【作者】 李金阳；

【导师】 夏虹；

【作者基本信息】 哈尔滨工程大学 ， 核能科学与工程， 2013， 博士

【摘要】 反应堆控制系统由大量的传感器、执行器等设备组成，它是保证核电站安全、稳定运行的重要系统之一。目前，核电站主要采用硬件冗余的容错控制技术以保证设备发生故障时系统仍能正常运行，这种技术不仅需要大量的资金投入，而且在设计和建造阶段也会带来大量的可靠性和安全性方面的问题。随着计算机技术、人工智能技术的发展，采用人工神经网络技术、模糊逻辑控制技术等智能控制技术对反应堆控制设备进行容错控制，将可以在很大程度上改善因采用硬件冗余方法所带来的缺陷。因此，研究反应堆控制系统的故障诊断与容错控制方法，对于提高核电站的安全性能、可靠性能和经济性能将具有十分重要的意义。本文以核电站反应堆控制系统作为研究对象，针对在不同的功率运行工况下设备发生故障的情况，对反应堆控制系统的容错控制技术和方法进行了分析和研究。首先，对反应堆控制系统进行了研究。分别对稳压器压力控制系统、稳压器水位控制系统和蒸汽发生器水位控制系统进行了分析，研究了控制系统的功能、特点，控制系统各项参数之间的相互关系，以及各项参数的变化规律等，同时研究了传感器和执行器典型的故障模型，分析了在这两个控制设备之间故障模型相互转换所需要的条件。其次，对几种常用的容错控制设计方法进行了研究。分析了被动容错控制方法和主动容错控制方法几种常见的类型。对基于Riccati型方程的容错设计方法、基于控制律重组的容错设计方法和基于状态反馈控制器的容错设计方法进行了研究。分析了这三种容错设计方法容错控制的设计过程，并指出其不足之处。在此基础上，研究了基于BP神经网络的容错控制方法。针对标准的BP神经网络算法具有收敛速度较慢、容易陷入局部极小点而得不到全局最优等缺点，在增加动量项、改进变换函数、改进误差信号和改变学习速率等几个方面，对BP神经网络算法进行了改进，通过仿真实验验证了改进后的BP神经网络算法能够有效的提高网络的收敛速度，并且更有可能获得全局最优。进而确定了容错控制方案，并建立了动态模型库，对于已知的故障，系统能够自动识别故障模型，同时启动控制律重构单元对其进行控制律重构；对于未知的故障，系统将对其进行模型辨识，并将新建立的模型储存到模型库中不断对其完善。文中首先以稳压器压力测量传感器作为研究对象，在三种功率运行工况下（功率处于100%的运行工况、功率由100%下降至90%的运行工况和功率由90%上升至100%的运行工况），对其发生卡死故障和恒偏差故障的情况进行研究，仿真实验结果表明，基于改进的BP神经网络算法的容错控制方法对于修正稳压器压力测量传感器的故障是有效的。其次以蒸汽发生器压力测量传感器作为研究对象对该方法进行验证，仿真实验结果表明，该方法对于修正蒸汽发生器压力测量传感器的故障也同样有效。最后，研究了基于模糊神经网络的容错控制方法。针对人工神经网络连接权值的物理意义不明确、知识解释困难的缺点，采用将人工神经网络与模糊逻辑控制相结合的模糊神经网络的方法对控制设备进行容错控制。以稳压器水位测量传感器作为研究对象，在上述三种功率运行工况下对其进行研究，仿真实验结果表明，采用该方法能够实时的获取到传感器的可信度，并且该方法对于修正稳压器水位测量传感器的故障是有效的。更多还原

【Abstract】 The reactor control system includes a large number of sensors and actuators. It is oneof the important systems which can ensure the nuclear power plant to secure and stableoperation. At present, the nuclear power plant mainly makes use of the hardwareredundancy fault-tolerant control technology to ensure that system can still run normally inthe control equipment failure occurs. This technology not only needs a lot of money, butalso will pruduce a lot of problems about reliability and safety in design and constructionstage. When the intelligent control technology such as artificial neural network technologyand fuzzy logic control technology is used to carry through fault-tolerant control for thereactor control system, it will greatly improve the defects of using the hardware redundancymethod. So studying the fault diagnosis and fault-tolerant control methods for the reactorcontrol system, it is the very vital significance that improve safety, reliability and economyof the nuclear power plant.In this paper, the reactor control system will be as the research object. Against thecontrol equipment fault in different power operating condition, the fault-tolerant controltechnology and method of the reactor control system will be analyzed and studied from thefollowing aspects.Firstly, the reactor control system is studied. Respectively on the pressurizer pressurecontrol system, the pressurizer water level control system and the steam generator water levelcontrol system are analyzed, and studied the function and characteristics of the control system,the relationship among various parameters and the change law of various parameters.Meanwhile, the sensor and actuator is studied in typical fault model, and analyzes the requiredconditions of the fault model mutual transformation between the two control equipment.Secondly, several kinds of fault-tolerant comtrol design methods are studied. Severalcommon types of the passive fault-tolerant control method and the active fault tolerantcontrol method are analyzed. It will respectively study the fault-tolerant design methodbased on the Riccati equation, the fault-tolerant design method based on the control ratereconfiguration and the fault-tolerant design method based on the state feedback controller.It mainly analyzes their application condition, the design process of fault-tolerant controland points out its shortcomings. On this basis, the fault-tolerant control method based on BP neural network is studied. Inview of the standard BP neural network algorithm has the slower convergence speed, easy tofall into local minimum point and can not get the global optimal, the BP neural networkalgorithm will be improved by increasing the momentum, improving the transform function,improving the error signal and changing the learning rate. Through simulation experiment, itis shown that the improved BP neural network algorithm can effectively improve theconvergence speed and obtain the global optimal. And then, it determines the fault-tolerantcontrol scheme and establishes the dynamic model library. For the known fault, the systemcan automatically identify fault model, at the same time carry through fault-tolerant control bythe control rate reconfiguration unit. For the unknown fault, the system will use the modelidentification method, and the new fault model will be stored into the dynamic model libraryto continuously improve on it. The pressurizer pressure measuring sensor will be as theresearch object in the simulation experiment. In three kinds of power operating condition (theoperating condition of power at100%，the operating condition of power from100%to90%and the operating condition of power from100%to90%), the simulation experimentsimulates the sensor fault of the locked and the constant deviation. The simulationexperimental results show that the fault-tolerant control method based on the improved BPneural network algorithm can effectively correct the pressurizer pressure measuring sensorfault. The steam generator pressure measuring sensor will be as the research object to validatethe method. The simulation experimental results show that this metnod is equally effective tocorrect the steam generator pressure measuring sensor fault.Finally, the fault-tolerant control method based on fuzzy neural network is studied. Inview of the physical meaning of the artificial neural network connection weights is not clear,and its knowledge to explain difficultly, the fuzzy neural network method will be used forfault-tolerant control by combining with the artificial neural network and fuzzy logic control.The simulation experimental will adopt the pressurizer water level measuring sensor as theresearch object. The simulation experimental results show that using this method can get thecredibility of sensor in real time, and the method can effectively correct the pressurizerwater level measuring sensor fault.更多还原