Based on the portable emissions measurement system(PEMS)

an actual road emissions test was performed on a certain type of heavy-duty diesel vehicle

and NOx emissions were fitted by the vehicle specific power(VSP) and vehicle traction force(VA) respectively. Using these two factors VSP and VA as input parameters

a double-hierarchical BP neural network improved by adaptive learning rate method was used to train and predict the NOx emissions. Compared with the original BP network prediction

the Pearson correlation coefficient between a predicted value and an actual value was increased by 0.1136

and the relative error was reduced by 0.6621%. The accuracy of the improved neural network prediction was promoted

and the generalization ability was strong. The improved neural network is applicable for the real-time prediction of the NOx emissions of this heavy-duty diesel vehicle

which has certain engineering application value.