CAJ | 학술논문

应用粗糙集理论中的决策表化简技术，提出了一种压缩人工神经网络(ANN)输入空间的方法，改善了ANN用于电力系统暂态稳定评估(TSA)时面临的数据训练瓶颈问题。由于训练样本是连续性的数据，采取了3种离散化方法：等频法、等距法和最大熵法。用10机39节点的新英格兰系统测试了该数据压缩方法的有效性。ANN初始输入变量为11个，利用粗糙集化简方法抽取出6个关键特征变量，样本集压缩了45.5%，而ANN稳定分类器的判别效果并没有因此受到影响。
응용조조집이론중적결책표화간기술，제출료일충압축인공신경망락(ANN)수입공간적방법，개선료ANN용우전력계통잠태은정평고(TSA)시면림적수거훈련병경문제。유우훈련양본시련속성적수거，채취료3충리산화방법：등빈법、등거법화최대적법。용10궤39절점적신영격란계통측시료해수거압축방법적유효성。ANN초시수입변량위11개，이용조조집화간방법추취출6개관건특정변량，양본집압축료45.5%，이ANN은정분류기적판별효과병몰유인차수도영향。
This paper proposes a rough-set-based approach for input dimension reduction in artificial neural network, which is used for power system transient stability assessment (TSA). Several discretization methods for the continuous data set are tested and evaluated. The 10-machine 39-bus New England system is used for simulation. Six out of the original 11 features are selected using rough set attribute reduction techniques. Comparison results show that the ANN classifier with the reduced input dimension is as effective as before, while the training data set compressed 45.5%.