Research On Dynamic Probability Model Of Unmanned Bicycle Based On Gaussian Process Regression

The unmanned bicycle system is a typical nonlinear,strongly coupled,underdriven system,and scholars at home and abroad have achieved fruitful results in unmanned bicycle dynamics models,equilibrium control strategies and trajectory tracking,etc.However,the existing dynamics models have many idealized assumptions,different unmanned bicycle structures lead to poor generalization ability of the models,and different road conditions put high demands on the environmental adaptability of the unmanned bicycle,which in turn brings challenges to the motion control of the unmanned bicycle.To address the above problems,this thesis is devoted to solve how to model and analyze the unmanned bicycle equilibrium motion data with Gaussian process regression,and validate the established dynamic probabilistic models,and finally give a reliable dynamic probabilistic model of the unmanned bicycle based on Gaussian process regression by predicting the unmanned bicycle equilibrium motion state under deterministic and uncertain inputs.This article main research contents are as follows:1)A linear parameter variation(LPV)model,which can respond to the dynamic effects of speed changes on the system,is introduced to build an unmanned bicycle balance control system by combining the relationship between the handlebar state and the lateral balance state of the unmanned bicycle.Three control algorithms,namely model predictive control,full-order sliding mode and partial feedback linearization,are introduced to realize the linear self-balancing motion of the unmanned bicycle using the above three controllers.2)The state data of the balanced motion of the unmanned bicycle are analysed to determine the type of state data required to build the dynamic probability model in order to construct the training and test samples.A suitable kernel function is selected for the derivation of the prediction distribution,and the conjugate gradient method is used to solve the hyperparameters introduced during the training of the model to build a dynamic probabilistic model based on Gaussian process regression for the unmanned bicycle in the simulation and experimental environments.The prediction of the unmanned bicycle motion state is then performed under the determined simulation and experimental test inputs respectively,and the prediction performance of the model is evaluated by different evaluation metrics.3)Based on the unmanned bicycle dynamic probability model obtained by training under simulation and experimental data,the test samples of simulation and experimental data are processed so that each test input dimension satisfies a certain probability distribution,when the test inputs are uncertain.The model is used to predict the unmanned bicycle motion state at uncertain simulation and experimental test inputs,respectively,and the model’s ability to predict the unmanned bicycle motion state at uncertain inputs is evaluated by different evaluation metrics.4)The mean and variance of the predicted output targets of the unmanned bicycle dynamic probability model with deterministic and uncertain inputs are compared and analyzed from the simulation and experimental environments,respectively,to summarize the differences in the prediction results of the unmanned bicycle dynamic probability model with deterministic and uncertain inputs.In this paper,the establishment of the dynamic probability model of the unmanned bicycle is studied.The Gaussian process regression method is used to train the model under the simulation and experimental data.The obtained model is used to predict the state of the unmanned bicycle under the deterministic and uncertain inputs.The experimental results show that the model has good prediction performance for both inputs.This result can provide a theoretical basis for the research of the dynamic probability model of the unmanned bicycle.