Optimization For Utilization Of Railway Carrying Capacity: An Approach Based On Uncertainty Analysis

In the past several years,the scholars have been building a comprehensive research system involving of theory,technology and methodology,for calculation,utilization,strengthening and reservation of railway capacity.As the crucial part of optimization of railway resource allocation,studies on optimization of capacity utilization have attracted a lot of academic attention.As the external social and economic environment of the railway changed,the focus of capacity utilization is not only on"quantity",but also on"quality",so that it is necessary to consider the uncertainty of the railway system in the capacity utilization optimization study.However,as a complex giant system with a complicated uncertainty structure and various origins,it is difficult to find and deal with uncertainties and to optimize capacity utilization plans.Although there have been many previous researches on theory of uncertainty analysis and capacity utilization optimization,there are still problems,on lacks of structure and hierarchies in the uncertainty analysis,data-driving forces in uncertainty analysis and utilization optimization technology,and optimization technology of capacity utilization against uncertainty applying to different data environments.Therefore,this study,based on existing railway data resources and the theory of uncertainty analysis and capacity utilization optimization,analyzed the static pattern,and dynamic evolution of uncertainty in railway,followed by optimization technology of capacity utilization against uncertainty in different data environments.Details are as follows,with a coal transportation heavy-haul railway in the North as the research background.?.Define concepts on uncertainty analysis and optimization of railway capacity utilization.The relationship among transportation organization elements of railway was analyzed on the basis of the concept of capacity utilization.Then by stratifying uncertainties and analyze structures,we found the distribution of uncertainty by granularities and the internal structure of subjective stochasticity and objective randomness,as the basic concept for designing methodology.Two analysis steps were come up with to comprehensively research uncertainty,namely static pattern analysis and dynamic evolution analysis.The static pattern analysis,based on pattern assumptions,measures uncertainty distributions by patterns obtained from micro-sequence data by data mining methods.The dynamic evolution analysis,based on the controlled-stochastic theory,analyzes dynamic processes of uncertainty from perspectives of‘execution'and‘control'as well as mechanisms of impacts of reserve capacity,the key element,on the dynamic evolution.At last,based on the proposed uncertainty analyses methods,a framework of capacity utilization optimization framework is designed.?.Propose methodology of static patterns for analyzing uncertainty.We used pattern analysis technique,a combination of unsupervised pattern discovery and supervised learning pattern recognition,to solve problems pattern recognition from large-scale data.In the unsupervised pattern discovery,pattern set were formed without manual intervention on the foundation of deep learning techniques by obtaining patterns from small sample data.After that,the pattern set refined by manual adjustments.In the supervised learning pattern recognition,related features to railway capacity utilization were extracted so as to recognize pattern sets in large-scale data,then the deterministic distribution was evaluated by the Dirichlet distribution.A heuristic method was come up with to improve micro-operability and assist the optimization of capacity utilization by the model of uncertainty dynamic evolution analysis.In numerical experiments,by inputting microscopic spatio-temporal trajectory data,the pattern and distribution on uncertainty in a certain section were analyzed.We analyzed the reasons for the formation of the model,test the performance of the method,and reveal the relationship between the model and subjective uncertainty and objective randomness.?.Propose methodology of dynamic evolution for analyzing uncertainty.According to the controlled-stochastic theory,we respectively established two finite-dimensional distribution family models of stochastic processes on uncertainty from perspectives of‘execution'and‘control',and built a mixing process model based on interlaced assumptions.Based on this,following the Markov property,we reformed the model by discrete state transition processes.Afterwards,we analyzed the information vectors that influenced the state probability,deduced uncertainty based on Bayesian networks so as to establish the dynamic evolutionary reasoning patterns.Next,by the use of the dynamic evolutionary reasoning model,we designed methods of collision detection and adjustments of reserve capacity,supporting finely tuning parameters of capacity utilization.At last,we analyzed the difference during the uncertainty evolution by numerical experiments,combining with results in the static pattern analysis.For example,If the delay is serious,the train would slow down in some sections?the delay mode is about 80%?,that is using block sections as temporary storage lines,proving the existence of"station-section"coordinated capacity utilization.?.Propose methods to optimize capacity utilization based on uncertainty analyses.On the foundation of the Event-Activity Network?EAN?,we established an ideal optimization model aiming at expected plan completeness,utilization plan cohesion,capability utilization feasibility,capability utilization robustness and capability utilization recoverability.According to the idea model,we established robustness models against uncertainty after analyzing uncertainty-related objective functions and constraints,and then discussed detailed procedures of robust optimization as follows.Firstly two uncertainty sets were established based on perturbation?₁?and data-driving?₂?.Then we designed a robust optimization technology including strong robustness,light robustness,feasible recovery robustness,optimal recovery robustness,which could be applied to various data environments.Next,we utilized the results of uncertainty analysis and designed an algorithm based on branch and bound to solve the model.This algorithm focuses on optimizing the relationship between passing and overtaking in a capacity utilization scheme.At last,we designed numerical experiments based on₁ and₂,respectively,and analyze the performances of each optimization model.?.An example verification is carried out for China's typical coal transportation heavy-haul railway.Applying the research content to actual cases,the following conclusions are obtained:1)Results proved the existence of the pattern and demonstrated the difference among pattern distributions in various sections.For example,in a certain section,the ratio of the early mode,punctual mode and delay mode was 25%:7%:68%,while in the key technology station front section,the corresponding ration was23%:58%:19%,which illustrated that internal auto-organization would happen in the railway operation system when lateness happened uncertainly.2)When different reserve capacity values are set and dynamic evolution is used to evaluate the effect on capacity utilization.For example,when 2%reserve capacity is allocated as a whole in the case,the maximum initial perturbation against average delay about 8 minutes in the case with2%of the total reserve capacity allocated.3)We revealed the fact that the allocation structure and total amount of reserve capacity jointly determine the merits and demerits of a plan against uncertainty.For example,because more reserve capacity is allocated to the key section,the scheme of optimal recovery robustness based on₂ would decrease the total reserve capacity by 6.4%compared with that of strong robustness optimization,but improve indices by 8.6%.This study focusing on optimizing the utilization of railway capacity based on analysis of uncertainty,from theory,models and algorithms.By the research framework of“Machine Learning+Optimization",we can adequately utilize historical data,analyze uncertainty from multiple dimensions of the static model and dynamic evolution.The analysis of uncertainty,which was directly involved into the construction of the uncertainty set in an optimization model,the selection of model parameters and the detection of the solution process conflict,will guide the optimization of multi-objective capability utilization schemes in different data environments.The results of this study can identify the features of uncertainty from the data,and appropriately optimize the capacity utilization methods through optimization models so as to improve the quality of capacity utilization schemes,which is beneficial to refined utilization of the railway capacity in China.