Machine Learning for Air Charge Estimation: A Residual Error Approach
(2025)Department of Automatic Control
- Abstract
- Improving the efficiency of combustion engines is crucial from an environmental perspective, serving as a complement to the transition towards the electrification of transportation. To optimize combustion efficiency, the amount of air in the cylinders needs to be estimated, which today is done by gray box-models based on traditional physics. This thesis examines the possibility of complementing these classical formulas with supporting machine learning models to improve the precision of the estimation. Both a Gaussian Process Regressor (GPR) and an Artificial Neural Network (ANN) are trained to estimate the airflow, and Bayesian optimization is performed to optimize the structure of the models. The results show reduced residual errors for... (More)
- Improving the efficiency of combustion engines is crucial from an environmental perspective, serving as a complement to the transition towards the electrification of transportation. To optimize combustion efficiency, the amount of air in the cylinders needs to be estimated, which today is done by gray box-models based on traditional physics. This thesis examines the possibility of complementing these classical formulas with supporting machine learning models to improve the precision of the estimation. Both a Gaussian Process Regressor (GPR) and an Artificial Neural Network (ANN) are trained to estimate the airflow, and Bayesian optimization is performed to optimize the structure of the models. The results show reduced residual errors for both models compared to the current methods. The study shows promise for utilizing machine learning models to improve combustion engine efficiency, and highlights that further studies for hardware modification and implementation are needed. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9222618
- author
- Christensson, Lukas
- supervisor
- organization
- year
- 2025
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6301
- other publication id
- 0280-5316
- language
- English
- id
- 9222618
- date added to LUP
- 2026-02-12 09:41:52
- date last changed
- 2026-02-12 09:41:52
@misc{9222618,
abstract = {{Improving the efficiency of combustion engines is crucial from an environmental perspective, serving as a complement to the transition towards the electrification of transportation. To optimize combustion efficiency, the amount of air in the cylinders needs to be estimated, which today is done by gray box-models based on traditional physics. This thesis examines the possibility of complementing these classical formulas with supporting machine learning models to improve the precision of the estimation. Both a Gaussian Process Regressor (GPR) and an Artificial Neural Network (ANN) are trained to estimate the airflow, and Bayesian optimization is performed to optimize the structure of the models. The results show reduced residual errors for both models compared to the current methods. The study shows promise for utilizing machine learning models to improve combustion engine efficiency, and highlights that further studies for hardware modification and implementation are needed.}},
author = {{Christensson, Lukas}},
language = {{eng}},
note = {{Student Paper}},
title = {{Machine Learning for Air Charge Estimation: A Residual Error Approach}},
year = {{2025}},
}