Adaptive Warehouse Task Allocation Using Stacked Surrogate Models and Active Learning with Simulation-Based Validation

Abstract

Modern warehouse operations face increasing order volumes, volatile demand, and shrinking lead times, challenging conventional static task dispatching policies such as FIFO or priority-based picking. This study proposes a closed-loop intelligent warehouse management framework that integrates task-level feature engineering, lightweight policy simulation, stacked surrogate modelling, and active learning retraining to enable adaptive, real-time decision support. The surrogate model combines an attention-enhanced TensorFlow multilayer perceptron, a LightGBM classifier, and a logistic regression meta-learner to predict optimal dispatching policies for dynamically generated warehouse tasks. Experiments were conducted using a publicly available logistics warehouse dataset comprising over 3,000 inventory records, from which realistic task batches were generated and evaluated across multiple dispatching strategies. Simulation-based validation was performed using a SimPy discrete-event model capturing contention between automated guided vehicles and human pickers under varying workload and resource conditions. The stacked surrogate achieved validation accuracy exceeding 98% with balanced precision and recall across all policy classes. Simulation results showed stable throughput, reduced average waiting times, and balanced resource utilisation compared with FIFO baselines. Additional robustness analyses demonstrated consistent model behaviour under demand variability and resource disturbances. The integration of an active learning loop enables continuous recalibration of the surrogate model using observed simulation KPIs, ensuring adaptability to changing operational contexts. The proposed framework provides a computationally efficient and interpretable approach for adaptive warehouse task allocation and demonstrates broader applicability to intelligent decision-making in cyber-physical systems aligned with Industry 4.0 principles.