D-QLSA: A Deep Q-Learning and Simulated Annealing Fusion Algorithm for IoT-based Laboratory Equipment Scheduling

Abstract

To address the dynamic and multi-objective optimization challenges in university laboratory equipment resource scheduling, this paper constructs a Deep Q-Learning-Simulated Annealing (D-QLSA) fusion algorithm based on IoT perception data. First, data is collected from equipment status sensors (such as vibration and temperature sensors) and usage demand sensors, then preprocessed using wavelet threshold denoising and Min-Max normalization to ensure data validity. Second, based on collaborative scheduling theory, two core optimization objectives are established: maximizing resource utilization and minimizing user waiting time. Finally, by integrating the dynamic decision-making capabilities of DQN with the global optimization capabilities of SA, three functional modules (perception data input, decision-making, and optimization) are designed to form a closed-loop scheduling system. In an experiment involving 80 devices (covering 12 categories including lathes, oscilloscopes, and servers) and 120 days of real-world usage data, the D-QLSA algorithm achieved a resource utilization rate of 89.2%—16.7% higher than the traditional Genetic Algorithm (GA) and 8.9% higher than the standard DQN algorithm. The average user waiting time was 2.3 minutes, which is 3.5 minutes shorter than GA and 1.4 minutes shorter than DQN. Additionally, the scheduling success rate reached 96.8%, 15.5% higher than GA and 6.3% higher than DQN. Among all tested algorithms, D-QLSA also had the lowest standard deviation for all key indicators, demonstrating its advantages in scheduling efficiency, stability, and adaptability to dynamic laboratory environments. This research provides technical support for the development of intelligent management systems for university laboratories, helping to reduce equipment idle waste and improve user experience.