unit cases. For both cases, the percentage of system load at each bus as well as the line data is given in [22]. The algorithm for solving the test systems is implemented on Matlab 6.5 platform and run on 1.1 GHz Celeron, 128-MB of RAM PC. 4.1. CASE 1

In this paper, the enhanced merit order and augmented Lagrange Hopfield network is simple and efficient for solving the unit commitment problem with transmission constraint. The proposed method obtains better solution quality than decomposition and co-ordination algorithms and three-phase algorithm scheme for the 24-bus IEEE reliability tests system for 26-unit and 32-unit cases, respectively.

REFERENCES
[1] N. P. Padhy, Unit commitment - A biography survey, IEEE Trans. Power Systems, vol. 19, pp. 1196-1205, 2004. F. N. Lee, The application of commitment utilization factor (CUF) to thermal unit commitments, IEEE Trans. Power Systems, vol. 6, pp. 691-698, 1991. C. L. Chen and S. L. Chen, Short-term unit commitment with simplified economic dispatch, Electric Power Systems Research, vol. 21, pp. 115-120, 1991. T. S. Dillon, K. W. Edwin, H. D. Kochs, and R. J. Taud, Integer programming approach to the problem of optimal unit commitment with probabilistic reserve determination, IEEE Trans. Power App. Syst., vol. PAS-97, pp. 2154-2166, 1978. A. J. Svoboda, C.-L. Tseng, C. A. Li, and R. B. Johnson, Short-term resource scheduling with ramp rate constraints, IEEE Trans. Power Systems, vol. 12, pp. 77-83, 1997. H. Sasaki, M. Watanabe, and R. Yokoyama, A solution method of unit commitment by artificial neural networks, IEEE Trans. Power Systems, vol. 7, pp. 974-981, 1992. A. H. Mantawy, Y. L. Abdel-Magid, and S. Z. Selim, A simulated annealing algorithm for unit commitment, IEEE Trans. Power Systems, vol. 13, pp. 197-204, 1998. A. H. Mantawy, Y. L. Abdel-Magid, and S. Z. Selim, A unit commitment by tabu search, IEE Proc. Gener., Transm. Distrib., vol. 145, pp. 197-204, , 1998. S. A. Kazarlis, A. G. Bakirtzis, and V. Petridis, A genetic algorithm solution to the unit commitment problem, IEEE Trans. Power Systems, vol. 11, pp. 83-92, 1996. K. A. Juste, H. Kita, E. Tanaka, and J. Hasegawa, An evolutionary programming

The unit characteristics and load demand in this case are given in [23]. The system spinning reserve is set to 5% of load demand. Table 1 shows the comparison of total production cost and computational time from the proposed method with those from DCA [14]. The proposed method obtains less expensive total production cost and faster computational time than those from DCA. Note the computational time obtained in DCA was from a Pentium-100 PC.
Table 1: Comparison of total production cost and CPU time for 26-unit test system

[2]

[3]

[4]

Method DCA [14] EMO-ALH

Total cost ($) 749,537.3 742,864.5

CPU time (s) 193 2.18 [5]

4.2.

CASE 2 [6]

Table 2: Comparison of total production cost and CPU time for 32-unit test system

Problem

Method

Total cost ($) 898,683 897,997 901,289

CPU time (s) 6.41 0.41 37.88

[7]

Neglecting transmission constraint With transmission constraint

TAS [13] EMO-ALH Indirect TAS [13] Direct TAS [13] EMO-ALH

[8]

[9] 900,206 899,240 19.86 2.53 [10]

solution to the unit commitment problem, IEEE Trans. Power Systems, vol. 14, pp. 14521459, 1999. [11] C. K. Pang, G. B. Sheble, and F. Albuyeh, Evolution of dynamic programming based methods and multiple area representation for thermal unit commitments, IEEE Trans. Power App. Syst., vol. PAS-100, pp. 12121218, 1981. J. J. Shaw, A direct method for securityconstrained unit commitment, IEEE Trans. Power Systems, vol. 10, pp. 1329-1341, 1995. C.-L. Tseng, X. Guan, and A. J. Svoboda, Multi-area unit commitment for large scale power systems, IEE Proc. Gener., Transm. Distrib., vol. 145, pp. 415-422, 1998. Q. Xia, Y. H. Song, B. Zhang, C. Kang, and N. Xiang, Effective decomposition and coordination algorithms for unit commitment and economic dispatch with security constraint, Electric Power Systems Research, vol. 53, pp. 39-45, 2000. V. N. Dieu and W. Ongsakul, Enhanced Merit Order and Augmented Lagrange Hopfield Network for Unit Commitment, Proc. of 15th Power System Computational Conference 2005, Liege, 2005. A. J. Wood and B. F. Wollenberg, Power generation, operation & control. 2nd ed., New York: John Wiley & Sons, 1996. W. Y. Ng, Generalized generation distribution factors for power system security evaluations, IEEE Trans. Power App. Syst., vol. PAS-100, pp. 1001-1005, 1981.

[18]

J. van den Berg and J. C. Bioch, Constrained optimization with a continuous HopfieldLagrange model, Technical report EUR-CS93-10, Erasmus University Rotterdam, Comp. Sc. Dept., Faculty of Economics, 1993. J. H. Park, Y. S. Kim, I. K. Eom, and K. Y. Lee, Economic load dispatch for piecewise quadratic cost function using Hopfield neural network, IEEE Trans. Power Syst., vol. 8, pp. 1030-1038, 1993. V. N. Dieu and W. Ongsakul, Augmented Lagrange Hopfield network for multi-objective fuel constrained economic dispatch, Proc. of Energy for Sustainable Development: Prospects and Issues for Asia, Phuket, Thailand, 2005. W. Ongsakul, and N. Petcharaks, Transmission and ramp rate constrained unit commitment using enhanced adaptive Lagrangian relaxation, Proc. 2005 IEEE St. Petersburg PowerTech, St. Petersburg, 2005. IEEE Reliability Test System Task Force, IEEE Reliability Test System, IEEE Trans. Power Systems, vol. PAS-98, pp. 2047-2054, 1979. S. J. Wang, S. M. Shahidehpour, D. S. Kirschen, S. Mokhtari,, and G. D. Irisarri, Short-term generation scheduling with transmission and environmental constraint using an augmented Lagrangian relaxation, IEEE Trans. Power Systems, vol. 10, pp. 12941301, 1995. C. L. Tseng, S. S. Oren, C. S. Cheng, C. A. Li, A. J. Svoboda, and R. B. Johnson, A transmission constrained unit commitment method, Proc. 31st Hawaii international conference on Systems science, Hawaii, 1998.

[19]

[12]

[20]

[13]

[21]

[14]

[22]

[15]

[23]

[16]

[17]

[24]

Вам также может понравиться