【摘要】 为提高地铁进站客流量预测精度,提出一种基于SARIMA模型和GA-BP神经网络的客流量组合预测模型,通过已有客流相关数据构建预测客流量的SARIMA模型和GA-BP神经网络模型作为组合模型的子模型,再利用拟合SARIMA模型的最大季节回归多项式个数确定组合模型因变量个数,之后结合季节周期和子模型的预测值确定组合模型的因变量,并基于子模型预测值的平均绝对百分比误差(MAPE)确定组合模型的因变量权重,最后进行实例验证。结果表明:当预测个数为5个时,组合预测模型的MAPE为3.11%,介于子模型之间但优于传统的线性组合模型;当预测个数为10个时其MAPE为5.13%,优于所有对比模型。更多还原

【Abstract】 To improve the prediction accuracy of subway passenger flow, this paper proposed a combined prediction model of subway passenger flow based on the seasonal auto-regressive integrated moving average(SARIMA) model and genetic algorithm and back-propagation(GA-BP) neural network. Firstly, the SARIMA model and GA-BP neural network model for predicting passenger flow are built through the related data of existing passenger flow as the submodels of the combined model. The number of dependent variables was determined by the maximum number of seasonal regression polynomials fitting the SARIMA model. Then, the dependent variables of the combined model were determined by the seasonal cycle and the predicted values of the submodels. The weight of dependent variables was determined based on the mean absolute percentage error(MAPE) of the predicted value of the submodels. Finally, an example was given for verification. The results show that MAPE is 3.11% for five predictions, which ranks among the submodels and is better than the traditional linear combination model. MAPE is 5.13% when it comes to ten predictions, which is better than all comparison models.更多还原