Estimation of rail capacity using regression and neural network

Railroads have been experiencing traffic demand growth and increasing capacity constraints. Effective capacity management is thus crucial to the successful operation of railroads. The initial step in capacity management is to measure and monitor capacity and congestion. This research established a process for constructing parametric capacity models for single- and double-track railroads with simulation results using both regression and neural network (NN) techniques. Experimental results show that the NN models outperform the regression models in terms of predicting both single- and double-track capacity. Aside from traffic volume, average train speed and siding/crossover spacing were identified as the most sensitive factors relative to rail capacity. The use of this capacity evaluation tool can determine efficiency of current operations and can provide an objective basis to assess the need for capital expansion projects.