To design advanced driver assistance systems with personalized service features, it is essential to delve into the diverse micro-driving styles of drivers. Traditional research often treats driving styles as fixed labels, yet drivers exhibit unique micro-driving behaviors and preferences in different traffic environments. Using data collected from a driving simulator, 68 different road geometries were analyzed to explore variations in driving characteristics under different conditions. Initially, dimensionality reduction was performed on evaluation metrics through feature extraction and selection. Subsequently, the K-means algorithm was employed for driving style clustering, with t-SNE visualization illustrating the differences between the two dimensionality reduction methods. Finally, genetic algorithms were used for adaptive parameter optimization, leading to the development of four identification models: GA-XGBoost, GA-LightGBM, GA-SVM, and GA-BPNN, enabling the discrimination of micro-driving styles without prior knowledge of drivers or driving scenarios. The results indicate that feature selection yields superior dimensionality reduction, retaining more feature information and providing a solid foundation for K-means clustering. Among the various micro-driving style identification models optimized by genetic algorithms, GA-BPNN demonstrated the best performance in predicting unknown data, with high accuracy and strong generalization ability. The research findings provide theoretical foundations and technical support for a deeper understanding of human driving behaviors and the personalized adjustment of autonomous vehicle assistance strategies.