To reduce traffic congestion, air pollution, and lost revenue due to wasted driving around looking for parking spaces, effective parking management has become a crucial issue in urban areas. Previous studies have shown that parking demand is related to the location of parking spaces/lots, parking fees, time spent looking for spaces, and various driver and environmental characteristics. However, there have been more studies of on-street rather than off-street parking. Also, the spatio-temporal variance between variables and turnover rate (such as the hours of operation for various types of Points of Interest (POIs) as well as the causes of bias estimators has been largely overlooked.

To solve the above problems, the Geographical and Temporal Weighted Regression (GTWR) model was utilized in this study to predict the hourly parking turnover rate in off-street parking lots. The flexible coefficient setting of this model can be used to capture the spatial and temporal variance. In addition to this model, the Geographically Weighted Regression (GWR) and the Ordinary Least Squares (OLS) models were used as a basis for comparison. The study included 346 public off-street parking lots in Taipei and 10 types of nearby POIs. The study period was for 24 h on August 31, 2021. Due to the dissimilar trend of the turnover rate, the researchers separated all parking lots into three groups for regression analysis: those that experience a daytime decrease, remain stable, and those with an increase during the day.

Our results show the following: (1) For the types of parking lots that experienced increases and decreases, the GTWR model performed best for the prediction of hourly turnover rate, while all models performed similarly for the prediction of the stable type parking lot. (2) With regard to related factors (the type 1 parking lot for example), parking lots in areas with more traditional markets, MRT stations, and with fewer schools, supermarkets, hotel bus stops, and lower parking fees tended to have higher turnover rates. (3) Incorporating a more precise POI dataset (the number and types of POIs as well as their business hours) and applying a suitable spatial temporal model improved the parking demand model’s performance. However, the GTWR model’s limitations may hamper its performance, especially with regard to analysis of the time-stable parking lots or variables. Future studies might include other datasets, such as nearby private parking lots or holiday data to improve the model fit.

为了减少交通拥堵、空气污染以及因浪费时间寻找停车位而造成的收入损失，有效的停车管理已成为城市地区的一个关键问题。先前的研究表明，停车需求与停车位/车位的位置、停车费用、寻找车位所花费的时间以及各种驾驶员和环境特征有关。然而，对路内停车的研究多于街外停车。此外，变量和周转率之间的时空差异（例如各类兴趣点（POI）的运营时间以及偏差估计器的原因在很大程度上被忽视了。

为了解决上述问题，本研究采用地理和时间加权回归（GTWR）模型来预测路边停车场每小时的停车周转率。该模型灵活的系数设置可用于捕获空间和时间方差。除了该模型之外，还使用地理加权回归（GWR）和普通最小二乘法（OLS）模型作为比较的基础。该研究包括台北市 346 个公共路边停车场以及附近 10 类 POI。研究时间为2021年8月31日24小时。由于周转率趋势不同，研究人员将所有停车场分为三组进行回归分析：白天减少的停车场、保持稳定的停车场以及白天周转率下降的停车场。白天增加。

我们的结果表明：（1）对于经历增加和减少的停车场类型，GTWR模型对于每小时周转率的预测表现最好，而所有模型对于稳定型停车场的预测表现相似。（2）从相关因素（以一类停车场为例）来看，传统市场、地铁站较多、学校、超市、酒店公交车站较少、停车费较低的地区的停车场营业额往往较高费率。（3）纳入更精确的POI数据集（POI的数量、类型及其营业时间）并应用合适的时空模型提高了停车需求模型的性能。然而，GTWR 模型的局限性可能会影响其性能，特别是在分析时间稳定的停车场或变量方面。未来的研究可能包括其他数据集，例如附近的私人停车场或假期数据，以提高模型的拟合度。