Designing online learning algorithms for electric fence placement in dockless bike-sharing system utilizing limited random parking strategies

Abstract

Dockless shared bicycles, parked and frequently misplaced throughout the city, infringe upon urban public spaces and incur public costs. To address this issue, existing electric fence technology has been employed; however, compelling users to park their bicycles in static electric fences with fixed locations undermines the convenience offered by the dockless bike-sharing system's flexible parking. In this study, we develop a novel, implementable online learning algorithm that dynamically updates the electric fence siting scheme based on users' parking needs. We introduce the concept of a dynamic no-parking area, termed a ‘limited random parking’ strategy, which significantly reduces the average walking distance for users to between 9% and 40% of the original distance. The dynamic electric fence introduced in this research enhances Shannon entropy reduction by 5–19% compared to static electric fences. Moreover, we discovered that deploying merely approximately 20% of the total electric fences as dynamic ones suffices to further regulate parking without notably extending users' average walking distance, while the remaining 80% can remain static. Additionally, the marginal benefit of Shannon entropy sharply declines once the number of electric fences reaches a certain threshold. Building on these findings, we also offer guidelines for determining the optimal number of specific dynamic electric fences, suggesting that establishing approximately 500 dynamic electric fences in Nanjing represents an optimal strategy.