Iuliana MARIN1, Diana-Alexandra CIUNGAN2, Dănuț-Nicolae ENEA3, Roxana Elena VASILIU4
Abstract. Heterogeneous monitoring of animals, environment and governance within precision livestock farming (PLF) requires interpretation, energy efficiency and auditability. To go beyond the classical CNN-LSTM based monitoring system, we extend the model suite to BIoTa-X, consisting of an extended federated temporal graph model (Fed-TGAT), a welfare risk estimating causal digital twin (CausalTwin) and a safe multi-objective control model (Safe-MORL) for a cattle smart-barn from Romania. The Fed-TGAT model represents all entities of interest, cow, barn, sensor and management events over time as a dynamic graph and learns their dependencies without export of farm raw data. CausalTwin enables the modeling of counterfactual scenarios to distinguish between intervention and correlation effects of time, animal identity, feed schedule and housing configuration. Safe-MORL is a constrained multi-objective reinforcement learning and predictive control model that recommends the best actions for ventilation, misting, shading and feeding to improve animal welfare, while being constrained by energy, and water. The data-governance layer of BIoTa-X consists of conformal uncertainty, explainable AI, federated model cards and blockchain. BIoTa-X is a solution for a transparent, adaptive and certifiable smart-barn intelligence system.
Keywords: precision livestock farming, graph attention networks, causal digital twin, blockchain, animal welfare.
DOI 10.56082/annalsarsciinfo.2026.1.28
1 Associate Professor, Ph.D., Faculty of Engineering in Foreign Languages National University of Science and Technology POLITEHNICA Bucharest, iuliana.marin@upb.ro
2 Ph.D. student, Faculty of Engineering in Foreign Languages National University of Science and Technology POLITEHNICA Bucharest, diana.ciungan@upb.ro
3 Lecturer, Ph.D., Faculty of Animal Productions Engineering and Management, University of Agriculture and Veterinary Medicine, nicolae.enea@usamv.ro
4 Ph.D. student, Faculty of Animal Productions Engineering and Management, University of Agriculture and Veterinary Medicine, roxana-elena.vasiliu@usamv.ro
PUBLISHED in Annals of the Academy of Romanian Scientists Series on Science and Technology of Information, Volume 19, No1