Stochastic Optimization on Parking Lots for Smart Parking Using Reinforcement Learning Methods

Various sciences have always considered the maximum use of limited resources and planning to optimize their use. Parking in urban spaces is regarded as a finite resource. This paper investigates the idea of introducing learning algorithms for parking guidance and information systems that employ a central server. Each driver engages in a systematic search during a cycle to identify the parking space with the highest perceived reward among all available options, the optimal parking is determined based on predefined tunings signifying the pursuit of identifying the best parking places to select. To provide estimated optimal parking searching strategies to travelers there are two options linear programming and dynamic programming. Several scenarios will be argued from basic linear programming to Reinforcement Learning methods. The main contribution of this paper lies in the identification of optimal parking spaces for drivers by considering various factors associated with the parking spaces, employing the MAB approach. The multi-armed bandit (MAB) model has been widely adopted for studying many practical optimization problems (network resource allocation, ad placement, crowdsourcing, etc.) with unknown parameters, as one of the reinforcement learning implementations used in this investigation. Each driver who played this MAB gained its reward with their probabilities. The reinforcement learning methodology is utilized for which both Greedy and ε-Greedy simulations are run considering stationary and nonstationary scenarios, meanwhile, each parking selection can change the environment, and regarding this change, scenario assumptions change which is a new scenario in the area of parking problems finally results of these conditions are analyzed while aspects of each scenario are discussed and compared to their efficiency.