For addressing the complex nonlinear relationship between NOx emissions from diesel engines of construction machinery in different altitudinal regions of plateau environments and actual operational conditions, a diminishing learning based hunting(DLH) search strategy to improve the grey wolf optimizer(GWO)-BP was proposed to optimize a BP neural network model for predicting NOx emissions. A portable emission measurement system(PEMS) was used to conduct emission tests on forklifts in plateau areas under various altitudinal operational conditions, and the experiment data were served as the dataset. Feature selection for the prediction model input was completed using the random forest algorithm. The results showed that the I-GWO-BP model outperformed both the BP and GWO-BP models in terms of evaluation metrics RMSE and R², with RMSE and R² values of 4.623 3 mg/s and 0.925 1 respectively. The model exhibited good prediction accuracy for NOx emissions from construction machinery at different altitudes in plateau areas.