**1. Introduction**

A prominent sharing economy business model, the bike-sharing systems, has emerged in recent years as a popular way of public transportation [1]. For society, a bike-sharing system meets the theme of sustainable development because of convenience, lower prices, and environmental protection [2,3]. Consequently, many bike-sharing systems are being established to satisfy the need. One example of a bike-sharing system is Citi Bikes, with more than 85,000 active users [4].

Distributing a suitable bicycle network structure can not only connect the system of urban traffic and commuting but reduce the greenhouse effect. However, constructing unwanted stations in a bike-sharing system will cause environmental damage and resource waste. The framework presented in the paper aims to assist the governmen<sup>t</sup> and planners in predicting bike demands at a macroscopic level in advance, i.e., evaluating and verifying whether new stations meet the needs of the public.

Research on bike-sharing systems has been widely studied in recent years. Some works [5–8] depend completely on station-based historical records and features, and their target is to make predictions for already established stations. The works of [9,10] aim to predict the demand in hours or only during rush hour. The work of [11] defines functional zones [12,13] and then predicts that the demand for bike expansion is the most relevant one to our work. Unfortunately, their mobility trip data in the expanded system is inapplicable for our long-term scenario, as it is also regarded as future data in the prediction stage. Different from previous works, we commit to long-term demand prediction, which is faced

**Citation:** Hsieh, H.-P.; Lin, F.; Jiang, J.; Kuo, T.-Y.; Chang, Y.-E. Inferring Long-Term Demand of Newly Established Stations for Expansion Areas in Bike Sharing System. *Appl. Sci.* **2021**, *11*, 6748. https://doi.org/ 10.3390/app11156748

Academic Editors: Agostino Marcello Mangini and Michele Roccotelli

Received: 12 June 2021 Accepted: 19 July 2021 Published: 22 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

with two challenges. First, mobility and meteorology data used in previous works are unavailable in expansion areas, for example, taxi usages, temperature, wind speed, etc. Moreover, we cannot directly apply the methodology of existing works, which focus on short-term demand prediction for a single station, since they usually have enough training data but lack future events [5]. Second, the real-world bike stations are mainly built-in batches for expansion areas. However, the different geographical characteristics between regions make the prediction task hard.

To tackle these challenges, we propose a robust framework called LDA (Long-Term Demand Advisor) to predict long-term (e.g., six months) demand in newly established bike regions. Apart from the short-term prediction, which is highly affected by emergencies and other temporal factors [6,7], the proposed long-term prediction can not only reduce inaccuracies resulted from unpredictable social events or traffic accidents but also advise decision-makers on where to build new stations. This framework aims to provide governments with a preliminary estimation of the amount of bike usages in the following periods (e.g., half-year) in the new regions of a city, given merely the locations of the bike stations. Our contributions are as follows:

