Electrical Power and Energy Systems 24 (2002) 583590

www.elsevier.com/locate/ijepes

Short term hydro scheduling using two-phase neural network

R. Naresh a,*, J. Sharma b
a

Department of Electrical Engineering, Regional Engineering College, Hamirpur, H.P. 177005, India b Department of Electrical Engineering, University of Roorkee, Roorkee 247667, India Received 24 January 1999; revised 27 June 2000; accepted 6 August 2001

Abstract This paper presents a solution technique based on two-phase neural network for short term scheduling of hydropower generations. The proposed method is based on solution of a set of differential equations realized from transformation of an augmented lagrangian energy function. Here the objective function is to minimize the nonhydraulic power production expenses over the scheduling period. A multireservoir cascaded hydroelectric system with a nonlinear power generation function of water discharge rate and storage volume is taken into account for implementation. The water transportation delay between cascaded reservoirs is considered. Results concerning this method are compared with those achieved from successive solution of augmented lagrangian function utilizing familiar conjugate gradient method. It is shown that proposed solution approach with suitable selection of control parameters is capable of generating a good optimal solution. q 2002 Published by Elsevier Science Ltd.
Keywords: Hydro scheduling; Nonlinear generation function; Two-phase neural network

1. Introduction The short term scheduling of hydro generations has been one of the principal and difcult optimization problems in economic operation and control of interconnected multireservoir power system. In such a system, the marginal cost of hydroelectric generation is insignicant. This suggests that the water at hand in system reservoirs should be utilized for hydropower generation in such a way so that overall cost of nonhydraulic power generations is minimized along the planning horizon while satisfying diverse constraints. The usual assumption is that target end point storage levels generally conform to water release schedule earlier set up by mid-term scheduling process in which long term river ow probabilities and demand predictions are taken into account. The short-term scheduler then allocates this water for hydro generation among the various time intervals along the scheduling period to minimize the production cost while trying to satisfy different constraints. The governing constraints of the short term scheduling problem are demandsupply balance, ow balance or continuity equation, bounds on reservoir storage, bounds on water release, limits on spillage and coupling constraint that puts a boundary condition on the initial and nal reservoir levels. Supplementary constraints such as ood control,
* Corresponding author. Tel.: 191-1972-24184; fax: 191-1972-23834. E-mail address: rnaresh@patra.recham.ernet.in (R. Naresh). 0142-0615/02/$ - see front matter q 2002 Published by Elsevier Science Ltd. PII: S 0142- 061 5( 01) 00069-2

irrigation, recreation, shing and wild life maintenance etc. could be dictated depending on contractual or legal obligations in respect to hydro system network. The common assumption is that load demand and river inows are known. A large number of researchers have extensively investigated the short-term hydro scheduling problem. The major computational techniques that have been employed are maximum principal [1], variational calculus [2,3], dynamic programming [48], functional analysis [9], network ow and linear programming [1012], nonlinear programming (NLP) [13], mathematical decomposition [1417], progressive optimality algorithm [18] and heuristics, expert systems, evolutionary programming, articial neural networks and genetic algorithm [1923]. These methods have one or the other limitation such as dimensionality difculty, large memory requirement and inability to handle nonlinear cost function. Hopeld neural network was used in Ref. [20]; however, heuristics were essentially required as the method did not yield a solution in the feasible region. The two-phase neural network is a time evolving dynamical system, which is described in continuous output space and is capable to provide optimal solution in the feasible region. The motion of the network is in the direction of lower computational energies exhibited by the system. This computational methodology contracts the probable solution space and yields efcient and correct solutions. The objective function and constraints of hydro scheduling problem are mapped into a closed loop two-phase

584

R. Naresh, J. Sharma / Electrical Power and Energy Systems 24 (2002) 583590

network. In phase 1, whenever constraint violation takes place, its magnitude and direction are fed back to adjust the states of the neurons. This brings the trajectory near the boundary of the feasible region. In the second phase, the directional vector of the constraints is slowly shifted to the corresponding lagrange multipliers and this moves the solution trajectory to the feasible region. In this way, the overall energy of the network is continuously decreasing until it acquires a minimum. Corresponding to this minimum energy, the states of the neurons are resolved to be a minimizer of the original problem. In this work, the states of the neurons are taken as scheduled discharge for each hydroplant. The network also furnishes lagrange multiplier associated with each constraint, automatically. It is emphasized that the purpose of the present work is not to devise the constrained nonlinear optimization technique, but rather to verify the optimal scheduling results obtained by the proposed method with those from conventional and rigorous nonlinear optimization procedure. Thus a constrained NLP based scheduling of multi-reservoir hydro system is introduced rst and then adaptation of twophase neural network for the constrained optimization is discussed, which will then lend itself to suitable hardware implementation on transputers. The outcome will be an extremely fast parallel processor for obtaining optimal solution. 2. Problem formulation Notation F Phtj xtj qtj utj Pd t Et composite cost function. power generation of jth hydro plant at time t. storage volume of jth reservoir at time t. water discharge rate of jth reservoir at t. turbine discharge rate at jth reservoir at t. load demand at time t. error between load demand and total hydro generation at time t. Ptl total transmission losses at hour t. vtj volume of spilled from jth reservoir at time t. ytj side inow of jth reservoir at time t. ztj inow from upstream plants to reservoir j at time t. Wj total available water in reservoir j over the complete time horizon. C j1 ; ; Cj6 power generation coefcients at jth hydro plant. tk water transport delay from reservoir k to reservoir j j. Nu upstream plant index. Nh number of hydro plants. T total time horizon. xj upper bound on storage of reservoir number j. xj lower bound on storage volume of reservoir number j.

qj qj uj  uj xbegin j xend j

upper bound on water discharge rate of reservoir number j. lower bound on water discharge rate of reservoir number j. maximum allowable water discharge rate through turbine at reservoir number j. minimum allowable water discharge rate through turbine at reservoir number j. initial storage volume of jth reservoir. storage volume of jth reservoir at the end of planning horizon.

The power system discussed here is composed of a cascaded multi-reservoir system and interconnection lines to the neighboring utilities through which electric power may be exchanged. A general system representation for jth reservoir is shown in Fig. 1. The problem of short term scheduling hydro generations in such a system can be stated as to nd out the water release from each reservoir and through each power house over all the planning time intervals so as to minimize the total cost of energy import or energy export while satisfying diverse hydraulic and load balance constraints. Here the energy import refers to the shortage in hydro generated energy to meet the load demand. On the other hand, the energy export is stated as the surplus of hydro generated energy over the load demand. Typically, the total planning period is one day or one week and time interval is one hour. This is a new contribution of our earlier work [25] wherein the objective was to minimize the cost of thermal generation explicitly and in Ref. [26], a multi-objective problem that maximizes hydropower generation and satises irrigation requirements as far as possible was considered. 2.1. Objective function and constraints The objective is to minimize the summation of the production cost of energy import or export over the scheduling period. The energy cost is a quadratic function [18] of the error between the demand and hydropower generation in any time interval t. Mathematically, the cost function is

Fig. 1. Reservoir representation.

R. Naresh, J. Sharma / Electrical Power and Energy Systems 24 (2002) 583590

585

expressed as F min
T t1

2.2. Hydropower generation characteristics 1 Hydropower generation is a function of net head and turbine discharge. Constant water head is generally assumed in short period scheduling formulation. However, this assumption is true only in case of large capacity reservoirs. Head variation cannot be ignored if there is strong relationship between inow and capacity. Since net head is a function of volume of stored water, hydropower generation can be written in terms of turbine discharge rate and storage, and the frequently used expression is [16,22] Phtj C1j xtj 2 1 C2j utj 2 1 C3j xtj utj 1 C4j xtj 1 C5j utj 1 C6j; 3. Problem reformulation The aim to reformulate the problem is to express the objective function (1) entirely in terms of reservoir variables and constraints in the scheduling problem considerably. The modeling of spill in Eq. (4) permits to use reservoir release qtj in place of utj in the hydropower generation function (7) Phtj C1j xtj 2 1 C2j qtj 2 1 C3j xtj qtj 1 C4j xtj 1 C5j qtj 1 C6j ; j [ Nh 8 j [ Nh 7

Et 2

subject to the following constraints. 2.1.1. System active load balance In a power system, the total power generated is equal to the total power demand including losses in each time interval. This equality interval t is expressed as
Nh i1

Phti

1 E Pd 1 Pl

2.1.2. Reservoir ow balance Reservoir ow balance or continuity equation is expressed as equality constraint. In this constraint, water transportation delays between reservoirs are taken into consideration. As is evident from Fig. 1 for jth reservoir, the ow balance equation relates the previous interval storage with the storage, net inow and net outow during the current time interval. Mathematically, the equality is expressed as t2tm t2tm xtj xt21 1 ytj 2 utj 2 vtj 1 um j 1 vm j 3 j
m[Nu

2.1.3. Spillage modeling Spillage is allowed only when water release from the reservoir exceeds the maximum discharge limit through the turbines. From Fig. 1, water spilled from reservoir during time interval t is written as follows ( t qj 2 uj ; if qtj . uj t 4 vj 0 otherwise 2.1.4. Physical constraints Reservoir storage volume and turbine discharge rates are bound between maximum and minimum by physical limitations. Net reservoir release may also be constrained between maximum and minimum limits. Thus, for reservoir number j, we have xj # xtj # xj ; uj # utj # uj ; qj # qtj # qj 5

That is, whenever reservoir release exceeds the maximum turbine discharge limit, qtj is replaced by uj in Eq. (8). Thus, it is evident from Eq. (8) that hydro generation is a function of storage volume and release of the reservoir. It is also easy to express storage entirely in terms of reservoir releases thus eliminating storage from Eq. (8). Using Eq. (3), the resulting expression for storage is given by H I t m l l2t d yj 1 xtj xbegin 1 qj j 2 qlj e zq 9 j
l1 m[Nu; l 2 t m .0 j

Substituting xtj from Eq. (9) into Eq. (8), we get  2   Phtj C1j zq2 1 C2j qtj 1C3j zq qtj 1 C4j zq 1 C5j qtj 1 C6j ; j [ Nh 10

Now using Eq. (2), the power (import or export) is expressed in terms of hydro generations and losses that are also a function of hydro generations, thus yielding the nonhydraulic power generation as a function of only release variables Et q Pd t 2 Plt q 2
Nh i1

2.1.5. Coupling constraint Terminal reservoir volumes are set earlier by midterm scheduling process. This constraint implies that the total quantity of available water should be used. Coupling constraint for reservoir number j is expressed as x0 xbegin ; j j xT xend j j 6

Phti q

11

Thus, the problem of reformulation yields an objective function entirely in terms of reservoir release variables and the load balance constraint is not treated explicitly but rather implicitly. The reformulated problem in compact form is

586

R. Naresh, J. Sharma / Electrical Power and Energy Systems 24 (2002) 583590

given by F min
T t1

where E q
t 2

12

_ Q

dQ 1 dQ2 dQ ; ; n dt dt dt

!T

subject to xj # xtj q # xj ; xT q xend j j In Eqs. (13) and (14), j 1; 2; ; Nh; t 1; 2; ; T: 4. Solution technique Consider a short-term hydroscheduling problem P reformulated in Section 3. min f Q subject to constraints gi Q # 0 hj Q 0 16 17 15 qj # qtj # qj 13 14

7f Q is a vector of the gradient of the objective function whose jth element is given by 2f =2Q j ; j 1; 2; n: 7g J Q is a matrix of the gradient of the inequality constraints whose elements are given by dgij 2gj ; i 1; 2; n; j 1; 2; r 2Qi

Similarly, 7hQ is a matrix of the gradient of the equality constraints. The expressions for calculating the gradient elements of the objective function and constraints can easily be derived from appropriate equations. Sign of h j for 1 # j # Nh may be either positive or negative. For t , t1 ; there is no contribution from subsystems within two large rectangles. When t $ t1 ; the dynamics of the network become _ 20 Q 27f 2 7g j sg1 1 l 2 7hsh 1 m J 1 _ l e sg J _ m esh 21 22

where Q Q1 ; Q2 ; Qn T is a vector of n variables of problem, i 1; 2; r; r 4TNh; j 1; 2; Nh and n TNh: The function f Q is a nonlinear function of variables. The vector Q corresponds to the discharge variables and is the transpose of q1 ; ; q1 ; q2 ; ; q2 ; ; qT ; ; qT : A 1 Nh 1 Nh 1 Nh two-phase neural network based on energy minimization concept [24] can be used to solve the problem P. Earlier this network has been successfully applied to short term hydro thermal scheduling [25] and hydro system scheduling [26]. The network is briey described here. The simplied block diagram of the neural network is shown in Fig. 2. The elements of the network are neurons such as integrators and constraint qualiers. This network is an extension of Refs. [27,28]. The energy function used here is the augmented lagrangian and is stated as follows [29]. La s; Q f Q 1 lT gQ 1 mT hQ  s 1 ig Qi2 1 ihQi2 1 2

where e is the controlling parameter of the network. The initial values of l s and m s are set to zero. When the network attains equilibrium, it is evident from Eqs. (21) and (22) that the

18

where l is a vector of Lagrange multipliers associated with inequality constraint vector gQ; m , a vector of lagrange multipliers associated with equality constraint vector hQ and s is constant penalty parameter. As can be seen from Fig. 2, the operation dynamics of the network are changed with a predetermined timing switch. For 0 # t # t1 ; the network operates according to the following dynamics, _ Q 27f Q 2 s 7g J Qg1 Q 1 7hQhQ 19 J

Fig. 2. Block diagram of two-phase neural network.

R. Naresh, J. Sharma / Electrical Power and Energy Systems 24 (2002) 583590

587

values of the constraints g1 Q 0; hQ 0; l . 0 and J 7f 1 7g J l 1 7hm 0 23 Eq. (23) satises the optimality conditions of the KT theorem, and thus the equilibrium point of the two-phase network is precisely a global minimizer to a convex program. The idea behind the two-phase network formulation is that in phase 1 t , t 1 ; for a sufciently large value of ~ parameter s, Eq. (19) converges to an equilibrium Q at 1 ~ ~ are very close to the correswhich sgi Q and shj Q ponding Lagrange multipliers l and m , respectively. In phase 2 t $ t1 ; the network shifts the directional vector sg1 Q slowly towards l , and shj Q towards m . Consei quently, Eqs. (2022) attain a minimum point of the augmented Lagrangian function stated in Eq. (18). The cost term f Q is minimized during the transient state, while the rest of the terms of Eq. (20) are needed to prevent the violation of the constraint relationships. 4.1. Generating the initial estimates for water discharge rates The control variables that are water discharge rates, require initialization in order to start the optimization operation by two-phase neural network method. As total available water is known a priori, an attempt is made here for its allocation in various hours of the planning period in such a way so that coupling constraints are satised as well as overall nonhydraulic generation cost is presumed to be minimum. These requirements are approximately met by considering hourly water discharge rate proportional to square of load demand in that hour. Dene a vector p such that t 2 Pd pt T   ; t 1; 2; ; T 24 j 2 Pd
j1

5. Calculate storage of the reservoirs using current values of water discharge rates. 6. Using water discharge rates and storage, determine hydro plant generations. 7. Calculate non-hydraulic power generations (import or export) using load balance equation. 8. Evaluate equality and inequality constraints using current values of discharges and reservoir contents. 9. Calculate gradients of both the objective function as well as equality and inequality constraints. 10. If t , t1 ; then compute new values of water discharge rates at time t t 1 Dt using phase 1 dynamics with the help of fourth order runge kutte step. 11. In case t $ t1 ; then calculate new values of both the water discharge rates as well as Lagrange multipliers using phase dynamics with the help of fourth order runge kutte step at time t t 1 Dt: 12. Check if t . t2 ; then calculate output values, save the solution and stop; otherwise go to step 5. 5. Example For substantiating the efcacy of the proposed method, a slightly modied test system [16,22,25] considered is shown in Fig. 3. The system consists of a multi-chain cascade of four reservoir type hydroplants. The scheduling period is 24 h. The hydro generation data used for the present work is supplied in Tables (13). Load demand data for 24 h is given in Table 1, while Table 2 provides the matrix of hydro generation coefcients. Bounds on reservoir storage volume, water discharge rates and boundary conditions on reservoir storage volume are presented in Table 3. In Table 3, the units of storage are 10 3 m 3, while units of water discharge rate are 10 3 m 3/h. The vector 10; 8; 1; 0T 103 m3 =h gives

qtj pt Wj ; t 1; 2; ; T; j 1; 2; ; Nh

25

In this way, all the available water is collected and hourly water discharge rates are proportional to the square of corresponding load demand. Even though some of the constraints may be violated initially, the two-phase neural network conveniently handles the infeasible state trajectory by way of penalty functions. 4.2. Algorithm 1. Scan input data and topology of interconnected hydro system. 2. Generate initial solution for discharge vector Q and assume parameters s, e , t1 and t2 : 3. Set vectors l 0 and m 0: 4. Set time t 0 and choose a suitable value for time step Dt:

Fig. 3. Test system.

588 Table 1 Load demand Hour 1 2 3 4 5 6 7 8 9 10 11 12 MW 135 130 130 135 135 140 150 160 170 180 200 180

R. Naresh, J. Sharma / Electrical Power and Energy Systems 24 (2002) 583590

Hour 13 14 15 16 17 18 19 20 21 22 23 24

MW 200 200 190 190 185 190 175 170 170 160 160 170

Fig. 4. Production cost trajectory.

reservoir hourly side inows. The water transportation delays considered are t1 1 h; t2 2 h; t3 2 h: 3 3 4 6. Results and discussion With the data furnished above, the proposed algorithm coded in C11 was simulated. A number of initial trial runs were made to ascertain the suitable values of network parameters. Based on this, two cases are considered. In case 1, s 20:0; e 10:0 while in case 2, s 20:0; e 5:0: The trajectories of the nonhydraulic production cost for case 1 and case 2 are shown in Figs. 4 and 5, respectively. It is worth noting the effect of parameters S and e on the convergence properties of two-phase neural network in relation to short-term hydro scheduling problem. For a specied value of integration time step Dt, larger is the penalty parameter S; closer is the phase 1 equilibrium point to the exact solution. This results in a better initial point for the second phase dynamics and consequently, rapid convergence. However, too large a value for S may lead to instability of numerical integration. In the present scheduling problem, largest
Table 2 Hydro generation coefcients Plant 1 2 3 4 C1 20.001 20.001 20.001 20.001 C2 20.1 20.1 20.1 20.1 C3 0.01 0.01 0.01 0.01 C4 0.40 0.38 0.30 0.38 C5 4.0 3.5 3.0 3.8 C6 230 230 230 230

possible value of S was found to be 20.0 with Dt 0:001: It may be observed from Figs. 4 and 5 that initial estimates of production cost are same in both the cases. Then, there is a sharp dip in the cost initially while it stabilizes at a constant value during succeeding stages. This is due to the initial estimates of control variables being far off from the optimal solution. A sharp hook may be noted near the bottom of the cost trajectories. This is because, phase 1 dynamics with nite penalty parameter S is not sufcient to accurately penalize violated constraints. The production cost so obtained is obviously less than the true optimal cost. Subsequently, phase 2 dynamics increases the appropriate lagrange multipliers and drives the cost trajectory towards the feasible region and the production cost is increased as expected. Finally, when equilibrium is attained, production cost remains constant. The nal production cost with case 1 is 1445$ while it is 1429$ with case 2. This is because in case 1, lagrange multipliers are slightly overestimated near the boundary of the feasible region due to larger e . However, the method can fail if e is too large. This is because the value of lj for some j may become larger than the corresponding lagrange multiplier before the reservoir release or storage trajectory reaches boundary of the feasible region. This is by virtue of Eq. (21), which only increases the value

Table 3 Characteristics of hydro stations Plant 1 2 3 4 x 80 60 100 70  x 150 120 240 160 q 5 6 10 13  q 15 15 30 25 xbeg 100 80 170 120 xend 120 70 170 120

Fig. 5. Production cost trajectory.

R. Naresh, J. Sharma / Electrical Power and Energy Systems 24 (2002) 583590

589

Fig. 6. Hourly reservoir release.

Fig. 7. Hourly reservoir release.

of l j and there is no way out to bring back the overestimated lagrange multiplier. As a result, the trajectory remains in the infeasible region and the system diverges. Thus, the performance of the two-phase neural network for hydro scheduling problem depends on the prudent choice of network parameters. The parameters should be large enough for fast convergence and small enough to maintain desired accuracy and numerical stability. We get reservoir releases as hydro scheduling network solution. It is also helpful to provide other quantities such as reservoir storages, hydro and nonhydraulic power generations during each hour. For the case 2, reservoir release and storage trajectories are shown in Figs. 6 and 7, respectively.
Table 4 Hydro generation schedule contrast Hour Two-phase neural network Plant1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 23.72 25.33 26.00 28.62 28.27 32.59 41.18 47.36 51.66 55.78 59.67 52.38 56.48 54.56 49.10 47.47 44.16 45.35 36.93 33.71 33.93 29.51 37.23 36.68 Plant2 16.76 17.41 17.83 18.37 18.89 19.41 19.91 22.31 26.16 29.34 35.00 28.05 31.97 31.09 27.40 26.67 24.66 25.85 21.84 20.87 21.76 20.71 13.48 20.40 Plant3 42.86 39.55 37.45 35.91 34.30 32.72 31.55 30.99 30.91 31.27 32.76 34.43 36.43 37.98 39.68 40.96 41.52 42.22 41.95 41.27 40.21 37.09 37.76 43.03 Plant4 45.62 41.60 44.01 46.05 47.63 49.11 50.38 51.50 52.53 53.54 59.40 55.57 63.46 65.26 64.90 66.59 67.36 69.26 68.64 69.14 69.36 68.52 67.09 63.65 E 6.04 6.10 4.71 6.05 5.91 6.18 6.98 7.84 8.74 10.06 13.17 9.58 11.66 11.11 8.92 8.32 7.30 7.33 5.63 5.01 4.75 4.17 4.43 6.24

In order to validate the results achieved with the proposed solution technique, we sequentially solve the augmented lagrangian function [18] of the short term hydro scheduling problem for the above mentioned test system by using Polack Ribiere [30] version of conjugate gradient method. The optimal results obtained for hourly hydro and nonhydraulic power generation schedules with two-phase neural network and conjugate gradient method are given in Table 4. As observed from the table, the set of schedules obtained using both the methods are in tandem with each other. The optimal production cost obtained by conjugate gradient method is 1430$, which is a marginal increase as compared with two-phase neural network with respect to case 2.

Conjugate gradient method Plant1 23.72 25.41 26.00 28.56 28.24 32.58 41.14 47.20 51.59 55.84 59.70 52.38 56.53 54.55 48.93 47.26 43.95 45.20 37.11 34.31 34.64 30.83 37.01 36.30 Plant2 16.75 17.30 17.84 18.37 18.90 19.41 19.93 22.46 26.18 29.32 35.07 28.03 31.93 31.05 27.35 26.63 24.66 25.88 21.99 20.97 21.68 20.43 13.49 20.59 Plant3 42.84 39.54 37.48 35.99 34.33 32.82 31.68 31.07 30.89 31.19 32.61 34.33 36.31 37.86 39.56 40.80 41.29 42.02 41.51 40.76 40.05 37.10 38.64 43.26 Plant4 45.61 41.60 43.99 45.98 47.55 49.03 50.30 51.43 52.52 53.58 59.46 55.65 63.56 65.43 65.24 66.98 67.80 69.58 68.81 69.00 68.94 67.55 66.53 63.63 E 6.07 6.15 4.70 6.09 5.98 6.16 6.95 7.84 8.83 10.07 13.16 9.60 11.68 11.12 8.93 8.33 7.29 7.32 5.59 4.95 4.69 4.09 4.33 6.21

590

R. Naresh, J. Sharma / Electrical Power and Energy Systems 24 (2002) 583590 optimization of variable head multi-reservoir power system for long term regulation. Water Resour Res 1986;22(60):8528. Brannud H, Bubenko JA, Sjelvgren D. Optimal short term operation of a large hydrothermal power system based on a nonlinear network ow concept. IEEE Trans PWRS 1986;1(4):7582. Wakamore F, Masui S, Mortia K. Layered network model approach to optimal daily hydroscheduling. IEEE Trans PAS 1982;101(9):33104. Xia Q, Xiang N, Wang S, Zhang B, Huang M. Optimal daily scheduling of cascaded plants using a new algorithm of nonlinear minimum cost network ow concept. IEEE Trans PWRS 1988;3(3):92935. Saha TN, Khapade SA. An application of a direct method for the optimal scheduling of hydrothermal power systems. IEEE Trans PAS 1978;97(3):97785. Wang C, Shahidehpour SM. Power generation scheduling for multiarea hydrothermal power system with tie line constraints, cascaded reservoirs and uncertain data. IEEE Trans PWRS 1993;8(3):133340. Pereira MVF, Pinto LMVG. Application of decomposition techniques to the mid and short term scheduling of hydro thermal systems. IEEE Trans PAS 1983;102(11):36118. Soares S, Lyra C, Tavares H. Optimal generation scheduling of hydrothermal power systems. IEEE Trans PAS 1980;99(3):110614. Kumar S, Sharma J, Ray LM. A nonlinear programming algorithm for hydro-thermal generation scheduling. Comput Elect Engng 1979; 6:2219. Turgeon A. Optimal short term hydro scheduling from the principle of progressive optimality. Water Resour Res 1981;17(3):4816. Liang RH, Hsu YY. Scheduling of hydroelectric generation units using articial neural networks. Proc IEE, Pt C 1994;141(5):4528. Liang RH, Hsu YY. Short-term hydro scheduling using hopeld neural network. Proc IEE, Pt C 1996;143(3):26975. Wong KP, Wong YN. Short-term hydro scheduling Part 1: simulated annealing approach. Proc IEE, Pt C 1994;141(5):497501. Yang PC, Yang HT, Huang CL. Scheduling short term hydrothermal generation using evolutionary programming techniques. IEE Proc Pt C 1996;143(4). Orero SO, Irving MR. A genetic algorithm modelling framework and solution technique for short term optimal hydrothermal scheduling. IEEE Trans PWRS 1998;13(2):50118. Maa CY, Shanblatt MA. A two-phase optimization neural network. IEEE Trans Neural Networks 1992;3(6):10039. Naresh R, Sharma J. Two-phase neural network based solution technique for short term hydrothermal scheduling. IEE Proc Gen Transm Distrib 1999;146(6):65763. Naresh R, Sharma J. Hydro system scheduling using ANN approach. IEEE Trans Pwr Syst 2000;15(1):38895. Tank DW, Hopeld JJ. Simple neural optimization networks: an A/D converter, signal decision circuit, and a linear programming circuit. JEEE Trans Circuits Syst 1986;CAS-33:53341. Kennedy MP, Chua LO. Neural networks for nonlinear programming. IEEE Trans Circuits Syst 1988;35:55462. Bertsekas DP. Constrained optimization and LaGrange multiplier methods. New York: Academic Press, 1982. Jacob DAH. The state of the art in numerical analysis. London: Academic Press, 1977.

7. Conclusions A novel method based on two-phase neural network has been developed to nd optimal schedule for hydro generations in an interconnected hydro power system. The advantage of the proposed technique is that it takes care of the concurrent interaction among all the water discharge rate variables of the problem. After the scheduling problem has been formulated in two-phase neural network framework, the only other issue, which requires resolution, is the setting of network parameters. Once the parameters of the two-phase neural network have been resolved, the solution to the hydro scheduling under diverse operating scenarios can be easily obtained. The results of the algorithm can be used in setting the hydro generation targets for each plant in a hydropower system. Comparison with conjugate gradient method has validated the results obtained by solving the short-term scheduling problem with proposed solution technique. Since the nature of the network is essentially similar to parallel processing, it provides a convenient way to realize its operation by the use of fast transputers. Such a hardware implementation will be bound to provide extremely fast computation for the optimal scheduling of large-scale hydropower systems. References
[1] Dahlin EB, Shen DNC. Optimal solution to the hydro steam dispatch problem for certain practical systems. IEEE Trans PAS 1996;85:437 58. [2] Chandler WG, Dandeno PL, Glimn AF, Kirchmayer LK. Short range economic operation of a combined thermal and hydroelectric power system. AIEE Trans PAS 1953;72(Part III):105765. [3] Drake JH, Kirchmayer LK, Mayall RB, Wood W. Optimum operation of a hydrothermal system. AIEE Trans PAS 1962;80:24250. [4] Chang S, Chen C, Fong I, Luh PB. Hydroelectric generation scheduling with an effective differential dynamic programming. IEEE Trans PWRS 1990;5(3):73743. [5] Turgeon A. Optimal operation of multi-reservoir power systems with stochastic inows. Water Resour Res 1980;16(2):27583. [6] Yang JS, Chen N. Short-term hydrothermal co-ordination using multipass dynamic programming. IEEE Trans PWRS 1989;4(3):10506. [7] Yeh WW-G, Becker L, Hua S-Q, Wen D-P, Liu J-M. Optimization of real time hydrothermal system operation. J Water Resour Plann Mgmt 1992;118(6):63653. [8] Wood AJ, Wollenberg BF. Power generation operation and control. New York: Wiley, 1996. [9] Soliman SA, Christensen GS. Application of functional analysis to [10]

[11] [12]

[13]

[14]

[15]

[16] [17]

[18] [19] [20] [21] [22]

[23]

[24] [25]

[26] [27]

[28] [29] [30]