A novel approach for railway maintenance budgeting using a hybridization of machine learning and integer programming

The balance between budgeting and retaining the optimum performance of railway infrastructure has gained significant importance in recent years. Due to the expansion of railway networks, high maintenance costs, and limited budgets, prioritizing maintenance operations takes time and effort. On the other hand, derailments are one of the most essential types of rail accidents worldwide. Derailments typically result from track conditions, machinery issues, human error, etc. One approach to railway track maintenance budgeting is to allocate the available budget with the purpose of minimizing the financial consequences of derailments considering the accidents' occurrence probability. In this regard, the probability of accident occurrence must be calculated, and the financial consequences of derailment must be estimated. Then, the maintenance plan could be determined by using optimization methods and considering related constraints. In this paper, to calculate the probability of track-caused derailment accident occurrence, the Track Performance Index (TPI) presented by Janatabadi et al. (2020) is employed. The TPI metric is calculated for each block based on the track geometric data of curves in the railway network. Then, various Machine Learning (ML) models are trained and tested in order to estimate the mentioned consequences and the best one with highest precision is selected. The proposed ML model estimates the financial consequences based on the blocks' characteristics with a Normalized Root Mean Square Error (NRMSE) of 18.2%. In the next step, the Branch and Bound (B&B) method is utilized to optimize railway track renewal budget allocation in order to minimize the total probable consequences of accidents in the railway network. To validate the efficiency of the presented method, it is implemented on Iran's railway network. Compared to other methods, the results indicated that the proposed method decreased the network loss more and saved an amount of the maintenance budget. • Developed a hybrid ML and optimization model for railway maintenance prioritization. • Proposed a framework to estimate financial consequences of derailment accidents. • Optimized maintenance budgets using Genetic Algorithm and Branch and Bound method. • Improved efficiency and reduced derailment risks in Iran's railway network. • Speed limit of railway blocks is the most important and feature in the model.