Vehicle Driving Risk Prediction Based on Markov Chain Model

Algorithm 1

The description of MNL-based Markov chain risk state prediction algorithm.

Initialization:

(1) Calculate and obtain the feature variable vector for time window at .

(2) Calculate the initial state probability by Eqn. (5)

(3) Obtain the corresponding independent variable vector for the initially observed time window:

MNL-based state transition probability estimation:

(4) For

(4.1) Calculate by Eqn. (7).

(4.2) Calculate the state probability distribution at by Markov property:

(4.3) Estimate the three-dimension feature variable vector by solving a set of three equations

according to Eqn. (5)

(4.4) Obtain the updated independent variable vector (assuming the driving mode remains unchanged).

(5) For

(5.1) Calculate by Eqn. (7).

(5.2) Calculate the state probability distribution at by Markov property:

(5.3) Estimate the three-dimension feature variable vector in the same way as step (4.3).

(5.4) Obtain the updated independent variable vector (same assumption as step (4.4)).

Outputs:

(6) Return the predicted state probability distribution for time window :