摘要
提出了一种最优FCM聚类分析和最小二乘支持向量机回归算法(LSSVR)相结合的电力系统短期负荷预测方法。在考虑电力系统负荷日周期性的基础上,运用基于改进划分系数最大原则的最优FCM聚类分析获取历史负荷样本的最优数据模式划分,并根据输入样本相似度选取LSS-VR训练样本。既强化了训练样本的数据规律,又保证了数据特征的一致性,从而提高了LSSVR训练速度,改善了预测效果。仿真实验表明:LSSVR点模型的平均预测精度约98%,而本文模型的平均预测精度达到了98.7%,证明了该方法的有效性和实用性。
A short-term load forecasting method using least squares support vector regression and FCM clustering analysis with the best clusters number is proposed. On the basis of the load changing periodical nature, the best FCM clustering based on maximizing improved division coefficient is used to obtain the best data pattern division of load samples in the history. The training samples of least squares support vector machine are chosen according to the similarity degree of input samples. The data regularity of the training samples is enhanced and the consistency of data characteristic is ensured. As a result, the suggested method not only speeds up the training process, but also improves the forecasting effect. The simulation results show the model's precision is about 98.7% compared with 98% of that of LSSVR hour model, which proves the efficiency and applicability of the proposed method.
出处
《现代电力》
2008年第2期76-81,共6页
Modern Electric Power
基金
973国家重点基础研究发展计划专项资助项目(2004CB217905)
关键词
短期负荷预测
最小二乘支持向量机
最优FCM聚类
相似度
电力系统
short-term load forecasting
least square support vector machine
FCM clustering with the best class number
similarity degree
power system