基于信息理论学习技术的随机分布控制方法研究

In this rapid development of information age, studying methods and techniques to design high precision, high performance control systems have important theory value and practical sense. Stochastic distribution control is proposed according to the control requirement of typical industrial process at the end of the last century, and has become an efficient tool for the research of data-driven modeling and control. This project aims to study on how to apply information theoretic learning technique to stochastic distribution control method, which is possible broad the theory scope of application. Based on the previous studies, new modeling method and control strategy for complex stochastic distribution system are developed under the situation that no prior knowledge of the stochastic input and output are needed. In addition, the control problems, such as stability, robustness and adaptability are also considered when the stochastic distribution system under consideration involves state time delay, external disturbances and parameter uncertainties. Furthermore, the visual simulation and test platform is designed to verify the feasibility of the proposed methods, and the results can provide theory basis and technical support for the application of the stochastic distribution control method based on information theoretic learning technique.

本项目围绕随机分布控制理论在工业生产过程中的未来发展需求，从信息理论学习角度出发，利用多种先进的建模技术及控制方法，基于数据驱动进行建模与控制。在目前研究的基础上，进一步发展神经网络逼近模型及非线性动态泛函模型。力求针对不同的实际工程对象提出既相对精确，又有利于分析和优化、易于工程实现的随机分布数学模型。针对典型的随机分布控制系统，考虑系统约束条件，设计可以工程实现而且符合性能要求的带有参数的固定结构控制器构造方法，研究输入输出统计特性未知情形下随机分布系统的广义跟踪\调节问题。另外，针对含有外部扰动、不确定性、时滞等的随机分布系统，进一步考虑系统的稳定性、鲁棒性、适应性等控制问题。设计可视化仿真演示平台，验证所研究控制方法的可行性，为未来基于信息理论学习技术的随机分布控制系统实现提供理论依据和技术支持。

本项目针对随机输入统计特性未知的非高斯随机系统，研究了基于非参数估计和统计信息集合优化准则的跟踪控制方法。针对分布泛函算子描述的自回归移动平均非高斯随机分布系统模型，通过构造辅助随机输出变量，设计了一种具有预测性能优化指标的线性随机分布跟踪控制方法。并且证明了当随机噪声服从高斯分布时，此方法等价于最小方差控制。针对含有随机噪声、可建模干扰和不确定性动态的随机系统，研究了基于干扰观测器的具有干扰抵消与抑制性能的复合分层随机鲁棒控制方法。