Reinforcement Learning-and Proximal Policy Optimization Framework for Adaptive Feeding and Health Management in Dairy Cattle

This study showed the viability of reinforcement
learning-based on a Proximal Policy Optimization (PPO) to
automate the tasks of adaptive feeding and health management in
dairy cows. Using combinations of multimodal sensor data,
environment data, and behavior, the system was found to be
highly accurate in the prediction of milk yield and their health
scores, and also increased feed efficiency and decreased their
operational expenses. The findings surpassed traditional RLbased benchmarks, including DQN, A2C, and the model-based
approaches, as well as confirming the real-time practicability
under low latency. This affirms that reinforcement learning may
be a viable real-world decision-support system in both enhancing
farm outputs and animal welfare. In the future, we will expand
on this structure to integrate explainable AI to increase
explanability to farmers, use federated learning to scale to more
than one farm, but not at the expense of data privacy, and lastly,
apply novel and complex multi-task GCN models to more fully
capture the complex interdependencies. The resultant
adaptability, strength, and sustainability of the strategy over the
long run of deployment in varying farm settings will also be
established.