两轮自平衡小车控制算法的研究与优化

发布时间：2018-03-22 03:25

本文选题：PID控制　切入点：遗传算法　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：两轮自平衡车是轮式移动机器人的一种,它可以感知到外部环境的变动并且根据此变动做出相应的动态决策。它的结构源自于倒立摆模型,因此它也包含倒立摆的不确定性、非线性和强耦合性等特点。可见,两轮自平衡车可以作为一个检验各种控制算法的平台,因而对它的研究具有非常重要的理论意义。两轮自平衡小车内置了陀螺仪和加速度计用于采集小车的倾斜角度和角加速度,小车采用PID控制算法,通过调整比例P、积分I和微分D三个系数来使小车保持平衡性。传统的控制过程通常是依靠人工经验来调整这三个参数的,不仅费时费力,而且较难接近理想参数。所以,本文的研究目的是采用一些智能算法来实现控制过程的自适应调整功能,提出结合遗传算法和神经网络来实现PID参数自整定。此外,通过分析Kalman滤波算法在单神经元PID控制过程中的作用,将其应用在小车的数据滤波中。遗传算法和神经网络相结合的方法不改变PID控制过程,而是优化PID参数的调整过程以替代人工经验调整。遗传算法先在有效的参数范围内生成一代染色体种群,然后依据PID控制效果来计算适应度函数,根据优胜劣汰法则,选取适应度较高的染色体继续进化。但计算适应度的过程需要被控对象的参与,如果采用实体小车作为被控对象不仅不方便而且耗时长,为解决这一问题,该方法充分利用了神经网络的模拟功能,用训练好的神经网络模拟被控对象,协助遗传算法计算适应度函数,继而在遗传算法进化过程中逐代优化PID控制器的参数。单神经元网络改变了传统PID控制过程,把控制参数的调整转变为网络的权值迭代。单神经元结构简单,易于计算,可以在闭环控制过程中不断地调整网络的权值,从而优化控制效果。针对实际的单神经元PID控制器包含输出噪声,从而导致控制性能降低的问题,本文提出采用Kalman滤波理论来改进单神经元自适应PID控制算法。在改进算法中,引入状态空间的概念,采用时域上的递推方法进行数据滤波,控制对象的输出值经过Kalman滤波算法处理后再返回闭环控制系统。本文对两轮自平衡小车进行了实验,以及做了大量的仿真实验。结果表明,遗传算法和神经网络相结合的PID参数整定方法通常能获得较好的控制性能,具有较强的鲁棒性和较快的响应速度,可以替代传统的人工调试。通过Kalman滤波算法在单神经元PID控制过程的方波位置跟踪实验中验证了Kalman算法的滤波性能,将其应用在平衡小车的数据融合中也获得明显的滤波效果。本文借助遗传算法和神经网络等智能算法提出了PID自适应控制解决方案,并通过实验验证了方案的可行性,为智能控制和参数自整定领域提供了一些借鉴和参考。
[Abstract]:Two wheeled vehicle is a wheeled mobile robot, it can perceive the changes of external environment and make changes according to the dynamic decision-making. Its structure is derived from the inverted pendulum model, therefore it also contains the inverted pendulum of uncertainty, nonlinear and strong coupling. Therefore, two wheeled car can as a platform for testing various control algorithms, so the research on it has very important theoretical significance. The two wheeled vehicle built-in gyroscope and accelerometer for tilt angle and angular acceleration acquisition car, car uses the PID control algorithm, by adjusting the proportion of P, I and D three integral differential coefficient to keep the car the balance of control. The traditional process is usually to adjust the three parameters based on artificial experience, not only time-consuming, and difficult to close to the ideal parameters. So, this research purpose Is the use of intelligent algorithm to realize the adaptive function of the control process is proposed to realize PID parameters self-tuning. Combining genetic algorithm and neural network in addition, through the analysis of the role of Kalman filtering algorithm in single neuron PID control in the process of its application in car number. According to the filtering method combining genetic algorithm and neural network the PID change control process, but a process of adjustment to optimize the parameters of PID to replace the artificial experience adjustment. Genetic algorithm generates a first generation of chromosome in the valid range of parameters, and then according to the PID control effect to calculate the fitness function, according to the law of survival of the fittest, selection of high fitness chromosome but the calculation continues to evolve. The fitness of the process requires the object involved, if the real car as the object is not only inconvenient and time-consuming, in order to solve this problem, the method of charging The simulation function of the neural network, the simulation object with the trained neural network, genetic algorithm to calculate the fitness function, then the parameters in the evolutionary process of genetic algorithm by optimizing PID controller. The single neuron network has changed the traditional PID control process, adjust the control parameters change as weights iterative network. The single neuron has the advantages of simple structure, easy to calculate, can be adjusted continuously in the closed-loop control in the process of network weights, so as to optimize the control effect. The single neuron PID controller contains the actual output noise, resulting in reduced performance control problems, this paper puts forward a improved single neuron adaptive PID control algorithm using Kalman filter theory. In the improved algorithm in the concept of state space, using recursive method for data filtering in time domain, the Kalman filtering algorithm the output of the control object value After return to the closed loop control system. The two wheeled vehicle was studied, as well as a large number of simulation experiments. The results show that the PID parameters of combining genetic algorithm and neural network tuning method usually can obtain better control performance, strong robustness and fast response speed, manual debugging instead of the traditional Kalman. Through filtering algorithm in single neuron PID control of square wave filter position tracking performance of Kalman algorithm is verified in the experiment and its application in the balance of the car also filter data obtained obvious effect of fusion. In this paper, by using genetic algorithm and neural network intelligent adaptive control algorithm is proposed for PID solutions and the feasibility of the scheme is verified by experiments for the intelligent control and the self-tuning domain provides some reference and reference.

【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP273

【相似文献】