抄録
This paper presents a new hybrid system model which is extended from the Hidden Markov Model (HMM) by specifying the state transition probability and the symbol output probability using a likelihood of Logistic Regression Model and the ARX model, respectively. By this extension, more complex behavior can be expressed with smaller number of states compared with the HMM. The parameter estimation algorithm for the proposed model is derived based on the EM algorithm with the weighted likelihood function. Then, the proposed algorithm is applied to identify some complex behavior, and the usefulness is confirmed.
