基于元模型全局最优化方法的含分布式电源配电网无功优化

Reactive Power Optimization of Distribution Network With Distributed Generation Using Metamodel-based Global Optimization Method

含分布式电源（distributed generation,DG）的配电网无功优化是保障配电网安全稳定、提高系统运行经济性的重要关键技术。含DG配电网的无功优化控制既要处理DG和静止无功补偿装置（static reactive power compensation device,SVC）的无功出力等连续控制变量,又要处理分组投切电容器组挡位、有载调压器（on-load voltage regulator,OLTC）分接头等离散控制变量,加之考虑DG、负荷的不确定性以及离散控制变量的投切次数约束之后,其模型的求解为一类复杂的非线性混合整数规划问题。该文采用概率场景的方法描述源荷的不确定性,并通过静态优化–离散变量优化–连续变量优化的多阶段方法解决电容器组和OLTC的投切次数约束,并提出一种基于元模型全局优化算法的无功优化求解方法,该方法采用拉丁超立方采样对复杂的无功优化目标函数进行合理抽样,并用克里金拟合建模方法在可能解的区域生成计算快及优化求解容易的元模型。在迭代寻优的过程中,用元模型估算目标值,并据此择优生成新的采样点来不断修正元模型,逐渐逼近复杂的无功优化目标模型并得出全局最优解。此方法由于无需反复大量调用原有的复杂目标函数,因而可有效减少计算负担和时间,提高全局最优解的搜索效率。以改造后的IEEE33节点和美国PG＆E 69节点系统为例进行了多种场景和多种算法的仿真对比测试,算例结果表明了文中提出方法的有效性和可行性。

Reactive power optimization of distribution network with distributed generation （DG） is an important technology to ensure network security and stability, improve power quality, and lower operation costs. The reactive power optimization control of distribution network with distributed generation should not only deal with the continuous control variables such as reactive power output of distributed generations and static reactive power compensation device （SVC）, but also the number of capacitor banks, on-load voltage regulator （OLTC） taps and other discrete control variables. After considering the uncertainty of DGs and loads, and the number of switching times of discrete control variables, the model is formulated as a class of complex nonlinear mixed integer programming problems with much difficult to solve. In this paper, firstly, the method of probability scenarios was introduced to describe the uncertainty of DGs and loads, and a multi-stage method was proposed to solve the daily switching times constraint of capacitor banks and OLTCs. Then a new method for the optimal control of reactive power for distribution network with distributed generation using metamodel-based global optimization was presented. The method uses Latin hypercube to efficiently sample data points on the complex and computation intensive reactive power control objective function, and build a Kriging metamodel of it in a region that most likely contains the global optimum. The method then uses the metamodel to approximate the complex objective function, selects new promising sample points to improve the metamodel, and identify the optimum of the region, before switching to the next most promising region and eventually locating the global optimum. The new method can effectively reduce the number of evaluations of the complex objective function through network reactive power simulation and computation time, thus improving search efficiency of the network operation control optimization. Using the modified IEEE33 node and PG＆E 69 node systems as test cases, the computational efficiency and robustness of this new method were compared with those of the conventional population based global optimization methods. Numerical results show the effectiveness and feasibility of the proposed new method.