基于PCA-RBF神经网络模型的城市用水量预测

发布时间：2018-02-12 20:22

本文关键词： 城市用水量预测主成分分析 RBF神经网络 BP神经网络主成分数量需水预测　出处：《水利水电技术》2017年07期 　论文类型：期刊论文

【摘要】：针对城市用水量影响因素众多、关联性较强以及BP神经网络收敛速度慢、易陷入局部极小值等问题,采用组合主成分分析(PCA)与RBF神经网络的方法预测城市用水量。利用主成分分析对用水量影响因素进行降维,消除多重共线性,选取能够替代原用水量影响因素的前三个主成分作为输入因子,选用学习和收敛速度快、模式识别能力强的RBF神经网络进行预测。研究结果表明,该模型的相对误差平均值在训练和预测阶段均最小,分别为0.165 4%和0.677 5%,学习和预测能力均优于RBF和BP神经网络模型,提高了收敛速度和预测精度;主成分数量从3个增加到5个,信息量累积贡献率从93.09%增加到98.37%,平均相对误差从0.250 7%降至0.206 0%,预测精度略有提高。对2015—2020年枣庄市用水量进行预测,总用水量先有小幅上升,后又下降,呈现"倒U型"增长。该模型对城市区域水资源规划具有参考价值。
[Abstract]:In view of many factors affecting urban water consumption, strong correlation and slow convergence speed of BP neural network, it is easy to fall into local minimum, etc. The method of combined principal component analysis (PCA) and RBF neural network was used to predict urban water consumption. The dimension of influencing factors of water consumption was reduced by principal component analysis (PCA), and the multiple collinearity was eliminated. The first three principal components which can replace the factors affecting the original water consumption are selected as input factors, and the RBF neural network, which has fast learning and convergence speed and strong pattern recognition ability, is selected for prediction. The average relative error of the model is the smallest in the training and prediction stages, which is 0.165 4% and 0.677 5, respectively. The learning and prediction ability of the model is better than that of the RBF and BP neural network models, and the convergence rate and prediction accuracy are improved. The number of principal components increased from 3 to 5, the cumulative contribution rate of information increased from 93.09% to 98.37, the average relative error decreased from 0.250 7% to 0.206 0, and the prediction accuracy was slightly improved. The total water consumption of Zaozhuang City in 2015-2020 was first slightly increased. The model has reference value for urban water resources planning.
【作者单位】：天津大学水利工程仿真与安全国家重点实验室;
【基金】：“十二五”国家科技支撑计划项目(2015BAB07B02) 国家自然科学基金创新群体基金项目(51621092);国家自然科学基金(51609166)
【分类号】：TV213.4

【相似文献】