Parking zones in a planar automated garage were divided based on vehicle dwell-time characteristics to improve retrieval efficiency and reduce customer waiting time. Actual vehicle dwell-time distribution, its probability density, and the total energy consumption of Rail Guided Vehicles (RGVs) and lifts during storage and retrieval operations were analyzed to partition the parking spaces. A Long Short-Term Memory (LSTM) network used the day of the week, vehicle arrival time, and weather conditions as inputs to predict vehicle dwell time, and each vehicle was then assigned to the corresponding zone according to the predicted dwell time. Within each zone, the final parking position was determined to minimize the total operation time of RGVs and lifts. Using an automated parking garage in Xi’an as a case study, results show that the proposed allocation strategy reduces the average total energy consumption by 24.65% and 7.90%, and shortens the average service time by 23.37% and 12.02%, compared with random and nearest-space allocation strategies, respectively. The strategy effectively enhances storage and retrieval efficiency and reduces customer waiting time.