Sei sulla pagina 1di 6

Fourth International Conference on Natural Computation

Short-Term Hydropower Optimal Scheduling of Multireservoir System Using a Decomposition Approach


Shuangquan Liu1 School of Hydropower & Information Engineering, Huazhong University of Science & Technology, Wuhan, P. R. China liushuangquqan@gmail.com Abstract
A decomposition approach combined with linear approximation method is applied to the short-term hydropower optimal scheduling (STHOS) problem of multireservoir system. The STHOS problem, which is a mixed, nonlinear and nonconvex optimization problem with discrete variables, is decomposed into a master programming problem and some sub programming problems in small scale, and the subproblems coordinate with each other through the master one. Then the STHOS problem is solved in a two-level optimization structure. The model developed in this study is capable of handling the complex objectives and constraints of STHOS, and also takes pump-storage plant into account. Simulation result demonstrates that the proposed approach is efficient.
1

Jinwen Wang* 1, 2 1 School of Hydropower & Information Engineering, Huazhong University of Science & Technology, Wuhan, P. R. China 2 Applied Mathematics & Industrial Engineering, cole Polytechnique de Montral, Montreal (QC), Canada Dr.Jinwen.Wang@gmail.com

Zhong Liu3 School of Energy & Power Engineering, Changsha University of Science & Technology, Changsha, P. R. China
3

1. Introduction
Operation optimization of multireservoir system is a challenging problem due to the fact that it is a large scale, highly nonlinear and nonconvex problem with significant potential economic, social and environmental impacts[1, 2]. A multireservoir system is usually operated with multiple purposes, including flood control, hydropower generation, water supply, navigation, etc. During the past decades, many optimization techniques were adopted in the field of multireservoir system operation and management. Piekutowski et al. formulated the optimal hydro scheduling problem as a linear programming problem[3], Franco et al. developed a network flow programming (NFP) model which applied a linear-quadratic penalty approach to enforcing the
* Corresponding author. Tel: 86-27-87556527

coupling between hydro and electric variables in a hydro-dominated power system[4]. Yang and Chen presented a technique that combined multipass dynamic programming technique with successive approximations to solve the daily hydrothermal coordination problem[5]. Ngundam et al. employed lagrangian multipliers to decompose a large scale hydrothermal problem into small independent subproblems[6]. Wu et al. proposed a diploid genotype based genetic algorithm (GA) to solve the STHOS problem, used a pair of binary strings with same length to represent a solution to the problem[7]. Yu et al. applied the particle swarm optimization (PSO) technique to solve the STHOS problem[8], and verified that a local version of PSO with inertia weight was more efficient compared with other evolutionary algorithms. Yet all these approaches are confined to a small application scope when it comes to multireservoir system. This paper describes a decomposition approach combined with linear approximation to solve the STHOS of multireservoir system. The proposed approach decomposes the nonlinear STHOS problem into a master programming problem and some sub programming problems in small scale, the master problem is solved by a decomposition approach combined with linear approximation method and the sub ones are solved by linear programming (LP) method. Then the schedule of operational regions for each plant is determined by a dynamic programming (DP) procedure. The constraints taken into account include the upper and lower bounds of reservoir storage, reservoir release, plant-based power and hydro systemwide power. Other involved elements include water traveling time, plant-based power generation characteristics, power ramp, minimum startup/shutdown time, maximum startup/shutdown frequency, prohibitive operational regions, and so on.

978-0-7695-3304-9/08 $25.00 2008 IEEE DOI 10.1109/ICNC.2008.468

565

2. Problem formulation
2.1. Notation
i, j, t hit Iit vit Qit Qitmin Qitmax qit qjt qjtqjt+ qj-,max qj+,max pj-(.) pj+(.) pit pjt pjtpjt+ pitmin pitmax pts,min pts,max split viini viend zitu , zitd qjd(k) qju(k) jt t respectively indexes of reservoir, plant and time interval average hydraulic head of reservoir i during t local inflow to reservoir i during t storage of reservoir i at the beginning of t release of reservoir i during t minimum release of reservoir i during t maximum release of reservoir i during t generation discharge of reservoir i during t generation discharge of plant j during t pumping flow of plant j during t generation discharge of plant j during t maximum pumping flow of plant j maximum generation discharge of plant j pumping power function of plant j power generation function of plant j power of reservoir i during t power of plant j during t pumping power of plant j during t power generation of plant j during t minimum power of reservoir i during t maximum power of reservoir i during t minimum hydro systemwide power during t maximum hydro systemwide power during t spillage discharge of reservoir i during t the beginning-of-study storage of reservoir i the end-of-study storage of reservoir i respectively forebay and downstream water levels of reservoir i at the beginning of t lower bound of k operational region of plant j upper bound of k operational region of plant j energy price of plant j during t energy purchase price in t transportation cost charged by electric net on plant j during t number of time intervals during the study horizon generation efficiency of reservoir i set of immediate upstream reservoirs of reservoir i set of plants belonging to reservoir i set of general plants in contrast to pump-storage plants startup/shutdown decision variable of plant j during t: 0shutdown, 1startup number of time intervals of keeping generating, pumping or shutdown of plant j during t startup number of plant j by t water traveling time between reservoir i and its

immediate downstream reservoir maximum power ramp of reservoir i

2.2. Objective function


Given the initial and the end-of-study storages of reservoirs[9], the STHOS problem is to allocate the water discharge among time intervals to maximize the generation revenue minus costs of transportation service and pumping water of reservoirs/plants during the study horizon while satisfying the involved constraints. The objective function is formulated as follows: max ( v, q + , q , p ,spl, z ) j j jt (1) + = max p j ( jt jt ) p t jt
j ,t j ,t

2.3. Constrains
The involved constraints include: water balance equations (2) vi ,t +1 = vi ,t + Qk ,t i + I it qit split
k i

where Qit = qit + split qit = q jt ji + q jt = q jt q jt q jt = 0 ( j ) + q jt q jt = 0 reservoir storage limits min max vit vit vit
vi 0 = v
ini i end i

(3)

(4) (5) (6) (7) (8)

jt
T i(h) i i wjt xjt yjt i

viT = v reservoir release limits min max Qit Qit Qit plant-based power limits min max pit pit pit where pit = p jt ji + p jt = p jt p jt p jt = 0 ( j ) reservoir-based power ramp limits pi ,t +1 pit i

(9)

(10)

hydro systemwide power limits

566

pts ,min p + p pts ,max jt jt


j j

(11)

min z = f jk jk
j =1

nk

(21)

plant-based power generation characteristics p + = q + i+ (hit ) ( j i ) jt jt

(12)

p = q i (hit ) ( j i ) (13) jt jt where u d hit = 0.5 ( ziu,t +1 + zit ) zit + + ,max (14) q jt q j (hit ) ( j i ) ,max q jt q j (hit ) ( j i ) plant-based operational region limits q dj (k ) ( z jt k ) q jt q uj (k ) ( z jt k ) (15)
k k 1

where (z) is a pulse function 1 ( z = 0) ( z) = (16) 0 else and plant-based minimum startup/shutdown time and maximum startup/shutdown frequency limits.

p nk R jk jk = b1 k =1 j =1 nk (22) jk = 1, k = 1, p j =1 0 jk where fjk=Ck xkj, Rjk=A1k xkj. And problem with (21) subject to (22), which is equivalent to that with (17) subject to (18) and (19), is called the master programming problem. Provide a basic feasible matrix B is derived, and let the associated multipliers be: = ( 1 , 01 , 02 , 0 p ) (23) where 1 is a row vector whose dimension equals to the number of equations (18), 0k is a value, the multiplier associated with the second expression of (22). Apparently, according to simplex method (SM), the condition that the derived matrix B is the optimal one is that the reduced cost of any xkj is nonnegative, that is: R f jk jk = (Ck 1 A1k ) x kj 0 k 0 (24) ek which is equivalent to that the following sub programming problems min (Ck 1 A1k )x k (25) A 2k x k = b 2 k , (k = 1, 2, , p) x 0 k have optimal solutions and the associated objective value of each sub problem is greater than 0k. If any of the sub problems of (25) has no feasible solution, the initial problem (17)(19) obviously has no feasible solution either. Otherwise, if there is optimal solution xks, and (Ck 1 A1k) xks<0k, then in the master problem a better solution can be derived by a pivot operation: A xs Ps = B 1 1k k (26) ek As a result, the initial problem is now decomposed into a master problem and p sub problems, with every sub problem coordinating with each other through the master one. With this approach, the nonlinear STHOS problem is to be solved in a two-level optimization structure.

3. Solution method
3.1. Mathematical preparation
With the successive linear programming approach[10], the continuous nonlinear programming problem can be solved by successively solving a series of linear programming problems, which are supposed to have the following structure: min z = C1x1 + C2 x 2 + + C p x p (17)
Subject to: A11x1 + A12 x 2 +

+ A1 p x p = b1

(18)

= b 21 A 21x1 = b 22 A 22 x 2 (19) A2 p x p = b2 p xp 0 x1 0, x 2 0, where Aik is matrix, xk, Ck, b1 and b2k are vectors with certain dimensions. This problem has been dealt with in the previous work by Dantzig and Wolfe[11]. Assume that S2k ={ xk |Aik xk =b2k, xk0} (k=1,2,,p) is a boundary closed set, and given all the basic feasible solutions of S2k (xk(1), xk(2), , xk(n )), the following expression is derived for any xkS2k: nk x k = jk x kj j =1 (20) nk = 1, 0 jk jk j =1 Substitute (20) into (17) and (18), we have:
k

3.2. Reservoir system operation


In this level, without considering the discrete variables such as the plant-based prohibitive operational regions,

567

startup/shutdown time and frequency, the nonlinearity comes from the reservoir and plant-based power generation characteristics. The nonlinear expressions are linearized by approximating them with the first order Taylor series expansion at current solution. With linear approximation, the initial problem is turned into a linear optimization problem with only linear constraints and can be decomposed into a master problem and several sub problems with a small scale of reservoirs with the proposed decomposition approach. SM is used to solve the sub problems, and then the solution of master problem is updated. The deviation from the initial problem caused by the linearization of objective and constraints is minimized by successive linear approximation. With respect to the variables with both lower and upper bounds involved in the sub problems, one of the feasible ways, which is computationally expensive nevertheless, is to incorporate artificial variables into the upper constraints and employ SM to get the optimal solution. Therefore SM is modified here to remove the explicit expression of these upper constraints, which is found to be efficient in decreasing the CPU time and energy management system (EMS) memory, detailed in [12].

repeated once for each plant until the derived solutions are convergent.

3.4. Coordination of two levels


Reservoir system operation (the first level) and schedule of plant-based operational regions (the second level) react on each other through coordination, in which the first level optimization directly impacts the second level while results of the second level optimization are fed back to the first level to improve the objective function.

4. Simulation result
In this study, we consider an example of 7-reservoirs system, which consists of 1 pump-storage plant and 6 general plants with a total of 33 units and an installed capacity of 4,170 MW. The initial (zini), end-of-study (zend), minimum (zmin) and maximum (zmax) forebay water levels and regulation characteristics of each reservoir are listed in Table 1, and the topology map of reservoirs is given in Figure 1. The study horizon is set to be 1 day with 96 time intervals. The proposed algorithm is implemented on a PC with Celeron 2.0 GHz and 512M RAM. According to the simulation result, there is no spillage discharge from any reservoir during the study horizon. The illustrative results are given in Figure 2Figure 4. As seen in Figure 2, the energy production in total of the multi-reservoir/plant system during the study horizon is significantly affected by the energy price, that is, more energy is generated during the energy price peak time than that during the energy price valley time to maximize the hydropower revenue. And the power generation of each plant is smooth due to the limits on power ramp and hydro systemwide power. Figure 3 illustrates that the pump-storage plant takes advantage of the low price energy to pump more water and then generates valuable energy at high energy price, and contributes to the hydropower revenue maximization. Figure 4 shows that the forebay water level of reservoir 1 barely changes during a day since it is a reservoir with multi-yearly live active storage. The pump-storage reservoir takes advantage of the lower price energy to elevate the forebay water level through pumping water and then produces more energy with a higher hydraulic head during the following high energy price time. The water level trends of reservoir 2, 3 and 4 are similar in that the water is kept to elevate the forebay water level during the lower energy price time in order to store the energy and then released to generate during the high energy price time. Reservoir 5 and 6 are reservoirs with daily-live active storage which are not capable of handling the water inflow that exceeds their storage capacities, so the release of reservoir 5 is adjusted according to the water inflow to

3.3. Schedule of plant operational regions


The optimal release of each reservoir Qit* is derived after the reservoir system operation optimization. Consequently, the hydraulic head hit* and the optimal generation discharge q+jt* or pumping flow q-jt* (pump-storage plant only) of each plant are preliminarily determined, and then the schedule problem of plant-based operational regions can be formulated as the following DP problem: Fjt ( x jt , y jt ) = max q + q + + Fj ,t +1 ( x j ,t +1 , y j ,t +1 ) (27) jt jt +
q jt , q jt

with the boundary condition: FjT ( x jT , y jT ) = 0 state transition equation: y j ,t +1 = y jt + w jt ( x jt , q + q + ) jt jt


jt + jt

(28) (29)

x j ,t +1 = g ( x jt , q + q ) (30) and subject to (3), (15), plant-based minimum startup/shutdown time, maximum up/down frequency, and * + q | Qit + (qmt qmt ) | ( j , m i ) (31) jt
+ q Q + (qmt qmt ) + jt

* it

( j , m i )

(32)

The preliminary q+jt* or q-jt* derived in reservoir system optimization is updated by solving this DP problem, and then the schedule of plant-based operational regions is determined. In case that one reservoir consists of more than one plant, the optimization procedure is

568

coordinate its hydro production with the energy price. As the last one of the cascaded reservoirs, there is a drawdown in reservoir 6 during the several beginning time intervals with lower energy price time in order to avoid the possible spillage during the late day. Table 1. Initial data of reservoirs zend zmin zmax zini active-live reservoir storage (m) 1 450 450 425 474 multi-yearly incomplete 2 240 240 238 254 seasonal 3 187 187 186 194 weekly 4 97 97 90 98 seasonal 5 65.5 65.5 65 72 daily 6 49.5 49.5 49 51 daily pump-storage 320 320 300 325 seasonal
p-s denotes the pumpstorage reservoir 1 3 2

(MW) 450 300 150 0 -150 -300 -450

($/MWh) 400 300 200 100 0

(m) 450.5 450 449.5


325 323 321 319
241 240 239

187.5 187 186.5 97.5 97 96.5

4 5 6 p-s

66 65.5 65

50 49.5 49

Figure 1. Topology map of reservoirs


(MW) 3000 2500 2000 1500 1000 500 0 ($/MWh) 400 300 200 7 100 0

Figure 4. Forebay water levels of each reservoir

5. Conclusion
In this paper, a decomposition approach combined with successive programming method is applied to the STHOS problem. The model developed in this paper is capable of handling the complex objectives and constraints of multireservoir system and taking pump-storage plant into account. The model is solved in a two-level optimization structure: reservoir system operation and schedule of plant-based operational regions, which are respectively solved by the decomposition approach and DP. Results of the second level depend on those of the first level and are fed back to the first level to improve the solution of STHOS. With an example of multireservoir system with pump-storage plant involved, it can be verified that the

Figure 2. Quarter-hourly schedules of hydropower system

0:00 1:30 3:00 4:30 6:00 7:30 9:00 10:30 12:00 13:30 15:00 16:30 18:00 19:30 21:00 22:30

plant 1 plant 3 plant 5 pump-storage plant

plant 2 plant 4 plant 6 energy price

569

0:00 1:30 3:00 4:30 6:00 7:30 9:00 10:30 12:00 13:30 15:00 16:30 18:00 19:30 21:00 22:30 0:00

0:00 1:30 3:00 4:30 6:00 7:30 9:00 10:30 12:00 13:30 15:00 16:30 18:00 19:30 21:00 22:30

Figure 3. Quarter-hourly schedules of the pump-storage plant

reservoir 1

pump-storage reservoir
reservoir 2

reservoir 3 reservoir 4 reservoir 5

reservoir 6

proposed algorithm is efficient to solve the STHOS problem.

6. Acknowledgements
This paper is supported by the National Natural Science Foundation of China under Grant Nos. 50539140, 50579022.

Reference
[1] W.W.-G. Yeh, "Reservoir Management and Operations Model: A State-of-the-Art Review", Water Resources Research, 1985, 21(12): pp. 1797-1818. [2] J.W. Labadie, "Optimal operation of multireservoir systems: State-of-the-art review", Journal of Water Resources Planning and Management, 2004, 130(2): pp. 93-111. [3] M.R. Piekutowski, T. Litwinowicz, and R.J. Frowd, "Optimal short-term scheduling for a large-scale cascaded hydro system", IEEE Transactions on Power Systems, 1994, 9(2): pp. 805-811. [4] P.E.C. Franco, M.F. Carvalho, and S. Soares, "Network flow model for short-term hydro-dominated hydrothermal scheduling problems", IEEE Transactions on Power Systems, 1994, 9(2): pp. 1016-1022. [5] J.-S. Yang and N. Chen, "Short term hydrothermal coordination using multi-pass dynamic programming", IEEE Transactions on Power Systems, 1989, 4(3): pp. 1050-1056. [6] J.M. Ngundam, F. Kenfack, and T.T. Tatietse, "Optimal scheduling of large-scale hydrothermal power systems using the Lagrangian relaxation technique", International Journal of Electrical Power and Energy System, 2000, 22(4): pp. 237-245. [7] Y.-G. Wu, C.-Y. Ho, and D.-Y. Wang, "A diploid genetic approach to short-term scheduling of hydro-thermal system", IEEE Transactions on Power Systems, 2000, 15(4): pp. 1268-1274. [8] B. Yu, X. Yuan, and J. Wang, "Short-term hydro-thermal scheduling using particle swarm optimization method", Energy Conversion and Management, 2007, 48(7): pp. 1902-1908. [9] J. Wang, X. Yuan, and Y. Zhang, "Short-term scheduling of large-scale hydropower systems for energy maximization", Journal of Water Resources Planning and Management, 2004, 130(3): pp. 198-205. [10] M. Frank and P. Wolfe, "An algorithm for quadratic programming", Naval Research Logistics Quarterly, 1956, 3: pp. 95-110. [11] G.B. Dantzig and P. Wolfe, "Decomposition principle for linear programs", Operation Research, 1960, 8(1): pp. 101-111. [12] Y. Yu, L. Yang, C. Deng, et al., Theory and algorithm of mathematical programming. 2nd ed. Press of Huazhong University of Sci. & Tech. Wuhan, China: 1993, pp. 134-140.

570

Potrebbero piacerti anche