摘要
由于能源消费内在的复杂性,传统的单尺度预测方法很难获得理想的预测效果.为提高能源消费量预测精度,提出了基于集合经验模态分解(EEMD)与最小二乘支持向量回归(LSSVR)的能源消费量多尺度预测模型.首先应用EEMD算法将能源消费量环比指数从高频到低频分解成若干结构更简单、变化更平稳、规律性更强、更易于预测的内在模态函数(IMF)和一个残差项;其次利用LSSVR对各IMF和残差项进行预测,并采用粒子群算法(PSO)选择最优的模型参数;然后将各分量的预测值直接加总求和重构出能源消费量环比指数的预测序列;最后通过逆环比化处理,获得原始能源消费量的最终预测值.利用该模型对1980-2013年广东省能源消费量进行实证分析,结果表明多尺度预测模型的确能够显著提高预测精度.
Due to the intrinsic complexity of energy consumption,traditional monoscale forecasting approaches fail to produce consistently good results.To improve energy consumption prediction accuracy,a novel multiscale ensemble learning paradigm integrating of ensemble empirical mode decomposition（EEMD）and least squares support vector regression（LSSVR）for energy consumption prediction is proposed.First,energy consumption link index is decomposed into several independent simpler structures,which are more stable,more regular,more predictable intrinsic mode functions（IMFs）and one residual from high to low frequencies,by using EEMD.Then,LSSVR,which is trained with particle swarm optimization（PSO）,is used to forecast every IMF and residual.Next,the forecasting values of all the IMFs and residual are summed as the final forecasted values of energy consumption.Taking energy consumption in Guangdong during 1980-2013 as a sample,empirical results show that the proposed multiscale prediction model can effectively improve forecasting accuracy.
出处
《内蒙古大学学报（自然科学版）》
CAS
北大核心
2015年第3期234-241,共8页
Journal of Inner Mongolia University：Natural Science Edition
基金
国家自然科学基金(71201010
71303174)
广东省自然科学基金(S2011010001591)
关键词
能源消费量
多尺度预测
集合经验模态分解
最小二乘支持向量回归
粒子群优化算法
energy consumption
multiscale forecasting
ensemble empirical mode decomposition
least square support vector regression
particle group optimization
