Previsão de demanda para controle de estoque: aplicação de redes neurais artificiais em séries temporais

Abstract

Nowadays, perform a demand forecasting that present the smallest error is a frequent challenge for organizations. In general, they have chosen to use qualitative models to predict demand and sales. However, inaccurate and biased results are found in the literature, as the method includes factors such as judgment, experience, opinions and intuition. In addition, many companies prefer to base their forecasts much more on their managers’ experience than on mathematical models. It is known that a forecast with high accuracy can lead to large savings, resulting not only a financial return, but also an increase in competitiveness and customer satisfaction. In this sense, this work proposed the use of artificial neural networks to solve the problem, because they present learning capacity, can process the variables, and through them, create a more assertive prediction about future demand and sales. This work studies the demand forecasting process in a commercial unit of the pet shop branch from Ponta Grossa-PR. The results obtained with different versions of the Perceptron Multilayer network show the viability of the proposal.