Beruflich Dokumente
Kultur Dokumente
Abstract—In this paper, it proposes a capacity expansion 𝛽𝑘𝑚,𝑦,𝑡 Dual variable for the disjunctive power flow
planning (CEP) considering the load uncertainty, using a Benders definition (positive) constraint.
decomposition method (BD) and power transfer distribution 𝛽𝑘𝑚,𝑦,𝑡 Dual variable for the disjunctive power flow
factors (PTDF). As antecedent, we have as a background a definition (negative) constraint.
disjunctive formulation, which has the mixed integer-linear
𝜋𝑔,𝑡 Dual variable for the disjunctive maximum power
programing (MILP) with stochastic characteristic, which involves
many variables and restraints. Due to the large size of this constraint of the generating unit g candidate for
problem, it is applied to the iterative BD algorithm and PTDF, investment in period t.
fractionating the complete formulation into a master investment 𝜋𝑔,𝑡 Dual variable for the disjunctive minimum power
problem and at operational subproblem. The first one only takes constraint.
care of the integer or binary variables, that is, in each iteration it 𝜆𝑘 Nodal price for the bus k.
delivers an investment plan, which is sent and evaluated in the 𝜆𝑆𝐿 Nodal price for the slack bus.
linear optimization problem. From this operation, the dual 𝜇̅𝑘𝑚 Dual variable for the power flow constraint
variables are obtained, which are used as economic indicators for (positive) by the k-m corridor of the formulation
the Benders cuts. About validation of the algorithm, this applies
two test systems, Garver and the National Electrical System of
with PTDF.
𝜇𝑘𝑚 Dual variable for the power flow constraint
Chile (SEN), with significant times reductions in solvers and
smaller formulations in the operational subproblems. (negative).
𝑒𝑥
𝜋𝑔 Dual variable for the maximum power constraint
Keywords—Benders Algorithm; Disjunctive Formulation; Power for existing generators.
Transmission Distributions Factors; Capacity Expansion Planning; 𝜋𝑔𝑒𝑥 Dual variable for the minimum power constraint.
Mixed Integer-Linear Programming.
Parameters
NOMENCLATURE 𝐶𝑘𝑚 Investment cost of the k-m broker [USD $]
Variables 𝐶𝐼𝑔 Investment cost of the generator g [USD $ /
𝑃𝑔,𝑠,𝑡,𝑒 Active power dispatched by the unit g in time MW]
block s, year t and scenario e 𝑂&𝑀𝐴𝑔 Cost of operation and maintenance of the
𝑟𝑔,𝑠,𝑡,𝑒 Power dispatched by the virtual unit g generator g [USD $ / MW]
𝑓𝑘𝑚,𝑠,𝑡,𝑒 Transmission power flow through the k-m 𝐹𝐶𝑔 Linear component of the generator cost function g
corridor [USD $ / MWh]
𝑓̂𝑘𝑚,𝑠,𝑡,𝑒 Transmission power flow through the investment ℎ𝑠 Duration in hours of time blocks [h]
candidate 𝑉𝑜𝐿𝐿 Value of load lost [USD$/MWh]
𝛿𝑘,𝑠,𝑡,𝑒 Voltage angle of bus k 𝑑𝑘,𝑠,𝑡,𝑒 Demand in bar k[MW]
𝑋𝑘𝑚,𝑦,𝑡 Binary variable that represents the number of y 𝐴𝑇 Transposed incidence matrix electric system
elements to build in corridor k-m 𝐴̂𝑇 Transposed incidence matrix of the electric
𝑛𝑔,𝑡 Integer variable (accumulated) that represents the system with investment candidate lines
decision to invest in a new generator g 𝑏𝑘𝑚 Susceptance of the k-m corridor
𝛼𝑘𝑚,𝑦,𝑡 Dual variable for the disjunctive maximum power 𝑀 Big M
flow (positive) constraint of corridor k-m and 𝑆𝑅 Percentage of system security reserve
𝑀
circuit y 𝑛𝑔,𝑡 Maximum investment amount
𝑀
𝛼𝑘𝑚,𝑦,𝑡 Dual variable for the disjunctive maximum power 𝑓𝑘𝑚 Maximum flow for the k-m corridor [MW]
flow (negative)
𝑀
𝑋𝑘𝑚,𝑦,𝑡 Maximum amount of circuit investment for in optimization terms with greater efficiency than in the non-
the k-m corridor linear case. This falls on a problem of great size, due to the
𝑀𝐴𝑋
𝑑𝑡,𝑒 Maximum load disjunctive constraints inclusion and power formulation by bus
𝛼𝑡 Monetary update factor angle voltage. In addition, if it is considered the integration of
𝑃𝑒 Probability of occurrence of the scenario e Renewable Energy Sources (RES) as solar and wind energy
𝑃𝑇𝐷𝐹𝑘𝑚,𝑘 Linear power distribution factor of the k-m with your respective intermittencies (uncertainties), it obtained
corridor, bus k an optimization problem with a size considerable. Then, the
𝑦𝑏 Diagonal matrix of the primitive admittances problem resolution it is very difficult with great memory
of the electrical system requirement from solvers, as it can check in [4].
𝐵𝑟−1 Reduced inverse bus admittance matrix In [5], [6] & [7] it applied decomposition techniques, with
𝑃𝑔𝑀 Maximum generator power [MW] purpose of interpret the optimization as two-stage problem,
𝑆𝐿 Bus slack specifically the BD technique. This division naturally induces
an iterative resolution algorithm in which the problems
Sets associated with the variables of the first stage (master problem)
𝑌 Index set of candidate transmission lines and those associated with the variables of the second stage
𝛺𝑙 Index set of existing circuits of system (subproblem) are solved consecutively. The Benders
𝛺𝑏 Index set of bus system decomposition optimality accord to author [8], just can reach a
𝑆 Index set of blocks time of the load near-optimal global solution. In this work the optimality is
modeling tested by disjunctive formulation.
𝑇 Index set of planning period Consistent in the previous discussion and literature review it
𝐸 Index set uncertainty scenarios is proposed the following contributions:
𝐺𝐵 Index set of candidate generators units
𝐺𝐸 Index set of existing generators units 1) The application of the Power Transfer Distribution Factors
𝐺𝑟 Index set of virtual generators (PTDF) widely used in literature [1], [4], [5] & [9], which
have been also used in GCEP & TCEP as an alternative
approach to eliminate the voltage angle from the power
I. INTRODUCTION flow formulation replacing by only one equation without
THE capacity expansion planning (CEP) problem is known sacrificing optimality. Accordingly, the size of the PTDF
as the co-optimization of generation and transmission capacity formulation in terms of number of variables and
expansion planning problem, and it is an issue of determining: constraints, becomes much smaller than that in the
(1) how much capacity to add, (2) when and what types of conventional angle-based formulation (disjunctive model).
generation and transmission are needed, and (3) where to locate 2) In this study, the uncertainty of the input variables will be
new generating units and transmission lines in order to supply modeled using the two-stage stochastic mixed-integer
adequately the load customers for a foreseen future, typically problem (SMIP) [10] based on the node-variable-
10-30 years. Because of the long-term horizon associated to formulation [11]. The first stage variables (investment
CEP, uncertainty is presented at many different levels in the decisions of new power units and new transmission
decision-making process: load forecasting, hydro energy elements) will be the decisions to make under uncertainty,
availability, fuel prices, emissions price, and renewable energy, and these variables are the same for all different
among other variables. Traditionally, the CEP problem has realizations of the uncertain parameter (in this case long-
been solved separately not only because of the infrastructure term load forecasting). The second stage variables (in this
involved in each type of decision is different, but also because case the power generation of each unit) must supply the
of the way each type of investment modifies the structure of the load customers for each scenario considering a DC-OPF
power flow formulation. While the generation capacity problem [12].
expansion planning (GCEP) [1] problem only modifies the 3) It is proposed to apply a Benders decomposition algorithm
supply curve locally at each electrical bus; i.e., local to solve the stochastic CEP problem (SC-CEP) using only
modifications in the operational problem, the transmission the PTDF to model both existing and candidate/future
capacity expansion planning (TCEP) problem [2] modifies the transmission power flows. Therefore, the main
admittance matrix of the associated power flow problems; i.e., contribution of this idea is related to obtain the Benders
global modifications in the operational problem. cuts for the candidate transmission lines and units
Widely used in literature for solve CEP, the disjunctive generators using the local marginal prices without
formulation was proposed for the first time in [2] and it present including the law voltage Kirchhoff (LVK) disjunctive
like to mixed integer-linear programing problem (MILP). This constraints; i.e., the Lagrange multipliers for the LVK
model allows to linearize the problem from the classical DC disjunctive constraints are computed based on the only
formulation [3], adding disjunctive restrictions that allow to application of the marginal price for the Slack bus and the
represent the operation of the system according to the PTDF.
investments in transmission and generation. Although, this
formulation presents an improvement in linearity and efficacy The rest of the paper is organized as follows. In Section II
the stochastic co-optimization expansion planning, disjunctive ∀𝑦, ∀𝑘𝑚, ∀𝑠, ∀𝑡, ∀𝑒
formulation is formulated. In Section III the Benders
decomposition is applied to disjunctive formulation CEP. In 0 ≤ 𝑃𝑔,𝑠,𝑡,𝑒 ≤ 𝑃𝑔𝑀 ∙ 𝑛𝑔,𝑡
𝑀
; ∀𝑔 ∈ 𝐺 𝐵 , ∀𝑠, ∀𝑡, ∀𝑒 ()
Section IV it is presented the proposed formulation. In the
section V the model proposed is tested in two power electric 0 ≤ 𝑃𝑔,𝑠,𝑡,𝑒 ≤ 𝑃𝑔𝑀 ; 𝐸
∀𝑔 ∈ 𝐺 , ∀𝑠, ∀𝑡, ∀𝑒 ()
system. The paper conclusions are summarized in Section VI.
II. STOCHASTIC CO-OPTIMIZATION EXPANSION PLANNING, |𝑓𝑘𝑚,𝑠,𝑦,𝑡,𝑒 | ≤ 𝑓𝑀 ; ∀𝑠, ∀𝑡, ∀𝑒 ()
𝑘𝑚
DISJUNCTIVE FORMULATION
In this study, it is assumed that (1) the electricity market is |𝑓̂ 𝑘𝑚,𝑠,𝑦,𝑡,𝑒 | ≤ 𝑋𝑘𝑚,𝑦,𝑡 ∙ 𝑓𝑀 ; ∀𝑦, ∀𝑘𝑚, ∀𝑠, ∀𝑡, ∀𝑒 ()
𝑦,𝑘𝑚
centrally operated -the Chilean situation- and (2) the generation
companies do not have the ability to accomplish local market 𝜋 ()
power. Therefore, the CEP optimization problem is formulated |𝛿𝑚,𝑠,𝑡,𝑒 | ≤ ; ∀𝑠, ∀𝑡, ∀𝑒
2
as the minimization of the following costs: (i) investment costs
𝑀
0 ≤ ∑ 𝑋𝑘𝑚,𝑦,𝑡 ≤ 𝑡 ∙ 𝑋𝑘𝑚,𝑦,𝑡 ; ∀𝑦, ∀𝑡 ()
and (ii) variable costs (operation) [13]. The investment problem
considers integer and binary decision variables. On the 𝑘𝑚∈Ω𝑙
contrary, the operational problem is modeled using an optimal 0 ≤ 𝑛𝑔,𝑡 ≤ 𝑡 ∙ 𝑛𝑔𝑀 ; ∀𝑔 ∈ 𝐺 𝐵 , ∀𝑡 ()
power flow (OPF) [12]. This optimization problem commonly
is formulated thought of the disjunctive model [2], as follows: 𝑋𝑘𝑚,𝑦,𝑡 ≥ 𝑋𝑘𝑚,𝑦,𝑡−1 ; , ∀𝑦, ∀𝑘𝑚, ∀𝑡 ∈ 𝑇/𝑡 > 1 ()
Subject to:
• Constraint (7).
• Constraints (15)-(18).
Benders cuts corresponding to each iteration (𝑙) are defined
as follows:
𝑙 𝑙
𝜃 + [(𝛼𝑘𝑚,𝑦,𝑡 + 𝛼𝑘𝑚,𝑦,𝑡 𝑀
)𝑓𝑘𝑚,𝑦 ()
𝑙 𝑇
𝑙
− (𝛽𝑘𝑚,𝑦,𝑡 + 𝛽𝑘𝑚,𝑦,𝑡 ) 𝑀] 𝑋𝑘𝑚,𝑦,𝑡
𝑙 𝑇
𝑙
+ (𝜋𝑔,𝑡 + 𝜋𝑔,𝑡 ) 𝑃𝑔𝑀 𝑛𝑔,𝑡
𝑙
≥ 𝑂𝐹𝑂𝑃𝐹
𝑙 𝑙 𝑀
+ [(𝛼𝑘𝑚,𝑦,𝑡 + 𝛼𝑘𝑚,𝑦,𝑡 )𝑓𝑘𝑚,𝑦
𝑙 𝑇
𝑙
− (𝛽𝑘𝑚,𝑦,𝑡 + 𝛽𝑘𝑚,𝑦,𝑡 𝑙
) 𝑀] 𝑋𝑘𝑚,𝑦,𝑡 Fig. 2 Benders algorithm.
𝑙 𝑇
𝑙 𝑙
+ (𝜋𝑔,𝑡 + 𝜋𝑔,𝑡 ) 𝑃𝑔𝑀 𝑛𝑔,𝑡
IV. PROPOSED FORMULATION
This paper aims to represent the transmission system by the
where 𝑂𝐹𝑙𝑂𝑃𝐹 is the operation SP function objective in the
PTDF matrix [15], eliminating the voltage angle from the
iteration l; on the other hand 𝛼𝑘𝑚,𝑦,𝑡 , 𝛼𝑘𝑚,𝑦,𝑡 , 𝛽𝑘𝑚,𝑦,𝑡 , 𝛽𝑘𝑚,𝑦,𝑡 , power flow formulation. In addition, it is included the Benders
𝜋𝑔,𝑡 , 𝜋𝑔,𝑡 are the dual variables from disjunctive constraints of decomposition applied to the whole problem already presented
the operational SP (constraints (13), (9) and (10) respectively. in the previous section. However, with this new formulation it
These multipliers are two for each constraint due to the is proposed to obtain the dual variables without the needed to
absolute value function). include the disjunctive transmission constraints.
B. Operation subproblem formulation A. Subproblem representation
The function objective can be formulated as follows: By the dual representation of the operational SP, the
following relationships are obtained with respect to the dual
() transmission and generation variables ((24) y (25)),
OFOPF = Min ∑ αt [∑ ∑ 𝑃𝑒 ∙ ℎ𝑠 ∙ 𝐶 𝑂𝑃𝐸 ] respectively:
t∈T e∈E s∈S 𝜆𝑘 − 𝜆𝑚 + 𝛼𝑘𝑚,𝑦 − 𝛼𝑘𝑚,𝑦 + 𝛽 − 𝛽𝑘𝑚,𝑦 = 0 ()
𝑘𝑚,𝑦
Subject to:
• Constraints (5) y (6). 𝐹𝐶𝑔 − 𝜆𝑘 + 𝜋𝑔 − 𝜋𝑔 = 0 ()
• Constraints (8)-(14).
It is worth to emphasizing that depending on the MP
The vector of dual variables must be obtained from each investment decisions and the saturation level in the new
iteration of the SP (𝑢(𝑣) : 𝛼𝑘𝑚,𝑦,𝑡 , 𝛼𝑘𝑚,𝑦,𝑡 , 𝛽𝑘𝑚,𝑦,𝑡 , 𝛽𝑘𝑚,𝑦,𝑡 , transmission lines, a new equivalence could be obtained for the
Benders cuts. In TABLE 1TABLE 1 is summarized the Con formato: Fuente: 10 pto
𝜋𝑔,𝑡 , 𝜋𝑔,𝑡 ). different cases for the transmission investments. For more
information, authors recommend to review in detail [5].
C. Convergence criterium
For the convergence, two bounds are defined, the upper and
the lower bounds:
TABLE 1: Benders cuts according to the investment and maximum ()
power flow for the candidate lines. ∑ 𝑃𝑇𝐷𝐹𝑘𝑚,𝑘 (𝑃𝑔,𝑠,𝑒 − 𝑑𝑘,𝑠,𝑒 ) ≥ −𝑓𝑀𝑘𝑚 ; ∀𝑘𝑚, ∀𝑠, ∀𝑒
𝑘∈Ω𝑏
Investment and
saturation case Dual multipliers − ∑ 𝑃𝑇𝐷𝐹𝑘𝑚,𝑘 (𝑃𝑔,𝑠,𝑒 − 𝑑𝑘,𝑠,𝑒 ) ()
Benders cuts
of the candidate equivalence 𝑘∈Ω𝑏
line 𝑀
𝑙 𝑇 ≥ −𝑓𝑘𝑚 ; ∀𝑘𝑚, ∀𝑠, ∀𝑒
𝛽𝑘𝑚,𝑦,𝑡 = 0; 𝜃 + [|𝜆𝑙𝑘 − 𝜆𝑙𝑚 |𝑓𝑘𝑚,𝑦
𝑀
] 𝑋𝑘𝑚,𝑦,𝑡
𝑇 0 ≤ 𝑃𝑔,𝑠,𝑒 ≤ 𝑃𝑔𝑀 ; ∀𝑔, ∀𝑠, ∀𝑒 ()
𝑙−1
𝑙
𝛽𝑘𝑚,𝑦,𝑡 = 0; + (𝜆𝑙𝑘
− 𝐹𝐶𝑔 ) 𝑛𝑔,𝑡 𝑃𝑔𝑀
𝑋𝑘𝑚,𝑦,𝑡 =0 𝑙
𝑙−1 𝑀 => ≥ 𝑂𝐹𝑂𝑃𝐹 On the other hand, if we take advantage again of the dual
𝑓𝑘𝑚,𝑦 < 𝑓𝑘𝑚,𝑦 𝑙 𝑙 𝑙 𝑀 𝑇 𝑙
𝑙
𝛼𝑘𝑚,𝑦,𝑡 − 𝛼𝑘𝑚,𝑦,𝑡 + [|𝜆𝑘 − 𝜆𝑚 |𝑓𝑘𝑚,𝑦 ] 𝑋𝑘𝑚,𝑦,𝑡 formulation of the subproblem but with the PTDF factors, we
𝑇
= |𝜆𝑙𝑘 − 𝜆𝑙𝑚 | + (𝜆𝑘 − 𝐹𝐶𝑔 ) 𝑃𝑔𝑀 𝑛𝑔,𝑡
𝑙
can reach the following expression:
𝑙
𝛼𝑘𝑚,𝑦,𝑡 = 0; 𝑇
𝑙
𝜃 + [|𝜆𝑙𝑘 − 𝜆𝑙𝑚 |𝑀] 𝑋𝑘𝑚,𝑦,𝑡
𝑙−1
𝑋𝑘𝑚,𝑦,𝑡 =1
𝛼𝑘𝑚,𝑦,𝑡 = 0;
+ (𝜆𝑙𝑘 − 𝐹𝐶𝑔 )
𝑇
𝑃𝑔𝑀 𝑛𝑔,𝑡 𝐹𝐶𝑔 − 𝜆𝑆𝐿 + 𝜇̅𝑘𝑚 ∑ ∑ 𝑃𝑇𝐷𝐹𝑘𝑚,𝑘 ()
𝑙−1 𝑀
=> 𝑇
|𝑓𝑘𝑚,𝑦 | < 𝑓𝑘𝑚,𝑦 𝑙
𝑙
𝑙
≥ 𝑂𝐹𝑂𝑃𝐹 + [|𝜆𝑙𝑘 − 𝜆𝑙𝑚 |𝑀] 𝑋𝑘𝑚,𝑦,𝑡
𝑙
𝑘∈Ω𝑏 𝑘𝑚∈Ω𝑙
𝛽𝑘𝑚,𝑦,𝑡 − 𝛽𝑘𝑚,𝑦,𝑡 𝑇
+ (𝜆𝑘 − 𝐹𝐶𝑔 ) 𝑃𝑔𝑀 𝑛𝑔,𝑡
𝑙 𝑒𝑥
= |𝜆𝑙𝑘 − 𝜆𝑙𝑚 | − 𝜇𝑘𝑚 ∑ ∑ 𝑃𝑇𝐷𝐹𝑘𝑚,𝑘 + 𝜋𝑔
𝜃 𝑘∈Ω𝑏 𝑘𝑚∈Ω𝑙
𝑙 𝑙
+ [(𝛼𝑘𝑚,𝑦,𝑡 + 𝛼𝑘𝑚,𝑦,𝑡 𝑀
)𝑓𝑘𝑚,𝑦 − 𝜋𝑔𝑒𝑥 = 0
𝑙
− (|𝜆𝑙𝑘 − 𝜆𝑙𝑚 | + 𝛼𝑘𝑚,𝑦 Now, if we use (25) to obtain an expression from which we
𝑙 𝑙 𝑇
𝛼𝑘𝑚,𝑦,𝑡 𝑜𝑟 𝛼𝑘𝑚,𝑦,𝑡 𝑙
− 𝛼𝑘𝑚,𝑦 )𝑀] 𝑋𝑘𝑚,𝑦,𝑡 can obtain the nodal price (𝜆𝑘 ) and replace it in (31), we can
≠0; 𝑇
𝑙−1
𝑋𝑘𝑚,𝑦,𝑡 =1 + (𝜆𝑙𝑘 − 𝐹𝐶𝑔 ) 𝑃𝑔𝑀 𝑛𝑔,𝑡 arrive to the following relation:
𝑎𝑛𝑑
𝑙−1
|𝑓𝑘𝑚,𝑦 𝑀
| = 𝑓𝑘𝑚,𝑦 𝑙
𝑙
𝛽𝑘𝑚,𝑦,𝑡 − 𝛽𝑘𝑚,𝑦,𝑡
𝑙
≥ 𝑂𝐹𝑂𝑃𝐹
𝜆 = 𝜆 − 𝜇̅ ∑ ∑ 𝑃𝑇𝐷𝐹 ()
𝑙 𝑙
+ [(𝛼𝑘𝑚,𝑦,𝑡 + 𝛼𝑘𝑚,𝑦,𝑡 𝑀
)𝑓𝑘𝑚,𝑦 𝑘 𝑆𝐿 𝑘𝑚 𝑘𝑚,𝑘
= |𝜆𝑙𝑘 − 𝜆𝑙𝑚 | 𝑙 𝑘∈Ω𝑏 𝑘𝑚∈Ω𝑙
− (−|𝜆𝑙𝑘 − 𝜆𝑙𝑚 | + 𝛼𝑘𝑚,𝑦
𝑇
𝑙
− 𝛼𝑘𝑚,𝑦 𝑙
)𝑀] 𝑋𝑘𝑚,𝑦,𝑡 + 𝜇𝑘𝑚 ∑ ∑ 𝑃𝑇𝐷𝐹𝑘𝑚,𝑘
𝑇
+ (𝜆𝑙𝑘 − 𝐹𝐶𝑔 ) 𝑃𝑔𝑀 𝑛𝑔,𝑡
𝑙 𝑘∈Ω𝑏 𝑘𝑚∈Ω𝑙
This demonstrates that it is possible to obtain the LMP of
Note that the dual variables associated with the disjunctive bus k using the PTDF and the Lagrange multipliers. Therefore,
constraints of generation are not included. According to [5] it is the Benders cuts can be obtained without the needed to
concluded that for any case of investment in generation and formulate the disjunctive constraints.
level of saturation in the dispatch, the Benders cuts does not V. STUDY CASES
present variations.
It can be said that if there is no saturation in the new The proposed formulation is evaluated using two electrical
transmission element independently of the building decision, power systems. The 6-bus Garver and the Chilean National
the dual multipliers could be computed using the difference of Electrical System (SEN) modified 156-bus. The mathematical
the nodal prices between buses k and m. In the same Table, it is formulation is modeled in MATLAB and the optimization
shown the proposed equivalence. problem is solved using Gurobi on a 64-bit operating system
Now the problem lies in how to obtain the dual variables computer with 128 GB RAM and an Intel Xeon 2.2 GHz
corresponding to the nodal prices. Because the formulation processor.
with PTDF only formulates a multinodal balance from which A. The Garver system
only the nodal price of the slack bus is obtained.
The Garver 6-bar [16] system was tested as the first case
B. Local marginal prices (LMP) using the PTDF formulation study. The transmission data are obtained from [13], while the
generator data are shown in TABLE 2TABLE 2.
According to [15], the PTDF formulation for the TABLE 2: Generator data for Garver system.
operational problem has the following mathematical Max. power CI OM&A FC
Bus
formulation: [MW] [USD/MWh] [USD/MW] [USD/MWh] Con formato: Fuente: 8 pto
1 3x30 - - 14.08
() 1 1x60 - - 22.11 Con formato: Fuente: 8 pto
FOOPF = Min ∑ αt [∑ ∑ 𝑃𝑒 ∙ ℎ𝑠 ∙ 𝐶 𝑂𝑃𝐸 ] 3 2x60 - - 25.95
Con formato: Fuente: 8 pto
t∈T e∈E s∈S 3 2x120 300 000 9 000 20.41
Subject to: 6 1x120 250 000 7 500 25.95 Con formato: Fuente: 8 pto
6 2x240 350 000 10 500 14.08
Con formato: Fuente: 8 pto
()
∑ 𝑃𝑔,𝑠,𝑒 + ∑ 𝑟𝑔,𝑠,𝑒 = ∑ 𝑑𝑘,𝑠,𝑡,𝑒 ; ∀𝑠, ∀𝑒 For all results, the reserve used is 20% and a VoLL cost of Con formato: Fuente: 8 pto
𝑔∈𝐺𝑟
10,000 [USD / MWh].
𝑔∈𝐺𝐸 𝑘∈Ω𝑏
Con formato: Fuente: 8 pto
In order to check, the performance of the proposed If we compare the statistics of the two formulations, there is
formulation (PF), we compare the results considering the a very important reduction in the number of variables and
disjunctive problem (CP). constraints close to 88% and 93% respectively applying the PF.
Also, it is been highlight the reduction in the solver time as
1) Static stochastic case: shown in TABLE 5TABLE 5. Con formato: Fuente: Sin Cursiva
Three uncertainty scenarios are formulated with their
respective probability of occurrence and demand factors. These TABLE 5: Simulation statistics for the multi-stage problem.
are shown in TABLE 3TABLE 3.
Solver
Con formato: Fuente: Sin Cursiva
Formulation Variables Constraints calculation
TABLE 3: Uncertainty scenarios. time [s]
CP 378 570 1 351 243 8093
Probability of
Scenarios Load factors Master PF 181 532 755 in 360
occurrence [%]
Subproblem PF 45 360 95 760 iterations
Low 0.9 30
Average 1 60
High 1.1 10 The convergence criterion for Benders algorithm of FP is
equal to 0.001%. While the PC solution is set to a gap of
The results are the same to the presented in [9] and [4] for 0.001%.
both formulations (CP & PF). The solution considers one On the other hand, the results in the investment plans have
circuit 3-5 and three circuits 4-6. The statistics in terms of differences between both formulations. In this case, it can be
computational simulation time are shown in TABLE 4. said that the PF presents optimal local solutions but not global
optimality, this is discussed in [18]. Contrary to what happens
TABLE 4: Results static stochastic Garver. with the solutions delivered by the CP. This problems about
optimality disclose the PF sensibility with respect to dual
Solver time variables in large scale optimization problems. TABLE 6 and
Formulation Variables Constraints
[s]
CP 132 269 0.47
TABLE 7TABLE 7 show these differences for investment in Con formato: Fuente: Sin Cursiva
Master PF 19 2 0.6 in 14 generation and transmission respectively.
Subproblem PF 7 27 iterations
TABLE 6:Generation investment multi-stage Garver system.
The proposed formulation (PF) decreases the number of
Gen/Year 1 2 3 4 5
variables and constraints is evident (81% & 90% respectively).
G3 120 MW 0 0 0 0 1
However, there is an increase in the simulation time with
CP
G6 120 MW 0 0 0 0 0
respect to the CP formulation. G6 240 MW 0 1 0 0 0
For compare both formulations, CP & PF in gap terms. The G3 120 MW 0 0 0 1 0
PF
G6 120 MW 0 0 0 0 1
F
G6 120 MW 0 0 0 0 1
average week type, as shown in ¡Error! No se encuentra el G6 240 MW 0 0 0 0 0
origen de la referencia.Fig. 3Fig. 3 This demand factor is
utilized in all buses from Garver system and it scaled by the
maximun value of every load. TABLE 7: Transmission investment multi-stage Garver system.
1
Circuit/Year 1 2 3 4 5
2-3 0 0 0 0 0
0.9
CP
3-5 0 0 0 0 0
0.8 4-6 0 2 0 0 0
Factor de Demanda [-]
2-3 0 0 0 1 0
PF
0.7 3-5 0 0 0 0 1
4-6 0 1 1 0 0
0.6
Circuit/Year 6 7 8 9 10
0.5
2-3 0 0 1 0 0
CP
0.4
3-5 1 0 0 1 0
4-6 0 0 0 0 0
2-3 1 0 0 0 0
PF
20
has to do with the reinforcement of the existing zonal system is
1.8
Crecimiento Demanda
18 the expansion of the Crucero-Chuquicamata, Crucero-
1.6 16 Encuentro and Domeyko-Sulfuro sections, due to the high
1.4 14
energy demand from mining companies, which have flat
Crecimiento [%]
Demanda [W]
1.2 12
0.4 4
San Luis-Quillota and Polpaico-Quillota stretches.
0.2 2 According to the results of the generation expansion, a high Con formato: No conservar con el siguiente
0
0 5 10
Año de planificación
15 20
0 RES penetration is foreseen for future years. To analyze the
consequences in the operation of the SEN in the future, Fig.
Fig. 4 Projection maximum load and annual growth for the SEN.
7Fig. 7¡Error! No se encuentra el origen de la referencia.
On the other hand, the modeling of demand is based on 24-time and Fig. 8Fig. 8¡Error! No se encuentra el origen de la
blocks. This corresponds to a type day of the year. This referencia. show the sum of the shipments of conventional
modeling is divided into two types for residential bars and units and RES for the 24 modeled time blocks. For the years
industrial / mining bars. This is observed in Fig. 5Fig. 5¡Error! 2026 and 2036 respectively.
No se encuentra el origen de la referencia. and Fig. 6Fig.
CEP), using the decomposition algorithm of Benders and
PTDF. The feasibility of the new formulation was validated in
small-scale problems, such as the static Garver. While in larger
problems such as multi-period SEN, optimal local solutions
were obtained, which presented higher costs than those
obtained from the CP. Based on this, it is concluded that the
Benders decomposition method is very sensitive in its
decisions, in relation to the dual variables that the subproblem
gives it. That is why for planning of many years and large
systems, the accuracy of the decomposition method is
deteriorated, obtaining feasible local solutions, but not optimal
Fig. 7 Conventional v/s RES dispatch year 2026. in terms of minimizing operating and investment costs. On the
other hand, there are notable improvements in the reduction of
the number of variables and restrictions, as well as in the
calculation times by the CF with respect to the CP.
Part of the future work that emerges from this work is
related to include uncertainty in the generation RES, including
restrictions of Unit Commitment (UC) considering the high
penetration RES in the near future and add hydrothermal
coordination and security analysis in the algorithm.
ACKNOWLEDGMENT