Nonlinear dynamic parking space allocation strategy for stereo garage parking systems based on fuzzy theory

To improve the service efficiency of planar automated multi-storey parking systems, a nonlinear dynamic slot allocation strategy is proposed, considering both vehicle dwell time and the balance of slot distribution. A long short-term memory (LSTM) neural network, with hyperparameters tuned by Bayesian optimization, is employed to predict the dwell times of vehicles parked in the system. Combining predicted dwell times with slot occupancy attributes, a fuzzy-theory-based partitioning rule is established, enabling nonlinear regional allocation based on dwell time and slot occupancy density. Additionally, a balance-detection algorithm and an adjacent-slot migration strategy are designed to assess the balance of slot utilization and dynamically adjust partitioned slot occupancy. Using the average running energy consumption and average service time of the rail-guided vehicle (RGV), which performs vehicle storage and retrieval operations, as system efficiency metrics, a tabu search algorithm is applied to optimize the selection of vacant slots within each partition. Results show that, compared with the nearest-slot allocation strategy commonly used in engineering practice, the proposed method reduces the average service time by 14.3%, the average energy consumption by 23.8%, and the standard deviation of slot-occupancy balance by 90.28%, demonstrating the effectiveness of the dynamic slot allocation strategy that accounts for vehicle dwell time and balanced slot distribution.