Driving State Recognition Based On BP Neural Network And Hidden Markov Chain

Firstly, this paper introduces the research results and the problems about driving state and automotive active safety technology both at home and abroad, and according to recent designing ideas of human-centered active safety systems the paper proposes the driver’s driving behavior prediction model based on BP neural network and Hidden Markov Chain in order to achieve human-centered automobile active safety warning and control system to improve the car’s active safety.Secondly,it analysis the reasons of the Hidden Markov Chain used widely in the driving state and introduces related parameters of Hidden Markov Chain through an example, such as the probability of transition matrix, the probability of state observation matrix and initial probability, etc, and gives a detail description and sample tests about his three algorithms:the forward-backward algorithm, Viterbi algorithm and Baum-Weich algorithm, corresponding to resolve issues: assessment problems, decoding problems and learning problems, the tests which verify the effectiveness of the algorithms.Then, it makes a analysis to the main composition factors of driving state and gets the initial decision vector of three driving states,this paper introduces the BP neural network theory, determining the number of layers of BP neural network, node number of input (out) layer, number of nodes in the hidden layer, the transfer function and learning algorithm on the basis of understanding, and then uses the MATLAB neural network toolbox to create the BP model and achieve BP neural network’s training by corresponding algorithms and training functions.Finally, the paper sets up the identified structure based on BP neural network and Hidden Markov Chain using Simulink in MATLAB, then making the BP neural network’s trained results as Hidden Markov Chain parameters. Using Viterbi algorithm to gain best output probability that achieve the prediction of driver’s driving state,the accuracy of the model and algorithm by using simulation platform to verify, achieve the purpose of improving automobile active safety.