To optimize traffic management and reduce environmental pollution through scientific methods, a method for constructing vehicle driving conditions based on an improved K-means++ clustering algorithm is proposed. Combined with Markov chain theory, this method analyzes and constructs vehicle driving conditions. The collected vehicle driving data are preprocessed, including data cleaning and feature extraction. Dimensionality reduction was performed using Principal Component Analysis (PCA), and a K-means++ algorithm based on cosine similarity is introduced. The optimal number of clusters is determined using the elbow method. The results show that four driving conditions effectively simulate real driving scenarios. The comparison of average silhouette coefficients from the clustering results demonstrates that the improved algorithm significantly outperforms traditional methods in clustering performance. Using the Markov chain model, the transition relationships between the driving condition states are validated, and the final vehicle driving conditions are constructed. According to the comparative results of the relative error of key characteristic parameters, the average relative error is only 4.726%, indicating that this method has high rationality and accuracy in simulating actual road conditions. This provides a solution for traffic data analysis and model construction in complex traffic environments.