The Research And Optimization On Control Algorithm Of Two-wheeled Self-balance Vehicle

Two-wheeled self-balance vehicle is a kind of wheeled mobile robots that can perceive the change of outside conditions and make corresponding dynamic decisions.It’s modal heads from inverted pendulum,so it possesses the characteristic of inverted pendulum such as nonlinearity,uncertainty and strongly-coupling.Therefore,research of the two-wheeled self-balance vehicle which is a platform for testing many control algorithms has great theoretical significance.Two-wheeled self-balance vehicle has built-in gyroscope and accelerometer that are used for collecting the tilt and angular acceleration of vehicle.The vehicle uses PID control algorithm.We can adjust three parameters to keep balance of the vehicle.In traditional control process,the three parameters are usually adjusted by manually,this method not only takes much time and energy,but also is difficult to approach ideal parameters.So,this paper is aimed at using some intelligent algorithms to implement adaptive tuning of control process.This paper proposes that combining genetic algorithm and neural network to tune PID parameters.In addition,analyze the effect of Kalman filtering algorithm in the control process of single neuron PID,and put it in the data filtering of the vehicle.Combining genetic algorithm and neural network does not change the PID control process,they just optimize the parameters tuning process.Genetic algorithm generates a population of chromosome within the valid parameters range firstly,and then calculates the fitness function based on PID control effect.The chromosome who has higher fitness could go on evolving under the evolution rule.But a controlled object is needed while calculating the fitness function,and if we use actual vehicle as the controlled object is inconvenient and time-consuming.To solve this problem,this paper makes the best of the training ability of neural network that could help to get the fitness function and gradually optimize the PID parameters during evolution.Single neuron network change the process of traditional PID control,it changes the control parameters adjustment to the network weight iteration.Single neuron has simple structure,is easy to calculate,and it could gradually adjust the network weighting in closed-loop control and optimize the control effect.Aiming at the problem that realistic single neuron PID controller had output noise and resulted in degradation of the control performance,this paper proposed an improved single neuron adaptive PID control algorithm based on Kalman filtering theory.The improved algorithm used the state space and recurrence method for data filtering.The output of the controlled object was filtered by Kalman algorithm and then returned to closed loop control system.This paper has done many experiments and simulation on the two-wheeled self-balance vehicle.And the results show that the PID parameters tuning method based on genetic algorithm and neural network could get better control performance generally and have stronger robustness and faster response speed.We can use it to replace traditional parameters tuning methods.In the square signal tracking experiment that Kalman filtering algorithm in the control process of single neuron PID,we can verify the filter performance of the Kalman algorithm,and its application in the two-wheeled self-balance vehicle also get obvious data filtering effect.This paper proposed PID adaptive control methods using genetic algorithm and neural network etc.their feasibility has been experimentally validated,and these methods could provide some reference both in intelligent control and parameter self-tuning.