Mathematical / Statistical and Physical / Meteorological Models for Short-term Prediction of Wind Farms Output

AUTOR(ES)
FONTE

IBICT - Instituto Brasileiro de Informação em Ciência e Tecnologia

DATA DE PUBLICAÇÃO

02/12/2005

RESUMO

(...)The rapid increase of the wind power penetration in the conventional grids all over the world has been requiring (from electricity companies and wind farm promoters) a special attention to the short-term prediction of wind farms output. Fomented by interest and necessity, companies and promoters are collaborating with universities and other research centres in order to develop a technology that is still not completely mature. The joint effort to the development of prediction models is justified by operational reasons (due to the fluctuations of the wind power) and competitiveness (with respect to the value to be added to the wind energy as a consequence of a \ precise\ estimation of the product to be offered: the energy). As a particular example, the Spanish government recently approved a Law (Real Decreto 436/2004, 12th March) to regulate the wind energy sales in the electricity daily market (spot), pressing the promoters towards an urgent search for the most precise models and the collaboration with recognized research groups to the development of models and computational tools in accordance with their necessities. Considering the described context, this Thesis tries to give a contribution to the development of a research field poorly explored yet: the \ true\ integration between two distinct and complementary research lines, mathematical/statistical and physical/meteorological models for short-term prediction of wind farms output. In this sense, one of the first attempts is being carried out in Denmark [Madsen et al., 2000a,b; Giebel et al., 2001; Landberg, 2001]. The aim of the present work is the development of a final model (and corresponding computational tools) integrating both mathematical and physical models for short-term prediction, according to the specific objectives detailed in the Section 3.2. For the purposes of this Thesis, \ short-term\ must be understood as a forecast horizon of 2 or 3 days ahead typically, with time-step of 1 or 3 hours, being the developed models and tools basically oriented to: management of systems with wind farms, optimising the planning for the other sources (e.g., thermal units schedule) in order to increase fuel saving, to decrease the spinning reserve etc.; requirements for the wind energy sales in the spot market; scheduling of some maintenance tasks in wind farms. The final model is composed of models with two different natures: mathematical/statistical and physica/meteorological. Both models (mathematical and physical), running simultaneously, produce forecasts of the total wind farm output (cf. Figure 1.1). The final output from the final model is then a function of these forecasts, a function modelled by a proposed model called here \ best intersection point tracking (b.i.tracking)\ for a \ real-time\ operation. Regarding the mathematical model, the goal is the evaluation of structures varying from the most simple to the most complex, as, for example, autoregressive models [Box and Jenkins, 1976; Bossanyi, 1985a; Ljung, 1987; Torres et al., 2005], fuzzy logic based models [Sugeno and Yasukawa, 1993; Kariniotakis et al., 1996] and neural networks [Haykin, 1994; Castillo et al., 1999]. The mathematical/statistical models (here, purely time series based models, without meteorological data as inputs) are used by their capacity of extracting information from the time series (on-line measurements) and (with this information) generating low error estimates in a shorter forecast horizon (up to 3 or 6 hours ahead). By convention and with a didactic purpose, the term \ mathematical/statistical models\ is always employed in plural in this text, meaning each one of the different structures (autoregressive, fuzzy logic, neural networks) tested for the block called mathematical model (cf. Figure 1.1). Regarding the physical model, the goal is the evaluation of three different structures (here, called as physmod1, physmod2 and physmod3) based on the downscaling of geostrophic wind estimates [Landberg, 2001]. All these structures make use of meteorological data as inputs and the main differences between them are related to the stability regime and the definition of the inputs to the power curve modelling. With respect to the stability, physmod1 is based on neutral stability, whereas physmod2 and physmod3 are thought for non-neutral conditions. Concerning the power curve modelling, physmodl and physmod2 are based on the conversion of wind forecasts at the site of an anemometrical reference mast into forecasts of the wind farm total output power. In its turn, physmod3 is based on the individual conversion of wind forecasts at the site of a wind turbine into forecasts of the turbine output power, being the wind farm total output power computed as the sum of all turbines. The physical/meteorological models are used by the wider range of their forecast horizon, typically up to 48 or 72 hours ahead. As in the case of the mathematical/statistical models, the term \ physical/meteorological models\ is always employed in plural in this text, meaning each one of the different structures (physmodl, physmod2 and physmod3) tested for the block called physical model (cf. Figure 1.1). In this work, as a first approach, only the outputs from the two structures presenting the best results (one from the three mathematical/statistical models proven and the other from the three physical/meteorological models proven) are considered as inputs to the final model. In a future approach, all the six structures will be proven as an ensemble. The present text deals with the above mentioned problem of development of a final model (starting from its two basic models: mathematical and physical), being structured in the following way: Chapter 2 presents some definitions concerning, for example, time/spatial prediction and long-term/short-term prediction; Chapter 3 makes a brief review on historical developments and the state-of-the-art on wind power prediction models. Also, presents the problem statement, i.e., describes the contributions (specific objectives) of this Thesis; Chapter 4 describes the mathematical/statistical models and discusses results from a case study (2 wind farms at the Northeast region of Spain); Chapter 5 describes the physical/meteorological models and discusses results from a case study (1 wind farm at the Northeast region of Spain); Chapter 6 describes the final model and discusses the final results; Chapter 7 remarks the outcomes from the work and the perspectives for future improvements.(...)

ASSUNTO(S)

sistemas híbridos eletrificação rural mecânica de fluidos engenharia mecanica energia eólica aerodinâmica aeroelasticidade wind power wind farms

ACESSO AO ARTIGO

Documentos Relacionados