| 基于残差融合和贝叶斯算法融合的ARIMA-LSTM混合模型及其在机场人流预测中应用 |
| 点此下载全文 |
| 引用本文：何剑,谭文安.基于残差融合和贝叶斯算法融合的ARIMA-LSTM混合模型及其在机场人流预测中应用[J].上海第二工业大学(中文版),2026,43(2):210-217 |
| 摘要点击次数: 297 |
| 全文下载次数: 10 |
|
|
| 中文摘要:针对机场卫生间人流量具有高度复杂的非线性特征, 而传统单一模型预测精度有限的问题, 本文提出一种改进的ARIMA-LSTM 混合模型。该模型采用残差融合策略: 自回归积分滑动平均模型(autoregressive integrated moving average, ARIMA) 提取线性趋势与季节性成分, 长短期记忆网络(long short-term memory, LSTM) 基于预测残差学习复杂的非线性波动模式。进一步引入贝叶斯优化算法, 采用高斯过程代理模型与期望改进采集函数, 通过50轮迭代自动化优化LSTM 的超参数。基于上海浦东机场12 个月小时级客流数据的实验表明, 该模型在平均绝对误差(mean absolute error, MAE)、均方根误差(root mean square error, RMSE)、决定系数(coefficient of determination,R2) 等指标上均优于单一的ARIMA、LSTM 模型以及其他ARIMA-LSTM 混合模型, 其验证集RMSE 降低18.75%,预测精度提升9.14%, 验证了残差融合与贝叶斯优化在提升复杂时间序列预测精度方面的有效性。 |
| 中文关键词:ARIMA-LSTM 混合模型 残差融合 贝叶斯优化 机场卫生间 人流预测 时间序列 |
| |
| An ARIMA-LSTM Hybrid Model Based on Residual Fusion and Bayesian Algorithm and Its Application in Predicting Pedestrian Flow |
|
|
| Abstract:To address the limited prediction accuracy of traditional single models for highly complex and nonlinear passenger flow in airport restrooms, this study proposes an improved ARIMA-LSTM hybrid model. The model employs a residual fusion strategy: autoregressive integrated moving average (ARIMA) extracts linear trends and seasonal components, while long short-term memory (LSTM) learns complex nonlinear fluctuation patterns based on prediction residuals. Bayesian optimization is introduced to automate the tuning of LSTM hyperparameters via a Gaussian process surrogate model and expected improvement acquisition function over 50 iterations. Experiments based on 12 months of hourly passenger flow data from Shanghai Pudong International Airport demonstrate that the model outperforms standalone ARIMA, LSTM models and other ARIMA-LSTM hybrid models in terms of mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2). The validation set RMSE is reduced by 18.75%, and prediction accuracy improves by 9.14%, validating the effectiveness of residual fusion and Bayesian optimization in enhancing complex
time series forecasting accuracy. |
| keywords:ARIMA-LSTM hybrid model residual fusion Bayesian optimization airport restroom pedestrian flow prediction time series |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|