Energy Prediction Of Freight Train Based On Ensemble Learning And System Design

The Chinese rail industry plays an increasingly important role to support economic and social progress.In particular,the rail freight has been a critical factor of economic development.However,the high-energy consumption,the low running efficiency and the complex order of freight transportation organization are difficult challenges of the increasing cargo demands for freight trains.The first task is the precise statistics and prediction of energy consumption to achieve the above goal.Meanwhile,the successful implementation is helpful to strengthen management of energy consumption,to improve transportation efficiency and to optimize transportation organization process.Concerning above challenges,this thesis investigates the topic energy prediction based on ensemble learning from a multi-disciplinary perspective,by integrating statistical theory,decision making theory and machine learning.The main work of this thesis is summarized as follows:(1)We propose an ensemble support vector machines(ESVM)via evidential rea-soning(ER)theory.Firstly,a number of base classifiers are established from training data set.Secondly,the weights of different base classifiers are determined by their AUC values on training data set.Thirdly,the fusion results of all base classifiers are obtained from their outputs and corresponding weights through ER theory.Finally,the feasibility and validity of the ESVM have been proved by comparison with conventional ensemble learning algorithms.(2)We propose a grid-based energy prediction algorithm for freight train.Firstly,a data set of energy consumption is established via the run data of freight train.Note that the average energy of every time slot is regarded as a sample by considering the temporal density of the obtained run data.Secondly,the corresponding learning model is selected based on the proposed ESVM from above data set.Thirdly,the average energy of certain section is the mean of the average energy values from all grids belonging to this section.Here,the section is divided into a number of uniform grids and the average energy level of every grid can be calculated from the ESVM.The simulation indicates that this algorithm can accurately predict energy with ample samples and can provide technical supports of optimal drive of freight train.(3)We design an analysis system of energy consumption of freight based on the run data and the proposed grid-based algorithm.Firstly,a series of techniques,i.e.,data cleaning,data conversion,data registration and data matching,are employed to construct a database from the run data and other sources.Secondly,an analysis system of energy consumption is built through synthetically considering the database and the grid-based energy prediction model.The system achieves the goal of the accurate statistics,the com-parisons from multiple dimensions and prediction of energy consumption.What's more,this system is helpful to solve the problems from practical application fields of freight trains.