【摘要】 提出了多层次细节分解的负荷聚类算法及其性能评估指标。该算法利用负荷序列间的差分序列均方差和欧氏距离形成交集优化判据;同时根据随机因素对负荷的敏感性加入对应参数要求来控制多层次细节分解聚类,对负荷曲线轮廓相似性细节程度聚类是提高预测精度的重要基础。笔者对所提出的聚类算法与一般欧氏距离聚类、Kohonen神经网络聚类算法进行了性能评估和比较,证明了该算法对季节性负荷具有高敏感性,对高温和气候因素与负荷之间的复杂相关性具有高识别能力,该聚类算法对提高负荷预测精度是有效的。更多还原

【Abstract】 A load clustering algorithm (CA) using multi-hierarchy and detailed decomposition based on data mining (DM) and its performance evaluation index are proposed, in which the Euclidean distances and mean square deviation of difference series of load series are used to construct optimal criterion of intersection, at the same time the multi-hierarchy and detailed decomposition clustering is controlled by adding requirements of corresponding parameters according to the sensitivity of random factors to loads, and it is an important base for improving load forecasting precision to cluster the detailed extent of load curve’s contour similarity. The performance of the proposed CA is evaluated and compared with that based on the Euclidean distance and that based on Kohonen neural network by simulation, and it is proved that the proposed CA possesses both high sensitivity to loads in different seasons and high recognition ability to complicated relativity of the load time series affected by high temperature and meteorological factors. So the presented CA can improve effectively the load forecasting precision.更多还原