Abstract: A resultant forecast model for prediction of hourly load of natural gas is proposed based on Haar wavelet transforming and ARIMARBF in this paper. Firstly, adopting Mallat fast algorithm and choosing Haar wavelet as mother wavelet, the gas hour load is decomposed, then the high frequency signals are predicted with ARIMA, and the low frequency is predicted with RBF. Secondly, the high frequency and the low frequency are reconstructed by Haar wavelet. Finally, taking gas hour load of a city for example, the effectiveness of prediction model is verified and compared with SOFM+MLP. The results indicate that the MAPE of the combination forecasting model is higher than 2.593 2%, the prediction accuracy is significantly improved in this paper, which provide a new useful reference for the shortterm forecasting in online engineering application.

Key words: Gas hour load,    ,  Haar wavelet transform,    , ARIMA,    , RBF,    ,  Forecast

摘要： 针对天然气时负荷预测问题, 提出了一种基于 H a a r小波变换和 AR I MA - R B F的天然气时负荷组合 预测模型。首先, 对天然气时负荷数据样本时间序列进行小波分解, 采用 M a l l a t快速算法, 母小波为 H a a r小波, 对 分解出来的高频分量进行 AR I MA预测, 低频分量进行R B F预测; 其次, 对高频分量预测结果和低频分量预测结果 进行 H a a r小波重构; 最后, 以某市实际采集的天然气时负荷为例进行研究, 并与自组织特征映射( S e l f - o r g a n i z i n g F e a t u r eM a p, S OFM) 网络和多层感知器( Mu l t i l a y e rP e r c e p t r o n, ML P) 网络( S OFM+ML P) 组合预测模型进行对比 分析。结果表明, 组合预测模型较S OFM+ML P预测模型的 MA P E值指标高出2. 5 9 32%, 预测精度显著提高, 为 实际工程的在线应用提供了有益参考。

关键词: 天然气时负荷, H a a r小波变换, AR I MA, R B F, 预测