应用共识PSO协同Trust-Tech方法的短期负荷预测  

Consensus-based PSO-assisted Trust-Tech Method for Short-term Load Forecasting

作　　者：张永峰[1] 崔凯[2] ZHANG Yongfeng CUI Kai(School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China State Power Economic Research Institute, Beijing 102209, China)

机构地区：[1]天津大学电气自动化与信息工程学院,天津300072 [2]国网北京经济技术研究院,北京102209

出　　处：《电力系统及其自动化学报》2017年第7期7-12,共6页Proceedings of the CSU-EPSA

基　　金：国家自然科学基金重点资助项目(51337007)

摘　　要：为了提高短期负荷预测的精度,基于共识粒子群算法协同Trust-Tech技术CPSOATT(consensus-based par-ticle swarm optimization-assisted trust-tech)的全局优化方法,该文提出一种新型神经网络预测器E-Elite。该预测器使用双层构架:底层使用CPSOATT方法设计一组具有不同最优结构的兼顾精度和多样性的子预测器;顶层选择子预测器作为隐含层神经元,设计基于神经网络结构的子预测器组合。顶层神经网络充分利用子预测器多样性和精度方面的性能优势,保证整体E-Elite预测器的高计算性能。最后使用E-Elite预测器对实际电力系统数据实现精确短期负荷预测,比较结果证明了该预测器的正确性和有效性。In order to improve the accuracy of short-term load forecasting, a novel neural network predictor, i.e., enhanced elite （E-Elite）, is proposed in this paper based on a global optimization method with consensus-based particle swarm optimization-assisted trust-teeh （CPSOATI＇）. The E-Elite adopts a double-stage architecture： on the bottom stage, CPSOATT is employed to design a set of accurate and diverse sub-predictors with different optimal structures; on the top stage, sub-predictors are selected as hidden neurons, and an ensemble of sub-predictors is designed based on the structure of neural network, which can ensure the high computation performance of E-Elite on the whole by fully taking advantage of the diversity and accuracy of sub-predictors. Finally, the accurate short-term load forecasting re- sults are realized by using E-Elite based on data from an actual power system, and the comparison result indicates the correctness and validity of the proposed predictor.

关 键 词：共识粒子群算法 人工神经网络 最优结构 短期负荷预测 

分 类 号：TM715[电气工程—电力系统及自动化]