Aiming at the problem of parameter selection and intention stage determination of intention representation in lane change intention identification， a new combined method is proposed. On the basis of the original parameters obtained from the driving simulator， the C4.5 decision tree algorithm and Pearson correlation analysis are used to obtain the parameter group which is composed of steering wheel angle， lane departure and yaw acceleration with high importance and low correlation. On this basis， K-means clustering is applied to the time series of steering wheel angle and lane departure to determine the driver's lane change intention stage. It is concluded that the length of intention stage is approximately linear related to the average speed， and the length of left lane change intention stage is larger than that of right lane change intention stage. Finally， the continuous Gaussian hidden Markov model is established， and the lane change intention recognition model and lane keeping recognition model are trained on the basis of intention representation parameter group and intention stage data. The average off-line recognition accuracy of the model is 90%. The driver's left lane change intention can be judged 1.5 s before the start of left lane change， and 1.4 s before the start of right lane change. The results show that the intention recognition model based on the intention representation parameters and intention stage can effectively identify the driver's lane changing intention， and the recognition accuracy is high and the timing is strong. This method can provide reference for intention parameter selection and intention stage determination in intention recognition research.