A variable weight combination prediction model for climate in a greenhouse based on BiGRU-Attention and LightGBM

• DOI: 10.1016/j.compag.2024.108818
• Publication Date: 2024-03-16T07:12:32Z
• Authors: Xiaojuan Mao; Ni Ren; Peiyu Dai; Jing Jin; Baojia Wang; Rui Kang; Decui Li

Accurate prediction of environmental changes in a greenhouse is crucial for precise control and promoting crop growth. However, the microclimate environment nonlinear, temporal, multivariate, strongly coupled, making it difficult to establish robust fitting model. To address these issues, this study proposed variable weight combination model based on attention mechanism optimized Bi-directional Gated Recurrent Unit (BiGRU-Attention) Light Gradient Boosting Machine (LightGBM). The Particle Swarm Optimization Algorithm (PSO) was employed optimize coefficients predicted values from BiGRU-Attention LightGBM models at different times. This optimization aimed enhance accuracy predicting air temperature, humidity, Photosynthetically Active Radiation (PAR) greenhouse. In predictions spanning time steps 30 120 min, demonstrated superior performance compared single equal BiGRU-Attention-LightGBM. At step coefficient determination R2 temperature 0.9586, humidity 0.9232, PAR 0.8066. indicated that (PSO-BiGRU-Attention-LightGBM) could more accurately predict future dynamic trends climatic factors