Abstract Photovoltaic (PV) integrated hybrid microgrid is inherently plagued by an intermittent power supply. Conventional solution is to maintain storage like batteries for grid restoration. However, collaboration development among multiple power sources is a formidable task. Grid contingencies like islanding and loading events further exacerbate the problem. In order to address these problems an efficient power management scheme is required. Machine learning based predictive tools are effective to forecast maximum available power of PV generator for any weather condition. In this study, historical climate data of Islamabad Pakistan is used to train Linear Support Vector Regression (LSVR), Matern 5/2 Gaussian Process Regression, and Rational Quadratic Gaussian Process Regression (RQGPR) based models. Root Mean Square Error (RMSE) is used as a key performance index for qualitative analysis of the trained models. RQGPR model returned lowest RMSE with slowest training time and LSVR returned vice versa. To maintain adequate battery storage level as well as grid power balance under varying climate conditions, a Power Scheduling Control (PSC) scheme aided by RQGPR controls power flow from PV. Grid frequency deviation from its nominal value of 50 Hz reflects the grid imbalance. Lastly, a set of outcomes are observed and discussed for a sample microgrid.

中文翻译：

---

基于机器学习的混合微电网有功功率控制加权调度方案

摘要 光伏（PV）集成混合微电网固有地受到间歇供电的困扰。传统的解决方案是像电池一样维持电网恢复的存储。然而，多个电源之间的协作开发是一项艰巨的任务。孤岛和加载事件等电网突发事件进一步加剧了问题。为了解决这些问题，需要一种高效的电源管理方案。基于机器学习的预测工具可有效预测光伏发电机在任何天气条件下的最大可用功率。在本研究中，巴基斯坦伊斯兰堡的历史气候数据用于训练基于线性支持向量回归 (LSVR)、Matern 5/2 高斯过程回归和有理二次高斯过程回归 (RQGPR) 的模型。均方根误差 (RMSE) 用作对训练模型进行定性分析的关键性能指标。RQGPR 模型以最慢的训练时间返回最低的 RMSE，反之亦然。为了在不同气候条件下保持足够的电池存储水平以及电网功率平衡，由 RQGPR 辅助的功率调度控制 (PSC) 方案控制来自 PV 的功率流。电网频率与其标称值 50 Hz 的偏差反映了电网的不平衡。最后，观察并讨论了样本微电网的一组结果。由 RQGPR 辅助的功率调度控制 (PSC) 方案控制来自 PV 的功率流。电网频率与其标称值 50 Hz 的偏差反映了电网的不平衡。最后，观察并讨论了样本微电网的一组结果。由 RQGPR 辅助的功率调度控制 (PSC) 方案控制来自 PV 的功率流。电网频率与其标称值 50 Hz 的偏差反映了电网的不平衡。最后，观察并讨论了样本微电网的一组结果。