TIME SERIES

Author: OCAÑA PEINADO FRANCISCO.
Year: 2004.
University: GRANADA.
Place of defense: CENTRO DE INSTUMENTACIÓN CINETÍFICA.
Place of preparation: FACULTAD DE FARMACIA.

Summary: The introduction in 1970 of ARIMA models in the work of engineers EPG Box and GM Jenkins, from work performed in the previous decade, was a new approach in the treatment of the time series, as they face the classic concept in which the series was consistent with a certain pattern or trend whose departures were at the heart of analysis, this new metología assumption that the basic tool for the modernization and the analysis were lals correlations between random variables serial. Despite the limitations of these models (linear structure, temporary dicreto, need to have long historical series, etc.), ARIMA been implemented in many scientific fields, such as engineering, economics, Behavioral Sciences, etc.. Being a tool for predicting effective in the short term. The basic formulation of an ARIMA is conceivable that the series under study conforms to a second-order stochastic process (with moments of first and second-order finite), which represents the answer or output of a linear system or a momentum that input if white noise, and the transfer function of the filter ratio of two polynomials of degree finite, ie type sound. In the text of Box and Jenkins raised a generation of these models univariate the case in which the variable response time can be explained by one or more random variables other than white noise, that is, proposing the extension of the model multiple regression to the event dynamic. This new formulation gives rise to the so-called role models of transfencia (MFT), or even primarily in the area econometric models regersión dynamics. The MFT consists of two terms additives, the first part contains the explicit by exogenous variables and the second, called the process of inertia, representing the random disturbance. Part explanatory arose initially with a structure of delays linalmente distributed from which they are derived most compact models in terms of delays rationally distributed (Pankratz, 1991. Otero, 1993). As for the process of inertia, making basic part of which follows a model ARIMA. Based on the constraints before the ARIMA commented before, this thesis aims to develop models more suitable to represent the party unexplained, thus allowing optimize their predictive capacity and make more general models. Specifically, the treatment is introduced is based on the decomposition of the process of inertia in terms of its major components, both from a functional and unobtrusive as, as shown in the memory, this approach provides a more accurate model predictive end. The dissertation has three chapters, with the first devoted the basic concepts MFT, ilustrándose with an application to the Behavioral Sciences, where both dynamic models are developed to explain symptoms of lupus erythematosus depending on the stress of a group of patients medical protocol on a numerical scale, and vice versa, which allows to classify them into three groups according to the order of the delay in part explained. The second chapter deals with the issues mentioned above is to model the process of inertia in terms of major components from its considerción as second-order stochastic process. This helps achieve a degree of approximation as high as desired as they incorporate new variables to their representation numerable. The effectiveness of the method is shown with two applications, one type of financial, which develops a model to predict the evolution of IBEX35 from the index SP5000 of the New York Stock Exchange, and another in the environmental field for predecri the concenración pollen cypress and olive pollen from two inputs alternative but correlados one another, temperature and hours of sunshine 8. The ca 38c pítulo third deals with the functional approach to the problem and, after introducing the basic principles models CARMA, devoting special attention to order one, is the version of the MFT - time continuum, in terms of linear stochastic systems. Here are dasarrolla the process of inertia in terms of major components such as number of Karhumen-loève obtained approximations time continuous optimization of this validándose modeling through a simulated example.

Author: ARAUJO DA COSTA ALEXANDRE.
Year: 2005.
University: POLITÉCNICA DE MADRID.
Place of defense: E.T.S. INGENIEROS INDUSTRIALES.
Place of preparation: E.T.S. ING. INDUSTRIALES.

Summary: The main objective of this thesis is to develop models and computational tools for the short-term prediction of the power output of wind farms. With regard to the purposes of this thesis, 'short-term' is meant as a prediction horizon until 2 or 3 days ahead, in steps of 1 to 3 hours (range of integration time step). The models and tools are developed primarily aimed at: Â operating systems integrated with wind farms, allowing better management of other sources (for example, the scheduling of thermal) and seeking to ensure the satisfaction of the claim; Â the requirements for the negotiation of the wind in the markets daily and intraday electric power; Â planning some maintenance work in windfarms. The final model developed is composed of two sub-models of various natures: models math / statistical models and physical / weather. The mathematical models are used by its ability to extract information from the time series (measured in real time) of wind power and wind turbines for with such information, generate estimates of high correlation in a very short horizon prediction, up 6 or 12 hours later. Here, the 'correlation' must be understood in the sense strictly statistical 'linear correlation', such that the higher the correlation between the output of a model and the actual data of a wind farm, the better adjustment in such a parametric model . The models mathematical / statistical tested are kind autoregresivo, fuzzy logic and neural networks. In turn, physical models are based on estimates of geostrophic wind (predictions of the meteorological centers for wind immediately outside the planetary boundary layer). These models use statistical corrections to the law of cortadura wind vertical layer Ekman, considering the complexity of the local orography and the degree of atmospheric stability. Thus, the estimates of geostrophic wind are 'moved' to the height of the towers anemométricas wind park with the lowest possible errors. In addition, the effects of local orography complex are calculated in greater detail for a high resolution mesh around the site of the wind farm. The physical models are used by the scope of their predictions, a horizon of up to 48 or 72 hours. Output from the final model is a function of the outputs of the submodels mathematicians and physicists. By adjusting the parameters of this function is performed on an adaptive (due to the dynamic variation of these parameters) through advanced statistical tools.