基于改进GM-LSSVR 模型的郑州市用水量预测

Forecast of Water Consumption in Zhengzhou City Based on Improved GM-LSSVR Model

准确的用水量预测是响应国家高质量发展的重要抓手,也是城市水资源优化配置的基础。针对用水量序列存在波动性、灰色模型与所需因素存在线性关系等问题,提出了一种基于HP滤波分解的GM-LSSVR预测模型,即先采用灰色关联分析法筛选合适的用水量影响因素,再利用HP滤波分解法将筛选的用水量及影响因素分解为长期趋势序列和短期波动序列,最后通过GM-LSSVR组合模型来预测用水量。以郑州市为例,使用该模型预测2001-2019年用水量,并与GM(1,N)模型、BP神经网络模型的预测结果进行对比。结果表明,基于HP滤波分解的GM-LSSVR预测模型预测精度大大提高,具有可行性与实用性,可以较好地应用于城市用水量预测研究中。

Accurate water consumption prediction is an important starting point for responding to the country’s high-quality development,and it is also the basis for the optimal allocation of urban water resources.In view of the volatility of the water consumption sequence and the linear relationship between the gray model and the required factors,this paper proposes a GM-LSSVR prediction model based on HP filter decomposition,that is,the gray correlation analysis method is first used to screen suitable water consumption influencing factors.And then the HP filter decomposition method is used to decompose the selected water consumption and influencing factors into a long-term trend sequence and a short-term fluctuation sequence,and finally the GM-LSSVR combined model is used to predict water consumption.Taking Zhengzhou City as an example,the model is used to predict water consumption from 2001 to 2019,and compared with the prediction results of the GM(1,N)model and the BP neural network model.The results show that the prediction accuracy of the GM-LSSVR prediction model based on HP filter decomposition is greatly improved,and it is feasible and practical,and can be better applied to the study of urban water consumption prediction.