Authors

Abstract

Accurate prediction of temperature and humidity variations in greenhouses is essential for optimizing crop growth conditions. To address the limited accuracy of existing prediction models in complex nonlinear environments, this study proposes an Improved Sparrow Search Algorithm-optimized Bidirectional Long Short-Term Memory (ISSA-BILSTM) model. The proposed approach enhances population diversity through Tent chaotic mapping initialization, employs an adaptive discoverer ratio mechanism to balance global exploration and local exploitation, and integrates a Lévy flight disturbance strategy to improve the ability to escape local optima. These enhancements effectively mitigate the convergence instability issues caused by the sensitivity of BILSTM hyper parameters. Validation results show that the proposed model achieves a temperature prediction R² of 0.9777 (MAE = 0.0159) and a humidity prediction R² of 0.9762 (MAE = 0.0213), significantly outperforming the standard BILSTM (temperature R² improved by 7.42%) and SSA-LSTM (temperature R² improved by 2.88%). These findings demonstrate that the ISSA-BILSTM model can accurately predict greenhouse temperature and humidity, enhance environmental control, optimize resource utilization, and effectively reduce production costs.

Abstract in Chinese

准确预测温室温湿度变化对优化作物生长环境至关重要。针对现有预测模型在复杂非线性环境中精度不足的问题，本研究提出改进麻雀算法优化双向长短期记忆网络（ISSA-BILSTM）模型。通过Tent混沌映射初始化增强种群多样性，自适应发现者比例机制平衡全局探索与局部开发，并引入Lévy飞行扰动策略提升跳出局部最优能力，有效解决了BILSTM超参数敏感导致的收敛不稳定问题。通过验证，该模型温度预测R²达0.9777（MAE=0.0159），湿度预测R²达0.9762（MAE=0.0213），性能显著优于标准BILSTM（温度R²提升7.42%）及SSA-LSTM（温度R²提升2.88%）。通过验证，本研究提出的ISSA-BILSTM温湿度预测模型能够准确预测温室内的温湿度，加强环境控制，优化资源利用，有效降低了生产成本。