基于卷积神经网络的电力系统小干扰稳定评估与预防控制  

Small-signal stability assessment and preventive control of power system based on convolutional neural network

作　　者：田芳 周孝信[1,2] 于之虹 TIAN Fang;ZHOU Xiaoxin;YU Zhihong(State Key Laboratory of Power Grid Safety,Beijing 100192;China Electric Power Research Institute,Beijing 100192)

机构地区：[1]电网安全全国重点实验室,北京100192 [2]中国电力科学研究院,北京100192

出　　处：《电气技术》2025年第3期1-6,14,共7页Electrical Engineering

基　　金：国家自然科学基金项目(U21666601)。

摘　　要：为了提升小干扰稳定预防控制措施制定的速度,本文提出基于卷积神经网络(CNN)灵敏度分析的小干扰稳定预防控制方法。针对系统中存在的若干弱负阻尼(阻尼比小于某一阈值)低频振荡模式,首先建立带小干扰稳定约束的优化模型,其次基于CNN阻尼比预测模型计算阻尼比相对于控制变量(可调发电机的有功功率)的灵敏度,通过灵敏度将小干扰稳定约束线性化,从而将优化模型转化为二次规划模型,最终得到发电机的有功功率调整量,通过多次迭代使阻尼比满足特定要求。WEPRI36节点算例分析结果表明,由CNN模型得到的控制措施十分有效,且较支持向量机模型更精准,控制措施制定的速度较传统特征值分析法快。本文研究思路也可用于暂态稳定预防控制。A small-signal stability preventive control method based on convolutional neural network(CNN)sensitivity analysis is presented in the paper,to improve the developing speed of small-signal stability preventive control measures.For poor or negative damping low frequency oscillation modes(i.e.,the damping ratios are smaller than a threshold),first,an optimization model with small-signal stability constraints is established;second,the sensitivities of the damping ratios with respect to control variables(the active power of adjustable generators)based on CNN model of damping ratio prediction are calculated and then the optimization model is transformed into a quadratic programming model by linearizing small-signal stability constraints through sensitivities;finally,the adjustment amounts of generator active power are obtained.Several iterations are needed to make the damping ratios meet specific requirements.Analysis results of WEPRI 36-node case show that the effective control measures can be obtained by the presented method,which is more precise than that of the support vector machine method.The computing speed of the presented method is faster than that of the traditional eigenvalue analysis method.The ideas presented in this paper can also be applied to transient stability preventive control.

关 键 词：卷积神经网络(CNN) 灵敏度分析 小干扰稳定 稳定评估 预防控制 

分 类 号：TM712[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]