摘要
在负荷预测中,历史负荷数据产生的复杂性和许多不确定因素影响的随机性,使观测到的数据既包含线性部分,又包含许多非线性部分,因此所建立的预测模型就必须综合考虑这2方面的因素。目前常用的预测技术很少能综合考虑这两方面的因素,预测精度达不到要求。本文提出了一种时序分析和神经网络结合的预测方法。由于时序模型中不同阶数的自回归移动平均适合线性预测,可利用自回归移动平均模型(ARM A)处理历史负荷数据中的线性部分;而神经网络模型适合非线性预测,可利用人工神经网络(ANN)模型处理历史负荷数据的非线性部分。这样所建立的模型有机地结合了历史负荷中的线性因素和非线性因素,利用不同模型的优势来处理数据的不同部分,使得预测结果更为准确。实证证明,ARM A-ANN组合预测能提高负荷预测的精度。
In area of load forecasting,because of the complexity produced by historical load and randomness influenced by a lot of uncertain factors,the observed data not only includes the linear part but also includes a lot of non-linear parts,so the forecasting model set up must consider two aspects of these synthetically.At present,the forecasting technologies cannot consider the two aspects synthetically and the precision cannot meet the requirement.So a combined model of ARMA-ANN was proposed in the text to deal with this problem.As time series is suitable for linear prediction,the ARMA model can be adopted to deal with the linear part.While Artificial Neural Network is suitable for nonlinear prediction,the ANN model can be adopted to deal with the non-linear part.So the model combined the linear factor and non-linear factor synthetically,and it can use the advances of different models to deal with the different part of data,which can make the result more accurate.Experimental results indicate that a hybrid ARMA-ANN model can improve the load forecasting accuracy.
出处
《中国电力》
CSCD
北大核心
2005年第7期11-14,共4页
Electric Power
基金
高等学校博士点专项基金资助项目(20040079008)
河北省自然科学基金资助项目(G2005000584)
