城市日供水量预测对供水部门具有十分重要的现实意义。为提高城市日供水量预测精度,以某市历史用水数据为基础,构建多粒度特征,并利用Pearson相关系数进行特征的筛选,基于XGBoost模型构建城市日供水量预测模型。本模型通过在训练集上进行训练和学习,在测试集上的平均绝对误差为70 571 t/d,平均相对误差为1.4%;传统的回归预测方法如随机森林法和支持向量机法,平均绝对误差分别为84 366 t/d和88 848 t/d。本模型预测精度更高,说明此模型可行、有效,具有一定的应用价值。
Abstract
Predicting the quantity of urban daily water supply is of great significance to water supply department in practice. To ameliorate the accuracy of predicting urban water supply, an XGBoost (eXtreme Gradient Boosting) integrated model of predicting the urban water supply is built based on the historical data of water supply with multi-granular features. Pearson correlation coefficient is used to select the optimal factor combination. Through training and learning on the training set, the results show that the average absolute error of this model is 70 571 t/d, and the average relative error is 1.4% on test set. Compared with traditional regression prediction methods such as random forest and support vector machine with the average absolute error amounting to 84 366 t/d and 88 848 t/d, respectively, the present method has higher prediction accuracy, indicating that the model is feasible and effective.
关键词
城市日供水量 /
多粒度特征 /
Pearson相关系数 /
XGBoost模型 /
预测精度
Key words
urban daily water supply /
multi-granularity feature /
Pearson correlation coefficient /
XGBoost model /
prediction accuracy
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 刘 洋,胡恩召,贾利民.组合优化的城市供水量预测模型[J].计算机仿真,2008(8):79-82,209.
[2] 白 云. 时间序列特性驱动的供水量预测方法研究及应用[D].重庆:重庆大学,2014.
[3] 孙晓婷,刘年东,杜 坤,等.混沌局域法与神经网络组合供水量预测[J].土木建筑与环境工程,2017,39(5):135-139.
[4] 王 盼,陆宝宏,张瀚文,等.基于随机森林模型的需水预测模型及其应用[J].水资源保护,2014,30(1):34-37,89.
[5] 郭冠呈,刘书明,李俊禹,等.基于双向长短时神经网络的水量预测方法研究[J].给水排水,2018,54(3):123-126.
[6] SHABANI S, YOUSEFI P, NASER G. Support Vector Machines in Urban Water Demand Forecasting Using Phase Space Reconstruction[J]. Procedia Engineering, 2017, 186: 537-543.
[7] BAI Y, WANG P, LI C, et al. A Multi-scale Relevance Vector Regression Approach for Daily Urban Water Demand Forecasting[J]. Journal of Hydrology, 2014, 517(2):236-245.
[8] 明均仁,肖 凯.基于R语言的面向需水预测的随机森林方法[J].统计与决策,2012(9):81-83.
[9] 鞠佳伟,宋良喜,马晓明.苏州市吴江区日供水量预测模型的建立与应用[J].中国给水排水,2017,33(23):141-144.
[10]MORADKHANI H, MEIER M. Long-lead Water Supply Forecast Using Large-Scale Climate Predictors and Independent Component Analysis[J]. Journal of Hydrologic Engineering, 2010, 15(10): 744-762.
[11]YIN Z, JIA B, WU S, et al. Comprehensive Forecast of Urban Water-Energy Demand Based on a Neural Network Model[J]. Water, 2018, 10(4):385.
[12]刘年东,周 明,杜 坤,等.基于相空间重构的BP供水量预测模型的改进[J].给水排水,2016,52(增刊1):289-292.
[13]肖 汉,曹 军,李凌霄.基于PCA的BP神经网络在城市供水预测中的应用[J].山东农业大学学报(自然科学版),2013,44(2):266-270.
[14]黄 卿,谢合亮.机器学习方法在股指期货预测中的应用研究:基于BP神经网络、SVM和XGBoost的比较分析[J].数学的实践与认识,2018,48(8):297-307.
[15]张家鸣,陈晓宏,叶长青.Pearson-Ⅲ型频率曲线对负偏水文序列的计算[J].水利学报,2012,43(11):1296-1301.
[16]蒋 辉,邓伟民,陈晓青.基于Pearson系数与多元核支持向量分类的葡萄酒分析[J].农业机械学报,2014,45(1):203-208.
[17]WILLIAMS S. Pearson’s Correlation Coefficient[J]. New Zealand Medical Journal, 1996, 109(1015): 38.
[18]王 可,王慧琴,殷 颖,等.基于Pearson关联度BP神经网络的时间序列预测[J].光学精密工程,2018,26(11):2805-2813.
[19]刘云翔, 吴 浩. 基于改进CART决策树建立水华预警模型[J]. 中国农村水利水电, 2018(1):26-28.
基金
国家自然科学基金面上项目 (71373123);中央高校基本科研业务费专项(NW2018004)
PDF(4810 KB)
