基于改进的Mogrifier LSTM算法水质浊度预测模型  

Water quality turbidity prediction model based on improved Mogrifier LSTM algorithm

作　　者：杨博韬 刘黎志[1,2] YANG Botao;LIU Lizhi(Hubei Key Laboratory of Intelligent Robot,Wuhan Institute of Technology,Wuhan 430205,China;School of Computer Science and Engineering,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]武汉工程大学智能机器人湖北省重点实验室,武汉430205 [2]武汉工程大学计算机科学与工程学院,武汉430205

出　　处：《环境工程学报》2024年第7期1958-1963,共6页Chinese Journal of Environmental Engineering

基　　金：智能机器人湖北省重点实验室创新基金资助项目(HBIRL202207);湖北省教育厅科学研究计划指导性资助项目(B2017051)。

摘　　要：水环境资源保护作为当下最重要的工作之一,为了提高水质模型预测精度并制定更加完善的水资源管理策略,现提出一种基于改进的Mogrifier LSTM算法水质浊度数据预测模型,实现对水质数据的精准预测。该模型首先利用CNN卷积神经网络对复杂的水质数据进行特征提取,有效解决了水质数据非线性不稳定的特点,并对传统Mogrifier机制进行优化,引入扩展系数,利用PSO粒子群优化算法对超参数进行寻优操作,通过改造后的Mogrifier机制对LSTM模型中不同时刻的上下文信息进行融合,增强了水质数据的信息的交互。将预测结果与众多传统模型进行对比,结果表明CNN-改进Mogrifier LSTM模型具有更好的预测效果。Water environment and resource protection represent crucial tasks in the present era.To enhance water quality model prediction accuracy and develop more comprehensive water resource management strategies,a water quality turbidity data prediction model based on the improved Mogrifer LSTM algorithm was proposed to achieve accurate prediction of water quality data.Firstly,the model employed CNN to extract features from complex water quality data,effectively addressing the nonlinear and unstable characteristics of such data.Additionally,the traditional Mogrifier mechanism was optimized by introducing an expansion coefficient and hyperparameters was optimized using PSO algorithm.Through modified Mogrifier mechanism,context information from different moments in LSTM models was fused together to enhance interaction among water quality data.Comparison with many traditional models showed that CNN-improved Mogrifier LSTM model yielded better prediction results.

关 键 词：水质预测 浊度 卷积神经网络 形变长短时记忆网络 

分 类 号：X832[环境科学与工程—环境工程]