面向电网调度自动化系统鲁棒性的验证用例生成方法研究

Research on the Verification Case Generation Method for Robustness of Power Dispatching Automation Systems

摘要:

电网调度自动化系统的稳定运行是保障电力系统安全的核心；对于电网调度自动化系统展开鲁棒性测试是保障系统面对异常、极端输入下性能良好的前提；对面向电网调度自动化系统鲁棒性的验证用例生成方法进行了综合分析；对电网调度自动化系统的特点与测试难点、现存电网调度自动化系统验证用例生成方法进行了分析，发现了现存方法在电网调度自动化系统鲁棒性测试方面的不适用性；研究了面向一般复杂系统鲁棒性测试的验证用例生成的方法框架；对于框架中涉及到的代理模型在不同维度下的拟合精度与速度进行了实验分析，经实验测试发现高斯过程分类模型适合于低维问题；支持向量机模型适合于中高维度问题；对于超过20维的问题，人工神经网络的拟合速度达到支持向量机模型的20倍，并且能达到令人满意的精度；详细分析了将鲁棒性测试验证用例生成框架用于电网调度自动化系统亟待解决的问题和可能存在的技术途径。

Abstract:

The stable operation of the power dispatching automation system is essential for ensuring power system security. Conducting robustness testing on such systems is a prerequisite for guaranteeing their performance under abnormal and extreme conditions. A comprehensive analysis was carried out on methods for generating verification cases aimed at evaluating the robustness of power dispatching automation systems. This study examined the characteristics and testing challenges of power dispatching automation systems, as well as existing methods for generating verification cases. The inadequacy of current approaches in robustness testing for power dispatching automation systems was identified. A methodological framework for generating test cases for testing other complex system robustness was investigated. The analysis focusing on fitting accuracy and speed of variant surrogate models was conducted. Experimental tests showed that the Gaussian process classification model suits low-dimensional problems. The support vector machine model fits medium to high-dimensional problems. For problems exceeding 20 dimensions, the artificial neural network's fitting speed is 20 times faster than the support vector machine model. It also achieves satisfactory accuracy. The unresolved issues and potential technical pathways for applying this framework to power dispatching automation systems was further analyzed.