GEVAZP – Scenario Generation Model for Energy and Water Inflows
Introduction The operation and expansion planning of hydrothermal systems is a complex optimization problem, in which decisions about hydro and thermal generation must be coordinated to meet the system’s demand, the hydraulic constraints and electrical constraints regarding energy exchanges. This problem is essentially stochastic due to the great uncertainty in the natural inflows to hydroelectric plants, which justifies the need to develop tools that are able to generate future scenarios of energy and water inflow.
Energy and inflow scenarios are also widely used in power generation planning studies, where the evaluatation of probabilistic criteria is necessary, such as supply adequacy that is based on risk indices, estimated from the simulation of the power system operation for a large set of hydrological scenarios (sequences).
The GEVAZP model was developed with the purpose of generating a scenario tree for the stochastic optimization problems solved by the models used in long, mid and shortterm operation planning (NEWAVE, SUISHI and DECOMP models), taking into account the characteristics of each solution method.
The only scenario available in practice (i.e., inflow records observed in the past, known as historical time series data) is insufficient to provide an appropriately large sample to estimate risk indices with acceptable uncertainty levels. However, the basic characteristics of the historical time series can be captured by stochastic models that can produce synthetic inflow series that are different but equally likely to the historical series.
Hydrologic series with a time interval shorter than one year, such as monthly series, are characterized by the periodic behavior of their probabilistic properties, such as the mean, variance, asymmetry and autocorrelation structure. The analysis of this type of series may be done using autoregressive formulations whose parameters have a periodical behavior. This class of model is commonly known as periodic autoregressive models. Such models are referenced as PAR (p) models, in which p represents the model's order, i.e., the number of autoregressive terms in the model.
Scenario Generation for Planning and Operation Models
The Brazilian power generation system is predominantly hydraulic and has temporal and spatial coupling. These characteristics turns the power operation planning of the whole country a large and complex task. For this reason, it must be divided into several stages, where each stage uses models with different degrees of detail to represent the system and hydrologic uncertainty over study periods with different horizons (medium term, short term and daily scheduling). The uncertainty is treated differently depending on the type of representation applied to model of the generation system, and the representation of possible inflow scenarios differs for each stage of the operation planning process.
The NEWAVE (Long and MediumTerm Operation Planning Model of Interconnected hydrothermal Systems) and SUISHI (Detailed Simulation Model of Power Plant Operation for Interconnected Hydrothermal Systems) models, developed for mediumterm operation planning, simulate a large number of hydrological series, calculating probabilistic performance indices of the system for each stage of the simulation. These streamflow scenarios are organized in a parallel structure (also called fan structure), as shown in Figure 1. The NEWAVE model also considers explicitly the uncertainty concerning streamflow in the calculation of optimal operation strategy through a scenario tree, represented in Figure 2.
Figure 1 – Scenarios (parallel or fan structure) applied to the policy simulation of NEWAVE and SUISHI models
Figure 2 – Scenario Tree applied to the modeling of the stochastic problem of NEWAVE model
In the DECOMP (Shortterm Operation Planning Model for Interconnected Hydrothermal Systems) model, developed for shortterm operation planning, the uncertainty regarding inflows to the hydropower plants is represented by a hydrological inflow scenario tree (Figure 3), with probabilities associated to each branch.
Figure 3 – Scenario Tree employed in the modeling of the stochastic problem in the DECOMP model
Aiming to achieve a good representation of the stochastic streamflow process with a reduced number of scenarios, the GEVAZP model uses the Selective Sampling (SS) method, which consists of applying clustering techniques to a large number of generated hydrological scenarios (original sample) to select a representative set from the original sample of scenarios.
The hydrological scenarios used in short and mediumterm operation planning are generated using the GEVAZP model, taking into consideration the preservation of the statistical characteristics of the original stochastic process, such as mean, variance, and temporal/spatial correlations.
GEVAZP can also calculate the streamflow to both artificialinflow and incrementalinflow gauge stations. Artificialinflow gauge stations have their own operating rules, inserted as data input into the model, generally associated with a naturalinflow gauge station. For cases where the streamflow travel (water delay) time between two hydropower plants is significant, the model allows the adoption of the incrementalinflow record for the hydropower plant that is affected by the travel time.
Additionally, there is an academic version of the GEVAZP model to be used exclusively for noncommercial activities in institutions of higher education.
Graphical Interface
Currently, the GEVAZP software is part of ENCAD system, which is a graphical interface allowing several procedures such as: importing and convertion of input data among models; friendly data edition; graphical visualization of results and output reports in text formats, etc. (Figure 4).
