基于集成多极限学习机的短期负荷预测

Short-Term Load Forecasting Based on Integrated Multi-Extreme Learning Machine

针对传统的电力负荷预测模型存在的预测精度不高和泛化能力不强的问题,提出一种基于Stacking集成学习、极限学习机(ELM)和最小二乘支持向量机(LSSVM)的短期负荷预测方法。以Calinski-Harabasz指标确定最佳簇数,采用模糊C均值(FCM)聚类筛选相似日样本集合,优选并减小样本规模。在Stacking集成学习框架下,建立多个ELM初级预测模型,以LSSVM作为元学习器,对多个预测结果重新融合,构成综合预测系统。算例分析表明,提出的方法较单一的ELM预测模型和传统的结合策略具有良好的预测精度和泛化性。

Aiming at the problems of low prediction accuracy and generalization ability of traditional power load forecasting model, this paper proposes a short-term load forecasting method based on stacking ensemble learning, extreme learning machine(ELM) and least squares support vector machine(LSSVM). The Calinski-Harabasz index was used to determine the optimal number of clusters, and the fuzzy C-means(FCM) clustering was used to select the similar day sample set to optimize and reduce the sample size. In the framework of stacking ensemble learning, elm was used to build multiple primary prediction models, LSSVM was used as a meta learner to re fuse multiple prediction results to form a comprehensive prediction system as a whole. The example analysis shows that the proposed method has better prediction accuracy and generalization than the single ELM prediction model and the traditional combination strategy.