基于复杂事件处理和深度学习的电压稳定协同感知与控制研究

As for today’s large-scale interconnected power grids in China, with the increasing dynamic loads and the complicated interaction between AC and DC power networks, the voltage stability problem caused by insufficient supports of local reactive power is becoming more and more pronounced. In terms of online situational awareness and control of voltage stability, there exist three main synergy problems: 1) the unified analysis of both mid- or long-term static and short-term transient processes; 2) the integration of temporal voltage/reactive power evolutions and the corresponding networked spatial information; 3) the combination of conventional simulation approaches and measured data-driven methods. Moreover, how to perform reliable real-time analysis and decision making for voltage stability is another great challenge. Faced with these problems, and based on the state-of-the-art technologies in artificial intelligence field, this proposal develops a systematic theoretical architecture and methodology for online synergic situational awareness and control of voltage stability. In particular, with the in-memory data grid technology, distributed simulation models for complex power grids are comprehensively constructed and further refined using actual measurement data. Based on deep learning approaches, basic voltage stability rules are adaptively mined from numerous operating scenarios. Guided by these rules, online synergic situational awareness of voltage stability is achieved via a complex event processing engine. After that, by systematically integrating multi-source yet heterogeneous information of networked spatial-temporal dynamics, an online synergetic voltage control scheme is designed on the basis of deep reinforcement learning. Furthermore, in order to enhance the adaptability and robustness of this control scheme, incremental learning using real-time small-size datasets is periodically performed. Finally, based on all these theoretical methods, a comprehensive application system for synergic situational awareness and control of voltage stability is developed and tested in real-world power grids for demonstration. The whole research is expected to achieve significant breakthroughs in both academic and industrial aspects.

目前中国复杂互联电网受动态负荷增加、交直流相互作用等因素影响，局部区域无功不足造成的电压稳定问题日益突出，在其感知与控制的研究和应用中主要存在三方面困难：中长期静态过程和短期暂态过程的协同、电压-无功时间演变和空间分布动态信息的协同、仿真计算和实测数据驱动方法的协同，此外，在线分析决策的实时性也是一个难点。针对这些问题，本项目借助人工智能技术的最新发展，研究基于复杂事件处理和深度强化学习的电压稳定协同感知和控制理论与方法，主要包括：基于内存数据网格建立分布式仿真和实测事件驱动模型；基于海量场景的深度学习，自适应挖掘电压稳定基本规则，通过复杂事件处理引擎，实现在线电压稳定协同感知；进一步有机融合多源异构的动态时空信息，利用深度强化学习实现在线电压协同控制，并针对实际电网实现小样本、增量式学习，提高适应性。最后，研究开发在线应用系统并在实际电网进行示范，取得理论创新和实际应用两方面的突破。

前中国电网由于动态负荷增加、交直流相互作用等因素，局部区域无功不足造成的电压稳定问题日益突出。电压稳定的感知与控制存在着中长期静态过程与短期暂态过程协同、电压稳定的时间演变和空间分布动态信息协同、仿真计算与实测数据驱动方法协同等难点。针对上述问题，本项目结合人工智能技术的最新进展，研究了基于复杂事件处理和深度强化学习的电压稳定协同感知和控制理论与方法。提出了基于面向对象建模技术与分布式内存数据网格技术的电力系统内存分析计算架构；考虑电压稳定时空分布特性，基于海量运行场景，利用shapelet方法与深度学习技术对电压稳定进行特征提取，自适应挖掘电压稳定的基本规则；提出了基于复杂事件处理技术的电网稳定态势感知方法，并开发了一种新的在线分析系统架构，实现了在线电压稳定协同感知；以前述的动态时空信息为输入，提出了基于深度强化学习技术的自治-协同低压减载方法与基于混合强化学习技术的静态电压协调控制方法，实现了在线电压协同控制；针对电力系统运行数据不断累积，提出了增量式电压稳定评估方案与基于迁移学习的增量式电压协同控制方案。最后，研究开发了相应的在线应用系统与软件，并在湖南省调和华中电网进行应用示范，实现了理论创新与实际应用的紧密结合。