| Autor: |
Paulo Hugo Espírito Santo Lima, João Fausto Lorenzato de Oliveira, Bruno Giublin, Luiz Felipe Vieira Verçosa, Luis Arturo Gómez Malagón, Vildson Rocha Borba |
| Rok vydání: |
2021 |
| Předmět: |
|
| Zdroj: |
Anais do 15. Congresso Brasileiro de Inteligência Computacional. |
| DOI: |
10.21528/cbic2021-31 |
| Popis: |
Time series forecasting has become an important task for several industrial processes. The employment of machine learning techniques also contributed to reduce costs and increase profits. In order to achieve accurate forecasts, it is important to perform a proper selection of parameters for the forecasting models. Hybrid sequential systems reduces model uncertainty through the employment of different models in the time series and residual series. In this work, an ensemble based hybrid sequential system is proposed, where a heterogeneous ensemble is used to perform residual forecasts. In this way, a pool of four models used in the industrial scenario are employed in order to achieve accurate forecasts. The experiments were conducted on real datasets from the industry in hourly averages and daily averages. The results show that the proposed system achieved promising results, and was able to improve the accuracy of the models in the pool in several cases. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
