The purpose of this study is improve calibration efficiency and obtain accurate diesel engine operating parameters, achieving improved diesel engine emissions and fuel efficiency. A PSO‐RBF (particle swarm optimization‐radial basis function) diesel engine performance prediction model combining an improved PSO (particle swarm optimization algorithm and an RBF neural network is proposed. A space‐filling experimental design method for diesel engine performance prediction is proposed based on the actual operating conditions of diesel engines. Training data are collected at the bench to build the RBF prediction model. An optimization PSO search method is proposed to improve the PSO optimization capability. An improved PSO algorithm is used to optimize the model and improve prediction accuracy. Then the BSFC (diesel brake‐specific fuel consumption), NOx ((Nitrogen Oxid), CO (Carbon Monoxide), and HC (Hydrocarbon) prediction models are constructed. Results show that the PSO‐RBF can find the global solution with good prediction accuracy and generalization ability during small amounts of data. The PSO‐RBF model fitting degrees of BSFC, NOx, CO, and HC are 0.9952, 0.9910, 0.9820, and 0.9870 respectively. Mean relative errors are 3.02%, 2.78%, 1.39%, and 2.01% respectively. Mean absolute percentage errors are 1.58%, 3.26%, 3.69%, and 2.96% respectively. The optimized model R2 (Model determination coefficient) is improved by 0.065, 0.102, 0.10, and 0.085, respectively.
In order to address the challenge of accurately predicting nitrogen oxide (NOx) emission from diesel engines in transient operation using traditional neural network models, this study proposes a NOx emission forecasting model based on a hybrid neural network architecture combining the convolutional neural network (CNN) and long short-term memory (LSTM) neural network. The objective is to enhance calibration efficiency and reduce diesel engine emissions. The proposed model utilizes data collected under the thermal cycle according to the world harmonized transient cycle (WHTC) emission test standard for training and verifying the prediction model. The CNN is employed to extract features from the training data, while LSTM networks are used to fit the data, resulting in the precise prediction of training NOx emissions from diesel engines. Experimental verification was conducted and the results demonstrate that the fitting coefficient (R2) of the CNN-LSTM network model in predicting transient NOx emissions from diesel engines is 0.977 with a root mean square error of 33.495. Compared to predictions made by a single LSTM neural network, CNN neural network predictions, and back-propagation (BP) neural network predictions, the root mean square error (RMSE) decreases by 35.6%, 50.8%, and 62.9%, respectively, while the fitting degree R2 increases by 2.5%, 4.4%, and 6.6%. These results demonstrate that the CNN-LSTM network prediction model has higher accuracy, good convergence, and robustness.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.