Intelligent cloud-based RAS management: integration of DDPG reinforcement learning with AWS IoT for optimized...

Area of Science:

• Aquaculture technology
• Artificial intelligence in agriculture
• Reinforcement learning applications

Background:

• Deep Deterministic Policy Gradient (DDPG) shows promise for aquaculture but faces scalability issues in commercial Recirculating Aquaculture Systems (RAS).
• Previous research focused on DDPG for feeding and energy management in controlled environments.
• Real-world commercial deployment of AI control systems in aquaculture is limited by infrastructure and scalability challenges.

Purpose of the Study:

• To present a novel cloud-edge hybrid architecture for deploying DDPG-based control systems in diverse commercial aquaculture operations.
• To address practical challenges in scaling AI control systems for real-world aquaculture environments.
• To establish a practical blueprint for commercializing DDPG in aquaculture management.

Main Methods:

• Developed a cloud-edge hybrid architecture integrating AWS IoT Core and AWS Greengrass.
• Implemented edge optimization techniques (16-bit quantization, architecture pruning) to reduce DDPG model size by 74%.
• Conducted field validation in a commercial facility with 108 tanks (3,132 m³ total volume).

Main Results:

• Optimized DDPG model achieved 47±8 ms latency, enabling real-time inference.
• Demonstrated exceptional scalability with only an 8.9% latency increase from small to large-scale operations.
• Achieved 99.97% IoT message delivery, 98.7% reliability in critical parameter control, and 72-hour network disruption tolerance.

Conclusions:

• The proposed cloud-edge architecture successfully enables scalable DDPG deployment in commercial aquaculture.
• The system demonstrates high reliability, resilience, and operational safety under challenging network conditions.
• This research provides a viable pathway for adopting AI-based control systems in the broader aquaculture industry.