摘要
为提高水产养殖中溶解氧预测的平稳性和时间依赖性,本研究提出了一种基于自注意力机制的双向长短期记忆神经网络(BiLSTM*)模型。通过引用自注意力机制,动态地调整不同时间点之间的权重,提取关键时间点特征;利用BiLSTM*模型建立溶解氧与其他水质指标之间的复杂交互关系;基于自注意力机制,构建包括自注意力层、模型结构层、模型优化层,加强模型对长序列数据的处理能力,提高序列间的依赖捕捉性能。结合水质检测水下机器人试验平台,在水槽和自然湖泊水域的初步验证实验的基础上,在海上养殖平台开展实际测试验证。结果表明本研究提出的溶解氧预测模型的平均绝对误差为6.2×10^(-3),均方根误差为9.49×10^(-3),与长短期记忆神经网络(LSTM)模型和BiLSTM*模型有明显提高,在长时序预测具有更好的精度和泛化能力。
In order to improve the stability and time dependence of dissolved oxygen prediction in aquaculture,this study proposed a bidirectional long short-term memory neural network(BiLSTM*)model based on selfattention mechanism.By referring to the self-attention mechanism,the weights between different time points were dynamically adjusted to extract the features of key time points.The complex interaction between dissolved oxygen content and other water quality indicators was established by using BiLSTM* model.Based on the selfattention mechanism,a self-attention layer,a model structure layer,and a model optimization layer were constructed to enhance the model's ability to process long sequence data and improve the dependence capture performance between sequences.Combined with the underwater robot test platform for water quality detection,based on the preliminary verification experiments in the flume and natural lake waters,the actual test verification was carried out on the offshore aquaculture platform.The results showed that the average absolute error of the dissolved oxygen prediction model proposed in this study was 6.2×10^(-3),and the root mean square error was 9.49×10^(-3),which was significantly improved compared with the long short-term memory neural network(LSTM)model and the BiLSTM* model,and had better accuracy and generalization ability in long time series prediction.
作者
高秀晶
姚海峰
洪黎丹
张碧雯
王芳
GAO Xiujing;YAO Haifeng;HONG Lidan;ZHANG Biwen;WANG Fang(School of Smart Marine Science and Technology,Fujian University of Technology,Fuzhou 350118,China;Fujian Provincial Key Laboratory of Marine Smart Equipment,Fuzhou 350118,China;Institute of Smart Marine and Engineering,Fujian University of Technology,Fuzhou 350118,China;School of Materials Science and Engineering,Fujian University of Technology,Fuzhou 350118,China;College of Mechanical and Transport Engineering,Hunan University,Changsha 410012,China)
出处
《环境工程学报》
北大核心
2025年第9期2300-2310,共11页
Chinese Journal of Environmental Engineering
基金
福建省财政厅教育和科研专项资助项目(GY-Z22011)
福建省财政厅教育和科研专项资助项目(GY-Z220232)
企事业委托横向资助项目(GY-H-23215)。
