On the Use of Artificial Neural Networks to Model In-Use Fuel Consumption and Tailpipe Emissions from Light-Duty Vehicles

摘要: In-use fuel consumption and tailpipe emissions from light-duty vehicles (LDVs) have been found to exceed those measured in chassis dynamometer tests. Emissions models, based on chassis dynamometer data, are hence likely to be biased low, which has implications for quantifying and mitigating the environmental impact from LDVs. There is a need to develop methods and numerical tools that will allow us to better track the fuel consumption and tailpipe emissions from LDVs during real-world use. In this work, we used a portable emissions monitoring system (PEMS) to measure in-use fuel consumption and tailpipe emissions of CO, NOX, HC (unburned hydrocarbons), and PM (particulate matter) from a light-duty gasoline and diesel vehicle. Artificial neural network (ANN) models were developed and evaluated using the PEMS data and considered the following as model inputs: velocity, acceleration, vehicle specific power, engine speed, time since ignition, intake air temperature, manifold air pressure, and mass air flow. The ANN models seemed to perform quite well in predicting the fuel consumption for both vehicles and tailpipe emissions for the diesel vehicle. For example, the ANN models for the diesel vehicle were able to predict the total fuel consumed or total NOX emissions over a given route to within 2% of the measured value. Pollutant emissions from the gasoline vehicle had large errors using the ANN models and we suspect that the poor model performance was due to the presence of a catalytic converter that decoupled the pollutant emissions from vehicle operation. We found that the ANN models performed better than …