基于GA-LSTM模型的雨水管道液位监测系统  

Rainwater pipe liquid level monitoring system based on GA-LSTM model

作　　者：巢凌逸 莫文洁 蔡燕[1,2] 唐慧 许鹏[1,2] CHAO Lingyi;MO Wenjie;CAI Yan;TANG Hui;XU Peng(School of Information Science and Technology,Nantong University,Nantong 226019,China;Nantong Advanced Communication Technology Research Institute Co.,Ltd.,Nantong 226019,China)

机构地区：[1]南通大学信息科学技术学院,江苏南通226019 [2]南通先进通信技术研究院有限公司,江苏南通226019

出　　处：《电子设计工程》2025年第7期1-6,共6页Electronic Design Engineering

基　　金：国家自然科学基金项目(62271271);江苏省高等教育教学改革研究项目(2019JSJG541);江苏省现代教育技术研究项目(2019-R-79606);江苏省大学生创新创业训练计划项目(202310304183H)。

摘　　要：针对城市排水体系建设的需求,构建了一种基于GA-LSTM模型的雨水管道液位监测系统,实时采集液位数据,通过物联网设备上传到云端,并采用遗传算法对超参数进行优化调整实现液位预测的边缘计算,避免在采用LSTM模型预测液位高度时因主观选择超参而陷入局部最优,实现精准预测雨水管道未来液位。结合对比BP神经网络模型和LSTM模型,结果表明,GA-LSTM模型是一种预测精度较高的模型,在雨水管道液位预测时误差为±0.2 cm,且当其装载在边缘设备前端时,也具备高精度预测的能力。To meet the demands of urban drainage system construction,a rainwater pipeline level monitoring system based on the GA-LSTM model was developed.This system collects liquid level data in real-time,uploads it to the cloud via Internet of Things(IoT)devices,and utilizes a genetic algorithm to optimize hyperparameters for edge computing of level predictions.This approach circumvents the issue of falling into local optima due to subjective hyperparameter selection when using the LSTM model to predict liquid levels,achieving precise predictions of future rainwater pipeline levels.In comparison with the BP neural network model and the LSTM model,the results indicate that the GA-LSTM model is a highly accurate predictive model.It exhibits an error margin of±0.2 cm in rainwater pipeline level predictions and maintains high predictive accuracy when deployed on edge devices.

关 键 词：液位监测 遗传算法 LSTM神经网络 边缘计算 

分 类 号：TN6[电子电信—电路与系统]