Rede neural artificial autoregressiva não linear aplicada a dados da estação de medição automática do INMET

Abstract

The installation of wind power systems connected to the electricity grid in Brazil, mainly in the Northeast region, has been expanding exponentially and to assess the feasibility of building a wind farm, one of the main information in the region is the wind, and through mathematical solutions it is possible to predict and estimate its energy potential. This article proposes to analyze and apply data from an automatic weather station in an artificial neural network (ANN) Nonlinear Autoregressive (NAR). The study will contribute to improving the efficiency of the wind farms, as well as the maintenance of wind turbines and also collaborate with the distribution of the energy matrix in the country, with prior wind information it will be possible to make scheduling for scheduled stops aimed at the maintenance of the wind farms and in times of greater energy production, obtain a good outflow of this energy. The database for the study of wind forecast will be used data from Instituto Nacional de Meteorologia (INMET) automatic measurement stations, which aims to collect field information, such as wind speed, temperature, wind direction and humidity at certain heights. From the history of these variables, which will be collected online in the databases of national agencies, it will be possible to train an artificial neural network (ANN) to trace the wind profile in the region for the seasons. For this, two different methodologies will be applied. In the first stage, with the support of Microsoft® Excel software, the data will be processed using statistical techniques, and in the second stage, the processed data will be submitted to an ANN-NAR. It is possible to notice that the ANN-NAR obtained a good response when implemented with INMET data, presenting a correlation above 90% cases and a root mean square error below 2.0 in the predominance of the tested networks. According to the results, it is conclusive that NAR networks are excellent options for wind forecasting