[1]王坤,孙新洋,黄显怀,等.基于BP神经网络的给水厂混凝剂投加量预测[J].中国给水排水,2025,41(9):53-58.
　WANGKun,SUNXin-yang,HUANGXian-huai,et al.Prediction of Coagulant Dosage in Water Supply Plants Using Back Propagation Neural Network[J].China Water & Wastewater,2025,41(9):53-58.
点击复制

基于BP神经网络的给水厂混凝剂投加量预测

中国给水排水[ISSN:1000-4062/CN:12-1073/TU] 卷: 第41卷 期数: 2025年第9期 页码: 53-58 栏目: 出版日期: 2025-05-01
Title:
Prediction of Coagulant Dosage in Water Supply Plants Using Back Propagation Neural Network
作者:
王坤1,2， 孙新洋1,2， 黄显怀1,2， 唐玉朝1,2， 伍昌年1,2， 尹翠琴3， 张良霄3
（1.安徽建筑大学 环境与能源工程学院，安徽 合肥 230601；2.安徽建筑大学 环境污染控制与资源化安徽省重点实验室，安徽 合肥 230601；3.合肥水务集团有限公司，安徽 合肥 230011）
Author(s):
WANG　Kun1，2， SUN　Xin-yang1，2， HUANG　Xian-huai1，2， TANG　Yu-chao1，2， WU　Chang-nian1，2， YIN　Cui-qin3， ZHANG　Liang-xiao3
（1. School of Environment and Energy Engineering， Anhui Jianzhu University， Hefei 230601， China； 2. Anhui Provincial Key Laboratory of Environmental Pollution Control and Resource Reuse， Anhui Jianzhu University， Hefei 230601， China； 3. Hefei Water Group Co. Ltd.， Hefei 230011， China）
关键词:
混凝剂投加量;  变分模态分解（VMD）;  遗传算法（GA）;  BP神经网络
Keywords:
coagulant dosage;  variational mode decomposition(VMD);  genetic algorithm (GA);  back propagation (BP) neural network
摘要:
选取合肥市某给水厂运行数据，使用相关系数法进行筛选后，选取流量、浊度、耗氧量、pH和水温作为预测模型的输入参数，利用多层变分模态分解（VMD）算法捕捉数据信息，通过遗传算法（GA）优化BP神经网络权重和偏置，建立基于VMD-GA-BP神经网络的给水厂混凝剂投加量预测模型。该模型评价指标平均绝对误差（MAE）、均方根误差（RMSE）、平均绝对百分比误差（MAPE）均低于单层VMD-GA-BP模型、输入参数未经多层VMD分解的GA-BP模型和未利用GA优化的VMD-BP模型，相比单层VMD-GA-BP模型，MAE下降37.26%、RMSE下降36.19%、MAPE下降2.44%；与GA-BP模型相比，MAE下降27.03%、RMSE下降23.94%、MAPE下降1.43%；与VMD-BP模型相比，MAE下降40.99%、RMSE下降41.47%、MAPE下降2.83%。结果表明，多层VMD算法与GA的参与提高了模型预测的准确性和稳定性，模型能有效拟合混凝剂投加量变化趋势。
Abstract:
The operational data of a water supply plant in Hefei were selected and screened using correlation coefficient method. Flow rate, turbidity, oxygen consumption, pH, and water temperature were identified as input parameters for the prediction model. Data characteristics were extracted using multi-layer variational mode decomposition (VMD) algorithm, while genetic algorithm (GA) was employed to optimize the weights and biases of the back propagation (BP) neural network. Consequently, a prediction model for coagulant dosage in a water supply plant based on VMD-GA-BP neural network was successfully established. The mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) of the model evaluation indices were significantly lower than those of the single-layer VMD-GA-BP model, the GA-BP model without multivariate VMD decomposition of input parameters, and the VMD-BP model without GA optimization. Specifically, compared to the single-layer VMD-GA-BP model, the MAE, RMSE and MAPE decreased by 37.26%, 36.19%, and 2.44% respectively. Compared with the GA-BP model, the MAE, RMSE and MAPE was reduced by 27.03%, 23.94%, and 1.43% respectively. Compared with the VMD-BP model, the MAE, RMSE and MAPE was reduced by 40.99%, 41.47%, and 2.83% respectively. The integration of multi-layer VMD algorithm and GA enhances both the accuracy and stability of model predictions, allowing the model to effectively capture the changing trend of coagulant dosage.
更新日期/Last Update: 2025-05-01