The enormous amount of data produced by the parking facilities, if properly analyzed, can change the way those facilities operate. The collected data points form patterns that, in most cases, would allow parking operators and property owners to better understand how to maximize revenue and decrease operating expenses and support the decisions such as setting specific parking policies (e.g. electrical charging only parking spaces for sustainable and eco-friendly parking).

However, there lacks an intelligent tool to assess the layout design and operational performance of parking facilities with respect to all aforementioned concerns. This research proposes an agent-based framework and system for off-street parking simulation, in which the elements such as drivers, pedestrians, aisles, spots, entrances, are modeled as agents to capture their behaviors. The agents are classified based on their measures and behaviors into two categories: entities (e.g. drivers and pedestrians) and locations (e.g. aisles and spots). The state transition processes of drivers and pedestrians are modeled as state machines. The consequent choice behaviors of the entities are modeled to replicate the spot choice and route choice within the parking facility. The model and system have the flexibility to be extended to incorporate modern special types of parking spots and intelligent parking management and guidance systems. Simulation outputs and statistics, such as utilization, turnover, occupied duration can be collected and analyzed to assess the potential outcome of the design and operational decisions. To investigate the performance of the proposed simulator, this study designs a Software as a Service (SaaS) platform to visualize simulation dynamics and outputs. The methodology for data collection, processing, and user behavior extraction in the parking system is also developed.

Application of the developed simulation system to a real-world case study demonstrates its capability to support decisions in designing, maintaining, operating parking facilities. In the case study, a Long-Short Term Memory (LSTM) neural network is used to replicate the arrival and departure of the vehicles with good credibility. The simulation heatmap validates the frequency of parking spot utilization to be consistent with reality. The online simulator can also calculate the active capacity of the parking system, which is defined in this study to be an affective measure to evaluate alternative parking facility layout designs. Numerical studies under various application scenarios are further conducted with quantitative results to demonstrate that: (a) the smart check-in device has marginal benefits in vehicle waiting time; (b) the flexible pricing policy may increase the average daily revenue if the elasticity of the price is not involved; (c) the number of electrical charging only spots have negative impact on the performance of the parking facility; and (d) the rear-in only policy may increase the duration of the parking maneuvers and reduce the efficiency during arrival rush hour.

In summary, the proposed parking simulator outperforms the state-of-practice simulators in the following aspects: (a) facilitates the customization of the discrete choice models and machine learning models for parking and pedestrian behavior modeling; (b) better captures low-speed mixed traffic dynamics with logic-based state machines rather than just adopting traditional road traffic modeling; and (c) offers capabilities and quantitative methods to aid the design of smart parking facility with multi-use, automated, shared spots, or electrical charging stations.