Application Of Gaussian Process Regression And Cognitive Learning In The Control Of Unmanned Bicycle

In the field of self-balancing control research in unmanned bicycles,many experts and scholars have achieved self-balancing control of unmanned bicycles based on a variety of traditional controllers,but unmanned bicycles are in a complex and changeable environment,the traditional controller’s self-adaptability to the unknown environment is poor.In recent years,machine learning has been in the ascendant,developing at a high speed in the field of intelligent control,and unmanned bicycles,as the product of combining bicycle mechanisms and intelligent control technology,can also apply machine learning methods to achieve their balanced control.Gaussian Process Regression(GPR)as a machine learning method,has strong learning and generalized ability,strict statistical theoretical foundation,can be used to tap the potential mapping relationship between the input and output of unmanned bicycle controllers The probability model of the unknown controller is established to predict and judge the unknown data.However,the output of the Gaussian process regression model is not necessarily the optimal value.This paper introduces an enhanced learning strategy-reaction cognitive learning system to optimize the output of the Gaussian process return model to improve the control performance of the probability model.Enhance the adaptability of unmanned bicycles to the environment.In view of the fact that the physical experiments of unmanned bicycle physical experiments require large outdoor venues,good weather conditions,and high restrictions on the cost of damage to bicycles,and considering that the inverted and unmanned bicycles are highly similar,the trial and error cost is reduced.This thesis first uses upside down as a research object to verify the above-mentioned research methods,and then conducts experiments from unmanned bicycle physical prototypes.The main research content of this thesis is as follows:(1)(1)modeling inverted pendulum and unmanned bicycle.For the inverted pendulum,this thesis uses Newton’s second law to analyze its dynamics,and establishes the dynamic model of the inverted pendulum.For unmanned bicycles,a dynamic second-order response model,which is widely used in the analysis of system characteristics and controller design of unmanned bicycles,is selected,namely,the linear variable parameter(LPV)mechanical model.(2)Establish a Gaussian process return simulation model for inverted and unmanned bicycles.Based on the establishment of inverted stippling and unmanned bicycle dynamic models,the Gaussian process returns to the learning model with the MATLAB/Simulink module to learn from the existing inverted and unmanned bicycle controllers to establish the probability model of an unknown controller.The results show that the use of Gaussian process regression learning model can realize the learning of inverted and unmanned bicycle controllers,and the probability model of the prediction value of the control volume is better.(3)Based on the controller probability model based on the regression of the Gaussian process,combined with cognitive learning to establish an optimized system for the probability model.Similarly,the MATLAB/Simulink module is used to build a cognitive learning system,and the control volume prediction values of the Gaussian process are optimized with inverted and unmanned bicycles as research objects.The results show that:Cognitive learning balance control optimization system can effectively improve the performance of the Gaussian process regression controller’s probability model.(4)Taking unmanned bicycle physical prototypes as the experimental platform,using the Visual Studio2010 to engineering the Gaussian process regression control strategy and cognitive learning balance control optimization strategy,and conduct physical prototype experiments of the Gaussian process regression model and cognitive learning optimization system.First,the experimental data of the traditional controller is constructed to the Gaussian process to return to the training sample set,and then select the square index coordinated differential function to establish a negative number margin-like function.Find the optimal super parameter,and finally get the training of the Gaussian process to regain the controller probability model.It acts its engineering to an unmanned bicycle physical prototype and conduct a balanced control experiment.Then,based on the results of Gauss process regression,the states of the system are classified and the corresponding automata are constructed.The automata are continuously updated according to the learning mechanism of the cognitive module,to optimize the output of the action signal,based on this to carry out the cognitive learning balance control optimization experiment.The results show that: Based on the Gaussian process,the controller probability model has a good control effect on the self-balance movement of unmanned bicycles;optimizing control systems established with cognitive learning can effectively improve the performance of the Gaussian process regression controller probability model.In this thesis,we study the application of Gauss process regression and cognitive learning in self-balancing control of unmanned bicycles.By establishing a Gauss process regression model,the potential mapping relationship between controller inputs and outputs is explored,and a probabilistic controller model is established,then the probability model is optimized by cognitive learning optimization system,and the validity and reliability of the learning results are verified by numerical simulation and physical prototype experiments.It can be used as reference for the application of machine learning and reinforcement learning in the field of unmanned bicycle research,and it has strong engineering significance for improving the self-adaptability of unmanned bicycle in unknown environment.