面向精细化管理的停车需求短时预测

停车诱导系统（PGS）是缓解交通拥堵的有效办法，但停车需求短时精准预测作为空余车位发布的关键技术并没有得到有效解决。利用停车需求时变特征曲线的线型稳定性，以及在周内各工作日间的振幅的显著差异性对数据进行分组，采用不仅具备记忆时间序列数据能力，同时有着更简洁的逻辑门控制结构的GRU（gated recurrent unit）模型对停车需求进行短时精准预测，发现相比于传统神经网络以及ARIMA模型，在考虑停车需求周内日间差异性并对数据进行分组后的GRU模型能提供更高的预测精度。

Short-term Prediction of Parking Demand for Parking Delicacy Management

Parking guidance system (GPS) is an effective way to alleviate traffic congestion, but as a key technology for releasing vacant parking spaces, the short-time accurate prediction of parking demand has not been effectively solved. Parking demand data were grouped based on the linear stability of the time-varying characteristic curves and the significant variability of the amplitudes among the working days. GRU (gated recurrent unit) Model was introduced to the accurate short-term prediction of parking demand. The model with a simpler logic gate control structure could memorize the time series data. Study results show that compared with the traditional neural network and ARIMA (autoregressive integrated moving average) Model, the proposed GRU model offers a satisfactory prediction accuracy.