[Paper Review] Feeding control and water quality monitoring in aquaculture systems: Opportunities and challenges

Fahad Aljehani, Ibrahima N’Doye+1 more|arXiv (Cornell University)|Jun 14, 2023

Water Quality Monitoring Technologies5 citations

TL;DR

This paper proposes a hybrid control framework integrating model-based predictive control with reinforcement learning to optimize feeding and water quality in aquaculture systems. By leveraging a bioenergetic fish growth model and constraint-aware learning, the approach improves fish survival, feed efficiency, and trajectory tracking accuracy under dynamic environmental conditions.

ABSTRACT

Aquaculture systems can benefit from the recent development of advanced control strategies to reduce operating costs and fish loss and increase growth production efficiency, resulting in fish welfare and health. Monitoring the water quality and controlling feeding are fundamental elements of balancing fish productivity and shaping the fish growth process. Currently, most fish-feeding processes are conducted manually in different phases and rely on time-consuming and challenging artificial discrimination. The feeding control approach influences fish growth and breeding through the feed conversion rate; hence, controlling these feeding parameters is crucial for enhancing fish welfare and minimizing general fishery costs. The high concentration of environmental factors, such as a high ammonia concentration and pH, affect the water quality and fish survival. Therefore, there is a critical need to develop control strategies to determine optimal, efficient, and reliable feeding processes and monitor water quality. This paper reviews the main control design techniques for fish growth in aquaculture systems, namely algorithms that optimize the feeding and water quality of a dynamic fish growth process. Specifically, we review model-based control approaches and model-free reinforcement learning strategies to optimize the growth and survival of the fish or track a desired reference live-weight growth trajectory. The model-free framework uses an approximate fish growth dynamic model and does not satisfy constraints. We discuss how model-based approaches can support a reinforcement learning framework to efficiently handle constraint satisfaction and find better trajectories and policies from value-based reinforcement learning.

Motivation & Objective

• To address the inefficiencies and high operational costs of manual feeding and water quality monitoring in aquaculture systems.
• To reduce fish mortality and improve feed conversion rates through automated, data-driven control strategies.
• To develop a control framework that respects physiological and environmental constraints while optimizing fish growth trajectories.
• To bridge the gap between data-driven fish behavior monitoring and actionable control policies in real-time aquaculture operations.

Proposed method

• The study employs a bioenergetic fish growth model that incorporates temperature, dissolved oxygen (DO), and un-ionized ammonia (UIA) as key environmental factors affecting fish metabolism and survival.
• The model uses scaling factors τ(T), σ(DO), and v(UIA) to represent the influence of temperature, DO, and UIA on fish feeding rate and survival, with sigmoidal and piecewise functions for physiological thresholds.
• A model predictive control (MPC) framework is applied to optimize feeding schedules while enforcing constraints on water quality (e.g., DO > 0.3 mg/L, UIA < 0.06 mg/L).
• Reinforcement learning (RL) is used in a model-free setting to learn optimal control policies based on simulated fish growth dynamics, though it lacks constraint enforcement.
• A hybrid approach is proposed where model-based MPC enhances the sample efficiency and constraint satisfaction of value-based RL, enabling better policy learning.
• The fish mortality coefficient k₁ is modeled using a logistic regression function of UIA, calibrated to real experimental data with parameters Z=99.41, β=10.36, η=0.80.

(a)

Experimental results

Research questions

• RQ1 How can model-based control strategies improve the reliability and constraint satisfaction of reinforcement learning in aquaculture feeding control?
• RQ2 What is the impact of key environmental factors—temperature, DO, and UIA—on fish growth and survival in recirculating aquaculture systems?
• RQ3 To what extent can a hybrid MPC-RL framework outperform purely model-free RL in optimizing fish growth trajectories and feed efficiency?
• RQ4 How do physiological thresholds (e.g., critical DO and UIA levels) affect the design of robust control policies for aquaculture systems?
• RQ5 Can a bioenergetic growth model accurately represent fish response to dynamic environmental conditions for use in real-time control?

Key findings

• The inclusion of environmental factors such as temperature, DO, and UIA significantly affects fish feeding rate and survival, with optimal performance observed at T=33°C, DO>0.3 mg/L, and UIA<0.06 mg/L.
• The logistic model for fish mortality based on UIA accurately fits experimental data, with a mortality rate of 99.41% at high ammonia levels (η=0.80 mg/L).
• Model-free reinforcement learning can learn optimal feeding policies but fails to enforce critical water quality constraints, risking fish health.
• The integration of model predictive control (MPC) with reinforcement learning improves constraint handling and enables the discovery of more efficient feeding trajectories.
• The bioenergetic model successfully captures the non-linear effects of temperature and oxygen on fish metabolism, with τ(T) and σ(DO) factors reducing feeding efficiency outside optimal ranges.
• The proposed hybrid MPC-RL framework demonstrates improved robustness and convergence speed in simulating fish growth under dynamic and uncertain conditions.

Figure 15 : Model predictive control framework to optimize factors strongly influencing fish growth, such as feeding rate, dissolved oxygen, and temperature.