基于聚类分析的滑动时均序列需水预测优化方法
Water Demand Prediction Optimization Method of Sliding Time-average Series Based on Cluster Analysis
-
摘要: 针对城市用水周期性及波动性特点,依据深圳市47个水厂及10个行政区2015-2019年逐月的供用水序列,考虑产业结构、人口特征及水厂供水的时间变化因子,提出KMeans聚类算法和季节性滑动平均自回归(seasonal moving average autoregressive,SMAAR)模型耦合方法,将水厂和行政区的时间序列进行聚类,分类别进行建模,预测2020年1-8月水厂及行政区的逐月供水数据,进而汇总出深圳市2020年1-8月的总供水数据,并与普通的自回归滑动平均(autoregressive moving average,ARMA)模型对比。结果表明:建模对象范围越小预测结果的RE较小。SMAAR的性能比ARMA有显著提升,且在长期预报中依旧表现出较强的泛化能力,254 d逐日预测结果的平均相对误差只有0.08。本研究方法可为城市需水预测和供水调度管理提供支撑。Abstract: In view of the cyclical and volatile characteristics of urban water use,based on the monthly water supply sequence of 47 waterworks and 10 administrative regions in Shenzhen from 2015 to 2019. Considering the industrial structure,demographic characteristics,and time change factors of water supply,a coupling method of the KMeans clustering algorithm and seasonal moving average auto regressive(SMAAR)model is proposed. The model can cluster the time series of waterworks and administrative regions and model by category to forecast monthly water supply data for waterworks and administrative regions from January to August 2020. Then the total water supply data of Shenzhen from January to August 2020 is summarized. The results show that the smaller the scope of the modeling object,the smaller the relative error(RE)of the predicted result. The performance of SMAAR is significantly improved compared to ARMA,and it still shows strong generalization ability in the long-term forecast. The average RE of the daily forecast results for 254 days is only 0.08. The method of this study can provide a reference for water demand forecasting and water supply scheduling management in other cities.