Analysis of Residual Autocorrelation in Forecasting Energy Consumption through a Java Program

Abstract

The study of forecasting of energy in Brazil is important for future planning, as the country has experienced crises of energy supply. And a model developed in java is an affordable and efficient tool to be used both in Brazil and in other countries. Time series analysis is highly important in many different application areas, for it allows description and modeling of a variable of interest’s behavior, thus enabling the forecasting of its future values, which serves as support for decision making. When the data used in regression analysis comprises time series, the dependency between the observations grants a dynamic quality to the regression model. In this situation, it is common to come across a problem known as residual autocorrelation, which invalidates the assumptions made about the term of error in the classical linear regression models. This paper presents a program created in Java by implementing the method of Cochrane-Orcutt for the correction of residual autocorrelation. And the application is made in the Brazilian energy final consumption forecasting.

Evaluation of Energy Distribution Using Network Data Envelopment Analysis and Kohonen Self Organizing Maps

“…According to a previous reference [22], "residuals in a regression are obtained from the difference between the observed value of the response variable and that forecast by the regression model". Therefore, in order to obtain adequate results from a regression and possibly to use it to forecast the results of new elements, it is important that the residues are as small as possible.…”