【作者】 王钢；

【导师】 陈杰；

【作者基本信息】 北京理工大学 ， 控制科学与工程， 2018， 博士

【摘要】 “中国制造2025”的核心是信息物理融合系统(Cyber-Physical Systems,简称为CPS)。随着传感技术、通讯技术、计算机技术以及信息处理技术等迅速发展,信息物理融合系统的控制理论与方法已成为国际控制理论学者的热点研究课题。为实现对信息物理融合系统中“物理”系统的自动和智能状态感知,先进的状态估计方法和理论变得尤为重要。智能电网被誉为20世纪最伟大的工程,毫无疑问是地球上最庞大的信息物理融合系统,也即所谓的信息物理融合能源系统。精确的系统状态估计对电网正常运行的一系列控制、监控与优化任务至关重要。本文在智能电网的背景下,建立了一套完整的信息物理融合能源系统的(静态、动态)状态估计方法和理论。现有智能电网的管理与控制主要依赖于数据采集与监控系统。由于该系统只能获取电网状态(即所有节点电压组成的向量)的非线性观测值,这造成了相应的(静态/动态)状态估计问题的非线性。本文首先从智能电网静态状态估计问题出发,设计了一系列快速高效的求解算法,并建立了严格的克拉美罗界用于分析所有无偏状态估计方法的性能。大多数情况下,很难精确建模复杂系统的动态变化,本文考虑动态模型精确建模和未知的情形,分别设计了基于在线凸优化和滚动时域的动态状态跟踪方法。另外,考虑到所有实时的电网测量数据都通过网络发送至电力系统调度中心的,为减轻网络传输的负担和降低数据传输丢包率和时滞,本文研究了带事件驱动和传输丢包的动态状态估计方法,并建立了严格稳定性理论,分析了丢包大小和事件驱动对估计系统稳定性的影响。具体的,全文从以下几个方面进行展开。本文首先研究了基于最小绝对误差的鲁棒电网状态估计问题。由于非线性SCADA观测数据,相应的静态状态估计问题等价于一非凸非平滑优化问题。巧妙地从复合优化的角度出发,本文设计了多种基于线性近端(proximallinear)优化方法的静态状态估计求解算法。进一步利用物理连接的稀疏结构,本文提出了基于小批量(mini-batch)测量处理的加速方法。在一定的条件下,该方法具有O(1)的算法复杂度(独立于电网系统的大小),因此非常适用于大规模信息物理融合能源系统。考虑到文献中仍缺失一种公平合理的性能评判标准。本文在高斯白噪声观测模型下,建立了智能电网状态估计问题的克拉美罗界(Cramer-Rao bound)。该理论适用于所有的无偏估计方法,可广泛用于比较不同状态估计器的性能。另外,在非凸优化的最新进展上,本文设计了可行解追踪求解算法。其通过求解一系列凸的带二次约束的二次规划问题,去逼近原始的非凸状态估计问题。针对信息物理融合能源系统的高度不确定性,本文考虑了动态模型精确建模和动态模型未知情形下的动态状态跟踪问题。在系统动态未知的情形下,本文利用机器学习领域的在线凸优化理论,设计了在线状态估计算法,并给出基于regret的性能保证。如果系统动态精确已知,设计了基于滚动时域的动态状态跟踪方法。为了减轻信息物理融合系统中的网络时延和丢包,从智能电网线性近似系统出发,本文设计了一类事件驱动的数据传输机制。大体上说,该方法可靠发送“重要的”数据,然而以一定的概率(可以为零)发送相对“不重要的”数据。为了保证估计算法的可扩展性,本文设计了顺序式的状态估计方法。在合理的近似假设下,证明了相应估计器在最小均方误差意义下的最优性。同时,通过严格的理论分析,给出了保证估计算法随机稳定的充分性和必要性条件。为了更好地逼近实际系统的非线性,本文进一步研究了带丢包传输的非线性信息物理融合能源系统的动态状态估计方法。在已有方法基础上,新方法能够在性能上和稳定性取得有效地平衡。通过引入和分析李雅普诺夫函数,本文得出了保证估计算法估计误差和估计误差协方差矩阵随机稳定的充分性条件。通过大量的仿真实验验证了本文提出的非线性估计方法的有效性以及相应理论的正确性。最后,对全文进行了归纳总结,并对未来的研究方向进行了展望。更多还原

【Abstract】 Cyber-physical systems(CPSs)are central to the "Made in China 2025" plan.With the advent of microsensor,communications,computer and computing,and information process-ing technologies,controlling and managing CPSs have recently gained substantial research interest.In particular,to intelligently sense the possibly dynamical physical process for CPSs in real time,advanced state estimation approaches and theory become a crucial prerequisite The Northern American power grids are regarded as the greatest engineering achievement in the 20th century,and is arguably the largest CPS on the earth,which is also known as a cyber-physical energy system,or CPES.Accurate monitoring of the grid’s state is central to several system control and optimization tasks.This thesis puts forth a algorithmic framework for static and dynamic state estimation of cyber-physical energy systemsThe supervision control and data acquisition(SCADA)system installed in current CPESs can provide nonlinear measurements of the system state,which renders the state estimation(SE)problem nonconvex,and generally difficult to solve.Beginning with the static SE,this thesis develops several fast,scalable,yet efficient solvers,as well as the fundamental Cramer-Rao bound(CRB)under the additive white Gaussian noise(AWGN)model.The lat-ter can be broadly invoked to benchmark the performance of all unbiased algorithms.Further,dynamic SE approaches under modeled and unmodeled dynamics are discussed,and subop-timal solvers are pursued.Given the bulk real-time data communicated via a network,an event-triggered transmission scheme is devised to mitigate the network congestion(i.e.,time delay,data drop).Under both linear and nonlinear dynamic models,(suboptimal)estimators are proposed,and the effect of event-triggered transmission together with the data drops on stability of the estimator is characterized.Specifically,the ensuring topics are pursuedIn the envisioned CPES context,the present thesis first revisits the least-absolute-value SE from the vantage point of composite optimization.A couple of efficient yet robust SE solvers relying on the proximal-linear method are proposed.By leveraging the sparsity nat-ural to energy systems,acceleration is made possible by means of carefully mini-batching the SCADA data.Under suitable conditions,the accelerated approach enjoys only O(1)complexity independent of the system size,hence well-suited for large-scale CPESsUnder the AWGN model,leveraging the Wirtinger’s calculus for complex analysis,this paper derives the fundamental CRB,which bridges the gap of missing judicious performance evaluation criterion between different SE algorithms in the literature.The established theory applies to and serves as a benchmark for all unbiased estimators.Building upon recent advances in nonconvex optimization,a new iterative algorithm called feasible point pursuit is advocated for nonlinear state estimation of CPESsDue to high penetration of renewables and wide participation of demand response pro-grams,future CPESs become increasingly uncertain and stochastic,rendering static SE insuf-ficient for tracking the fast-evolving system states.In this context,dynamic SE approaches are studied.With unmodeled dynamics,an online convex optimization based dynamic SE scheme is developed,which comes with strong performance guarantees.By explicitly mod-eling system dynamics,a moving-horizon estimation based SE method is devisedTo mitigate the network congestion in CPESs,this paper devises an event-triggered data transmission protocol for a linear approximate model.Intuitively,the "important" data will be reliably sent to the control center,while "less important" ones are sent with certain(or zero)probability.To ensure the scalability,a sequential Kalman filter variant is developed,whose near-optimality is proved under standard Gaussian approximation assumptions.Suffi-cient and necessary conditions ensuring the stability of the developed estimator are derivedTo capture the nonlinearity of CPESs,state estimation of nonlinear dynamic systems with data drops is investigated.A new extended filter is suggested,which trades of the stability and performance among existing approaches.By analyzing a suitably constructed Lyapunov function,the stochastic stability of the estimation error and the error covariance matrix is establishedComprehensive simulated tests corroborate the merits of the developed SE approaches as well as the correctness of the established theory.At last,this thesis is concluded with meaningful future directions.更多还原