Deep reinforcement learning for intelligent decision-making in smallholder beef cattle farming: model innovation and industrial practice

This study addresses the long-standing challenge of decision-making in smallholder beef cattle farming, where reliance on experience often results in low efficiency and limited resilience to risk. We explore the application of deep reinforcement learning (DRL) to develop an intelligent decision support system tailored for beef cattle production. The research first reviews the current adoption of reinforcement learning in agriculture and identifies key factors influencing farmers’ decisions. Specifically, eight macro-level state variables—such as production cost, market price, and animal health—together with five micro-level state variables—including land resources and financial capacity—were defined as model inputs. A decision-making model was constructed based on the Deep Q-Network (DQN), incorporating a composite reward function that accounts for baseline state rewards, seasonal adjustments, stability penalties, risk modifiers, and long-term planning incentives. Model training employed ε -greedy exploration and experience replay mechanisms. Experimental results demonstrate that after 40 training episodes, the model achieved an average reward of 203.85, while the average loss decreased to 0.557, indicating stable convergence of the learned strategy. In a farmer-level case analysis, the dominant decision action accounted for 71.5% of the outcomes. Sensitivity analysis, combined with local interpretability using LIME and global feature importance via SHAP, revealed that risk preference, livestock health, financial status, and production cost were the most critical determinants of decision outcomes, while seasonal variation and labor availability also played significant roles. These findings confirm that DRL can provide precise decision support for farmers, thereby enhancing resource allocation, management priorities, and overall production resilience. The study ultimately offers actionable insights to promote the digital and intelligent transformation of the beef cattle industry.

Authors

Institutions

Publication Details

Journal
Frontiers in Sustainable Food Systems
Published
2025-12-08
DOI
https://doi.org/10.3389/fsufs.2025.1715568
Primary Topic
Animal Behavior and Welfare Studies
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Deep reinforcement learning for intelligent decision-making in smallholder beef cattle farming: model innovation and industrial practice

Chunrong Guo, Pan He, Xiaodong Zhang
Frontiers in Sustainable Food Systems
Animal Behavior and Welfare Studies
article

Deep reinforcement learning for intelligent decision-making in smallholder beef cattle farming: model innovation and industrial practice

Chunrong Guo, Pan He, Xiaodong Zhang
article en

Abstract

This study addresses the long-standing challenge of decision-making in smallholder beef cattle farming, where reliance on experience often results in low efficiency and limited resilience to risk. We explore the application of deep reinforcement learning (DRL) to develop an intelligent decision support system tailored for beef cattle production. The research first reviews the current adoption of reinforcement learning in agriculture and identifies key factors influencing farmers’ decisions. Specifically, eight macro-level state variables—such as production cost, market price, and animal health—together with five micro-level state variables—including land resources and financial capacity—were defined as model inputs. A decision-making model was constructed based on the Deep Q-Network (DQN), incorporating a composite reward function that accounts for baseline state rewards, seasonal adjustments, stability penalties, risk modifiers, and long-term planning incentives. Model training employed ε -greedy exploration and experience replay mechanisms. Experimental results demonstrate that after 40 training episodes, the model achieved an average reward of 203.85, while the average loss decreased to 0.557, indicating stable convergence of the learned strategy. In a farmer-level case analysis, the dominant decision action accounted for 71.5% of the outcomes. Sensitivity analysis, combined with local interpretability using LIME and global feature importance via SHAP, revealed that risk preference, livestock health, financial status, and production cost were the most critical determinants of decision outcomes, while seasonal variation and labor availability also played significant roles. These findings confirm that DRL can provide precise decision support for farmers, thereby enhancing resource allocation, management priorities, and overall production resilience. The study ultimately offers actionable insights to promote the digital and intelligent transformation of the beef cattle industry.

Frontiers in Sustainable Food SystemsVol. 9
Inner Mongolia Agricultural University (CN), Ningbo University of Technology (CN)
Natural Science Foundation of Inner Mongolia, Inner Mongolia Agricultural University, National Office for Philosophy and Social Sciences
Openalex Percentile: Top 56%
Animal Behavior and Welfare Studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.