【作者】 王兴志；

【导师】 严正；

【作者基本信息】 上海交通大学 ， 电力系统及其自动化， 2011， 博士

【摘要】 我国电网作为世界上最复杂的人造工业网络系统之一,其安全稳定性的研究一直是电力科技工作者们关注的热点。本文针对动态电力系统提出了复杂电网的动态安全评估系统模型,并在网格计算模式下进行并行化数值仿真。本文的主要工作可概括如下:1.提出了一种广域电力系统建模和参数辨识方法。广域电力系统建模包括输电线路、发电机和负荷等三个重要部分。首先,基于复功率平衡性指标建立了输电线路数学模型,并采用修改的扩展卡尔曼滤波技术和在线核学习技术进行不良数据处理和线路参数快速辨识;然后,对发电机采用随机微分方程进行建模,并结合最大似然估计法进行参数估计;最后,给出了电力负荷静态和动态模型,以及其相应的参数辨识方法。算例研究表明了应用本文输电线路建模方法可以对不良数据进行快速动态检测和输电线路参数在线并行化辨识,验证了结合广义卡尔曼滤波的改进参数辨识方法在发电机快速参数辨识和静态负荷模型/动态负荷模型参数辨识的有效性。这对于未来基于广域电力系统测量信息进行快速建模具有重要价值。2.研究了基于网格计算资源静态调度的电力系统并行化暂稳计算方法。该方法基于多区域电网分解技术建立相对通用的并行化暂稳计算模型,并使用一阶Adams多步法和改进一阶Adams多步法对边界协调系统参数进行预测、基于混合法求解代数方程组等策略提高了计算效率。不同规模电网算例测试表明新算法具有较高的算法收敛性和计算速度,多步法参数预测策略的采用可以节约高达14.8%的计算时间,结合数据文件预处理等方法可以获得近似线性加速比。3.提出一种基于网格计算资源适应性调度的并行化暂稳计算新方法。该方法基于电网“高内聚、低耦合”和网格计算资源异构且动态多变等特点进行电网区域分解适应性任务调度,并基于核学习的大电网快速分区算法、适应性指令级并行、适应性数值积分步长调整和提前终止仿真、适应性多时步协调计算等方法提高了并行化暂稳计算效率。不同规模电网算例测试表明新算法可以更好的挖掘暂稳计算的并行化能力。采用适应性分区、指令级并行、适应性数值积分步长调整和多时步协调计算等性能改进措施对华东电网实际系统进行测试,当处理器数目为8时,改进算法的并行计算时间比传统方法缩短了约54.2%,而改进算法的总计算时间比传统方法节省了约40.17%。4.建立了考虑三维协调的并行化安全评估算法。该算法通过电网多区域并行化方法来保证区域电网交换功率和计划断面交流潮流的收敛性,引入随机扰动和蓄意攻击等策略来生成预想故障集合,在面向网格计算服务框架下进行多时段并行化安全评估,并对筛选后比较严重的故障进行辅助决策和并行化裕度评估。通过仿真结果验证了并行化安全评估计算方法的可行性和高效性,可作为大电网安全评估并行化计算的一种研究参考。5.提出了一种基于网格计算模式的动态安全评估框架和平台实现参考。建立了基于Globus进行定制的电力网格平台原型系统,分析了原型系统基本结构和计算任务处理流程,提出了改进的基于.NET的动态安全评估网格计算平台系统构架,开发了基于.NET的计算环境,在单机/共享内存机/集群机等计算资源下自动并行化执行计算任务,实现了网格计算模式下的动态安全评估示范应用。更多还原

【Abstract】 With the rapid growth of national power demands, the national power grid has become one of the largest man-made industry complex networks. The development of complex system-based modeling and simulation procedure can help planners and operators to cope with large power system security assessment situations. The main object of the research described in this thesis is to include these complex system-based modeling methods and grid computing techniques. The main contributions of the dissertation include the following.1. A wide-area power system identification method is present for large scale power system parallel modeling. The modified Kalman filter based PMU bad data monitoring and system identification methods are firstly used to identify the line parameters of the power system model. Secondly, stochastic differential equations and maximum likelyhood estimation based methods are introduced for the purpose of modeling generators.Thirdly, the parameters of static load models or dynamic load models are estimated using system identification methods. PMU expermental results of different scale power systems show that the proposed system identification methods can model the power system accurrately and efficiently.2. A grid computing resource static scheduling based parallel transient simulation method of power system is present. In order to improve the parallel computational efficiency, the following speedup techniques are used such as first-order Adams boundary system parameter prediction, improved first-order Adams multistep prediction, hybrid methods to solve the algebraic equations, and other strategies. Different scal experimental results show that the proposed algorithm can be used to solve the transient simulation problems of large scale power systems. To consider the improved multi-step parameter prediction strategy, the simulation time of up to 14.8% can be saved. And the near linear parallel speedup can be obtained.3. The grid computing resource adaptive scheduling based parallel transient stability simulation is studied. Considering the characteristics of“high cohesion-low coupling”and the heterogeneous of dynamic grid computing resources, an adaptive scheduling strategy is introduced. The stategies for improving the parallel simualtion of large scale power systems are as follows. Firstly, the multilevel kernel learning-based grid partition method is present. Secondly, the adaptive instruction level parallelism is used to improve the parallel effiency of multiple processes. Thirdly, the strategies of adaptive step size adjustment and early termination of transient stablitliy simulation are produced for the propuse of saving calculation tasks. And finaly, the adaptive multistep simulation method is introduced to consider the convergency differences of multiareas. Grid computing expermental results of different scale power systems show that the proposed adaptive method can improve the computation efficiency.4. To consider the three dimentional coordinate and grid computing technique of transient simulation, a parallel security asseement algorithm is present. In the algirithm, the domain partitioning based parallel method is firstly proposed to ensure the regional exchange plans and the convergence of power flow calculating. The random failure multiple time parallel analysis method of N-x and the attack multiple time parallel analysis method of vulnerability are secondly introduced for the static security assessment. And finally the feature-based functional parallelism method is used for the purposes of decision-making support and margin assessment. The simulation results of large scale power system show that this algorithm is feasible and has a high computational efficiency. It provides a viable approach for the security assessment with parallel transient stability analysis in grid computing research.5. A grid computation-based dynamic security assessment framework and demonstration platform is present. Firstly, taking into account the industrial standard of grid computation implementation, a globus toolkit-based prototype is introduced and analysed. To overcome the shortcomings of traditional globus-based simulation systems, an improved .NET-based grid computing platform for dynamic security assessment has been developed. In the proposed .NET grid computing environment, different computing resources, such as PCs, shared memory machines and clusters, are organized with high efficiency. Data-level parallelism tasks or functional parallelism tasks can be executed in different software operation systems. A demonstration application to Shanghai power grid shows that the proposed grid computing platform has many dynamic security assement functions. Predictably, more and more dyanmice security assessment simulation applications will be included in this type of grid computing enviorment.更多还原