一种结合互补集合经验模态分解和小波核极限学习机的短期电力负荷预测模型

A SHORT-TERM POWER LOAD FORECASTING MODEL COMBINING COMPLEMENTARY ENSEMBLE EMPIRICAL MODE DECOMPOSITION AND WAVELET KERNEL EXTREME LEARNING MACHINE

电力系统的管理和调度对精确的负荷预测模型有着极高的要求。为全面提高负荷预测模型的性能,提出一种新型的结合互补集成经验模态分解(CEEMD)和小波核函数极限学习机(WKELM)的短期电力负荷组合预测模型。首先通过CEEMD将历史电力负荷数据自适应地分解为一系列相对平稳的子序列,对各分量建立小波核极限学习机的预测模型,预测各分量的负荷值并对其进行求和得到最终预测结果。用四种预测模型对真实的负荷数据进行训练预测,算例表明新模型在预测精度和效率上都具有一定优势,同时克服了传统EMD中容易出现的模态混叠问题以及ELM中存在的过拟合等缺陷,具有一定的实际应用潜力。

Power system management and scheduling has extremely high demand on power load forecasting model. In order to comprehensively improve the performance of load forecasting model, we proposed a novel combination forecasting model for short-term power ioad, which combines the complementary ensemble empirical mode decomposition （CEEMD） and the wavelet kernel extreme learning machine （WKELM）. First, it adaptively decompose the historical power load data into a series of relatively stable sub-sequences by CEEMD, then it builds the forecasting models of WKELM on each decomposed component to forecast the load values of each component and makes summation on them to gain the final forecasting result. Using four kinds of forecasting models we carried out the training forecast on real load data, numerical examples showed that the new model has certain advantages in both accuracy and efficiency, at the same time it overcomes the defects of mode mixing easily arisen in traditional EMD and the over fitting in ELM, thus has certain practical applied potential.