Intelligent optimization and resource allocation system for railway dispatching empowered by the Internet of Things

To address the deficiencies of existing railway scheduling schemes—namely, insufficient intelligence, ineffective utilization of multi-source heterogeneous IoT sensor data, and weak real-time dynamic resource allocation capability—this paper reviews the literature and identifies that conventional algorithms suffer from low computational efficiency, poor generalization, and a lack of IoT-oriented multi-agent fusion mechanisms. Accordingly, an improved weighted multi-source IoT adaptive data preprocessing pipeline integrating entropy weight and coefficient of variation combined weighting is constructed based on a holistic IoT perception architecture. A DDQN-based deep reinforcement learning collaborative scheduling optimization framework is further designed. A multi-objective mathematical model is established with dual minimization objectives total system operational cost and train arrival/departure delay penalties. The railway scheduling problem is formulated as a Markov decision process, with complete definitions of the observation state, discrete scheduling actions, and a multi-dimensional reward function incorporating nonlinear penalty terms. After preprocessing raw IoT data through standardization, noise removal, missing-value interpolation, weighted fusion, and PCA dimensionality reduction, the processed data are fed into the network. Comparative experiments under multiple disturbance conditions are conducted on the OpenAI Gym simulation platform. Simulation results demonstrate that the proposed DDQN algorithm exhibits stronger sample experience reuse capability and achieves stable convergence to the optimal scheduling policy after multiple iterations, effectively mitigating train delays. Four mainstream models—rule-based scheduling, genetic algorithm, vanilla DQN, and improved particle swarm optimization—are selected as baselines. The proposed algorithm achieves a single-solution time within 1 s across various delay scenarios, with a global maximum solution error of only 0.1%, representing an approximately 3.3-fold improvement in solution accuracy over the 3.4% error of traditional rule-based scheduling. Comprehensive experimental results indicate that the DDQN optimization framework based on multi-source IoT fused data enables integrated collaborative configuration of dynamic railway scheduling and station resources. Equipped with a built-in safety constraint validation mechanism, the framework balances model interpretability and operational safety, offering significant theoretical research value and promising engineering application prospects.