Previsão da produtividade de arroz: uma aplicação de redes neurais recorrentes LSTM

Abstract

Rice, responsible for supplying the population with calories and protein, occupies a prominent position from the social and economic point of view. It is an essential product in the basic basket of Brazilian consumers. In this context, this work present an LSTM (Long Short-Term Memory) model for forecasting rice productivity in the state of Rio Grande do Sul. The database, obtained by the Rio Grandense Rice Institute (IRGA), presents a historical series, of rice productivity, of the harvests between 1921 and 2020. The forecasting model, based on LSTM Neural Networks, was implemented through the Pytorch machine learning library. The results obtained for the 2017/18, 2018/19 and 2019/20 harvests show that the forecast model provided reliable estimates for rice productivity in Rio Grande do Sul.