Exploring the application of neural networks for the prediction of vehicle emissions
In this project, data from the EPA's fuel efficiency and emissions tests was used to train neural networks so that they could predict vehicle emissions. Vehicle characteristics were taken directly from the database, driving cycles were analyzed using their velocity and acceleration profiles in...
- Autores:
-
Rodríguez Llorente, Diego
- Tipo de recurso:
- Trabajo de grado de pregrado
- Fecha de publicación:
- 2019
- Institución:
- Universidad de los Andes
- Repositorio:
- Séneca: repositorio Uniandes
- Idioma:
- eng
- OAI Identifier:
- oai:repositorio.uniandes.edu.co:1992/45206
- Acceso en línea:
- http://hdl.handle.net/1992/45206
- Palabra clave:
- Consumo de combustible
Gases de combustión
Gases de escape en automóviles
Redes neurales (Computadores)
Ingeniería
- Rights
- openAccess
- License
- https://repositorio.uniandes.edu.co/static/pdf/aceptacion_uso_es.pdf
| Summary: | In this project, data from the EPA's fuel efficiency and emissions tests was used to train neural networks so that they could predict vehicle emissions. Vehicle characteristics were taken directly from the database, driving cycles were analyzed using their velocity and acceleration profiles in order to calculate some defining characteristics, and these two groups were used as inputs to predict HC, CO, CO2, and NOx. For most of the project only HC was predicted to simplify the process with the idea that if a model was successful, then the output changed to another gas until a different model was successful with the predictions and so on. By the end of the project Multi-Task learning was applied to predict all four emissions simultaneously hoping that information could be shared between those tasks to improve the predictions. Feature engineering was applied to analyze the input variables and understand if one or more were introducing significant errors into the data and ruining the predictions. The error for the predictions was calculated using Mean Squared Percentage Error (MSPE) to compare the performance of different models. This error was consistently above 10E7% and many predictions were negative, rendering this exploration of neural networks unsuccessful. At the end of the project a phenomenon known as Emissions Deterioration was analyzed based on three different studies to understand how emissions deteriorate with the age of the vehicles. It was found that this phenomenon is much more complex than initially thought. |
|---|