Moisés Lima de Menezes

HYDROELECTRIC ENERGY FORECAST

The aim of this paper is to propose a new methodology for hydroelectric energy forecasting. A new approach for selection of the number of eigenvalues in SSA is also proposed. In this paper it is proposed the hierarchical clustering associated to PCA and integrated to ARIMA models. The proposed approach is applied to forecast the affluent flow in a hydroelectric plant located at Parana River Basin, Brazil. As a matter of fact, modeling such series is quite important for the optimal dispatch of the energy generation in Brazil due to the heavy participation of hydro plants in the country (over 85% of the generated energy comes from hydro plants).

RESIDENTIAL ELECTRICITY CONSUMPTION FORECASTING USING A GEOMETRIC COMBINATION APPROACH

The forecasting of electricity consumption and demand plays a pivotal role in electric power systems planning. This paper proposes the combination of forecasts from two approaches with the aim of improving the forecasting accuracy, in order to make the best use of the installed transmission and generating capacity. In the first approach, the consumption time series is decomposed by wavelet analysis and a Box-Jenkins model is fitted to each wavelet component, following which the individual components forecasts are added to compute the total consumption forecast. The alternative approach, uses the Singular Spectrum Analysis technique to model the consumption time series in order to shrink the noise level. Thereafter, the Box-Jenkins model is used to forecast the filtered time series, producing a second forecast for the consumption series. Eventually, the two forecasts are combined geometrically in order to minimize the mean square error. The proposed methodology is illustrated by a computational experiment with the time series of residential consumption of electricity in Brazil.

ARTIFICIAL NEURAL NETWORK AND WAVELET DECOMPOSITION IN THE FORECAST OF GLOBAL HORIZONTAL SOLAR RADIATION

This paper proposes a method (denoted by WD-ANN) that combines the Artificial Neural Networks (ANN) and the Wavelet Decomposition (WD) to generate short-term global horizontal solar ra- diation forecasting, which is an essential information for evaluating the electrical power generated from the conversion of solar energy into electrical energy. The WD-ANN method consists of two basic steps: firstly, it is performed the decomposition of level p of the time series of interest, generating p + 1 wavelet orthonormal components; secondly, the p + 1 wavelet orthonormal components (generated in the step 1) are inserted simultaneously into an ANN in order to generate short-term forecasting. The results showed that the proposed method (WD-ANN) improved substantially the performance over the (traditional) ANN method.

Combining singular spectrum analysis and PAR(p) structures to model wind speed time series

Singular spectrum analysis (SSA) is a technique that decomposes a time series into a set of components, such as, trend, harmonics, and residuals. Leaving out the residual components and adding up the others, the time series can be smoothed. This procedure has been used to model Brazilian electricity consumption and flow series. The PAR(p), periodic autoregressive models, has been broadly used in modelling energy series in Brazil. This paper presents an approach of this decomposition method, by fitting the PAR(p), considering its multivariate version known as multivariate SSA (MSSA). The method was applied to a vector of two wind speed series recorded at two locations in the Brazilian Northeast region. The obtained results, when compared to the univariate decomposition of each series, were far superior, showing that the spatial correlation between the two series were considered by MSSA decomposition stage.

Introducing DBSCAN in the singular spectrum analysis procedure

This work proposes using DBSCAN to recognition of noise components of eigentriples in the grouping stage of SSA. The DBSCAN is a modern (revised in 2013) and expert method at to identify noise through regions of lower density. The hierarchical clustering method was the last innovation in noise recognition in SSA approach, implemented on package RSSA. However, it repeated in the literature that the hierarquical clustering method is very sensitive to noise, is unable to separate it correctly, and should not be used in clusters with varying densities and neither works well in clustering time series of different trends. Unlike, the density based clustering methods are effective in separating the noise from the data and dedicated to work well on data from different densities. This work shows better efficiency of DBSCAN over the others methods already used in this stage of SSA, because it allows considerable reduction of noise and provides better forecasting. The result is supported by experimental evaluations realized for simulated stationary and non-stationary series. The proposed combination of methodologies also was applied successfully to forecasting a real series of winds speed.