基于聚类的支持向量回归模型在电力系统暂态稳定预测中的应用

Clustering Based on Support Vector Regression Model and Its Application in Power System Transient Stability Prediction

提出了一种电力系统暂态稳定实时预测方案,通过相量测量单元(PMU)获得扰动后短时间内的发电机功角相量,利用支持向量回归(SVR)模型可以快速而准确地预测发电机相对功角的变化趋势,从而可以判断电力系统的暂态稳定性。为了提高预测的精度和降低SVR的训练负担,利用自组织特征映射(SOFM)网络对训练样本集进行聚类分析,对聚类后的每一类样本训练SVR,由于每类样本具有相似性,所以对每类样本单独训练SVR可以更好地提高训练精度;又由于分类后的子类样本数目相对较小,所以可以克服全体训练样本对SVR训练时间过长的缺点。结合新英格兰10机系统对基于多种不同样本的SVR从训练时间和预测精度进行对比说明该方法的有效性。

This paper proposes a real time prediction method that may be applied in power system transient stability forecasting which can predict the future behavior with SVR （support vector regression） and the data coming from PMUs （Phasor Measurement Units）. With a view to improving the training efficiency of SVR and the prediction accuracy, the proposed method is based on self-organizing feature map （SOFM） that can discover the similar input data and cluster them into several classes in which input data have approximate trend. Then, the similar data is used as input data for a SVR predictor. Because the SOFM extracts similar data from learning data as a preprocessor, which decreases the sample set for one SVR, and also reduces the mutual influence of other learning data that is not related to the similar data of one class, the method not only enhances the training speed but also can forecast with high accuracy under different conditions. Forecasting results of simulation on New England 10 generators system prove the feasibility of this model.