基于Time-Causality模型的供热用气量预测分析

FORECASTING ANALYSIS OF HEATING GAS CONSUMPTION BASED ON TIME-CAUSALITY MODEL

目前关于时间序列预测的特征选择一直是研究的热点,但很少有学者分析多时间尺度下不同特征对预测的差异。提出基于Granger关系的Time-Causality预测模型,利用Granger关系进行特征选择,引入时间维度作为输入维度,并利用LSTM模型进行实验,在多时间尺度下分析预测供热用气量的特征。实验结果表明:Time-Causality模型能筛选到更有助于用气量预测的特征;从不同的时间尺度预测,所选取的特征不同;每个特征的预测作用也可能会随时间尺度的变化而变化。这为长期和短期预测提供理论和实践支持。

At present,feature selection of time series prediction has always been a hot topic of research,but few scholars have analyzed the difference of different features to the prediction at multiple time scales.This paper proposes a Time-Causality prediction model based on Granger causality.The Granger causality was used for feature selection,and the time dimension was introduced as the input dimension.And the LSTM model was used for experiments,and we analyzed and predicted the characteristics of heating gas consumption at multiple time scales.The experimental results show that the Time-Causality model can select the characteristics that are more helpful for gas consumption prediction.From different time scales,the selected features are different,and the prediction effect of each feature may also vary with time scales.It provides theoretical and practical support for both long-term and short-term forecasting.