Semi-autonomous control is now the focus of intelligent wheelchair researches, and User-adapted Plan Recognition has presented its advantages in this field. Based on user plan recognition, we are developing a controller which is not only capable for estimating users' low-level intent (e.g. get to a point), but also high-level intent (e.g. open a door), for a manipulator-mounted intelligent wheelchair. Representation of user plan and user plan recognition is discussed in this paper, including several key points: (1) a detailed classification and corresponding representation of user intent (2) a method to connect different types of user intent, (3) A method to change user operation data from continuous form (not suitable for reasoning) to sequence form (suitable for reasoning), (4) a plan recognition method for manipulator-mounted intelligent wheelchair.