基于粒子群和LSTM模型的变区间短时停车需求预测方法

停车信息是智能停车诱导系统得以成功实施的关键与基础, 被广泛认为能够有效解决当前停车难问题。鉴于停车信息在解决停车问题中的重要性, 研究了基于粒子群和LSTM模型的变区间短时停车需求预测方法。为充分发挥数据在提高模型预测精度的作用, 提出了以马尔可夫生灭过程为基础概率转移模型, 将停车到达率、离开率量化车随时间变化的停车需求, 通过标定实际的停车到达率和离开率, 确定预测模型的动态预测间隔与时段; 采用LSTM网络作为基础预测模型, 并利用粒子群优化算法优化网络参数。以吉林大学南岭校区停车场为研究对象, 按工作日与非工作日分别对停车数据进行预测并与其他预测模型进行对比分析。结果表明: 提出的停车需求预测模型在工作日的预测平均绝对误差为2.53辆, 均方误差为11.89辆; 非工作日的预测平均绝对误差为2.32辆, 均方误差为10.89辆。 

A Prediction Method for Short-term Parking Demands in Variable Interval Based on Particle Swarm Optimization and LSTM Model

An intelligent parking guidance system is widely considered to solve the problem of difficult parking at present, providing management, traffic participants, and parking operators with immediate and future parking information. A prediction method for short-term parking demands in variable interval based on particle swarm optimization and LSTM model is studied due to the importance of parking information. Based on the birth and death of the Markov process, the characteristics of the temporal parking demand are analyzed. It is formulated as a combination of the arrival rate and departure rate of parking calibrated by the temporal parking quantity. Dynamic prediction intervals are determined according to the calibrated arrival rate and departure rate. The improved LSTM network is used as the basic prediction model, and the network parameters are optimized by the particle swarm optimization algorithm. The parking lot in the Nanling campus of Jilin University is selected as a research object, and its parking data are predicted and compared with other prediction models. The results show that the MAE and MSE of the proposed parking demand prediction model are 2.53 vehicles and 11.89 vehicles in working days, respectively. For non-business days, the MAE is2.32 vehicles and the MSE is 10.89 vehicles. Therefore, a predictable prediction model of parking demands proposed in the work can predict the real-time and future parking demands, providing a reliable reference for management, traffic participants, and parking operators.