Assessment of Real-Time Emissions and Fuel Consumption of Light-Duty Vehicles Using Artificial Neural Network Technique: for Addis Ababa Urban Road Conditions

Abstract

Urbanization And Rising Vehicle Ownership Contribute To Traffic Congestion And Negatively Impact The Environment. Since 2017, Europe Has Used Portable Systems For Real-Time Vehicles Emissions Testing, But These Systems Are Cumbersome, Expensive, And Inconsistent, Making Them Unsuitable For Real-Time Monitoring In Resource-Limited Countries Like Ethiopia. Furthermore, Existing Emissions Models Do Not Account For How Varying Drive Cycle (Dc) Characteristics Affect Vehicle Emissions. This Study Aims To Develop Artificial Neural Network (Ann) Models For Real-Time Assessment Of Exhaust Emissions And Fuel Consumption (Fc) Of Light-Duty Vehicles, Tailored To The Road Conditions Of Addis Ababa (Aa) Including The Addis-Adama Expressway. Initially, A Representative Dc Was Developed Using A Neural Network (Nn) And Principal Component Analysis, Reflecting The Traffic Flow Characteristics Of The Study Area. Real-Time Driving Data From Five Vehicles Were Collected Using Gps Devices, And Two Vehicles Were Tested For Emissions And Fc On A Chassis Dynamometer Setup In Uppsala, Sweden. Next, Emissions Of Five Primary Gases(Co, Co2, Nox, Pm, And Voc) And Fc Were Determined Using The Copert Model For The Dc?�?S Developed In This Study And The Worldwide Harmonized Light Vehicles Test Cycle (Wltc).To Reduce Co2, Hc, And Co Emissions Concurrently, A Full Factorial Design And Desirability Function Analysis (Dfa) Were Employed To Assess The Effects Of Vehicle Speed And Road Slope On Emissions, With Data Gathered From Two Vehicles In Aa Using A Portable Emissions Analyzer. Finally, An Ann Simulation Model Was Developed In Matlab To Simulate Real-Time Vehicle Exhaust Emissions. The Model, Trained With Bayesian Regularization And Back-Propagation, Predicted Velocities For Individual Route Segments With An Overall R-Value Of 0.99. Compared To Collected Data, The Nn-Derived Dcs Exhibited A Relative Difference Of 0.056, While The Micro Trip Methods Showed A Relative Difference Of 0.111. The Findings Of This Study Reveal That Driving In The Congested Urban Areas Of Aa Contributes To 56.25% Of Co, 37.28% Of Co2, 38.19% Of Nox,58.25% Of Voc, And 29% Of Both Pm2.5 And Pm10 Emissions, As Well As 37.29% Of The Fc From The Overall Amount. Utilizing Dfa, Was Found That The Most Effective Speed For The Concurrent Reduction Of Co, Hc, And Co2 Emissions Was 40 Km/H On A Level Road And 30 Km/H On A Road With A 2?? Road Slope, Achieving Composite Desirability Indexes Of 0.83 And 0.72,Respectively. The Findings Revealed That The Dc Developed In This Study Resulted In Significantly Higher Levels Of Emissions And Fc Compared To The Wltc. The Trained Ann Model Was Effective In Predicting Hc, Co, Nox, And Pm Emissions (Mse <2.438x10-5, R2 Of 92-99.9%). Except For The Number Of Stops, The Predictors Used As Input To The Ann Model Proved To Be Essential In Predicting Real-Time Exhaust Emissions. The Ann Model Outperformed The Copert Model And Experimental Chassis Dynamometer In Estimating Co, Nox, And Fc. The Findings Of This Study Shall Be Used To Develop Effective Strategies For Reducing Vehicular Emissions And Fc In Aa