Neural Network Decision And Control Of Lane Keeping System Based On Driver Behavior

The lane-keeping system is a new generation of driver assistantsystem, when the vehicle depart from the running road, it canautomated correct the heading, make the vehicle return to the safetyroadway, thereby effectively avoid the accident.The departure of vehicle is the main reason of road accidents.How to improve the automobile running safety, reduce the casualtyand loss are the problems automobile field and the whole worldconcentrated on. So the intelligent vehicle and driver assistant systemgot more and more attention of researchers. Nowadays the building ofintelligent transportation system just at the beginning, and the cycle offoundation establishment will be very long, therefore to develop theintelligent vehicle and driver assistant system, via the improvement ofvehicle intelligence to realize safety driving is the best choose.Meanwhile settle the application foundation in the self-containedintelligent transportation condition.From the point of control algorithm, along with the developmentof computer technology and the perfection of intelligent controlalgorithm, the intelligent vehicle control more and more close tointelligentize, the fuzzy logic, neural network are both used in theintelligent vehicle control, and got a good effect.The purposes of this paper are: 1,Refer to the artificial potentialfield method, using the neural network to decide the optimal trajectorydynamically;2,Using the self-adaptability of neural network to adjustthe control parameters, solve the strong nonlinearity of vehicledynamics, process the control emendation on different roads.This paper introduced the constitute, the decision and controlalgorithms in detail. In all, we can divide the paper into four parts:1,Combining the driver model of our laboratory, as the modelingfoundation, separately introduced the preview and following theoryand steady preview and dynamically emendation hypothesis,optimalpreview model, and optimal acceleration decision model. From thepoint of simulation of real driver's manipulate characteristics,introduced the structure of lane-keeping system.2,On the base of preview and following model and optimalacceleration preview theory, aim at the characteristic of intelligentvehicle direction control, fully considering the strong nonlinearity andtime-vary of automobile dynamics and the control behavior of driver,presented the lane-keeping system based on neural network. Refer tothe artificial potential field, using the neural network to decide theoptimal trajectory and control emendation. The paper divided thedecision and control model into two parts : optimal preview decisionand control emendation.(1),In the optimal preview decision, combined the artificialpotential field, adopted the neural network to define the road punishfunction at preview points, made it the security index of optimalpreview decision. With the manipulate index, used the synthesisenergy function to decide the optimal preview point.(2),In the control emendation, separately established PIDlane-keeping controller based on single neural network, BP neuralnetwork and CMAC neural network. Utilizing the self-adaptive ability,solved the strong nonlinearity and time-vary characteristic, achieved agood trajectory following effect. And reflect the driver manipulatebehavior characteristic, from the point of apery to simulate the realdriver's behavior, and make our lane-keeping system close to thebehavior of real driver model.3,To validate the rationality and veracity of the theory,established the seven degree of freedom dynamic model, combine theseventeen degree of freedom model of our lab, imitated driverdecision and control behavior of several typical condition. Validate thevalidity of this lane-keeping system.The innovation of this paper are:introduction of the neuralnetwork into lane-keeping system, made the self-adaptive preview andcontrol emendation, simplified the calculation and solved thenon-linearity problem.