ML Self-Sufficient Sustainable Energy Resiliency Management System: Outage Forecasting, Classification and Restoration with Maintenance Indicators for All Types of Power Outages
ML Self-Sufficient Sustainable Energy Resiliency Management System: Outage Forecasting, Classification and Restoration with Maintenance Indicators for All Types of Power Outages
复制标题
DOI：
10.1109/dmc55175.2022.9906471
复制
发表时间：
2022-09
期刊：
影响因子：
--
通讯作者：
Susan Oluropo Adedokun;Zhenhua Luo;Patrick Luk;N. Balta-Ozkan;Mohammad Farhan Khan;Xin Zhang
中科院分区：
文献类型：
--
作者：
Susan Oluropo Adedokun;Zhenhua Luo;Patrick Luk;N. Balta-Ozkan;Mohammad Farhan Khan;Xin Zhang
电力系统弹性研究主要集中在极端事件和自然灾害造成的停电后恢复关键负载的运行规划、优化和控制策略，这些事件具有高影响力、低概率事件的特点。对其他事件的弹性研究存在缺陷，包括对技术故障进行分类的高影响、高概率的停电。然而，最大比例的停电是由于设备故障和技术相关故障造成的。很少有机器学习研究涵盖停电预测和恢复，包括所有类型停电的弹性方法。这项研究提出了一个弹性管理系统框架，其中包含维护指标，适用于不同事件造成的所有类型的停电，特别是在发展中国家，高达 60% 的停电与技术相关。应用了具有机器学习分类和回归的新颖框架。该模型通过尼日利亚四个州的真实历史负荷流和停电中断进行了验证。结果显示，由于不同地点的不同原因，造成了复杂的多次停电。继电目标指示率为91.8%，停电类型分类准确率为85%，启动时间回归（R）值为1，意味着可以准确预测所有类型停电的发生，包括指示可以应用自给自足、可持续能源来增强电力系统弹性的维护目标。
Power systems resiliency studies focus largely on operational planning, optimization, and control strategies to restore critical loads, after blackouts from extreme incidents, and natural disasters, which characterize high-impact, low-probability events. There is a lacuna of resiliency studies of other events, including blackouts with high-impact, high-probability, which classify technical faults. However, the highest percentage of blackouts are from equipment failure technical related faults. Few ML studies cover both outage forecasting and restoration, including resiliency methods for all types of power outages. This study presents a resiliency management system framework, incorporating maintenance indicators, for all types of outages from different events, particularly in developing countries, where up to 60% of blackouts are technical related. A novel framework, with machine learning classification and regression is applied. The model is validated with real historic load flows and outage interruptions of four Nigeria states. Results reveal complex multiple power outages due to different causes at different locations. A relay target indication of 91.8%, an outage type classification accuracy of 85%, and a start time regression (R) value of one, signify that the onset of all types of power outages can be predicted accurately, including indication of maintenance targets where self-sufficient, sustainable energy resources can be applied to enhance power system resilience.