基于电压相量和深度学习的电力系统暂态稳定快速评估

发布时间：2018-11-05 11:46

【摘要】：近些年以来,中国、美国、印度等国家发生了多起大面积停电事故,造成了巨大的经济损失和社会影响。可再生能源、电力负荷的进一步增长、电力系统中电力电子化的广泛应用等增加了电网运行的不确定性和复杂性,电网安全稳定运行面临着严峻的挑战。随着人工智能、深度学习的兴起,为研究人员从大数据、大样本、概率学的角度分析电网信息物理系统的安全稳定性,提供了坚实的理论基础。本文采用深度卷积网络(CNN)技术,通过构建可视化的全信息电网图,将CNN方法引入到电力系统暂态安全稳定性分析中。并在总结分析前人工作的基础上,结合支路势能的方法,解释了系统暂态可视化过程中电网“撕裂”特征的机理,并提出快速判断电力系统薄弱断面以及暂态稳定性快速判别的方法。基于可视化以及理论分析的成果,构建了深度卷积神经网络对系统暂态稳定进行分析,实现电力系统暂态稳定快速评估。本文主要包括以下内容:(1)构建了动态可视化的电网图。建立了基于电压复平面的电力系统映射平面,以IEEE 10机39节点为例,获取其仿真数据以及已有拓扑连接关系,基于Echart实现该系统节点信息在电压复平面上动态展示。(2)建立了基于电压复平面动态信息的电力系统暂态稳定分析模型。从薄弱断面的角度,分析了电压相量距离和振荡中心落点支路的关系,给出快速识别薄弱断面的方法。在此基础上,进一步分析了薄弱断面领先机群侧母线的相轨迹特征,提出了基于薄弱断面两端母线相轨迹特征的暂态稳定性判别方法。(3)开发了样本批量制造程序,为深度学习提供足够的样本。借鉴深度学习方法在人工视觉领域处理数据的方法,以处理图片像素矩阵的角度,将深度学习的方法应用于电力系统暂态稳定性评估。通过实验的方法,确定了合适的模型参数,提出了多窗口滑动识别电力系统暂态稳定性的方式,减少了模型的参数并提升了判断准确率。
[Abstract]:In recent years, China, the United States, India and other countries have had a number of large-scale power outages, resulting in huge economic losses and social impact. With the further growth of renewable energy, power load and the wide application of electronization in power system, the uncertainty and complexity of power grid operation are increased, so the safe and stable operation of power network is facing severe challenges. With the rise of artificial intelligence and deep learning, it provides a solid theoretical basis for researchers to analyze the security and stability of power grid information physical systems from the perspective of big data, large samples and probabilities. In this paper, the deep convolution network (CNN) technique is used to construct a visual full information grid diagram, and the CNN method is introduced to the transient security stability analysis of power system. On the basis of summing up and analyzing the previous work, combined with the method of branch potential energy, the mechanism of power grid "tearing" in the process of system transient visualization is explained. A fast method for judging the weak section and transient stability of power system is proposed. Based on the results of visualization and theoretical analysis, a deep convolution neural network is constructed to analyze the transient stability of power system. The main contents of this paper are as follows: (1) A dynamic visual grid diagram is constructed. The power system mapping plane based on voltage complex plane is established. Taking 39 nodes of IEEE 10 machine as an example, the simulation data and the existing topology connection are obtained. The node information of the system is dynamically displayed on the voltage complex plane based on Echart. (2) the power system transient stability analysis model based on the voltage complex plane dynamic information is established. From the point of view of weak section, the relationship between the voltage phasor distance and the branch of the oscillation center drop point is analyzed, and the method to identify the weak section quickly is given. On this basis, the phase locus characteristics of the side bus of the weak section leading cluster are further analyzed, and a method of judging transient stability based on the phase trace characteristics of the weak section is proposed. (3) A batch manufacturing program is developed. Provide enough samples for deep learning. The depth learning method is applied to power system transient stability evaluation by using depth learning method for data processing in artificial vision field to deal with image pixel matrix. Through the experimental method, the suitable model parameters are determined, and a multi-window sliding identification method is proposed to identify the transient stability of power system, which reduces the parameters of the model and improves the accuracy of judgment.
【学位授予单位】：中国电力科学研究院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM712

【参考文献】