DeepMAP : deep modular attention for time-series prediction in multi-station environments
We propose model based on deep neural networks for time-series prediction at a specific site using information from multiple measuring stations. The key aspects of this model is the presence of an attention mechanism that dynamically determines the importance of the information provided by the stati...
- Autores:
-
Roncancio Pinzón, Jonathan Steven
- Tipo de recurso:
- Fecha de publicación:
- 2021
- Institución:
- Universidad de los Andes
- Repositorio:
- Séneca: repositorio Uniandes
- Idioma:
- spa
- OAI Identifier:
- oai:repositorio.uniandes.edu.co:1992/53459
- Acceso en línea:
- http://hdl.handle.net/1992/53459
- Palabra clave:
- Redes neuronales (Computadores)
Calidad del aire
Análisis de series de tiempo
Ingeniería
- Rights
- openAccess
- License
- https://repositorio.uniandes.edu.co/static/pdf/aceptacion_uso_es.pdf
| Summary: | We propose model based on deep neural networks for time-series prediction at a specific site using information from multiple measuring stations. The key aspects of this model is the presence of an attention mechanism that dynamically determines the importance of the information provided by the stations to conduct the prediction process and a structure that allows for the implementation of an end-to-end learning scheme and that can be interpreted after training. Through experiments in air-quality prediction and solar irradiance forecasting, we show that the proposed model is simple but effective to solve time-series prediction problems in multisation environments compared with other data fusion techniques. |
|---|