Abstract: The prediction of car operation rate is of great significance for traffic managers to formulate precise control plans, implement coordinated management strategies, and regulate the scale of car ownership in advance. This paper proposes a car operation rate prediction method based on the combination model of hunter-prey optimizer algorithm and bi-directional long short-term memory neural network (HPO-BiLSTM). The key influencing variables of the car operation rate are analyzed, and 17 feature influencing factors are extracted. Combined with the reconstructed time series after normalization, the importance of variables is evaluated based on random forest algorithm, and the optimal feature set is selected as the input of the prediction model. Then, a combined prediction model that fuses the hunterprey optimizer (HPO) and the bidirectional long short-term memory (BiLSTM) is established to solve the problem that the neural network algorithm is prone to fall into local extrema. It utilizes the exploration-exploitation mechanism of the HPO algorithm to realize the dynamic construction and precise parameter adjustment of the BiLSTM framework.The model performance is then verified by combining the data set of car operation rate in the central urban area of Beijing. The results indicate that: compared with the classic algorithms such as auto- regressive integrated moving average model, grey model, convolutional neural network model, long short-term memory model and bidirectional long short-term memory model, the HPO-BiLSTM model is more effective in predicting the car operation rate. The mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) are reduced by 23.85% to 54.38% , 20.67% to 57.40% , 27.48% to 59.32% , and the mean relative error is - 1.57% . The proposed algorithm shows high prediction accuracy and practical performance.

Key words: urban traffic, vehicle operation rate prediction, bi-directional long short-term memory, hunter-prey optimizer algorithm, deep learning

摘要： 汽车出车率预测对于交通管理者预先制定精准化管控方案、实施协调化统筹调度，以及调控汽车保有量规模具有重要意义。为此，本文提出一种基于猎人猎物优化算法与双向长短时记忆神经网络组合模型(HPO-BiLSTM)的汽车出车率预测方法。首先，分析汽车出车率的关键影响因素，提取出17个特征影响因子，结合标准化处理后的重构时间序列，基于随机森林算法进行变量的重要度评估，筛选出最优特征集合作为预测模型输入；其次，为解决神经网络算法容易陷入局部极值的难题，建立一种融合猎人猎物优化算法(HPO)与双向长短时记忆神经网络(BiLSTM)的组合预测模型，利用HPO的探索-开发机制，实现BiLSTM框架的动态化搭建与精细化调参；最后，结合北京市中心城区的汽车出车率数据集进行模型性能的测试与检验。结果表明：与自回归差分移动平均模型、灰色模型、卷积神经网络模型、长短时记忆神经网络模型以及双向长短时记忆神经网络模型等经典算法相比，HPO-BiLSTM 模型在汽车出车率预测中的平均绝对误差(MAE)、平均绝对百分比误差(MAPE)和均方根误差(RMSE)分别降低了23.85%~54.38%、20.67%~ 57.40%、27.48%~59.32%，平均相对误差为-1.57%。说明本文提出的混合深度学习算法具有较高的预测精度与实用性能。

关键词: 城市交通, 汽车出车率预测, 双向长短时记忆神经网络, 猎人猎物优化算法, 深度学习