Emanuel Pimentel Barbosa

Short-term forecasting of industrial electricity consumption in Brazil

This paper presents short-term forecasting methods applied to electricity consumption in Brazil. The focus is on comparing the results obtained after using two distinct approaches: dynamic non-linear models and econometric models. The first method, that we propose, is based on structural statistical models for multiple time series analysis and forecasting. It involves non-observable components of locally linear trends for each individual series and a shared multiplicative seasonal component described by dynamic harmonics. The second method, adopted by the electricity power utilities in Brazil, consists of extrapolation of the past data and is based on statistical relations of simple or multiple regression type. To illustrate the proposed methodology, a numerical application is considered with real data. The data represents the monthly industrial electricity consumption in Brazil from the three main power utilities: Eletropaulo, Cemig and Light, situated at the major energy-consuming states, Sao Paulo, Rio de Janeiro and Minas Gerais, respectively, in the Brazilian Southeast region. The chosen time period, January 1990 to September 1994, corresponds to an economically unstable period just before the beginning of the Brazilian Privatization Program. Implementation of the algorithms considered in this work was made via the statistical software S-PLUS. Copyright

State-space analysis of soil data: an approach based on space-varying regression models

The assessment of the relationship among soil properties (such as total nitrogen and organic carbon) taken along lines called transects is a subject of great interest in agricultural experimentation. This question has been usually approached through standard state-space methods by some authors in the soil science literature. Important limitations of the mentioned procedures used in practice are pointed out and discussed in this paper, specially those related to the model parameters, meaning and practical interpretation. In the standard state-space approach, based on an autoregressive structure, it does not present any parameters that express the variables relationship at the same point in space, but only at lagged points. Also, its model parameters (in the transition matrix) have a global meaning and not a local one, not expressing more directly the soil heterogeneity. Therefore, the objective here is to propose an alternative state-space approach, based on dynamic (space-varying parameters) regression models in order to avoid the mentioned drawbacks. Soil total nitrogen and soil organic carbon samples were collected on a Typic Haplustox. Samples were taken along a line (transect) located in the middle of two adjacent contour lines. The transect samples, totaling 97, were collected in the plow layer (0-0.20 m) at points spaced 2 meters appart. Results show the comparative advantages of the proposed method (based on an alternative state-space approach) in relation to the standard state-space analysis. Such advantages are related to a more adequate incorporation of soil heterogeneity along the spatial transect resulting in a better model fitting, and greater flexibility of the models building process with an easier interpretability of the local model coefficients.

Soil-plant interaction evaluated by the state-space approach.

The interaction soil-plant was evaluated using a state space approach (dynamic model) comparatively to a static regression model using both, standard and sequential estimations. Experimental soil data consisted of bulk density, macroporosity, microporosity and porosity of a dark red latosol, and plant data of root density in length per unit volume, of a forage-oat crop. Among these, only soil porosity had a good correlation with the root system density, which is the response variable of this study. A static regression model written in the state space form with a sequential estimation, gave a R2 coefficient of 0.69, comparatively to a conventional (non-sequential) regression model, which gave a R2 coefficient of only 0.59. This soil-plant relation was better described by a dynamic regression model, which gave a R2 coefficient greater than 0.98. These results indicate the advantage of the state space approach in relation to the other more conventional regression methods.

Range Control Charts Revisited: Simpler Tippett-like Formulae, Its Practical Implementation, and the Study of False Alarm

This article presents simpler alternative formulae and procedures of implementation to deal with the relative range statistic used in the range control chart for process dispersion monitoring. The chart performance is assessed considering false alarm comparison between our exact limits charts versus normal approximated ones, which shows the serious drawbacks of such misplaced control limits. These much simple tools introduced here, we believe, will permit the presentation of R charts more transparently and without unrealistic normal approximations, avoiding the serious limitations of such “ad hoc” practice.

A multiplicative seasonal growth model for multivariate time series analysis and forecasting

This paper is devoted to a model for analysis and forecasting of vector time series and the corresponding procedure of Bayesian sequential estimation. This model can also be viewed as a multivariate extension of the (univariate) seasonal growth multiplicative model(Harrison, 1965; Migon, 1984). The basic structure of this multivariate model consists of a locally linear trend component for each individual series and a shared multiplicative seasonal component, common to all marginal series.The procedure of sequential estimation is based on analytic approximations to obtain a conjugate analysis and represents a nonlinear extension of the algorithm presented by Barbosa and Harrison (1992). Details of the proposed procedure and its practical implementation are shown, and two numerical examples are provided.

Estimação de Funções de Pedotransferência: Redes Neurais Recorrentes de Pesos Variáveis no Espaço para Predição de Propriedades do Solo

Resumo—O objetivo do artigo é propor um novo modelo de regressão que relacione variáveis do solo de medição dif́icil ou complexa com outras variáveis de medição mais fácil e menos dispendiosa, visando a predição da primeira com base em dados das demais. As medições são tomadas ao longo de linhas do solo chamadas transeções espaciais ou transects. O estudo dessas relações (funções de pedotransferência), entretanto, apresenta a complexidade da presença simultânea de 3 elementos: dependência espacial dos dados, não-homogeneidade do solo e não-linearidade da relação. Os principais modelos usualmente considerados na literatura para tais relações (principalmente espaço de estado linear e redes neurais feedfoward) têm a limitação de expressarem apenas duas das três caracteŕisticas do problema. De modo a superar essa limitação, é proposto aqui um modelo de regressão para a função de pedotransferência baseado em rede neural recorrente (a realimentação ajuda a expressar melhor a dependência espacial), mas, ao contrário de uma rede padrão de pesos fixos, com pesos variando suavemente ao longo do espaço, de modo a incorporar a nãohomogeneidade do solo. O algoritmo desenvolvido para estimação do modelo e predição é baseado em extensão não-linear de 2a ordem do filtro de Kalman, o que corrige substancialmente as deficiências do FK estendido de 1a ordem. As vantagens comparativas do modelo proposto em relação aos outros modelos propostos na literatura é mostrada, considerando-se diferentes medidas de performance de predição para os extremos do transect.

Approximation and quantiles of the distribution of the modified likelihood ratio criteria for covariance matrix testing and monitoring

Sugiura [11] gives an asymptotic expansion of the modified likelihood ratio criteria for testing the hypothesis that a covariance matrix is equal to a given matrix. An improvement of this expansion is presented here. Numerical comparisons via simulation with the original approximation to the criterias distribution confirm the superiority of our expansion. This enable us to use the proposed method in usual hypotheses testing and in applications where extreme tail quantiles are necessary, as for instance, for monitoring dispersion in multivariate processes quality control charts.