Node Document


Stability-constrained Optimal Power Flow for Power Svstems:Model,Algorithm and Parallelization

【Author in Chinese】 耿光超

【Supervisor】 江全元

【Author's Information】 浙江大学, 电气工程, 2014, 博士

【Abstract in Chinese】 稳定约束最优潮流是电力系统运行与控制决策中的重要研究课题,它能够在最小化系统运行成本的同时,通过调整稳态运行点提升系统受扰后的动态性能,包括系统的暂态稳定性和短期电压稳定性。稳定约束最优潮流在数学上属于动态优化问题,即含有微分代数方程组约束条件的非线性规划问题,在涉及含复杂模型的大规模电力系统、长仿真时间窗口和多预想故障时,其求解过程计算时间长、耗用内存多,计算复杂性是该问题研究的主要理论和技术障碍。本文着重研究了基于数值优化理论和高性能计算技术高效求解稳定约束最优潮流问题的优化算法及其并行化实现,主要研究内容及其学术成果包括:1)提出了统一考虑电力系统暂态稳定和短期电压稳定约束的稳定约束最优潮流模型,给出了其基于动态优化问题的数学模型。同时针对复杂电力设备元件的动态模型集成问题,基于面向对象设计和自动微分技术,提出了应用于稳态和暂态分析的系统化复杂模型集成方法,进而设计并实现了应用于稳定约束最优潮流的模块化框架,提升了其算法实现的灵活性,拓展了该优化模型的应用前景。2)针对动态优化问题的两个算法阶段,即微分代数方程组的转化阶段和非线性规划问题的求解阶段,提出了基于直接多重打靶法和简约空间内点法的两阶段数值优化算法。与已有研究成果相比,该优化算法能够充分利用稳定约束最优潮流的问题特点和结构性质,从而显著提高优化算法的收敛性能和计算效率。通过一系列大规模电力系统算例的数值实验,验证了所提出两阶段优化算法的有效性。3)对于稳定约束最优潮流问题的优化求解过程,在不同的算法层面提出了可组合使用的四种并行分解策略,即预想故障分解策略、矩阵分块分解策略、打靶区间分解策略和轨迹灵敏度参数分解策略。能够充分利用基于多核CPU的计算集群、对称多处理平台和图形处理器(GPU)等多种高性能计算平台的计算资源,实现了问题求解的多层并行化,有效提高算法执行的计算效率,拓展能够求解的计算规模。

【Abstract】 Stability-constrained optimal power flow (SOPF) is an important research topic in power system operation and control. It is able to minimize system operating cost, while enhancing system dynamic performance after disturbance, including transient stability and short-term voltage stability, by means of adjusting steady-state operating conditions. Mathematically, SOPF belongs to the category of dynamic optimization, which is essentially a nonlinear programming problem with differential algebraic equation constraints. The computational process of dynamic optimization is commonly complicated, especially for large-scale power systems with realistic dynamic component models, long simulation intervals and multiple contingencies, which leads to high CPU and memory usage. Computational complexity is the major theoretical and technical barrier in the research of SOPF. In this thesis, efforts have been made on the development of an optimization method and its corresponding implementation based on numerical optimization theory and high performance computing techniques. Major outcomes of this research work include:1) A unified SOPF model with both transient stability and short-term voltage stability constraints is proposed, which is described as a dynamic optimization formulation. Also, in order to consider various dynamic model of realistic power system components in SOPF problem, a systematic approach is developed for model integration for power system steady-state and transient-state analysis, based on object-oriented design and automatic differentiation technique. A modular-based framework is designed and implemented for SOPF, in order to increase the algorithm flexibility and application potential of this approach.2) An combination of direct multiple shooting method and reduced-space interior point method is proposed for the two stages of dynamic optimization solution, i.e. the conversion stage for differential algebraic equations and the solving stage for nonlinear programming problems. Compared with previous investigations in the literature, the proposed algorithm is able to utilize the characteristics and structures of SOPF problems, which greatly increases the convergence performance and computational efficiency. The effectiveness of the proposed two-stage optimization algorithm is verified by numerical experiments on a series of large-scale power system test cases.3) Four parallel decomposition strategies are proposed for different algorithmic aspects in the solving process of SOPF, including contingency decomposition, elemental decomposition, shooting interval decomposition and sensitivity parameter decomposition. These parallel approaches are able to fully utilize various high performance computing platforms, including multi-core CPU based computing clusters, symmetric multi-processing platforms and graphics processing units. Multi-level parallelization is achieved, in order to speed-up and scale-up the proposed SOPF approach.

  • 【Contributor】 浙江大学
  • 【Year of Internet Publish】201507
Snapshot search of full-text: