Scholarlink

<p>Dissolved oxygen (DO) management is crucial for the health and survival of fish in aquaculture. Maintaining optimal DO levels requires accurately predicting DO trends and developing appropriate control strategies. However, the dynamic and complex interactions among multiple water quality parameters make accurate multivariate time series prediction of DO challenging. Furthermore, traditional DO control methods often rely on intricate, site-specific DO mathematical models, complicating operational procedures and limiting their adaptability. To address these problems, we propose a novel prediction-control framework called IBM-MPC, which integrates a sophisticated predictive model with a neural network model predictive control (MPC) system. The predictive model incorporates a wrapper feature selection model, an improved snake optimization algorithm (ISO), and a hybrid deep learning model that combines bidirectional long short-term memory (BiLSTM) networks with a multi-head self-attention mechanism (MHSA). This configuration captures complex nonlinear relationships in multivariate time series data, enhancing prediction accuracy. The neural network MPC system leverages accurate DO predictions to optimize control strategies, reducing reliance on extensive physical data and computational complexity. Experimental results demonstrate the superior performance of the IBM-MPC model, achieving a root mean square error (<i>RMSE</i>) of 1.72%, a mean absolute error (<i>MAE</i>) of 1.05%, and a mean absolute percentage error (<i>MAPE</i>) of 12.57%, surpassing other baseline models in prediction accuracy. Furthermore, the neural network MPC achieves a reduction in the integral time absolute error (<i>ITAE</i>) by 5.2% compared to traditional MPC system, representing a significant advancement in aquaculture water quality management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A two-stage hybrid model for dissolved oxygen prediction and control in aquaculture

  • 期刊论文

  • 发表日期：

    2025年1月2日

查看全文
  • Ziang Chen,
  • Huiting Hu,
  • Shuangyin Liu,
  • Zhuhong Che,
  • Xinmiao Wang,
  • Zhuhua Hu,
  • Tonglai Liu,
  • Meng Cui,
  • Longqin Xu

摘要

Dissolved oxygen (DO) management is crucial for the health and survival of fish in aquaculture. Maintaining optimal DO levels requires accurately predicting DO trends and developing appropriate control strategies. However, the dynamic and complex interactions among multiple water quality parameters make accurate multivariate time series prediction of DO challenging. Furthermore, traditional DO control methods often rely on intricate, site-specific DO mathematical models, complicating operational procedures and limiting their adaptability. To address these problems, we propose a novel prediction-control framework called IBM-MPC, which integrates a sophisticated predictive model with a neural network model predictive control (MPC) system. The predictive model incorporates a wrapper feature selection model, an improved snake optimization algorithm (ISO), and a hybrid deep learning model that combines bidirectional long short-term memory (BiLSTM) networks with a multi-head self-attention mechanism (MHSA). This configuration captures complex nonlinear relationships in multivariate time series data, enhancing prediction accuracy. The neural network MPC system leverages accurate DO predictions to optimize control strategies, reducing reliance on extensive physical data and computational complexity. Experimental results demonstrate the superior performance of the IBM-MPC model, achieving a root mean square error (RMSE) of 1.72%, a mean absolute error (MAE) of 1.05%, and a mean absolute percentage error (MAPE) of 12.57%, surpassing other baseline models in prediction accuracy. Furthermore, the neural network MPC achieves a reduction in the integral time absolute error (ITAE) by 5.2% compared to traditional MPC system, representing a significant advancement in aquaculture water quality management.