基于数据驱动的控制与故障检测及其应用

发布时间：2018-06-07 21:15

本文选题：数据驱动控制 + 故障检测　；参考：《华东理工大学》2014年博士论文

【摘要】：太阳能发电技术的快速发展,为新能源取代传统石化类能源提供了条件。在光伏发电技术中,除了光伏电池组件的材料技术与生产工艺,光伏逆变器的控制技术也十分重要,控制器性能的好坏直接决定了光伏并网发电系统的稳定性、输出效率、安全性等性能。此外,根据太阳能发电系统的安全性要求,太阳能发电间歇性等特点,以及太阳能分布式发电的发展趋势,需要对太阳能发电系统的故障检测技术、分布式太阳能发电系统控制技术进行研究,以保证太阳能并网发电系统的稳定,设备的安全,以及在分布式发电中,光伏发电渗透率较高时电网的稳定和优化。对于太阳能发电系统这样的复杂系统来说,传统的基于模型的方法需要通过繁琐耗时的建模过程来得到数学模型。针对这样的缺点,伴随着计算机技术和数字控制技术的发展,控制领域出现了一类基于数据驱动的控制器设计方法,通过采用输入输出数据直接设计控制器,这一类方法具有理论上独特的优势和应用价值。本文围绕基于数据驱动的子空间预测方法进行理论研究,设计数据驱动的控制和故障检测方法,并以太阳能发电系统为应用目标,分析了太阳能并网逆变器的原理,分类和结构,明确了并网逆变控制系统的性能要求和实际中存在的一些难题,并针对这些问题采用数据驱动的方法来予以解决。本文的主要工作包括：在子空间辨识方法的基础上对基于子空间方法的预测控制进行研究。从离线的输入输出数据直接得到子空间预测方程,并在此基础上设计控制器。在数据驱动子空间预测控制方法的基础上,结合H∞混合灵敏度控制的框架,针对太阳能并网逆变系统中输出电流受外界干扰如电网阻抗变化影响的问题,为了满足逆变输出电流控制中对跟踪性能和鲁棒性的要求,提出了一种基于数据驱动的H∞混合灵敏度控制算法,通过在太阳能并网逆变器中的仿真应用,验证了这种数据驱动子空间预测控制的有效性及其在太阳能发电系统中的适用性。通过与传统PI控制器和LQG控制器的对比仿真验证,所设计的控制器能够很好地抑制逆变器并网运行过程中电网阻抗变化特别是感性变化的干扰,证明了控制器的鲁棒性。论文对太阳能并网的分布式发电形式进行研究,分析当一个局部的微电网中含有多个分布式太阳能发电单元的情况时,光伏发电的间歇性等特性给微电网的功率调节以及稳定性带来的威胁。设计一个两层的控制结构,上层采用分布式协同控制让所有的光伏发电单元在通讯条件下按照相同的功率输出比率来发电,并为每个光伏单元给定参考电流；下层针对多个光伏逆变器运行时逆变器之间的相互作用给控制带来的困难,利用子空间预测控制方法对于处理多输入多输出控制系统方面的优势,设计了数据驱动子空间方法的分布式控制方法。最后通过数字仿真和SimPowerSystems仿真进行验证。在子空间预测器的基础上,本文还设计了一种数据驱动的故障检测方法,保证并网发电系统的安全,避免太阳能并网逆变器的故障对用电设备造成的损坏和对电网稳定性构成的威胁。利用子空间预测器对输出电流的预测,根据预测信号与实际电流的差值,通过自适应故障检测滤波器来得到残差信号以指示故障是否发生。并设计了利用在线数据更新故障检测滤波器的快速算法。通过仿真实验对常见三种故障的进行检测,验证了方法的有效性。
[Abstract]:The rapid development of solar power generation technology provides the conditions for the replacement of the traditional petrochemical energy. In the photovoltaic power generation technology, in addition to the material technology and production technology of the photovoltaic cell components, the control technology of the photovoltaic inverter is also very important. The direct connection of the controller performance determines the stability of the photovoltaic grid connected power generation system. In addition, according to the safety requirements of the solar power generation system, the intermittent characteristics of solar power generation, and the development trend of the solar energy distributed generation, it is necessary to study the fault detection technology of the solar power generation system and the distributed solar power generation system control technology to ensure the grid connected power generation of solar energy. The stability of the system, the safety of the equipment, and the stability and optimization of the grid in the distributed generation with higher penetration of photovoltaic power.
For a complex system such as solar power generation system, the traditional model based method needs to get the mathematical model through the time-consuming modeling process. With the development of computer technology and digital control technology, a kind of controller design method based on data driven is developed in the control field. Using input and output data to design controllers directly, this method has unique theoretical advantages and application value.
This paper studies the theory of subspace prediction based on data driven, designs data driven control and fault detection methods. The principle, classification and structure of solar grid inverter are analyzed with solar power generation system as application target. The performance requirements of grid connected inverter control system and some existing problems are clarified. Problems are solved and data driven methods are used to solve these problems.
On the basis of subspace identification method, the prediction control based on subspace method is studied. The subspace prediction equation is obtained directly from the off-line input and output data. On the basis of this, the controller is designed. On the basis of the data driven subspace prediction control method, combined with the framework of H infinity hybrid sensitivity control, the solar energy is applied to the solar energy. In an inverter system, the output current is affected by external interference, such as the change of electrical network impedance. In order to meet the requirements of tracking performance and robustness in the control of inverter output current control, a H infinity hybrid sensitivity control algorithm based on data driven is proposed. Through the simulation application in Solar Grid inverter, this number is verified. According to the effectiveness of the driver space prediction control and its applicability in the solar power generation system, through the comparison with the traditional PI controller and the LQG controller, the designed controller can effectively restrain the disturbance of the electrical network impedance change, especially the sensitivity change during the operation of the inverter, and prove the robustness of the controller. Sex.
In this paper, the distributed generation of solar energy grid is studied. When there are multiple distributed solar power units in a local microgrid, the intermittent characteristics of the photovoltaic power generation are the threat to the power regulation and stability of the microgrid. A two layer control structure is designed, and the upper layer is distributed. The cooperative control allows all photovoltaic units to generate electricity at the same power output ratio under the same power output ratio and give a reference current for each photovoltaic unit. The lower layer is difficult to control the interaction between the inverters when the multiple photovoltaic inverters run, and the subspace predictive control method is used to deal with the multiple input. The distributed control method of the data driven subspace method is designed for the advantages of multi output control system. Finally, it is verified by digital simulation and SimPowerSystems simulation.
On the basis of the subspace predictor, a data driven fault detection method is also designed to ensure the security of the grid connected power generation system, avoid the damage caused by the solar grid connected inverter and the threat to the stability of the power grid. The difference between the actual current and the actual current, through the adaptive fault detection filter to get the residual signal to indicate whether the fault occurs or not, and designs a fast algorithm to update the fault detection filter by on-line data. Through the simulation experiment, the common three kinds of faults are detected, and the effectiveness of the square method is verified.
【学位授予单位】：华东理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TM615;TP273

【参考文献】