Article: Environmental variable regulation and optimisation strategy in forest seedling cultivation based on reinforcement learning Journal: International Journal of Environment and Sustainable Development (IJESD) 2026 Vol.25 No.6 pp.63 - 86 Abstract: Addressing reinforcement learning's failure to model high-dimensional dynamic interactions in seedling cultivation, this paper proposes an intelligent regulation strategy integrating species embedding and proximal policy optimisation (PPO) for precise, dynamic environmental management of diverse seedlings. The study constructs a 12-dimensional state space encompassing initial biological traits, real-time environmental parameters, and species encoding, designs a 243-dimensional discrete action space encompassing temperature, humidity, light, water, and CO<SUB align="right"><SMALL>2</SMALL></SUB>, and introduces a multi-objective reward function to co-optimise growth rate, survival rate, resource efficiency, and stress avoidance. The strategy improves the average daily growth rate, with a final survival rate of 95.2%, an average reduction in electricity consumption per unit biomass of approximately 0.4 kWh/g, and an average reduction in stress events of 19.86. Furthermore, a transfer learning mechanism enables fine-tuning for new tree species in just seven days. This study provides a generalisable, efficient, and personalised regulation paradigm for intelligent seedling cultivation. Inderscience Publishers - linking academia, business and industry through research

Title: Environmental variable regulation and optimisation strategy in forest seedling cultivation based on reinforcement learning

Authors: Xianghong Liu

Addresses: Qingyang Forestry and Grass Seedling Station, Qingyang Forestry and Grassland Bureau, Qingyang, 745000, Gansu China

Abstract: Addressing reinforcement learning's failure to model high-dimensional dynamic interactions in seedling cultivation, this paper proposes an intelligent regulation strategy integrating species embedding and proximal policy optimisation (PPO) for precise, dynamic environmental management of diverse seedlings. The study constructs a 12-dimensional state space encompassing initial biological traits, real-time environmental parameters, and species encoding, designs a 243-dimensional discrete action space encompassing temperature, humidity, light, water, and CO₂, and introduces a multi-objective reward function to co-optimise growth rate, survival rate, resource efficiency, and stress avoidance. The strategy improves the average daily growth rate, with a final survival rate of 95.2%, an average reduction in electricity consumption per unit biomass of approximately 0.4 kWh/g, and an average reduction in stress events of 19.86. Furthermore, a transfer learning mechanism enables fine-tuning for new tree species in just seven days. This study provides a generalisable, efficient, and personalised regulation paradigm for intelligent seedling cultivation.

Keywords: reinforcement learning; forest tree seedling cultivation; environmental variable regulation; environmental optimisation strategies; proximal strategy optimisation.

DOI: 10.1504/IJESD.2026.154269

International Journal of Environment and Sustainable Development, 2026 Vol.25 No.6, pp.63 - 86

Received: 29 Oct 2025
Accepted: 11 Mar 2026
Published online: 18 Jun 2026 *

Title: Environmental variable regulation and optimisation strategy in forest seedling cultivation based on reinforcement learning

Keep up-to-date