智能电网最优潮流计算方法及其收敛性研究

发布时间：2018-10-09 08:39

【摘要】：在智能电网的环境下,潮流计算作为其基本的分析工具有着非常重要的地位。为了将电力系统经济性,安全性以及电能质量三方面要求与潮流计算完美的结合起来,人们提出了最优潮流计算。最优潮流深受电力系统规划设计人员和运行调度人员的青睐,在电力系统规划、运行、分析和控制中起着重要的作用。本文首先针对牛顿法潮流计算的初值敏感性问题,提出了最优初值选取方法。解决了由于初值选取不当造成的潮流计算不收敛问题。其次,针对内点法,简化梯度法等方法在最优潮流计算中出现的初值敏感问题以及计算量过大等问题,提出了基于Fisher函数的广义梯度投影最优潮流算法。最后,为了使最优潮流获得更快的收敛速度,更方便的处理离散变量,运用双Hopfield神经网络来求解最优潮流。文中的主要工作如下：(1)提出了牛顿法潮流计算的收敛性判据,来判断初值能否使潮流方程得到收敛解。若初值可行,根据所提出的最大迭代次数估计判据,对潮流计算的迭代次数进行初步估计。通过以上两个判据可以对所选初值能否使潮流方程收敛进行一个初步判断,从而避免了初值任意选取造成的冗余计算。针对于牛顿法的初值敏感性问题,结合所提出的两个判据,利用遗传算法提出了最优初值的选取方法。IEEE标准节点系统和通辽电网的实例仿真结果可以证明所提出的最优初值选取方法的有效性。(2)提出了基于Fisher函数的广义梯度投影最优潮流算法。该方法与简化梯度法相比,不必每次迭代都求解潮流方程,大大减少了计算量。并运用动态选取罚函数技术,避免了由于罚函数选取不当导致的病态条件数。与内点法相比,扩大了初值的选取范围,避免了初值选取不当导致收敛过程的不稳定或使寻优进展缓慢。IEEE标准节点系统的仿真结果验证了该算法的有效性,并将简化梯度法,内点法以及广义梯度投影法最优潮流计算分别运用到通辽电网实例计算中,作出对比分析,仿真结果验证了基于Fisher函数的广义的梯度投影法的快速收敛性和稳定性。(3)运用双Hopfield神经网络来进行最优潮流计算。双Hopfield神经网络分为两个部分：首先,利用一个网络来优化罚函数项,使解落在可行解子空间中。另一个网络用来优化目标函数,朝着目标函数的可行下降方向进行求解。两个网络是相互独立的,交替运行。双Hopfield神经网络避免了Hopfield网络求解最优问题时既需满足约束条件又需得到高质量的解之间的矛盾,并且该算法极大的提高了计算速度。IEEE标准节点系统的仿真结果验证了双Hopfield神经网络方法与Hopfield神经网络相比取得了更好的优化效果。
[Abstract]:In the environment of smart grid, power flow calculation as its basic analysis tool has a very important position. In order to combine the three requirements of power system economy, security and power quality with the power flow calculation perfectly, people put forward the optimal power flow calculation. The optimal power flow is favored by the power system planners and dispatchers, and plays an important role in power system planning, operation, analysis and control. In this paper, an optimal initial value selection method is proposed to solve the sensitivity problem of Newtonian power flow calculation. The problem of non-convergence of power flow calculation caused by improper selection of initial value is solved. Secondly, a generalized gradient projection optimal power flow algorithm based on Fisher function is proposed to solve the initial value sensitivity problem and the excessive amount of calculation in the optimal power flow calculation by the interior point method and the simplified gradient method. Finally, in order to get faster convergence speed of optimal power flow and more convenient to deal with discrete variables, a double Hopfield neural network is used to solve the optimal power flow. The main work of this paper is as follows: (1) the convergence criterion of Newtonian power flow calculation is proposed to determine whether the initial value can converge the power flow equation. If the initial value is feasible, the iterative number of power flow calculation is estimated preliminarily according to the proposed criterion of maximum iterative degree estimation. Through the above two criteria, we can make a preliminary judgment on whether the selected initial value can make the power flow equation converge, thus avoiding the redundant calculation caused by the random selection of the initial value. In view of the sensitivity of the initial value of Newton's method, combined with the proposed two criteria, The simulation results of IEEE standard node system and Tongliao power network show that the proposed method is effective. (2) the generalized ladder based on Fisher function is proposed. Degree projection optimal power flow algorithm. Compared with the simplified gradient method, the proposed method does not have to solve the power flow equation every iteration, thus greatly reducing the computational complexity. The technique of dynamic selection of penalty function is used to avoid the ill-conditioned condition caused by improper selection of penalty function. Compared with the interior point method, the range of initial value selection is enlarged, and the instability of convergence process caused by improper initial value selection or the slow progress of optimization are avoided. The simulation results of IEEE standard node system verify the effectiveness of the algorithm and simplify the gradient method. The interior point method and the generalized gradient projection method are used to calculate the optimal power flow in Tongliao power network. The simulation results verify the fast convergence and stability of the generalized gradient projection method based on Fisher function. (3) double Hopfield neural network is used to calculate the optimal power flow. The dual Hopfield neural network is divided into two parts: first, a network is used to optimize the penalty function term, so that the solution falls in the feasible solution subspace. Another network is used to optimize the objective function and solve the problem in the direction of feasible descent of the objective function. The two networks are independent of each other and run alternately. The double Hopfield neural network avoids the contradiction between satisfying the constraint condition and obtaining the high quality solution when the Hopfield network is used to solve the optimal problem. The algorithm greatly improves the computation speed. The simulation results of IEEE-based standard node system verify that the dual Hopfield neural network method has better optimization effect than that of Hopfield neural network.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM744

【相似文献】