Scholarlink

<p>Dissolved oxygen (DO) represents a critical water quality parameter in aquaculture environments, where its accurate prediction and effective control play a determining role in ensuring the health and productivity of cultured organisms. Conventional prediction methods struggle to capture the nonlinear dynamic characteristics of DO variations, while traditional control strategies rely on complex physical models with limited adaptability. This study proposes a hybrid model combining long short-term memory (LSTM) networks with Kolmogorov–Arnold networks (KAN), optimized by a quantum-inspired electric eel foraging algorithm (QEFA), for DO prediction and control systems. The proposed approach leverages probability representation and evolution mechanisms from quantum computing, enhancing global search and local refinement capabilities through classical computational strategies that simulate quantum superposition, entanglement, and interference effects. This study employed IoT-based continuous monitoring at 5-min intervals from June 23 to July 31, 2022. After quality control, 10,934 valid data points were obtained, measuring key parameters including dissolved oxygen, water temperature, pH, ambient temperature, humidity, and atmospheric CO<sub>2</sub> concentration. The experimental findings reveal that the QEFA-LSTM-KAN model exhibits superior performance relative to current methodologies when evaluated using various assessment criteria. The model demonstrates exceptional accuracy with a root mean square error (RMSE) value of 0.0244, a mean absolute error (MAE) measuring 0.0197, and a mean absolute percentage error (MAPE) recorded at 0.3054%. Building upon this model, a deep learning model predictive control (NNMPC) framework was developed, which delivers precise dissolved oxygen (DO) control capabilities. This system shows a notable improvement in performance, with the integral time-weighted absolute error (ITAE) decreasing by 5.17% when benchmarked against traditional model predictive control (MPC) approaches. This research provides a novel technological pathway for intelligent water quality management in aquaculture, with significant implications for improving farming efficiency and promoting sustainable aquaculture practices.</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 QEFA-optimized LSTM-KAN hybrid model for dissolved oxygen prediction and control in aquaculture

  • 期刊论文

  • 发表日期：

    2025年12月3日

查看全文
  • Zhuhong Che,
  • Longqin Xu,
  • Bohao Zhang,
  • Weiwei Zhang,
  • Xinmiao Wang,
  • Huiyuan Pang,
  • Tonglai Liu,
  • Shahbaz Gul Hassan,
  • Liang Wang,
  • Shuangyin Liu

摘要

Dissolved oxygen (DO) represents a critical water quality parameter in aquaculture environments, where its accurate prediction and effective control play a determining role in ensuring the health and productivity of cultured organisms. Conventional prediction methods struggle to capture the nonlinear dynamic characteristics of DO variations, while traditional control strategies rely on complex physical models with limited adaptability. This study proposes a hybrid model combining long short-term memory (LSTM) networks with Kolmogorov–Arnold networks (KAN), optimized by a quantum-inspired electric eel foraging algorithm (QEFA), for DO prediction and control systems. The proposed approach leverages probability representation and evolution mechanisms from quantum computing, enhancing global search and local refinement capabilities through classical computational strategies that simulate quantum superposition, entanglement, and interference effects. This study employed IoT-based continuous monitoring at 5-min intervals from June 23 to July 31, 2022. After quality control, 10,934 valid data points were obtained, measuring key parameters including dissolved oxygen, water temperature, pH, ambient temperature, humidity, and atmospheric CO2 concentration. The experimental findings reveal that the QEFA-LSTM-KAN model exhibits superior performance relative to current methodologies when evaluated using various assessment criteria. The model demonstrates exceptional accuracy with a root mean square error (RMSE) value of 0.0244, a mean absolute error (MAE) measuring 0.0197, and a mean absolute percentage error (MAPE) recorded at 0.3054%. Building upon this model, a deep learning model predictive control (NNMPC) framework was developed, which delivers precise dissolved oxygen (DO) control capabilities. This system shows a notable improvement in performance, with the integral time-weighted absolute error (ITAE) decreasing by 5.17% when benchmarked against traditional model predictive control (MPC) approaches. This research provides a novel technological pathway for intelligent water quality management in aquaculture, with significant implications for improving farming efficiency and promoting sustainable aquaculture practices.