Prediction of Diesel Engine NO<italic> <sub>x</sub> </italic> Transient Emissions Based on a Combined Model with Convolutional Neural Network–Efficient Channel Attention–Bidirectional Gated Recurrent Unit

<div>Widely used as power equipment, diesel engines emit NO<i> <sub>x</sub> </i>, which significantly threatens the well-being of both the ecosystem and individuals. The SCR system, which is employed to reduce NO<i> <sub>x</sub> </i> emissions from diesel engines, relies on precise control of the NO<i> <sub>x</sub> </i> emission levels. Addressing the challenge that traditional NO<i> <sub>x</sub> </i> emission prediction methods struggle to accurately forecast the emissions under transient operating conditions, this article introduces a deep learning model that integrates CNN, ECA, and BIGRU.</div> <div>The model’s necessary experimental data were collected during the hot phase of the WHTC, and input parameters were screened through correlation analysis. The model employs a CNN for feature extraction, integrates an ECA module to refine key feature processing, and utilizes BIGRU to capture temporal dynamics and dependencies, yielding predictive outcomes. Additionally, the model employs the Adam optimizer and combines it with BWO to adjust hyperparameters, thereby elevating the model’s accuracy for predicting transient NO<i> <sub>x</sub> </i> emission.</div> <div>Comparative analysis with existing CNN, LSTM, and CNN–LSTM models revealed that CNN–ECA–BIGRU model had a notable reduction in MAE, RMSE, and MAPE, along with an enhanced <i>R</i> <sup>2</sup> value. These improvements highlight the model’s superior predictive accuracy and its robust nonlinear fitting ability.</div>