忆阻器件/CMOS混合结构的智能PID控制器及应用研究

With the rapid development of modern industry and increasingly high complexity of the control object, higher requirements have been put forword to the automatic control technology and the limitations of conventional PID controller emerge. This project utilizes the valuable research results of the electric technology and life science to explore a new generation of high-performance controller. In order to solve the difficult problem of the parameter adjustment in conventional controller, a robust and miniaturized memristive PID controller with nanometer memristive device/CMOS hybrid architecture is structured. Based on the perious work, two types of intelligent controllers are constructed: (1) The memristive single cell adaptive PID controller is designed fistly. Further, by taking advantage of the STDP learning rule and designing proper memristive synapse circuit, the memristive neural network adaptive PID controller is constructed. It possesses obvious advantages such as fast learning, correcting speed, small error and overshoot, which can enhance the control ability to deal with complex issues. (2) Aiming at the difficult to describe the control object with precise mathematical model, a memristive neuro-fuzzy intelligent PID controller is proposed. Where, the new neuro-fuzzy inference system is based on memristive crossbar array, which improves the integration and intelligent level of the controller. At last，two typical applications from industrial control, that is, servo motor control and temperature control are considered to verify the effectiveness of the proposed schemes. The research results of designed memristive controllers will provide significant theoretical and experiential supports for the development of novel intelligent PID controllers.

随着现代工业的飞速发展，控制对象复杂度不断增加，对自动控制技术提出了更高要求，传统PID控制器的局限性逐渐显现。本项目利用电子技术、生命科学的研究成果，研究新一代高性能控制器。设计忆阻器件/CMOS混合单元，构建鲁棒的、微型化的忆阻PID控制器，解决传统控制器的参数不易调节问题。在此基础上，构建两类智能控制器：(1) 研究单细胞自适应忆阻PID控制器，引入仿生智能STDP规则建立忆阻突触电路，构建具有学习修正速度快、误差小、超调量小等优势的忆阻神经网络自适应PID控制器，提升控制器处理复杂问题的能力。(2) 针对难以用精确数学模型描述的控制对象，建立忆阻模糊推理系统，结合交叉阵列，构建忆阻神经模糊智能PID控制器，提高集成度和智能化水平。最后，以电机伺服控制和温度控制为例，验证提出方案的有效性。本项目提出的控制器将为开发新型智能控制器提供重要的理论依据和实验支撑。

我们的工作主要按照原定计划进行，比较圆满地完成了预期的研究任务，超额实现了预期的研究目标。在项目执行期间，研究忆阻器的工作原理，分析忆阻器的基本特性，构建两类忆阻器件数学模型，结合现代CMOS工艺，设计具有鲁棒性和易于集成化的混合结构，完成M-PID控制器的设计。利用仿真软件对系统电气性能进行分析和功能测试，验证其稳定性、有效性及控制能力。分析仿真结果，并与传统PID控制器进行比较，验证了M-PID的优越性。设计单细胞自适应控制单元和忆阻器件突触转换器、基于STDP学习规则的忆阻器件突触电路。选择合适的网络结构，建立基于忆阻神经网络或神经形态系统的自适应PID控制器(NNM-PID)，并进行电路设计和仿真。利用现代CMOS工艺技术，构建忆阻交叉阵列，通过对阵列的训练，将模糊规则及参数整定方法存储在阵列中。结合传统PID控制器，构建NFM-PID控制器，对整个系统进行仿真，并分析该系统的基本性能指标。以电机伺服控制验证智能I型（NNM-PID）—神经网络自适应忆阻PID控制器的有效性。以温度控制验证智能II型（NFM-PID） — 忆阻神经模糊PID控制器的有效性。将重要的研究成果和实验方案，在国内外知名期刊上发表学术论文86篇，其中SCI检索62篇，EI检索52篇，出版专著4部，申请专利30项，其中发明专利28项，已经获授权7项。