基于增强拓扑神经网络的自平衡车研究-机械工程专业论文.docxVIP

西南科技大学硕士研究生学位论文第 西南科技大学硕士研究生学位论文 第 PAGE IV 页 西南科技大学硕士研究生学位论文第 西南科技大学硕士研究生学位论文 第 PAGE III 页 Abstract Balancing vehicle is a new kind of urban transport, with the feature of simple structure, small size, flexible operation, green and so on. With the increasing congestion in the city today, the birth of balancing vehicle can effectively alleviate the pressure on urban traffic. With the similar structure with the pendulum, the control system of it has nonlinear, strong coupling, time-varying characteristics and so on. The high security driving performance of self balancing vehicle challenged the traditional control methods. This paper gives a detailed analysis of theory of the self balancing vehicle and puts forward two solutions on the main problems of the control method. Firstly, the traditional control method need to simplify the process of model parameters and linear process in the dynamic modeling and establishing state equations. This operation may lead to the problems of modeling accuracy, small angle range model ultimately controllable, and poor control effect. This paper proposed the use of neural networks as a balancing vehicle controller and the use nonlinear control method to balance the car. Secondly, due to the use of a fixed control parameters in traditional control method, the car cant solve the changes of the structural parameters and cant find a solution from the cycle of error. For example due to the difference of the drivers height and weight, control systems may often remain volatile, even failure to control the situation. This paper presented a control self-balancing vehicle algorithm based on evolving neural networks through augmenting topologies (NEAT). With the use of NEAT method, it can optimize neural network in the interaction with the environment and change with the environment and the structure parameters. Continuous evolution can benefit the autonomy learning and exploration. According to the needs of the evolutionary algorithm, the