Power System Stability Analysis and Control Using Statistical Machine Learning Techniques

基于传统离线模型的仿真方法在电力系统稳定分析和控制设计中得到了广泛的应用，是保证系统稳定的有效工具。由于电网环境的不断演变，这些稳定性分析和设计方法的有效性正在稳步下降，这些变化在很大程度上归因于潮流变化的增加以及难以获得各种基于电力电子设备的精确离线模型。信息和通信技术的进步促进了大量数据的真实的时间传输，并意味着先进的真实的时间监测系统的更广泛应用的机会，允许采集电力系统的各种组件的真实的时间条件和动态的数据。这使得整个系统更具可观察性。与此同时，数据驱动的方法，如统计机器学习技术，近年来得到了显着发展，并已成功应用于各个领域。因此，使用统计机器学习技术的真实的时间稳定性分析和控制已经成为一个重要的研究方向，因为它旨在通过真实的时间数据直接感知系统的运行情况，并提供对最佳操作和控制的见解。这有可能解决当使用离线模型时的偏置参数的问题，在大多数情况下，离线模型不适合电网中的真实的时间操作条件。这一研究的意义促使本研究计划将联合收割机统计机器学习与电力系统领域知识相结合，使其适用于真实的电力系统的稳定分析和控制。该研究计划的长期目标是开发电力系统稳定分析的新方法和有效的在线无模型和自优化控制策略。为了实现这一最终目标，短期目标是（i）开发新的方法预测，控制和优化的电力系统，以解决传统的电力系统仿真中使用的离线模型的偏差问题;和（ii）应用本计划中提出的一般方法，以各种问题有关的电力系统稳定性和开发新的在线控制策略相同。本研究的成果不仅将成为电力系统稳定性分析和控制的里程碑，而且将有助于未来更可靠和稳定的电力系统的发展。

Simulation methods based on conventional offline models have been used widely in power system stability analysis and control design and they constitute effective tools for ensuring system stability. Effectiveness of these stability analysis and design methods is declining steadily because of the constant evolution of the power grid environment, the changes being largely attributable to increased variations in power flow and the difficulties in acquiring accurate offline models for various power-electronics-based devices. Advancement in information and communications technologies have facilitated transfer of massive data in real time and implies an opportunity for wider applications of advanced real time monitoring systems, allowing the acquisition of data of real time conditions and dynamics of various components of power systems. This makes the whole system more observable. Meanwhile, data-driven methods such as statistical machine learning techniques have developed significantly in recent times and have been successfully applied in various areas. Therefore, real time stability analysis and control using statistical machine learning techniques has become an important research direction since it aims to perceive the system's operational situation directly through real time data and provide insights into optimal operations and controls. This has the potential to resolve the problems of biased parameters when using offline models which, in most cases, do not fit real time operating conditions in the power grid. The significance of this research motivates this research program to combine statistical machine learning with domain knowledge in power systems and make them applicable to stability analysis and control in real power systems. The long-term goal of this research program is to develop new approaches for power system stability analysis and effective online model-free and self-optimization control strategies. To achieve this ultimate goal, the short-term goals are (i) to develop new approaches for prediction, control and optimization of power systems to resolve the problem of bias in offline models used in the conventional power system simulation; and (ii) to apply the general approaches proposed in this program to various problems related to power system stability and develop new online control strategies for the same. The outcomes of this research are expected to not only constitute milestones in power system stability analysis and control, but also contribute to the development of a more reliable and stable power system in the future.