合肥工业大学 博士学位论文 基于CMAC的多驱动系统协调控制 姓名：李鑫 申请学位级别：博士 专业：电力电子与电力传动 指导教师：张崇巍 2010-09 摘摘 要要 对于由分散的多个电机驱动的长距离带式运输机和由多个机器人进行的提升和装配操 作，协调控制用于解决这些复杂运动控制系统中各子系统之间负载、功率、速度和张力的平 衡，平稳启动和停车，以及机器人的快速跟随等问题是必不可少的这些系统由于具有非线 性和时变特性而难以用一般的解析方法来辨识、建模和控制，但是这些问题往往可以用人工 智能导出的方法来加以解决从控制的角度看，人工智能算法也就是控制器 通过对有关工程背景和一些用于协调控制的人工智能方法的学习，作者认为小脑模型关 联控制器（CMAC）方法是一种可有效地应用于协调控制的手段，并针对一些典型的复杂运动 控制系统开发了一些基于 CMAC 的算法,应用于工程实践 作者提出了一种 DCMAC+P 控制器 该控制器使用一个主 CMAC 粗略地逼近控制对象的 逆，同时使用一个辅助 CMAC 来补偿由于不够精确的逆所导致的残差针对一个由两个伺服 系统组成的复杂系统，其中两个子系统具有相同的周期性变化的速度给定，但是承担不同的 周期性变动负载，系统仿真验证了 DCMAC+P 控制策略；该算法可使两个子系统达到速度同 步，其快速性优越于常用的 CMAC+P 控制器。
ASE/ACE 算法具有与 CMAC+P 算法不同的机制且也可用于复杂运动控制系统的协调， 其 特点是通过两个互相协作的单元来实现的一种强化学习机制，其中一个单元提供控制输出， 而另一个单元则通过对控制效果的评价来修正前者的参数为了改善这种机制的实时性，作 者提出将 CMAC 引入这两个单元从而形成 CMAC-ASE/CMAC-ACE 控制器并将这种控制器应 用于上述复杂运动控制系统，仿真验证了独立控制的两个伺服系统快速而自动地实现了同步 基于上述由 CMAC 改进和增强的控制器，作者试图将其应用于典型的或在工程项目中的 复杂运动控制系统在一个机器人跟踪系统中，作者使用两个基于 CMAC 的 ASE/ACE 控制 器分别控制一个跟踪机器人的相对于逃逸机器人的距离和角度相对于另一个采用一般 CMAC 算法的跟踪机器人，该算法在快速响应和控制精度上占优 在一个由三台交流电机驱动的带式运输机中，三台电机会因为回路电阻或供电电压的差 别而导致功率不平衡，为此采用了 DCMAC+P 算法用于自动功率调节，仿真结果验证了这种 算法的有效性及其相对于一种模糊控制器的优越性 论文中介绍了控制算法在带式输送机上的应用，该长距离带式输送机由分布安置的多台 电机驱动。
为了保障工程成功，在预研中对系统进行了深入细致分析；对系统的主要部件如 传送带、滚筒、变速箱以及具有双闭环的驱动系统进行了建模，采用有限元法对传送带的粘 弹性进行了处理由于各驱动子系统通过传送带而互相耦合，工程中采用基于 CMAC 的 ASE/ACE 算法进行功率平衡仿真证明该算法可以获得良好系统性能并实现功率平衡 论文最后对于研究内容进行了总结，描述了后续的研究工作 关键词：关键词：复杂运动控制系统，协调控制，小脑模型关联控制器，强化学习，关联策略单元/自 适应评价单元算法，长距离带式运输机 Abstract Coordination in complex motion control systems, such as long-distance belt conveyor driven by multiple geographically distributed electric drives or a lift and assembly task performed by multiple cooperative robots，is mandatory to the problems such as the static and dynamic balance of load, power, speed and tension among the subsystems, the smooth startup and shutdown of the whole system, the quick catch-up of robot tracking etc. Those systems characterized of nonlinear and time-varying properties are quite hard to be identified, modeled and controlled by commonly used analytical methods but it happens that those problems might fortunately be solved by the methods deduced from artificial intelligence. In the sense of control, AI algorithm is a synonymous of controller. With the preliminary study on the engineering background and some AI methods used for coordination, the author found CMAC to be a useful measure for system coordination and developed algorithms for typical complex motion control systems in his engineering project, as well as for the preparation of his dissertation. The author proposed a DCMAC+P controller that uses a main CMAC to roughly approximate the inverse of the object and an auxiliary CMAC is used to fetch up the residue error caused by the roughness of acquired inverse. This strategy is verified superior to CMAC+P controller by simulating a difficult task of a complex system composed of two servo subsystems that are required to reach speed synchronization with same periodical speed set-value but different periodically varying loads. The ASE/ACE algorithm is a different mechanism from CMAC+P and may also be used for coordinated control of complex motion control systems. It is characterized of reinforced learning that is implemented by two mutually assisted elements that the first delivers control action and the second yields a judgment used to refine the parameters of the former. To improve real-time ability of such a mechanism, CMAC is introduced into the ASE and ACE respectively to form CMAC-ASE/CMAC-ACE controllers. The simulation is performed by applying such an algorithm to the two servo subsystems of the complex system mentioned above with the results that the two independently controlled subsystems can reach synchronization quickly and automatically. With the acquired knowledge of the CMAC improved and enhanced controllers the author tried to apply them to some typical complex motion control systems as well as the systems met in the engineering project. In a robot tracking system, two CMAC based ASE/ACE controllers are applied to the tracking robot to respectively control the distance and angle relative to the escaping robot. This algorithm is verified superior in response and accuracy to common CMAC algorithm running on a reference tracking robot performing the same tracking task. In solving the power unbalance among the three AC motors cooperatively driving a long belt conveyer, DCMAC+P algorithm is used to perform APR (Automatic Power Regulation) to achieve balanced load currents of the motors with unequal operating conditions such as circuit resistances and supply voltages. Simulation validates the effectiveness of the controller, as well as the better performance than that of a fuzzy controller. An important chapter of the dissertation is concerned with the application to long distance belt conveyers driven by multiple distributed electric drive systems. To secure the success of the projects, more thorough and detailed analysis has been done in the preliminary research. Main system components such as conv。