基于贝叶斯框架和回声状态网络的日最大负荷预测研究.pdfVIP

第 36 卷 第 11 期 电 网 技 术 Vol. 36 No. 11 2012 年 11 月 Power System Technology Nov. 2012 文章编号：1000-3673 （2012 ）11-0082-05 中图分类号：TM 731 文献标志码：A 学科代码：470·4054 基于贝叶斯框架和回声状态网络的 日最大负荷预测研究 嵇灵，牛东晓，吴焕苗 （华北电力大学经济与管理学院，北京市 昌平区 102206 ） Daily Peak Load Forecasting Based on Bayesian Framework and Echo State Network JI Ling, NIU Dongxiao, WU Huanmiao (School of Economics and Management, North China Electric Power University, Changping District, Beijing 102206, China) ABSTRACT: To overcome the pseudo-regression in neural 后来发展到了神经网络和系统动力学方法。神经网 network, the standard echo state network is improved and 络作为一种非线性动力学方法，具有强大的逼近非 generalization ability of standard echo state network is 线性映射能力，特别适合存在非线性和时变性的复 enhanced by Bayesian framework. To verify the availability and adaptability of the proposed method, in the analysis on 杂系统时间序列的建模和预测。在短期负荷预测 empirical example actual load data and related climatic data of 中，神经网络是最受欢迎的研究方法之一。各种改 a certain region are used as input variables to forecast the daily 进优化的神经网络用来提高神经网络的预测精度， peak load of this region. Comparison of the forecasted daily 如支持向量机、蚁群算法、模糊推理等[1-6] 。 load curve with actual daily load curve shows that the 递推神经网络在实际运用中数学分析困难，传 forecasted load curve by improved echo state network is more accurate than those forecasted by BP neural network and 统的学习算法耗时、计算量大。作为新的循环神经 standard echo state network, and the generalization ability of 网络，回声状态网络(echo state network，ESN)在稳 the improved echo state network is more stronger. 定性、全局最优性、局部最小问题和训练过程复杂 KEY