基于改进麻雀搜索算法优化LSTM网络的城市轨道交通短时客流预测

Optimization of Short-Term Passenger Flow Forecast Based on Improved Sparrow Search Algorithm in LSTM Network

摘要: 准确预测城市轨道交通短时客流，对于制定客流组织方案、合理分配运输资源等具有重要意义。针对短时客流预测精度问题，提出一种基于改进麻雀搜索算法(ISSA)优化长短时记忆神经网络(LSTM)的短时客流预测方法。首先，在麻雀搜索算法的基础上，引入混沌映射与柯西变异，提出ISSA算法，用于提升麻雀搜索算法的全局搜索与局部寻优能力，克服标准麻雀算法易陷入局部最优值的问题。其次，利用ISSA算法对LSTM神经网络的学习率、迭代次数、第一隐层节点数和第二隐层节点数这四个关键参数进行寻优。最后，以寻优结果重构LSTM神经网。利用广州某地铁站自动检票系统(AFC)采集的客流数据，对ISSA-LSTM模型的有效性进行验证。实验结果表明：该模型的均方根误差、平均绝对误差，平均绝对百分比误差和决定系数分别为4.8306、5.1435、1.5445%、99.1117%。ISSA-LSTM模型的预测效果均优于MLP模型、LSTM模型和SSA-LSTM模型。本文提出的ISSA-LSTM模型可有效降低神经网络参数对于预测精度的影响，提高预测精度，同时也对城市轨道交通运营管理服务精度进行提升。

Abstract: Accurately predicting short-term passenger flow in urban rail transit is of great significance for formulating passenger flow organization plans and reasonably allocating transportation resources. A short-term passenger flow prediction method based on improved Sparrow Search Algorithm (ISSA) optimized Long Short Term Memory Neural Network (LSTM) is proposed to address the accuracy issue of short-term passenger flow prediction. Firstly, based on the sparrow search algorithm, chaotic mapping and Cauchy mutation are introduced to propose the ISSA algorithm, which is used to improve the global search and local optimization capabilities of the sparrow search algorithm and overcome the problem of standard sparrow algorithms easily falling into local optima. Secondly, the ISSA algorithm is used to optimize the four key parameters of the LSTM neural network: learning rate, number of iterations, number of nodes in the first hidden layer, and number of nodes in the second hidden layer. Finally, the LSTM neural network is reconstructed based on the optimization results. This paper verifies the effectiveness of the ISSA-LSTM model using passenger flow data collected by the Automatic Fare Collection System (AFC) of a subway station in Guangzhou. The experimental results show that the root mean square error, average absolute error, average absolute percentage error, and determination coefficient of the model are 4.8306, 5.1435, 1.5445% and 99.1117%, respectively. The prediction performance of ISSA-LSTM model is superior to MLP model, LSTM model, and SSA-LSTM model. The ISSA-LSTM model proposed in this article can effectively reduce the impact of neural network parameters on prediction accuracy, improve prediction accuracy, and also improve the accuracy of urban rail transit operation and management services.