摘要
文章选取某地区不同时段的共享单车需求数据,采用BP神经网络算法构建了基于Tanh函数、Logistic函数、Relu函数、Identiey函数四种不同激活函数下的需求预测模型,并应用RMSE、MSE、MAE、R2四种模型评估指标对模型进行评估,从而选出最优需求预测模型进行共享单车的需求预测。结果表明:无论在检验训练数据下还是检验测试数据下,精度最高的模型为Tanh函数下的BP神经网络模型,其测试集下的标准误差(RMSE)比训练集下的标准误差(RMSE)略有上升,测试集下的拟合优度(R2)比训练集下的拟合优度(R2)略有下降,说明该模型具有强泛化性能。
This article selects the shared bicycle demand data in different regions in a certain period of time and uses BP neural network algorithm to construct the demand forecasting model based on four different activation functions of Tanh function,Logistic function,Relu function and Identiey function,and applies RMSE and MSE.The MAE and R2 model evaluation indicators to evaluate the model,so as to select the optimal demand forecasting model for the demand forecasting of shared bicycles.The results show that:both under the test training data and under the inspection test data,the model with the highest precision is the BP neural network model under the Tanh function,the standard error(RMSE)under its test set is slightly higher than the standard error(RMSE)under the training set,and the goodness of fit(R2)under the test set is slightly lower than the goodness of fit(R2)under the training set,indicating that this model has strong generalization performance.
作者
杨军
赵继新
易安军
周佳慧
YANG Jun;ZHAO Ji-xin;YI An-jun;ZHOU Jia-hui(Department of Management Engineering,Guangxi Transport Vocational and Technical College,Nanning,Guangxi,530023;Neimenggu Mengtie Petroleum Co.,Ltd.,Baotou,Neimenggu,014100;Beijing Luojiesite Technology Development Co.,Ltd.,Beijing,100020)
出处
《西部交通科技》
2019年第2期155-158,共4页
Western China Communications Science & Technology
关键词
共享单车
BP神经网络算法
需求预测模型
Shared bicycle
BP neural network algorithm
Demand forecasting model
