Abstract

Optimal control of power converters to avoid voltage instability in cases such as system loading or faults is one of the most studied nonlinear problems that affect energy quality in power systems. The optimization problem related to converter control becomes more difficult with the inclusion of renewable energy systems while trying to fulfill power system constraints and providing an adequate amount of energy. In this paper, a simple approach based on artificial neural networks (ANNs) has been proposed and applied to photovoltaic-fed high-voltage DC and high-voltage AC systems interconnection consisting of PI-controlled power converters. By using the proposed method, converter control parameters are optimized for different cases to improve steady-state and dynamic voltage stability while also avoiding any kind of system faults. In order to implement hybrid control methodology by using ANN and PI control, the network should be well trained with samples including not only global best values but also the whole possible system characteristic. For this reason, a novel optimization algorithm, differential search algorithm, is used to sample solution space and train ANN by using random and localized samples. Obtained and presented results of the proposed approach show that due to robust and fast response, ANNs can be successfully used to overcome nonlinear security and optimization problems concerning power system stability.

中文翻译：

---

混合人工神经网络的光伏馈电直流和交流互连的最优变流器控制

摘要

功率变换器的最佳控制以避免系统负载或故障等情况下的电压不稳定，是影响功率系统电能质量的研究最多的非线性问题之一。在试图满足电力系统约束并提供足够量的能量的同时，随着包含可再生能源系统，与转换器控制相关的优化问题变得更加困难。本文提出了一种基于人工神经网络（ANN）的简单方法，并将其应用于由PI控制的功率转换器组成的光伏供电的高压DC和高压AC系统互连。通过使用所提出的方法，针对不同情况优化了转换器控制参数，以提高稳态和动态电压稳定性，同时还避免了任何类型的系统故障。为了通过使用ANN和PI控制来实现混合控制方法，应该对网络进行充分的训练，以使样本不仅包括全局最佳值，还包括整个可能的系统特性。因此，一种新颖的优化算法（差分搜索算法）用于对解决方案空间进行采样并通过使用随机和局部采样来训练ANN。所提出的方法的获得和提出的结果表明，由于鲁棒性和快速响应性，人工神经网络可以成功地用于克服非线性安全性和与电力系统稳定性有关的优化问题。差分搜索算法用于对解决方案空间进行采样并通过使用随机和局部采样来训练ANN。所提出的方法的获得和提出的结果表明，由于鲁棒性和快速响应性，人工神经网络可以成功地用于克服非线性安全性和与电力系统稳定性有关的优化问题。差分搜索算法用于对解决方案空间进行采样并通过使用随机和局部采样来训练ANN。所提出的方法的获得和提出的结果表明，由于鲁棒性和快速响应性，人工神经网络可以成功地用于克服非线性安全性和与电力系统稳定性有关的优化问题。