Standard driving cycles (DCs) and real driving emissions (RDE) legislation developed by the European Commission contains significant gaps with regard to quantifying local area vehicle emission levels and fuel consumption (FC). The aim of this paper was to review local DCs for estimating emission levels and FC under laboratory and real-world conditions. This review article has three sections. First, the detailed steps and methodologies applied during the development of these DCs are examined to highlight weaknesses. Next, a comparison is presented of various recent local DCs using the Worldwide Harmonized Light-Duty Test Cycle (WLTC) and FTP75 (Federal Test Procedure) in terms of the main characteristic parameters. Finally, the gap between RDE with laboratory-based and real-world emissions is discussed. The use of a large sample of real data to develop a typical DC for the local area could better reflect vehicle driving patterns on actual roads and offer a better estimation of emissions and consumed energy. The main issue found with most of the local DCs reviewed was a small data sample collected from a small number of vehicles during a short period of time, the lack of separate phases for driving conditions, and the shifting strategy adopted with the chassis dynamometer. On-road emissions measured by the portable emissions measurement system (PEMS) were higher than the laboratory-based measurements. Driving situation outside the boundary conditions of RDE shows higher emissions due to cold temperatures, road grade, similar shares of route, drivers’ dynamic driving conditions, and uncertainty within the PEMS and RDE analysis tools.

This study aimed to develop the Addis Ababa Driving Cycle (DC) using real-time data from passenger vehicles in Addis Ababa based on a neural network (NN) and principal component analysis (PCA) approach. Addis Ababa has no local DC for automobile emissions tests and standard DCs do not reflect the current scenario. During the DC’s development, the researchers determined the DC duration based on their experience and the literature. A k-means clustering method was also applied to cluster the dimensionally reduced data without identifying the best clustering method. First, a shape-preserving cubic interpolation technique was applied to remove outliers, followed by the Bayes wavelet signal denoising technique to smooth the data. Rules were then set for the extraction of trips and trip indicators before PCA was applied, and the machine learning classification was applied to identify the best clustering method. Finally, after training the NN using Bayesian regularization with a back propagation, the velocity for each route section was predicted and its performance had an overall R-value of 0.99. Compared with target data, the DCs developed by the NN and micro trip methods have a relative difference of 0.056 and 0.111, respectively, and resolve the issue of the DC duration decision in the micro trip method.

Urban transit systems are known to operate under the influence of dynamic elements and uncertainties. Static schedules and meta-heuristic approaches are too rigid to account for the changing dynamics caused by the variability of demand and other factors. In this regard, this work presents a novel approach based on the implementation of Deep Reinforcement Learning (DRL) in conjunction with a Digital-Twin (DT) of metropolitan urban transit systems. Specifically, a microscopic DT model constructed with Open Street Map (OSM) and GTFS datasets is used alongside the Deep Q-Network agent in an adaptive loop. A 260-dimensional microscopic digital twin is combined with a Deep Q-network agent that selects one out of 27 actions. Multi-objective rewards were defined in order to optimize for reduced waiting times, reduced emission levels, and reduced irregularity (bunching). Experiments conducted using the developed framework have proved the superiority of the proposed methodology over the conventional methods in terms of the reduction of total CO2 emissions by up to 28.3%, as well as achieving an increase in regularity level by 50% while remaining resilient to extreme changes in demand levels. Furthermore, sub-millisecond inference delays indicated the suitability of the model for applications in real-world scenarios within smart cities. Results revealed the applicability of the proposed framework to the issue at hand.

—Real-driving emissions from cars measured using portable systems are cumbersome, expensive, and produce inconsistent results. Previous studies have not investigated the role of varying drive cycle characteristics in estimating vehicle emissions, apart from variables related to vehicle speed, acceleration, and drive modes. In this study, real-time vehicle emissions were modeled using a technique that combines Simulink with artificial neural networks in the Addis Ababa drive cycle. Using a specific set of inputs, an emissions model for predicting passenger vehicle emissions was then created. The trained model was effective in predicting emissions (mean square error less than 2.438 x 10-5, R-square of 92-99.9%). Except for the number of stops, the predictors used as input to the model proved to be essential. Thus, the developed model overcomes many of the limitations of conventional emissions models and an experimental chassis dynamometer in estimating emissions and fuel consumption.,Index Terms— Addis Ababa, artificial neural network, drive cycle, Tail pipe vehicle emissions.

Passive suspension systems (PSS) use fixed spring and damping characteristics and cannot adapt to changing road disturbances and operating conditions. Active suspension systems (ASS) address this limitation by generating additional control forces, but controller design must balance ride comfort, road holding, model uncertainty, and control effort. This study develops a four-degree-of-freedom (4-DOF) longitudinal half-car model and evaluates an Adaptive Neuro-Fuzzy Inference System (ANFIS) controller against PSS and PID-controlled ASS under double-bump and ISO-classified random Type A and Type C Road profiles at 10 and 20 m/s. The ANFIS controller is trained using PID-generated input-output data with two inputs, five Gaussian membership functions per input, 25 Takagi-Sugeno rules, and 150 training epochs. For the double-bump case at 10 m/s, ANFIS reduces body vertical displacement from 0.0435 to 0.0084 m (80.6%), vertical acceleration from 2.40 to 0.77 m/s ² (67.9%), and front suspension travel from 0.0652 to 0.0090 m (86.2%) relative to PSS. Compared with PID, the corresponding reductions are 53.6%, 49.3%, and 43.8%, respectively. Settling time is reduced by 38.8% for body vertical displacement and 38.6% for pitch acceleration relative to PSS. The result meets all the qualifications for riding comfort of the ISO 2631 standard. The study identifies limitations related to PID-supervised training, linear modeling, and actuator-energy assessment by employing simulation. Experimental evaluation is required to establish robustness to parameter uncertainty, sensor noise, actuator delay, and implementation energy requirements.