抗御随机干扰的自适应学习控制理论与应用

Deterministic learning theory is a new adaptive learning theory, which is put forward in recent years. It has been paid more attention by scientists and engineers not only for its high intelligence,rapid convergence, wide applications but also for that it has more challenging problems in it. Among them, how to reduce or stand against the influence of random disturbance in adaptive learning process, and to ensure of learning the unknown dynamics of systems within limited time is one of the difficulties. Clearly, the convergence speed of learning is the key problem. If the learning process is in accordance with the exponential convergence,it will certainly have strong resisting ability to random interference. When making use of auxiliary filters, disturbance observers and adaptive output regulation theory and tools to improve the adaptive learning ability and convergence speed furtherly, we are supposed to achieve better effect afterwards. Based on this idea mentioned, this project intends to give deeper considerations into it and at the same time takes advantage of its adaptive ability to random disturbance to push forward the improvement of learning process. That is to say, optimize the adaptive learning process according to the characteristics of random interference and change the random interference into possitive fators to the systems. By doing this, we can both weaken the strength of random interference and decrease the energy paid by the systems to withstand random disturbance. Thus, the learning convergence speed will certainly be propelled. On this basis, by using the data-based modeling and control theory, and the powerful computing ability and rapid response capability of DSP(digital signal processing), the objective of this project is to develop a more advanced disturbance resisting adaptive learning algorithm and apply it to control of practical motor systems.

确定学习理论是近几年新兴的自适应学习理论，由于其学习方法具有智能性高、收敛速度快、应用广泛、挑战性问题多等特点，备受科学家与工程师们的关注.其中，具挑战性的问题之一就是如何减小或抗御随机干扰对自适应学习效果的影响，确保在有限时间学习到系统的未知动态.显然，学习过程的收敛速度是关键，如果能依指数收敛，系统必具有较强的抗御随机干扰能力，借助辅助滤波器、干扰观测器、自适应输出调节等理论及工具，进一步提高自适应学习能力与收敛速度，可望获得更佳效果.项目拟基于上述思路进行深入探讨，同时融合学习过程对随机干扰的自适应性，即根据随机干扰的特性优化自适应学习过程，变随机干扰为增强自适应学习能力的辅助能量，不仅可以降低随机干扰强度，还可以减少自适应学习过程为抗御随机干扰所付出的能量，学习过程收敛速度势必增高.项目还利用数据的建模和控制思想及DSP的计算与响应能力，实现自适应学习在电机系统中的控制.

申请人带领研究团队，按照项目计划书开展了相关研究，项目进展比较顺利，在规定的时间内完成了项目计划书拟定的研究任务，取得了较好的结果，达到了预期目标。主要结果总结如下：.1、成功构造了两类新型Chua电路，研究了这两类电路的Lagrange指数稳定性以及平衡点镇定控制；在系统参数未知的情况下，研究了系统的自适应学习控制，引入新的持续激励条件判别法，得到系统参数的指数收敛结果，从而实现了参数的在线学习。.2、研究了两类基因调控网络的Lyapunov渐近稳定性与指数稳定性的系列充分必要条件。由于研究的系统具有类似Lurie系统的特征，我们应用研究Lurie系统稳定性类似的理论，研究基因调控网络的渐近稳定性和指数稳定性，得到一系列充分必要条件。在此基础上，我们利用比较原理和Dini导数，给出了这两类系统的渐近稳定与指数稳定的若干充分条件。.3、研究了无刷直流永磁电机的线性反馈控制器设计和自适应反馈控制器设计，所设计的线性反馈控制器结构简单，容易实现；自适应控制器在控制的过程中，能实现在线学习功能，具有较好的参数收敛效果。.4、将自适应学习控制的研究成果应用到某军用发电机故障的分类、检测与保护控制上，得到较好的效果，形成了产品，获得国家专利。.5、将系统的稳定性与控制的研究成果应用于大功率数字直流电源的稳定与控制中，取得较好的效果，形成了产品，获得国家专利。