World Aquaculture 2023

May 29 - June 1, 2023

Darwin, Northern Territory, Australia


Yanqing Duan*, Hui Li, Yingyi Chen, Yang Wang, Guoping Lian, Tao Chen


*University of Bedfordshire,

Park Square, Luton LU1 3JU

United Kingdom



Fish feeding optimization contributes to the balance of controlling fish growth and ensuring water quality in aquaponic system. The fish feeding control aims to minimize the gap between the target and actual individual fish weight at harvest time while minimizing feeding amount. The aim of our research is to develop a new fish feeding control strategy in aquaponic system based on reinforcement learning (RL) technology informed by model predictive control (MPC). We employ the fundamental RL based feeding control method using Q learning to solve feeding optimization problem. A fish feeding control method based on RL informed by MPC is proposed. As shown in figure 1, it integrates the feeding control decision of MPC into RL. In this process, MPC is firstly applied on the known aquaponic system model to obtain feeding strategy. Then, the obtained feeding control trajectory is considered as the background information of aquaponic system to determine initial Q table. The proposed methods are validated by simulated aquaponic system model. Different feeding control methods are compared to show performances with respect to final individual fish weight and feed conversion ratio (FCR). The simulation results show that the proposed feeding control method based on RL informed by MPC can make effective feeding decisions in terms of optimised fish growth, water quality of fish and plant tanks. The introduction of background knowledge improves the performances of fundamental RL based feeding control method, which would decrease 7.19% in FCR and improve 41.89% in final individual fish weight. Our work offers a simple way to solve feeding optimization method based on RL. The proposed approach can be applied to the feeding decision optimization in aquaponic site based on the specific production data collected