Swarm-Optimized BiLSTM-Attention Model for Adaptive Irrigation Scheduling in Precision Agriculture

Efficient irrigation scheduling is essential to optimize crop yield while conserving water in precision agriculture. Traditional static or rule-based methods often fail to capture the dynamic, nonlinear nature of agro-environmental conditions. To overcome these limitations, we propose a hybrid PSO-BiLSTM Attention model that combines Particle Swarm Optimization for hyperparameter tuning with Bidirectional Long Short-Term Memory networks for bidirectional temporal sequence learning and an attention mechanism for feature importance weighting. This approach improves the modeling of irrigation demand based on variables such as crop stage, soil moisture, temperature, and humidity. Compared to baseline models such as LSTM, BiLSTM, PSO-LSTM, and SSA-LSTM, the proposed model demonstrates improved predictive accuracy, enhanced precision, and significantly greater training efficiency per epoch. These results underscore its effectiveness in delivering accurate, efficient, and robust predictions, making it well-suited for resource-aware irrigation control in smart agriculture environments.