Modeling And Optimization Of Seat Inventory Control For Passenger Railway

In recent years, with the rapid construction of China’s railway, the railway capacity has been significantly improved. The short supply of railway capacity has been alleviated. Nowadays, the utilization of capacity on different sections of some trains is uneven. In this background, by considering the characteristics of the railway industry,this paper studies how to control the capacity of train seats during the pre-sale period, so as to maximize the train revenue. The thesis consists of four parts: research on modeling methods, research on demand mining, research on control problem of single train and control of multiple train. The research work and conclusions of each part can be summarized as follows.(1) The modeling method of railway seat control is studied. First, we describe the problem of railway seat control and analyze the unique characteristics of railway problems compared with traditional inventory control problems. Then, the continuous,discontinuous state, and convex and nonconvex states of train seat capacity are defined respectively, and the methods of booking-limit control and seat allocation for railway seat control are studied. Finally, the general model of railway seat control modeling is established. The basic properties of the model and the proper algorithm are analyzed,which can be used to provide the basic modeling framework for the subsequent modeling research.(2) The actual demand mining methods based on historical sales data are studied.First, from the perspective of independence demand, the censored passenger flow data is used to mine the probability distribution of a single product’s demand. The mining results show that the estimation and demand of historical passenger flow data directly lead to underestimation of demand. Then, from the perspective of the overall market, we build a non-linear model of passenger volume on the transport capacity and passenger demand, as a basis for mining the overall demand for multiple products. Finally, from the perspective of passenger choice behavior, using the real-time sales data and availability data during the pre-sale period, we propose the EM algorithm to explore the actual preference of passengers.(3) We study the control method, model, algorithm and applicability of the single train. First of all, according to the current precise control and fuzzy control method of China Railway, we construct the model and design the optimization algorithm respectively, and verify the validity of the model and algorithm by experiments. In addition, a method of associated booking-limit control is proposed, and its model and algorithm are constructed and analyzed experimentally. Finally, through a large number of simulation experiments, we study the applicability of the various control models.(4) The control method, model and algorithm are studied for multiple trains. First of all, the control method for multiple trains is designed. Then, the optimization model for the method is studied, including the sampling process of the multi-train and the design of simulation steps for the problem. The simultaneous perturbation stochastic approximation algorithm with computational feasibility is studied. Finally, the validity of the model and algorithm is tested by two experiments.The main innovations and research significance of this paper can be summarized as follows:(1) The existing literature cannot extract the core difference between passenger railway and traditional RM industry. In order to make up for this deficiency, this paper puts forward the method of seat control modeling which meets the characteristics of China’s railway industry, including the continuous seat restraint of homogeneous resources, the definition of capacity state, the feasible conditions of partitioned booking limits and the method of seat allocation.(2) Most of the existing railway literature uses the historical sales data as the demand input of the seat control model, ignoring the historical data can only reflect the satisfied demand. This results in the deviation of the demand estimates, thus affecting the seat control effect. To overcome this deficiency, this paper constructs the demand mining method from the perspective of independence demand, overall market demand and passenger choice behavior.(3) Most of the existing railway seat control modeling did not take into account the dynamic arrival process of the booking demand during the pre-sale period, and also failed to consider the impact of the seat allocation process on the capacity state when the product is sold. In this paper, four kinds of seat control models of P-JQ, P-MH,P-LD and P-XT are constructed, and the feasible algorithm are designed respectively.Among them, P-JQ and P-MH correspond to the existing precise control and fuzzy control methods of China’s passenger railway, which can be directly applied in practice.The P-LD model is better than P-JQ and P-MH under certain conditions. The P-XT model can be applied to the problem of multiple train problem under the passenger’s choice behavior.