PESWS-LGBM: A Hybrid Swarm Optimization and LightGBM Framework for Real-Time Production Scheduling in Smart Factories

Abstract

The smart factory sector has been a frontrunner in adopting machine learning (ML) technologies to enhance production systems and enable real-time decision-making. However, challenges remain in translating raw sensor data into actionable scheduling strategies for fully autonomous operations. To address this issue, this research proposes a novel hybrid framework, the production-based Elephant Swarm Water Search-Driven Light Gradient Boosting Machine (PESWS-LGBM), which integrates metaheuristic optimization with predictive modeling for real-time production scheduling. The model not only identifies key process parameters (e.g., vibration, power usage, cycle time) but also dynamically generates optimized scheduling decisions, including task sequencing, job-to-machine assignments, and time allocations. These decisions are guided by a dual-stage process in which the LGBM component predicts scheduling performance, and the PESWS component refines scheduling configurations to minimize makespan, reduce job tardiness, and maximize machine utilization. The dataset based on real-time industrial sensor data was preprocessed using noise filtering, missing value imputation, and feature scaling. Experimental results show that PESWS-LGBM significantly improves scheduling outcomes, lowering downtime and material loss while increasing Overall Equipment Effectiveness (OEE). The proposed model achieved strong performance metrics, including accuracy (0.96), Precision (0.91), recall (0.98), and F1-score (0.94). These findings validate the effectiveness of hybrid intelligent systems in enabling adaptive scheduling and improving operational efficiency in smart manufacturing environments.