Deep Learning Based Distributed Meta-Learning for Fast and Accurate Online Adaptive Powertrain Fuel Consumption Modeling

Xin, Ma; Shahbakhti, Mahdi; Chigan, Chunxiao

doi:10.1109/tvt.2023.3239943

Cited by 1 publication

(2 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…ML algorithms have been explored in automotive-related research recently, given their ability to analyze data for pattern recognition, whether analyzing driver [36] or vehicle behavior [10], mainly because of their relevance in developing fuel consumption models based on real-world driving data, which can improve the accuracy of engine-related parameters and reduce the need for physical tests [26,7,25]. Moreover, task offloading emerges as a promising strategy for addressing the computational demands of vehicular applications, given the constraints imposed by limited onboard computing capabilities [39].…”

Section: Related Workmentioning

confidence: 99%

“…ML has several strengths over traditional methods, including its ability to handle large amounts of data and complex relationships between variables [45]. ML algorithms can also provide a more accurate prediction of real-world fuel consumption based on a broader range of variables than traditional linear regression models [26] and can reduce the need for physical testing, without the need for new calibrations [25].…”

mentioning

confidence: 99%

See 1 more Smart Citation

Machine Learning for Real-Time Fuel Consumption Prediction and Driving Profile Classification Based on ECU Data

Canal,

Riffel,

Gracioli

2024

IEEE Access

View full text Add to dashboard Cite

Data extracted directly from a vehicle's electronic control unit (ECU) play a crucial role in the automotive industry because they contain valuable information from the engine and electronic parts. These data have the potential to enable compliance analysis, detect faults and errors, and guarantee driver and car safety as well as product quality. Among the possible uses of the data from the ECUs, driving profile analysis and fuel consumption prediction stand out, which enable analyses for insurers and transportation companies, and help to reduce fuel consumption and greenhouse gases, in addition to providing feedback to the driver. In this work, we apply machine learning algorithms to real data from an engine ECU to examine the driver's driving behavior and accurately classify their fuel efficiency. Moreover, we develop regression models that predict fuel consumption for vehicles in operation. To ensure the effectiveness of our models, we carefully select variables strongly correlated with fuel consumption using a feature selection process. Compared to related works, both our profile classification results in precision, recall, and accuracy metrics, and our regression models result in the metrics of mean square errors, mean absolute error, and coefficient of determination, which are superior or similar. Notably, our algorithms exhibit lower computational costs and enable real-time analysis by utilizing a cloud server.INDEX TERMS Machine learning, driving profile analysis, fuel consumption prediction, electronic control unit (ECU).

show abstract