With the developing world, cities have begun to become smarter. Smart parking systems, with the ever-increasing number of vehicles, are among the important matters in smart cities. The reason for this is that the search for parking spaces that are already insufficient, brings along a serious cost, air pollution and stress issues. In this study, a new approach that attempts to forecast the parking lot occupancy rate in the short- and medium-term with its deep learning-based Gated Recurrent Units (GRU) model was proposed. Initially, data belonging to 607 carparks located in the city of Istanbul in Turkey, and weather data have been collected, and a multivariate time series data set has been created. In the second stage, to forecast the parking places that would be available in the short- and medium-term, the GRU model was used in the system proposed. To show the effectiveness of the model, the results obtained through the 27 different models were compared by means of the Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM), which were some other sequence models. According to the experimental results made on the weather data obtained from İSPARK dataset and AKOM, the our proposed GRU model achieves 99.11% accuracy gave the best results with 0.90 MAE, 2.35 MSE and 1.53 RMSE metric values. Experimental results obtained with various hyperparameters clearly demonstrate the success of the GRU deep learning model in prediction parking occupancy rates.

随着世界的发展，城市已经开始变得更加智能。随着车辆数量的不断增加，智能停车系统是智慧城市的重要事项之一。其原因是寻找已经不足的停车位，带来了严重的成本、空气污染和压力问题。在这项研究中，提出了一种新方法，该方法试图利用基于深度学习的门控循环单元 (GRU) 模型来预测短期和中期停车场的占用率。最初，已经收集了属于土耳其伊斯坦布尔市 607 个停车场的数据和天气数据，并创建了一个多元时间序列数据集。在第二阶段，为了预测中短期内可用的停车位，所提出的系统中使用了 GRU 模型。为了展示模型的有效性，通过循环神经网络 (RNN) 和长短期记忆 (LSTM) 等其他一些序列模型，比较了通过 27 个不同模型获得的结果。根据对从 İSPARK 数据集和 AKOM 获得的天气数据所做的实验结果，我们提出的 GRU 模型实现了 99.11% 的准确度，并以 0.90 MAE、2.35 MSE 和 1.53 RMSE 度量值给出了最佳结果。使用各种超参数获得的实​​验结果清楚地证明了 GRU 深度学习模型在预测停车占用率方面的成功。根据对从 İSPARK 数据集和 AKOM 获得的天气数据所做的实验结果，我们提出的 GRU 模型实现了 99.11% 的准确度，并以 0.90 MAE、2.35 MSE 和 1.53 RMSE 度量值给出了最佳结果。使用各种超参数获得的实​​验结果清楚地证明了 GRU 深度学习模型在预测停车占用率方面的成功。根据对从 İSPARK 数据集和 AKOM 获得的天气数据所做的实验结果，我们提出的 GRU 模型实现了 99.11% 的准确度，并以 0.90 MAE、2.35 MSE 和 1.53 RMSE 度量值给出了最佳结果。使用各种超参数获得的实​​验结果清楚地证明了 GRU 深度学习模型在预测停车占用率方面的成功。