基于ARIMAX-GA-XGBoost组合模型的景区客流量分析——以象鼻山景区为例
Analysis of Tourist Flow in Scenic Areas Based on the ARIMAX-GA-XGBoost Combined Model—A Case Study of Elephant Trunk Hill Scenic Area

旅游作为绿色经济推动了地区经济社会的发展。本文以象鼻山景区为例，利用百度指数分析游客对该景点的网络关注度，并针对单一模型对景区日客流量预测精度不足的问题展开研究。提出将ARIMAX模型与GA-XGBoost模型采用残差法进行组合，将数理统计模型和机器学习模型组合，实现优势互补，提高预测精度；首先使用ARIMAX对数据进行预测分析，称预测结果为$y_1$ ，再把ARIMAX模型的残差放入XBGoost模型进行学习，基于GA算法对XGBoost的超参数进行优化，解决了ARIMAX模型难以对非线性数据预测的问题，GA-XGBoost的预测结果为$y_2$ ，组合模型的最终预测结果$y=y_1+y_2$ 。最后，根据预测误差评价指标对多个模型进行对比。研究结果表明，ARIMAX-GA-XGBoost组合模型预测精度更高，适应性及泛化能力更强，可为旅游相关管理部门的科学决策提供必要的参考，具有很高的经济效益与实际意义。
Tourism, as a green economy, drives regional socioeconomic development. Taking Xiangbi Mountain Scenic Area as a case study, this paper analyzes the network attention of tourists towards this attraction using Baidu Index. To address the issue of insufficient prediction accuracy of single models for daily tourist flow forecasting in scenic areas, a hybrid modeling approach is proposed. By integrating the ARIMAX model with the GA-XGBoost model through residual combination methodology, this study combines mathematical-statistical modeling with machine learning techniques to achieve complementary advantages and enhance prediction accuracy. Specifically, the ARIMAX model is initially employed for data prediction (denoted as $y_1$ ), followed by feeding its residuals into the XGBoost model for learning. The genetic algorithm (GA) optimizes XGBoost’s hyperparameters, effectively resolving ARIMAX’s limitations in handling nonlinear data prediction (GA-XGBoost prediction denoted as