基于解耦机制的短期母线负荷预测方法

准确的短期母线负荷预测是实现节能降耗与调度精细化管理的基础，提出了一种基于解耦机制的预测方法。首先研究划分样本集最优簇结构的AFS（AP，FCM，Silhouette）聚类算法。利用AP聚类（affinity propagation clustering）计算样本集聚类数的搜索区间；从大到小排列各样本点的密度指标，得到初始化矩阵；通过Sil—houette指标进行有效性检验，获取最优聚类结果。将预测过程分为负荷水平预测和标幺曲线预测两部分，并制定适应其各自特点的预测策略。采用改进的灰色关联分析计算各日特征相关因素关联负荷水平的权值，并将该权值赋予相似选择的目标函数，由最小二乘支持向量机训练相似集进而做出预测；划分标幺曲线样本集的最优簇结构，利用逐步判别分析建立的Bayes判别函数将目标日归类，并根据相似度加权平均该类历史标幺曲线。实例分析验证了该预测机制及模型的优越性。

Short-time Bus Load Prediction Method Based on Decoupling Mechanism

Correct short-time bus load prediction is the basis for delicacy management on realizing energy saving and cost re- duction and dispatch. This paper proposes a kind of prediction method based on decoupling mechanism which firstly studies AFS clustering algorithm for dividing optimal cluster structure of sample set. By using AP clustering, it is able to calculate searching region of number of cluster of the sample set. According to density index of various sample points arranged from large to small, it is feasible to acquire initial matrix. On the basis of effectiveness check by using Silhouette index, it is able to obtain optimal clustering results. This paper divides prediction process into two parts including load level prediction and per unit curve prediction and formulates related prediction strategies suitable for each characteristic. Using improved grey corre- lation analysis, it studies calculation on weight of associated load level of characteristic relevant factor of each day which is given to similar selective obiective function. By using least square support vector machine to train the similar set, it is able to make prediction. Dividing the optimal clustering structure of per unit curve sample set and classifying the objective day by making use of Bayes discrimination function established by means of stepwise analysis, it is passable to weighted average this kind of historic per unit curve according to similarity. Practical example verifies superiority of this prediction mechanism and modcl