Previsão de séries temporais por meio de métodos estatísticos e neurais: aplicação em uma indústria de bebidas
Given the importance of beverage industry (especially breweries) for the Brazilian economy, forecasting techniques are useful to support production planning and decision making in these companies. Through production forecasting, it is possible to estimate the resources required for raw materials tra...
| Autor principal: | Cunha, Marina Moreira |
|---|---|
| Formato: | Trabalho de Conclusão de Curso (Graduação) |
| Idioma: | Português |
| Publicado em: |
Universidade Tecnológica Federal do Paraná
2020
|
| Assuntos: | |
| Acesso em linha: |
http://repositorio.utfpr.edu.br/jspui/handle/1/12925 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
| Resumo: |
Given the importance of beverage industry (especially breweries) for the Brazilian economy, forecasting techniques are useful to support production planning and decision making in these companies. Through production forecasting, it is possible to estimate the resources required for raw materials transformation into finished products and to optimize their use, as well as to get more assertiveness about the consumer’s preferences, increasing profitability and reducing financial losses. Therefore, the present work aims to obtain a three-month-ahead forecast of a beverage industry’s weekly production, through the use of statistical methods and the training of an Artificial Neural Network. For this, the company’s historical production data was used. The methodology consists of an applied nature study, with a quantitative approach, modeling procedures and descriptive goals. With the application of the statistical methods, it was possible to verify that the Moving Average model has not fitted correctly the time series and, consequently, did not provide a trustworthy forecast, which is the reason it was not presented in this study. Also, it was found that the Simple Exponential Smoothing model (with ∝ = Ͳ,ʹʹ77) and the ARIMA (2,1,6) model presented satisfactory data adjustments, however predicted, respectively, constant values and near to an average value. Finally, the Artificial Neural Network forecasts were better adjusted to real data (used for comparison), which demonstrates its forecasting process is more assertive and reliable. The comparison between all the tested models was based in the RMSE value, calculated for the predicted values, being that the Artificial Neural Network had the lowest error, followed by the Box-Jenkins and the Exponential Smoothing models, which confirms the first method’s better efficiency in forecasting the analyzed time series and enables its use as a support for decision making processes related to the production planning. |
|---|