In order to accurately reflect the actual kinematic characteristics of vehicles and open up new directions for driving cycle acquisition, generation and recognition, a new energy vehicle driving cycle generation and recognition method was proposed based on unit driving conditions. Kernel principal component analysis and Gaussian mixture model clustering were used to classify the kinematic fragments of pretreated driving cycle. Based on the Markov chain, the fragment state division and characteristic condition construction were carried out for various unit condition groups, and the customized unit driving condition was constructed by the operating condition generation model according to various characteristic parameters. Using the aforementioned method as the training sample source, a driving condition recognition model was established based on a backpropagation neural network. The results show that the constructed driving conditions exhibit an average error of only 2.69% compared to the feature parameters of original sample data. Customized unit conditions show an average error of merely 1.85% relative to the corresponding category sample data feature parameters. The neural network driving condition recognition model achieves an overall accuracy of 93.8%.