基于机器学习的河湖底泥机械脱水效果试验  

Experiment of Mechanical Dewatering Effectiveness of River and Lake Sediments Based on Machine Learning

作　　者：曾嘉辰 白鹤 王盛 郭兵 严晓威 郝宇驰 ZENG Jiachen;BAI He;WANG Sheng;GUO Bing;YAN Xiaowei;HAO Yuchi(CCCC National Engineering Research Center of Dredging Technology and Equipment Co.,Ltd.,Shanghai 200082,China)

机构地区：[1]中交疏浚技术装备国家工程研究中心有限公司,上海200082

出　　处：《净水技术》2023年第11期159-165,205,共8页Water Purification Technology

基　　金：国家重点研发计划(2021YFB2601100)。

摘　　要：为研究河湖底泥含水率、加药量和泥浆污泥比阻对河湖底泥脱水固化的影响,文章分别利用BP神经网络和符号回归方程方法,建立了泥浆含水率、加药量、泥浆污泥比阻表达的泥饼含水率之间的预测模型。结果发现:两种机器学习方法得到的预测模型相关性良好,均能够达到80%以上,基于4种常用的误差评价指标(MAE、MRE、MSE、RMSE)比较之下,BP神经网络预测结果准确度更优,误差均小于4%,且两种模型中输入参数中泥浆含水率、污泥比阻对最终泥饼含水率贡献程度相似且较大,占比均能够达到80%以上。研究建立的相关模型为河湖底泥机械脱水固化提供了可靠的预测和分析工具。In order to study the effects of water content of river and lake sediment,dosage and specific resistance of mud and sludge on dehydration and solidification of river and lake sediment,a prediction model between mud cake water content expressed by mud water content,dosage and specific resistance of mud and sludge was established by using BP neural network and symbolic regression equation.The results showed that the prediction models obtained by the two machine learning methods had good correlation accuracy,could reach more than 80%.Based on the comparison of four commonly(MAE,MRE,MSE,RMSE)used error evaluation index tables,the prediction accuracy of BP neural network was better,error was less than 4%,and the contribution of specific resistance of mud and sludge in the input parameters of the two models to the final mud cake moisture content was similar and large,could reach more than 80%.The established correlation model can provide a reliable prediction and analysis tool for the dehydration and solidification of river and lake sediment.

关 键 词：河湖底泥 机器学习 BP神经网络 符号回归方程 底泥脱水 

分 类 号：X703[环境科学与工程—环境工程]