預測共享自行車系統之站點空滿程度

Predicting Occupancy Rates of Stations in Bike-Sharing Systems

陳妍妡(Yen-Hsin Chen)

指導教授 ： 李瑞庭

國立臺灣大學/管理學院/資訊管理學研究所/碩士(2022年)

https://doi.org/10.6342/NTU202201883

摘要

在共享自行車系統中，使用者可以在鄰近的車站借用自行車，並在目的地附近的車站歸還。使用者可能會面臨沒有自行車可以借用或沒有空位可以停靠的情況，從而導致客戶滿意度降低。如何有效管理自行車的配置成為一個重要的議題。過往相關研究旨在預測每個站點的進、出流量需求，但預測每個站點的佔有率對於自行車管理員制定自行車分配計劃上會更為直觀。就我們所知，目前並未有研究根據共享自行車系統站點的佔有率進行預測。因此，我們提出了一個深度學習架構，捕捉複雜的空間和時間相關性，以預測每個站點的佔有率。我們提出的研究框架包含兩個階段，首先，我們利用圖卷積網路學習有向加權圖中站點間的空間相關性，以導出每個站點的特徵向量。然後，運用所學習的特徵向量，我們提出一個站點佔有率預測模型，利用門控循環單元學習站點間的時間相關性，進而預測每個站點的佔有率。實驗結果顯示，我們所提出的方法在平均絕對誤差、均方根誤差和正確率方面均優於現有的方法。我們的研究框架可以幫助自行車管理者事先分配和平衡自行車，亦可幫助使用者更方便租借與歸還自行車，進而提高客戶滿意度。

關鍵字

共享自行車系統 ； 站點佔有率預測 ； 圖卷積網路 ； 門控循環單元 ； 注意力機制

並列摘要

A bike-sharing system allows customers to check out bikes at adjacent stations and return them to stations close to their final destinations. However, they could be confronted with a situation where there was no bike at nearby stations to check out or no docks to return, resulting in lower customer satisfaction. How to manage bikes pre-allocating becomes a key issue in bike-sharing systems. Most previous studies predicting pick-up and drop-off demands for each station; however, predicting the occupancy rate of each station is more intuitive for bike administrators to develop bike allocation plans. To the best of our knowledge, there is no study dedicated to forecasting the occupancy rates in a bike-sharing system. Therefore, we propose a deep learning framework for capturing the complicated spatial and temporal correlations to predict the occupancy rate of each station. There are two phases in the proposed framework. First, we derive the feature vector of each station by applying the graph convolutional networks to learn spatial correlations among stations in the directed weighted graph. Next, based on the derived features from previous time intervals, we propose an Occupancy Rate Prediction model (ORP) by using the gated recurrent units to capture the temporal correlations among stations for predicting the occupancy rate of each station. The experiment results show that the proposed framework outperforms the state-of-art methods in terms of mean absolute error, root mean squared error, and accuracy. Our framework can support bike administrators to effectively pre-allocating and balancing bikes in each station, and help users easier to rent and return a bike, which results in higher customer satisfaction.

並列關鍵字

bike-sharing system ； occupancy rate prediction ； graph convolutional network ； gated recurrent unit ； attention mechanism

參考文獻

Chai D, Wang L, Yang Q (2018) Bike flow prediction with multi-graph convolutional networks. Proceedings of the ACM on International Conference on Advances in Geographic Information Systems. 397–400.

Chen L, Zhang D, Wang L, Yang D, Ma X, Li S, Wu Z, Pan G, Nguyen TMT, Jakubowicz J (2016) Dynamic cluster-based over-demand prediction in bike sharing systems. Proceedings of the ACM International Joint Conference on Pervasive and Ubiquitous Computing. 841–852.

Cho K, van Merrienboer B, Gulcehre C, Bahdanau D, Bougares F, Schwenk H, Bengio Y (2014) Learning phrase representations using RNN encoder–decoder for statistical machine translation. Proceedings of the International Conference on Empirical Methods in Natural Language Processing. 1724–1734.

Feng S, Chen H, Du C, Li J, Jing N (2018) A hierarchical demand prediction method with station clustering for bike sharing system. Proceedings of the IEEE International Conference on Data Science in Cyberspace. 829–836.

Guo R, Jiang Z, Huang J, Tao J, Wang C, Li J, Chen L (2019) BikeNet: Accurate bike demand prediction using graph neural networks for station rebalancing. Proceedings of IEEE Symposia and Workshops on Ubiquitous, Autonomic and Trusted Computing. 686–693.