应用时效函数对天然气需求进行预测

用线性回归、人工神经网络、灰色系统等预测模型对天然气消费需求量进行预测后，为综合利用各种方法提供的信息，避免单一预测模型丢失有用信息的缺陷，减少随机性，提高预测的准确性,采用了优化组合预测模型和变权重的自适应递推优化组合预测模型对天然气消费需求量进行动态预测,取得了较好的预测效果。但经过翔实的分析,发现变权重的自适应递推优化组合预测模型计算出的权系数,假定了“移动样本数据”的权重向量为常值,权重没有充分体现出数据的时效性。依据信息论“远小近大“的观点，引入时效函数对自适应优化组合预测模型进行改进。结果表明,融入时效函数的自适应递推优化组合预测模型比采用其它预测模型预测的结果更好。

Time Effect Function Applied to Predict Natural Gas Demand

The gas demand is predicted with the linear regression forecast model; the artificial neutral network forecast model; and the gray system forecast model etc. respectively. To comprehensively utilize the information from different forecast model; avoid dropping the useful information, which may happen when using only one forecast model; decrease randomness and improve the forecast accuracy, the dynamic forecast is conducted for the gas demand by the optimized combination forecast model and the self adapt optimized combination forecast model with weight variation. And good forecast results are achieved. But with careful analysis, it is found as for the weight coefficient calculated by the self adapt optimized combination forecast model with weight variation, the weights don’t reflect the aging property of the data sufficiently since assuming the weight vector of “moved sample data” as a constant. According to the view of “the farer it is small, the closer it is big” by the information theory, the aging function is introduced to modify the self adapt optimized combination forecast model. The results show the self adapt optimized combination forecast model covering the aging function is better than the other forecast models for gas demand forecast.