关键词: NOX排放, 小波降噪, 特征选择, 双向GRU, 注意力机制, 贝叶斯优化

Abstract: In view of the complex nonlinear relationship between NOX emissions and actual operating conditions of diesel engines in agricultural vehicles, a datadriven NOX emission prediction model is proposed. The actual operating data was collected based on vehicle OBD, and the nonsmoothness of the original data was reduced by wavelet threshold denoising. The integrated feature selection algorithm was used to complete the selection of model input features. The BiGRU-Attention model was constructed by integrating BiGRU and the attention mechanism, and bayesian optimization was also used to select the model hyperparameters. Based on the experimental analysis of the actual vehicle road test dataset, compared with the LSTM, GRU, and BiLSTM-Attention models, the prediction Rsquare of the model in this paper was increased by 7.65%, 3.26%, and 4.09%, respectively. The mean absolute error of the model was maintained at 0.001 4 g/s. The prediction Rsquare was maintained at 85% on different vehicle datasets. This model can effectively predict NOX emissions with high accuracy in real scenarios and provides data support for NOX emission prediction control of diesel engines in agricultural vehicles.

Key words: NOX emissions, wavelet denoise, feature selection, bidirectional GRU, attention mechanism, bayesian optimization