An Exact Multi-Plant Hydro Power Production Function for Mid/Long Term Hydrothermal Coordination André Luiz Diniz Electric Energy Research Center Rio de Janeiro, Brazil [email protected] Ana Lucia Saboia Electric Energy Research Center Rio de Janeiro, Brazil, [email protected] Abstract – The generation of a hydroelectric power plant is a nonlinear function of its storage and turbined outflow. Nonlinear, mixed-integer or linear programming models have been proposed to approximate this function in the context of hydrothermal coordination, usually with one function for each plant. This paper proposes a new exact “Multi-plant hydro production function (MHPF)”, which depends on the storage and discharge of a given reservoir power plant and comprises not only its generation but also the power output to all subsequent run-of-the-river plants until the next reservoir. Based on this function we build a piecewise linear model that can be embedded in stochastic linear programs for optimal hydrothermal coordination. We illustrate the proposed MHPF approach for real sets of cascaded hydro plants of the Brazilian System. Keywords: Hydroelectric power generation, linear programming, power generation planning. I. INTRODUCTION The generation of a hydroelectric power plant is a nonlinear function of its state (reservoir storage) and its operation (turbined outflow and spillage) and is called in the sequel as “individual Hydro Production Function” (IHPF). Approximation models for this function for the hydrothermal coordination problem have been applied under different formulations, such as nonlinear [1], mixed-integer [2] or linear programming [3]. In the latter case, this function is usually approximated by individual piecewise linear functions, one for each plant, and embedded into the linear programs as a set of linear inequalities for each hydro plant and time step. The model above allows to accurately consider the dependency of hydro plants generation on the water head [3]. However, for long term hydrothermal planning with a very long time horizon and a huge scenario tree, the individual representation of each hydro plant - and as consequence, of its production function - may lead to impractical CPU times to solve the problem. For this reason, equivalent energy reservoirs [4], [5] have been considered so far to model the hydro cascaded in such models [6]-[10]. Nevertheless, we note that in a hydro cascade the system operation is defined only by the operation of hydro plants with storage regulation capability (which are simply called in the sequel as “reservoirs”), since the run-of-the-river hydro plants will tend to release (pref- R. M. Andrade COPPE / UFRJ Rio de Janeiro, Brazil [email protected] erably by turbining rather than spilling) all incoming water, either from natural inflows or coming from the operation of upstream plants. In this sense, this paper proposes a new model for the hydro production function, where we only build functions for reservoirs. This so called “Multi-plant hydro production function (MHPF)” comprises not only each reservoir but also all cascaded downstream run-of-the river plants until (but not including) the next reservoir. Such function gives the "exact"1 generation for this set of plants, without approximations or pre-established operation assumptions, based on the release of upstream reservoir plant and the vector of natural inflows to runof-the-river plants. We also present how to extend the piecewise linear model proposed in [3] for individual hydro plants to the MHPF proposed in this paper. Such MHPF model is useful to reduce the dimension of a deterministic or a (more usual) stochastic mid-long term hydrothermal coordination problem, since only hydro plants with reservoirs need to be represented. We illustrate the proposed approach for real sets of cascaded hydro plants of the Brazilian System and perform an analysis of the accuracy and number of constraints obtained with such function. The results show that we are able to produce as accurate values of hydro generation as if an individual hydro production function had been used, with a smaller number of constraints. II. INDIVIDUAL HYDRO PRODUCTION FUNCTION (IHPF) The generation of a given hydro plant can be expressed through expression (1) of its exact hydro production function : = = ( , , )= ℎ ( )−ℎ ( + )−ℎ (1) where = 9.81 × 10 is a constant related to gravity and unit conversion, is the overall hydro plant efficiency2, , and are the storage volume, turbined and spilled outflows, and the term in brackets consists in the net water head ℎ, which is the difference between up1 Throughout this paper, we mean by "exact" generation the values obtained by the analytical expressions (1)-(3). 2 In unit commitment problems, it is important to use a more accurate representation of each turbine and generation efficiency factors. stream (ℎ ) and downstream levels (ℎ ) after subtracting head losses ℎ . Such levels may be represented as polynomial functions3 of and the total release = + [3]: ℎ ( )= + + + + (2a) ℎ ( )= + + + + (2b) As explained in [3], for most of the plants the HPF turns out to be a composition of a nearly concave function on variables and , and a convex function on , as shown in Fig. 1. For modeling purposes in linear programming models, the first behavior is desirable while the second one forces us to apply a linear approximation to ensure convexity. plants, which are represented as triangles and circles, respectively, in Fig. 2. The operation of all hydro plants in each of these sets can be determined based solely on the operation of the upstream reservoirs, since system optimization will favor the run-of-the-river plants to turbine all incoming water up to its maximum turbined outflow. As a result, we can compute a unique "multiplant hydro production function (MHPF)” for each set, as explained next. We note that such aggregated model is more suitable for mid-long term planning, where time discretization is longer (e.g., weekly/monthly) and some aspects of hydro generation such as forbidden operating zones and hill curves for turbine efficiency factors do not need to be considered in detail. MULTI HYDRO PLANT run-of-thereservoir river plants Fig. 2. Aggregation of hydro plants in a cascade, where a multiplant hydro production function can be defined. (%) ( / ( / Fig. 1. Illustrative example of the hydro production function and its dependence on and (left) and (right). Several models have been proposed to model the HPF of a hydro plant, for a broader review we refer to [3]. We are particularly interested in convex formulations for this function that are suitable to be used for mid-long term power generation planning [6]-[11]. For this reason, the benchmark for this work is the piecewise linear formulation proposed in [3]. We note that due to the large size of such problems, multistage stochastic linear formulations are preferred and decomposition approaches such as nested Benders decomposition (NBD) [12] or Stochastic dual dynamic programming (SDDP) [13] are employed to solve the problem. Even though the model in [3] provides a high accuracy to represent the hydro plants generation function, it requires an individual representation of all plants in a river basin. However, most of the recent applications of NBD or SDDP to long term hydrothermal planning in real large scale systems with an accurate representation of uncertainties in the water inflows still make use of modeling of equivalent reservoirs [4], in order to reduce the problem dimensionality and allow the problem resolution in a reasonable CPU time [6]- [10]. In this sense, the main motivation of this work is to propose a so called "multi-plant hydro production function", which allows a more concise representation of the river basins, while still representing the individual characteristics of the hydro plants and avoiding their aggregation in equivalent reservoirs. III. MULTI-PLANT HYDRO PRODUCTION FUNCTION (MHPF) Based on the hydro basins topology, it is possible to identify several sets of cascading plants composed of a reservoir followed by one or more run-of-the river 3 Such functions are obtained based on topographical studies and should be periodically updated. However, "V-shaped" configurations are very common in large river-basins, where a given plant has more than one upstream reservoir. It is also possible to define multi-plants in such case, which will contain more than one upstream reservoir. Fig. 3 illustrates a reduction of a small cascade with 13 hydro plants into a cascade with only 5 multi-plants. Fig. 3. Aggregation of a 13-hydro plant cascade into a set of 5multi-plant cascade. We can compute the exact hydro production of a multi-plant as a function of the discharge and storage of the upstream plant, by summing up the generation of all plants in the set, with the assumption that each plant will only spill when the maximum turbined outflow has been reached. A. MHPF for a single upstream reservoir We first consider a configuration such as "C" in Fig. 3, which is composed of 1 upstream reservoir and run of the river plants. We use index 1 for the reservoir and indices {2, … , + 1} for the run-of-the-river plants. Let and denote the volume and release of the reser( = 2,…, + 1) be the natural inflow voir, and , and water intakes due to irrigation, evaporation4, etc, in each run of the river plant , as in Fig. 4(a). The turbined and spilled outflows of the reservoir can be determined as follows: 4 Since the reservoir surface is constant for run-of-the-river plants, their evaporation can be computed a priori. { , } = (3) = − . By applying the usual assumption that run-of-the river plants spill only when the maximum turbined outflow has been reached, their generation output are given by: = , , , (4) with: ( )= , +∑ −∑ ( )= +∑ −∑ − (a) ( ), (b) Fig. 4. Variables that impact the generation of the multi-plant, with only one (a) or several (b) upstream reservoirs. where is a constant value for such plants and the term under brackets accounts for the incoming water to each plant, taking into account the release of upstream plants, natural inflows and water intakes. The total generation of the multi hydro plant is the sum of the outputs of all the plants, which can be expressed as a function of only storage and release of the upstream reservoir, as follows: = , ( ( , ) == ), ( ) + + ∑ ( (5) ), ( ) B. MHPF for several upstream reservoirs A similar procedure can be applied to obtain the production of the multi-plant in the more general case where it is composed of upstream reservoirs followed by run-of-the-river plants, as shown in Fig. 4(b): = =∑ +∑ (( , ), ( , ( )+ ( ,…, , ), … , ( ), , ( )) = ,…, (6) ) , where the control variables for each plant are: ( ,…, ) = min ( ,…, +∑ , +∑ ( − )=∑ + ( − ) − ,…, ∑ ) ). (7a) (7b) Again, the generation of the multi-plant is a function of only variables related to the upstream reservoirs. B. Remarks We make the following important remarks regarding the proposed multi-plant hydro production function: • the MHPF gives the exact generation (without approximations) for the set of plants, based on the storage and the release of the upstream reservoirs. We emphasize that the only (quite reasonable) assumption is that spillage occurs after maximum discharge is reached; • The vector of natural inflows to run-of-the-river plants is supposed to be known, whether in a deterministic approach or as a possible inflow scenario in a stochastic setting. Therefore, if a stochastic problem is considered, the MHPF will be different for each scenario. In this sense, 'warm start' procedures to solve the stochastic programming subproblems should take such differences into account; • The use of a MHPF in an optimal hydrothermal dispatch problem does not exclude the possibility of imposing individual operation constraints for the reservoir or run-of-the-river plants, such as minimum/maximum outflow and power generation. They can be modeled by proper conditions when building the MHPF; • The MHPF approach is also interesting for long term time horizons, as a competitive and more accurate model as compared to the usual approach of equivalent reservoirs. In such context, it is necessary to build a different model of all multi-plants in all inflows scenarios of each time step. However, this imposes only an increase in memory, since in terms of CPU time these functions (and corresponding piecewise linear models) are computed very fast and only once, before starting a NBD or SDDP solving procedure. The drawback of the proposed MHPF is that it cannot be applied when some aspects of hydrothermal scheduling are considered, as for example: • in network constrained problems [14], where each hydro plant injects power in a different bus of the electrical network. In such case, line flow limits or additional security constraints may constrain the maximum generation of one or more plants in the set, and spillage may occur before a plant reaches its maximum outflow, thus violating the basic assumption of the model; • in hydro unit commitment problems [2], [15], where the generation of each unit of a plant is optimized. Actually, in that case even the individual hydro production function (1) would not be appropriate, since it neither takes into account the units status (on/off) nor allow a proper representation of ramping constraints. • when water delay times are considered [16], because the release of the upstream plant reach the run-of-theriver plants in further time steps. However, all these aspects are typical only for shortterm or real-time hydrothermal scheduling problems, while the focus of this paper is the mid/log term hydrothermal planning, where the consideration of such constraints are of less interest. IV. PIECEWISE LINEAR APPROXIMATION FOR THE MHPF The MHPF described previously is nonlinear and probably non-concave in part of the ( × ) domain. Even though nonlinear programming approaches [1] can be employed to solve this problem, a linear program- ming formulation is suitable in view of the large developments achieved so far in the stochastic linear programming literature [17]. In this sense, we also propose in this paper an extension of the piecewise linear model (PWL) that had been proposed in [3] for the individual hydro production function (1) to the MHPF formulation (5) and (6), in order to provide a convex approximation of the feasible region below such functions. Due to space limits, we do not provide details of how to build this convex approximation here but rather refer to section IV of our previous work [3]. However, there are some differences in the procedures to obtain a PWL approximation for the multi-plant hydro production function proposed in this paper, in comparison with the procedures to approximate an individual function. • the MHPF is a function of (besides storage) the total release of the upstream reservoir, instead of individual turbined outflow and spillage variables. As a consequence, the PWL model for the case of a single upstream reservoir (section III.A) has only three dimensions and no secant approximation is necessary to model spillage; • the MHPF is non-differentiable at those points where the maximum discharge of each plant in the set is reached (see Fig. 7 later). In this sense, besides the uniform grid that is usually employed to generate discretization points of the true HPF (section IV.B of [3]), it is useful to include at most ( + ) additional points related to maximum discharge of those plants5. • if there is more than one upstream plant (section III.B), the PWL approximation of the MHPF lies in the space. Own experience by the authors has shown ℜ that computation of a convex hull becomes impractical6 for ≥3. Fortunately, such cases are not so common and appear only three times in the real Brazilian hydro topology. In such case we may apply an individual model for the HPF of each upstream reservoir and a MHPF for a set including only the run-of-the-river plants. After applying all these procedures, we end up with a set of linear inequalities (8) that approximate the generation for each multi-plant, as a function of decision variables and for the upstream reservoirs: ≤ ( ) +∑ ( ) ( ) +∑ , (8) = 1, … , , which should be used together with the box constraint: ∑ ≤ ≤∑ . (9) A. Recovering the generation of each plant The aim of using a MHPF is to allow the reduction of the dimensionality of a deterministic or stochastic hydrothermal coordination optimization problem, by representing only variables associated to reservoirs with regulation capability. However, once the optimal power 5 If the sum of water inflows up to plant is already greater than its maximum discharge, such point will not be necessary. 6 for example, if only 3 discretization points are used for and variables, up to 2.1×1016 hyperplanes may be evaluated. output of a given multi-plant is obtained, it is necessary to extract the individual generations for each plant that composes such function. We note that would be obtained by the approximated model (8), with the storage and release for all upstream reservoirs. Based on the values of such , = variables, we can compute the exact generation 1, … , by expressions (6) and (7), as well the overall generation of the multi-plant. We propose the following algorithm to split the deviation Δ = − as evenly as possible among the set of all = + plants of the multi-function, respecting their individual power generation limits: = (initial generation of each Initialization: plant); = |Δ | (amount of deviation to be allocated). If Δ > 0 do algorithm 1, else do algorithm 2. Algorithm 1. Update index set Ω = { − > 0} of plants that can still increase its generation. Update each generation by = min{ , + /|Ω|}. Update = −∑ . If = 0, stop, otherwise repeat Algorithm 1. Algorithm 2. Update index set Ω = { − > 0} of plants that can still decrease its generation. Update each generation by = max{ , − /|Ω|}. Update = ∑ − . If = 0, stop, otherwise repeat Algorithm 2. Constraints (9) ensure that both algorithms yield a feasible point { , = 1, … } with ∑ = . V. NUMERICAL RESULTS We present numerical experiments of application of the proposed multi-plant function for real hydro plant data of the Brazilian system. A. Detailed Analysis of a specific MHPF First we perform a detailed analysis of the MHPF function proposed in this paper for a set of four hydro plants (one reservoir and three run-of-the-river plants), as shown in Fig. 5. We list in Tables I and II the main hydro plant and reservoir data, respectively, to compute the production function of this set of plants. Each line in Table II shows one coefficient of the forebay or tailrace level polynomial (depending on the column) of each reservoir. Storage limits in the reservoir are = 1540 hm3 and = 5040 hm3. M. Moraes ( = ) Estreito ( = ) Jaguara ( = ) Igarapava ( = ) Fig. 5. Set of cascaded plants in the illustrative example. The graph of the exact hydro production function for this set of plants as a function of storage and outflow of the upstream reservoir is shown in Fig. 6. We also present in Fig. 7 a sensitivity analysis of the generation in this set as the inflow in each run-of-the-river plant is increased, for fixed pairs ( , ) and natural inflow values for the other plants. In each case, the nondifferentiable points appear when the discharge of any of the remaining downstream plants achieves the value related to its maximum generation or turbined capacity. Beyond such point, an increase in inflow will cause spillage in that plant, thus sharply reducing the local efficiency of the multi-plant. TABLE I. POWER GENERATION DATA - ILLUSTRATIVE EXAMPLE. Turbine Efficiency (MW/(m3/s)) Plant (% or ) ’ ⁄ ) ( ) ( =1 =1 0.0083 0.80% 1328.0 478.0 0.0088 1.50% 2028.0 1104.0 =2 0.0089 1.34% 1076.0 424.0 =3 0.0090 0.30 m 1480.0 210.0 - - 5912.0 2216.0 Total TABLE II. Coef POLYNOMIAL DATA - ILLUSTRATIVE EXAMPLE (2a) =1 (2b) =1 =2 =3 =4 0 6.42e-2 6.19e+2 5.57 e+2 5.10e+2 1 8.09e-03 1.73e-3 1.22e-3 1.77e-3 7.82e-4 2 -3.70e-07 -4.89e-8 -7.51e-8 -7.40e-8 -1.04e-7 3 -7.11e-11 - 2.16e-12 1.12e-11 - 4 9.12e-15 - - -4.80e-16 - 4.94e+2 not incur in a significant loss of accuracy as far as the detailed operation of each plant is concerned. TABLE III. Turbined Outflow Q (% maximum) V 25% 50% 0% -0.30 0.07 25% 0.22 0.12 50% 0.64 0.05 75% 0.64 0.05 100% 0.64 0.05 min. absolute deviation max. absolute deviation Hydro generation (MW) 4 1 5 6 2 7 16 2 3 In Table III we show the deviations between the exact function and the proposed piecewise linear model for this MHPF function, along a discretization grid on the storage × outflow domain for the upstream reservoir. The results show that the proposed MHPF model yields very accurate generation values, as if each individual production function (IHPF) had been considered. These results illustrate that the aggregate model indeed does 26 11 47 17 48 18 156 72 6 38 8 39 27 49 19 169 77 28 12 9 32 9 10 40 42 31 13 21 249 15 50 43 51 12 172 20 22 173 33 61 16 71 78 82 174 52 10 11 155 14 15 4 8 Fig. 7. Sensitivity analysis of the multi-plant output as the inflow in each run-of-the-river plant increases. 37 14 7 Natural inflows (m3/s) 75% 100% 150% 200% 0.07 0.07 0.07 0.07 0.11 0.11 0.18 0.16 -0.04 0.15 0.25 0.56 -0.04 0.19 0.27 0.76 -0.04 0.25 0.44 1.03 0.89 MW 5.73 MW (V = 0%, Q= 10%). B. Overall analysis for several cascades Now we perform a broader analysis taking into account 51 cascaded hydro plants, which compose part of the current Brazilian system. We split “V-shaped” configurations into two or more parallel cascades in order to avoid having more than one upstream reservoir for the multi-plant. This was done because the extension of our 3-dimensional convex hull algorithm of [3] to five or more dimensions does not present satisfactory results so far. As shown in Fig. 8, such plants are aggregated in 23 multi-plants, where the number of plants in each set depends on the type of the hydro plants (reservoir or run-of-the river plants). We note that whenever there are consecutive reservoirs in a given cascade, we have a multi-plant with only one reservoir, which is similar to the individual HPF described in section II and presented in [3], except for the model of turbined outflow and spillage as a single release variable . The plants ID are the official ones used in the centralized planning of the Brazilian system, which is performed by the Independent system operator with the optimization models described in [6]. All the characteristic data for those plants can be accessed through the ISO´s web page (http://www.ons.org.br). 1 Fig. 6. Hydro generation of the set (M. de Moraes, Estreito, Jaguara and Igarapava) as a function of operation of M. Moraes. DEVIATIONS BETWEEN THE MHPF AND THE PWL APPROXIMATION OF THIS FUNCTION 111 23 112 62 113 63 114 Fig. 8. Overall hydro cascade for part of the Brazilian system. In order to allow an assessment of our model for different system conditions, we considered twelve different inflow data for the run-of-the river plants (which are necessary to compute the MHPF) corresponding to their monthly long-term average values extracted from the historic record. Therefore we built 12 different MHPFs as well as 12 piecewise linear models to approximate them, for each multi-plant. As mentioned previously, the MHPF described in section III provides the exact generation of the set of plants as a function of the release of the upstream reservoir. However, in order to allow taking into account this function in a hydrothermal dispatch optimization problem, the piecewise linear approximation described in section IV was computed for all multi-plants. For comparison purposes, we also computed the piecewise linear approximation for each individual hydro production function (IHPF) [3]. The number of discretization points for storage was 5 in both MHPF and IHPF models, with a window width of 20% around the initial storage (see Table V in [3]). For the turbined outflow we considered 5 discretization points for the IHPF, which was shown in [3] to provide a good trade-off between accuracy and CPU time to solve an underlying hydrothermal scheduling problem. As for the MHPF we used only 3 points in the grid for two reasons: (i) additional discretization points will automatically be included due to maximum discharge limits, as explained in the second bullet of section IV; (ii) since the MHPF tends to present more points in its convex hull as compared to the IHFP (due to being a composition of different individual functions), using the same number of breakpoints in both models will lead to a much higher number of constraints in the MHPF. 1) Assessment of the deviations between the piecewise linear model and the exact functions. We then randomly selected a number of “operation points” ( , ) for the upstream reservoir in each set and computed the exact generation levels and the approximated generation yielded by the piecewise linear functions, both for the aggregated and the individual case. We show in Table IV the relative deviations of the approximated model as compared to the values of the exact function, for both the original IHPF and the proposed MHPF functions. The deviation of the IHPF was computed based on the sum of the exact and approximated generations for all individual plants in the set. TABLE IV. TABLE IV. MEAN RELATIVE DEVIATION OF THE MHPF AND IHPF MODELS, AS COMPARED TO THEIR RESPECTIVE EXACT FUNCTIONS. Multi% dev. % dev. Multi% dev. % dev. plant ID (MHPF) (IHPF) plant ID (MHPF) (IHPF) 1 14 0.35 0.40 1.27 0.62 2 15 0.77 0.73 0.14 0.06 3 16 0.29 0.54 0.07 0.00 4 17 0.73 0.62 0.24 0.31 5 18 5.54 4.67 8.93 10.25 6 19 0.13 0.33 3.33 3.20 7 20 0.31 0.07 0.36 0.20 8 21 0.00 0.00 1.37 0.49 9 22 1.21 0.99 0.37 0.42 10 23 0.23 0.12 0.21 0.82 11 MEAN 1.12% 1.22% 0.66 0.73 12 MAX 8.27% 10.15% 0.02 0.08 13 1.55 0.00 It can be seen that the proposed MHPF model provides similar results and in some cases even outperforms the individual model. In the overall average anal- ysis taking into account all multi-plants, deviations of the MHPF are have a quality only around than 10% poorer than the IHPF model. It is important to note that the purpose of the MHPF proposed in this paper is NOT to obtain a more accurate model than [3], but rather to obtain a level of accuracy as close as possible to it, but modeling the cascaded hydro plants multi-plants, in order to reduce the size of the problem. 2) Assessment of the number of constraints of the PWL model Probably the main goal of the MHPF is to provide a more concise set of variables and constraints for the hydrothermal coordination problem in comparison with the individual model, which is widely used in short/mid term planning but becomes prohibitive for long-term planning with an extended time horizon and a detailed modeling of uncertainties. Therefore we compare in Table V the average number of constraints (linear inequalities) in both MHPF and IHPF models, for each of the 23 multi-plants. TABLE V. Multiplant ID 1 2 3 4 5 6 7 8 9 10 11 12 NUMBER OF CONSTRAINTS OF THE MHPF AND IHPF MODELS. # constraints # constraints Multi- # constraints # constraints plant ID (MHPF) (IHPF) (IHPF) (MHPF) 6.0 7.0 10.2 13.3 7.0 9.0 9.0 5.0 11.0 7.0 12.0 11.3 11.0 11.0 31.0 19.0 14.0 15.0 28.0 11.0 15.0 11.0 19.0 15.0 13 14 15 16 17 18 19 20 21 22 23 MEAN 7.0 10.5 9.3 11.1 9.0 7.0 7.0 8.7 8.8 11.1 8.4 8.94 14.0 18.0 27.0 19.0 29.0 11.0 11.0 11.0 19.0 19.0 26.0 17.57 We can see that the proposed MHPF reduces the number of constraints to represent the hydro production function by roughly 50%, in relation to the individual model. Therefore one major advantage of this approach is the drastic reduction in the size of the resulting linear programming hydrothermal dispatch problem, with a very small loss of accuracy, as seen in Table IV. We note that additional reductions in the hydrothermal dispatch problem by using the multi-plant representation of the hydro cascade are possible, since hydro balance equations and some operation constraints can also be represented by multi-plants. Finally, we mention that if we had used 5 breakpoints for turbined outflow in our MHPF model (plus the maximum discharge breakpoints), the average number of constraints would have jumped to 21.51, which would turn the aggregated less favorable. This is the reason why a reduced number of breakpoints is recommended to such aggregated model. VI. CONCLUSIONS This paper presented a new multi-plant model for the production function of hydro plants (MHPF) that comprises several cascaded reservoirs in a single function. This approach is able to represent the individual charac- teristics of the hydro plants while providing a reduction in the number of linear constraints to approximate the function as compared to the traditional approach of having individual models for each plant (IHPF). Due to the reduced number of constraints, the use of such multi-plant models tends to yield lower CPU times when solving an underlying hydrothermal coordination problem. Numerical results with real data for a large number of hydro plants of the Brazilian system also show that our MHPF does not lead to an extra loss of accuracy in representing the generation of the hydro plant as compared to the IHPF model. In particular, the proposed approach is suitable for long-term planning, as a more accurate approach to represent hydro plants as compared to usual equivalent energy reservoir model. It is important to note that environmental constraints have forced most of the new hydro plants to be run-of-the river plants, making the use of multi-plant hydro production function models even more suitable. Future work by the authors will comprise the inclusion of the MHPF developed in this work in the stochastic long-term hydrothermal planning problem for large systems, as well as adjustments in the modeling of several constraints of the problem to adapt them to this multi-plant framework. In this sense we can verify the practical advantages of this multi-plant model from the computation point of view of solving large-scale stochastic hydrothermal planning problems. REFERENCES [1] J.P.S.Catalão, S.J.P.S. Mariano, V.M. Mendes, L.A.F.M. Ferreira, “Scheduling of head-sensitive cascaded hydro systems: A nonlinear approach”, IEEE Transactions on Power Systems, v.24, n.1, pp. 337-346, 2009. [2] Borghetti, A., D´Ambrosio, C., Lodi, A., Martello, S., “An MILP approach for short-term hydro scheduling and unit commitment with head-dependent reservoir ”, IEEE Transactions on Power Systems, v. 23, n.3, pp. 1115-1124, Aug. 2008. [3] A.L. Diniz, M.E.P. Maceira, “A four-dimensional model of hydro generation for the short-term hydrothermal dispatch problem considering head and spillage effects”, IEEE Transactions on Power Systems, v. 23, n.3, pp. 1298-1308, Aug. 2008. [4] N. V. Arvantidis, J. Rosing, “Composite representation of multireservoir hydroelectric power system”, IEEE Transactions on Power Apparatus and Systems, v. 89, n. 2, pp. 319-326, Feb. 1970. [5] M.E.P. Maceira, V. S. Duarte, D.D.J. Penna, M.P. Tcheou, “An approach to consider hydraulic coupled systems in the construction of equivalent reservoir model in hydrothermal operation planning”, 17th PSCC - Power Systems Computation Conference –, Stockholm, Sweden, 2011. [6] M.E.P. Maceira, V.S. Duarte, D.D.J. Penna, L.A.M. Moraes, A.C.G. Melo, “Ten years of application of stochastic dual dynamic Programming in official and agent studies in Brazil –Description of the NEWAVE program”, 16th PSCC - Power Systems Computation Conference, Glasgow, SCO, July 2008. [7] A. Helseth, B. Mo, G. Warland, “Long-term scheduling of hydro-thermal power systems using scenario fans”, Energy Systems, v.1, n.4, pp. 377-391, Dec. 2010. [8] V.L. de Matos, E.C. Finardi, A computational study of a stochastic optimization model for long term hydrothermal scheduling, Int. Journ. Of Electrical Power and Energy Systems, v. 43, n.1, pp. 14431452, Dec. 2012. [9] A. Shapiro, W. Tekaya, J.P. Costa, M.P. Soares, “Risk neutral and risk averse Stochastic Dual Dynamic Programming method”, European Journal of Operational Research, v. 224, n.2, pp. 0375-0391, Jan. 2013. [10]A. Gjelsvik, B. Mo, A. Haugstad " Long- and Medium-term Operations Planning and Stochastic Modelling in Hydro-dominated Power Systems Based on Stochastic Dual Dynamic Programming", S. Rebennack et al. (eds.), Handbook of Power Systems I, Energy Systems, Springer, 2010. [11]S. Rebennack, B. Flach, M.V.F. Pereira, P. M. Pardalos, “Stochastic Hydro-Thermal Scheduling under CO2 Emissions Constraints”, IEEE Transactions on Power Systems, v.27, n.1, pp. 58-68, Feb. 2012. [12]J.R. Birge, “Decomposition and partitioning methods for multistage stochastic linear programs”, Operations Research, v.33, n.5, pp. 989-1007, 1985. [13]M. V. F. Pereira, L. M. V. G. Pinto, “Multi-stage stochastic optimization applied to energy planning”, Mathematical Programming, v. 52, n.1-3, pp. 359375, May 1991. [14]T. N. Santos, A. L. Diniz, “A Dynamic Piecewise Linear Model for DC Transmission Losses in Optimal Scheduling Problems”, IEEE Transactions on Power Systems, v.26, n.2, pp. 508-519, May 2011. [15]E. C. Finardi, E. L. da Silva, “Solving the hydro unit commitment problem via dual decomposition and sequential quadratic programming”, IEEE Transactions on Power Systems, v. 21, n. 2, pp. 0835-0844, May 2006. [16]A. L. Diniz, T. M. Souza, “Short-Term Hydrothermal Dispatch With River-Level and Routing Constraints”, IEEE Transactions on Power Systems, v.29, n.5, pp. 2427 – 2435, Sep. 2014. [17]P. Kall, J. Mayer, “Stochastic linear programming: models, theory and computation”, John Wiley & Sons, 2005.
© Copyright 2026 Paperzz