摘要
单一模型在铁路客流量预测中很难同时捕获到数据序列的线性、非线性和周期性等多种特征,从而无法取得很好的预测效果。针对该问题提出基于机器学习的ARMA-LSTM组合模型预测方法。对原始数据进行分析和特征提取;训练LSTM(Long Shot-Term Memory)模型和ARMA(Autoregressive moving average model)模型,分别得到两模型预测结果;通过BP神经网络算法对两模型进行权重优化,得到预测结果。将ARMA-LSTM组合模型与ARMA、LSTM、灰色模型、GM-LSTM组合模型进行对比,预测效果明显优于其他单一模型,预测结果误差(MAPE)降至3.10%~10.73%,验证了ARMA-LSTM组合模型在铁路客流量预测中有更高的准确性和更好的适用性。
It is difficult for a single model to capture the linear,nonlinear and cyclical characteristics of the data series in the prediction of railway passenger flow,so that it can not achieve good prediction results.Aiming at this problem,an ARMA-LSTM combined forecasting method based on machine learning is proposed.The original data was analyzed and the feature was extracted;we trained LSTM model and ARMA model to get the prediction results,and then optimized the weight of the two models by BP neural network;we got the prediction results.Compared with ARMA,LSTM,grey model and GM-LSTM combined model,the prediction effect of ARMA-LSTM combined model is better than that of other single models,and the prediction error(MAPE)decreases to 3.10%~10.73%.It is proven that the combined model of ARMA-LSTM has higher accuracy and better applicability in railway passenger flow prediction.
作者
孙越
宋晓宇
金莉婷
刘童
Sun Yue;Song Xiaoyu;Jin Liting;Liu Tong(College of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,Gansu,China)
出处
《计算机应用与软件》
北大核心
2021年第12期262-267,273,共7页
Computer Applications and Software
基金
国家自然科学基金项目(61262044)。