Research On The Application Of New Machine Learning Methods In Urban Highway Freight Volume Forecasting

With the rapid development of China's economy,the links between cities have become more and more close,and the freight industry market has become larger and larger.As the most important mode of cargo transportation in China,road freight transportation can accurately predict the traffic volume of roads by using effective methods,which can help improve urban traffic conditions,improve transportation efficiency and promote rational deployment of resources.Firstly,it analyzes and summarizes the research status of freight volume forecasting and machine learning at home and abroad.The results show that using machine learning method to predict freight volume can have obvious advantages.Then through the qualitative analysis method,comprehensive consideration from various aspects,determine the factors affecting the road freight volume used in predictive modeling.At the same time,the traditional road freight volume forecasting method and the machine learning method are compared and analyzed theoretically: due to the complex correlation between the relevant influencing factors in the urban road freight volume forecasting research and the abnormal values in the statistical data,etc.Traditional time series smooth prediction method and related factors principle prediction model performance is not good,and the introduction of machine learning method into prediction research can improve the accuracy of prediction results.In-depth research on machine learning in recent years has shown that new machine learning methods in the field of deep learning and integrated learning have a wider range of applications,and model performance is better than traditional machine learning methods.In this paper,two new machine learning methods are adopted: the Stacking method in the integrated learning field and the noise reduction self-encoding method in the deep learning field to predict the urban road freight volume.Noise reduction self-encoding reduces the risk of local extremes and gradient dispersion in traditional BP neural network models during training.In addition,the random noise added in the self-encoding structure training can improve the robustness and generalization ability of the road freight forecasting network,and it is not easy to over-fitting.Stacking can choose a model that has good performance for solving specific problems as a primary learner,thus improving the efficiency of model training.Constructing a secondary training set by cross-validation can make the final generated model have higher prediction accuracy.Taking the forecast of Shenzhen highway freight volume as an example,the application research is carried out according to the above two machine learning methods,and the prediction results of the traditional method are used to compare and verify the feasibility and effectiveness of the new machine learning method.It shows that the noise reduction self-coding model and Stacking have good precision,which are better than the traditional prediction method.Among them,Stacking is better in predicting accuracy and building model efficiency,and can be applied to the forecasting of urban road freight volume.This paper proposes two new machine learning methods to provide reference and reference for freight volume forecasting research in other regions.