【作者】 刘宝；

【导师】 丁永生；

【作者基本信息】 东华大学 ， 控制理论与控制工程， 2006， 博士

【摘要】 在现代工业过程控制中，随着要求产品质量越来越高，对生产过程的控制效果提出了更高的要求。同时，在现代复杂信息环境下，出现了越来越多的复杂控制复杂控制系统。因此，需要开发研究智能化程度更高、实用性更强的智能控制算法。本文基于神经内分泌免疫生物网络系统的多种生物调节机制，对相关智能控制算法进行了研究。首先，对人工神经网络、人工免疫系统和人工内分泌系统等人工生物智能理论及各种智能控制技术发展进行了综述，指出了目前发展存在的问题以及将来的发展方向。接着，对神经系统、内分泌系统和免疫系统的生理基础及其调控机理或网络模型进行了介绍，为本文各种智能控制算法的研究设计奠定了生物理论基础。然后，基于NEI的调节机制，立足于解决过程控制中的实际问题，结合传统控制理论技术，从智能控制、学习控制、解耦控制、优化控制和网络控制等方面，对相关的智能控制技术进行了深入研究。在智能控制方面，首先基于下丘脑—垂体—睾丸素内分泌调节回路模型，设计了一种双层结构控制器。该控制器包括一级控制单元和二级控制单元两部分。一级控制单元根据控制偏差的大小，动态调整二级控制单元的设定值输入，从而能够迅速、稳定地消除控制偏差。然后，基于内分泌系统超短反馈调节机制，设计了一种超短反馈智能控制器。该控制器的传统控制单元的输出信号反馈给超短反馈处理单元，然后超短反馈处理单元按照激素调节分泌规律进行处理，处理后的信号与原传统控制单元输出信号叠加，从而构成一种非线性控制算法并提高控制效果。最后，通过仿真实验分别对两种智能控制器的控制性能进行了验证，结果表明其控制性能均优于传统PID控制器。在学习控制方面，基于免疫系统的初次—再次应答机制，设计了一种新颖的增强学习智能控制器(RLIC)。该RLIC具有学习、记忆、和进化能力，能够在消除控制偏差的过程中自动地形成控制抗体。当控制偏差再次出现时，RLIC能够结合传统控制算法，快速、稳定地消除控制偏差。控制偏差消除后，新的控制抗体即形成。这样随着控制器消除偏差次数的增多，其学习能力和响应速度变得越来越强。仿真实验表明，该学习控制算法不但优于传统控制算法，也优于传统的Q增强学习控制算法。在解耦控制方面，基于内分泌生长激素双向调节原理，设计了一种仿生双向解耦控制器和一种逆控制解耦控制器，并分别给出了相应的解耦算法。这两种解耦控制器分别根据不同的解耦算法，通过协调控制相应的执行器，从而消除不同控制回路之间的耦合影响。与其它解耦控制技术相比，这两种解耦控制算法比较实际，且更容易实现。通过仿真实验，分别将两种解耦控制算法与传统控制算法进行比较，实验结果表明智能解耦控制算法的解耦效果优于传统控制算法。在优化控制方面，基于内分泌激素调节规律，提出了一种自适应遗传算法(HGA)，该算法的收敛速度、搜索精度均优于标准遗传算法。并在此基础上，基于不同的神经内分泌免疫系统调节机理，先后设计了两种非线性优化智能控制器。第一种是基于NEI系统的整体调节机制的非线性优化控制器(NOIC)。根据免疫提呈机制，NOIC的提呈单元首先对实时控制偏差进行预处理，然后抗体控制单元通过调整抗体控制实体的数目来消除控制偏差。主控单元调节或协调提呈单元和控制抗体单元的控制作用，优化单元和辨识单元优化实时控制参数，从而提高NOIC的控制性能。第二种是基于肾上腺激素调节机制的智能优化控制器(ALIC)。根据实时控制偏差和激素调节规律，ALIC的主控制单元动态调整副控制单元的控制参数；利用HGA，优化单元和辨识单元可以优化主控制单元和副控制单元的控制参数，从而提高控制性能。通过仿真实验表明，这两种智能优化控制器比传统优化控制器均具有更好的控制性能。在网络控制方面，首先基于神经、内分泌和免疫三大系统整体调节机制，提出了一种新颖的分布式网络控制体系架构。然后提出了基于HGA的远程网络辨识算法和远程网络优化控制算法。最后利用一种6自由度微型操平台模型，对提出的网络辨识和优化控制算法进行了验证。最后，对全文研究内容进行了总结，指出研究工作中存在的不足，明确了下一步的研究方向。更多还原

【Abstract】 During the process control of modern industry, product quality is required higher and higher, which demands control performance with more efficiency. And, more intelligent and practical control algorithms are required by more and more complex control systems at the environment of the modern involuted information. Based on some bio-regulation mechanisms of the neuroendocrine-immune system, some intelligent control algorithms are studied in this thesis.First, we investigate the development of artificial bio-intelligent technologies, including artificial neural network, artificial immune system, artificial endocrine system, and the others intelligent control technologies. And their difficulties and further developments are summarized. Furthermore, some relative physiological theories and modulation mechanisms or models of neural system, endocrine system, and immune system, are introduced briefly. That provides the bio-base for the intelligent control algorithms studied in this thesis. Then, based on some bio-regulation mechanisms of the neuroendocrine-immune system, some intelligent control algorithms including intelligent control, learning control, decoupling control, optimized control and networked control, are studied respectively.For intelligent control, a two-level controller is first presented based on the hypothalamo-pituitary-adrenal model. The two-level controller includes the master control unit and the secondary one. The master control unit can adjust dynamically setpoint of the secondary one according to the real-time control error. Consequently, the controller can eliminate control error quickly and stably. Next, an ultrashort feedback intelligent controller is presented based on the ultrashort feedback mechanism of endocrine system. The output of the conventional control unit (CCU) is first fed back to the ultralshort-feedback unit (UFU), where the output of CCU is processed according to the hormone regulation law. Then the output of UFU is added to the output of CCU. Thus a nonlinear control algorithm is built. Consequently, the control performance is improved. Finally, the control performances of both controllers are examined via simulation experiments, whose results demonstrate the control performance and adaptation of both controllers are better than that of the conventional PID controller.For learning control, a novel reinforcement learning intelligent controller (RLIC) is presented based on the primary-secondary response mechanism. The RLIC has the abilities of learning, memory, and evolution, and can learn and produce the control antibodies (CABs) automatically during the period of eliminating the control error. When the control error appears again, the RLIC can eliminate it rapidly and stably, combined with the conventional control algorithm. After the control error is eliminated, a new CAB is produced and stored. Repeating the above process, the RLIC’s learning ability and response rate become stronger and stronger. Consequently, the control performance of the RLIC can be improved. Simulation results demonstrate that response ability and stability of the RLIC are better than those of the conventional PID controller, and also better than the Q-reinforcement learning control.For decoupling control, a bio-imitated decoupling controller and an inverse decoupling controller are presented respectively, based on the bi-regulation mechanism of the growth hormone (GH) in endocrine system. And the corresponding decoupling control algorithms are also provided. Both decoupling controllers can eliminate the coupling influence between different control loops via adjusting actuators harmoniously, according to their related decoupling algorithms. Compared with the others decoupling control technologies, both the decoupling controller are more practical and implemented more easily. The results of simulation indicate that the schemes of both decoupling controller can completely eliminate the coupling influence and show better control performance.For optimized control, a novel adaptive genetic algorithm (HGA) is first presented based on the regulation law of hormone in endocrine system. The convergence rate and search precision of HGA are better than that of the standard genetic algorithm (GA). Then, two optimized controllers are presented respectively according to HGA and based on the different modulation mechanism of neuroendocrine-immune system. The first one is a novel nonlinear optimized intelligent controller (NOIC) based on the modulation mechanism of neuroendocrine-immune system. Also, a method to optimize and adjust the control parameters dynamically is provided, as thus the control performance is improved. According to the presentation mechanism of immune system, the presentation unit (PU) first pretreats real-time control error dynamically, and then the antibody control unit (ACU) can regulate the number of antibody control entities (ACEs) to eliminate control error. The main control unit can regulate the control action of PU and ACU. Furthermore the optimum unit (PU) and identification unit (IU) can optimized the real-time control parameters. Thus, the control performance of NOIC is improved. The second one is an intelligent optimized controller based on the regulation mechanism of adrenalin (ALIC) in endocrine system. The method to optimize and adjust the control parameters dynamically is also provided, and thus the control performance of ALIC is improved. The simulation results demonstrate that the control performances of the NOIC and ALIC are better than that of the conventional PID controller.For networked control, a novel architecture of distributed networked control system is first presented. And then the model identification and optimized control methods via remote network are presented respectively according to HGA. Finally, the network identification and optimized algorithms are applied in the micro-motion platform mechanism with 6 DOF.At last, a summary of the thesis is made, and the deficiency in the project and the further development are narrated respectively.更多还原