Search

The Nitrogen Oxides (NOx) from engines aggravate natural environment and human health. Institutional regulations have attempted to protect the human body from them, while car manufacturers have tried to make NOx free vehicles. The formation of NOx emissions is highly dependent on the engine operating conditions and being able to predict NOx emissions would significantly help in enabling their reduction. This study investigates advanced method of predicting vehicle NOx emissions in pursuit of the sensorless engine. Sensors inside the engine are required to measure the operating condition. However, they can be removed or reduced if the sensing object such as the engine NOx emissions can be accurately predicted with a virtual model. This would result in cost reductions and overcome the sensor durability problem. To achieve such a goal, researchers have studied numerical analysis for the relationship between emissions and engine operating conditions. Also, a Deep Neural Network (DNN) is applied recently as a solution. However, the prediction accuracies were often not satisfactory where hyperparameter optimization was either overlooked or conducted manually. Therefore, this study proposes a virtual NOx sensor model based on the hyperparameter optimization. A Genetic Algorithm (GA) was adopted to establish a global optimum with DNN. Epoch size and learning rate are employed as the design variables, and R-squared based user defined function is adopted as the object function of GA. As a result, a more accurate and reliable virtual NOx sensor with the possibility of a sensorless engine could be developed and verified.

Environmental issues such as air pollution and abnormal climate are global concerns. To overcome these problems, the automobile industry is prioritizing the development of eco-friendly vehicles that reduce greenhouse gas emissions Among these, fuel-cell electric vehicles (FCEVs) use hydrogen as a fuel and do not emit exhaust gas and their higher mileage and shorter fuel charging time compared to electric vehicles make them promising next-generation eco-friendly vehicles. However, conventional energy management strategies have not effectively implemented both real-time capability and optimal energy distribution in FCEVs. To address these issues, a powertrain utilizing multi-energy sources is utilized, and a real-time energy control strategy based on long short-term memory (LSTM) is proposed. The training data for LSTM is obtained from the results of dynamic programming, utilizing six-city bus driving cycles, and the Braunschweig city driving cycle is chosen for test simulation. As a result, the LSTM prediction performance is evaluated, resulting in the development of an effective algorithm for real-time energy management of multi-energy sourced FCEVs. [ABSTRACT FROM AUTHOR]

In the previous research, we had developed a Nitrogen Oxides emissions (NOx) prediction model using deep learning methodology. However, it is necessary to develop a lightweight model for practical application. We developed a machine learning model for predicting NOx using Random Forest method to reduce the negligible input features. For comparison, a Base model was developed with all features. The Shapley Additive Explanations method, which can show the influence of the input features, and the Pearson Correlation Coefficient method, which can show the proportional relationship between the output and the input values, were used to choose the dominant features. The final model was determined using the Shapley Additive Explanations method. The final model shows similar prediction performance to the base model, though it has only 11 features (30 % of the Base model). The final model shows 15.76 for the root mean square error and 0.965 for R2 with the test data. By extracting the dominant features influencing NOx prediction, we could develop a lightweight and accurate NOx prediction model. [ABSTRACT FROM AUTHOR]

Deep-learning models were developed and evaluated for predicting the engine-out emission of NOx—one of the main pollutants emitted from diesel engines—under transient conditions, that is, the Worldwide Harmonized Light Vehicles Test Procedure (WLTP). Phenomena of transient conditions are difficult to predict accurately via the conventional modeling approaches. Two algorithms—the deep neural network (DNN) and long short-term memory (LSTM)—were evaluated regarding the accuracy and calculation time. Training was performed using measured data, and the results indicated that the LSTM model (R 2 = 0.9777, RMSE = 20.6 ppm) was more accurate than the DNN model (R 2 = 0.9671, RMSE = 25.5 ppm). However, the DNN model had a significantly higher computation speed (0.36 s) than the LSTM model (1381.0 s). Data preprocessing was performed to insert time-related information into the data; the DNN model trained with the measured data lacked this feature. Data were preprocessed by the calculation of the weighted average of previous timestep data to the current-timestep data. The weighted average was calculated according to various ratios, for example, 7:3, 6:4, 5:5, 4:6, and 3:7. By applying a 7:3 weighted average for training the DNN model, the accuracy of the DNN model was achieved to an R 2 value of 0.9741, and RMSE 22.8 ppm (only R 2 value 0.0036 smaller and RMSE 2.2 ppm larger than the LSTM model) without sacrificing the calculation speed. The results of this study suggest that data preprocessing of the DNN model is an effective method for achieving accuracy as high as that of the LSTM model. The developed DNN model for the NOx emission prediction can be used as a virtual sensor for real-time prediction owing to its accuracy and computation time. [ABSTRACT FROM AUTHOR]