Rail freight volume trend prediction faces challenges due to data fuzziness, complexity and nonlinearity, and traditional deterministic prediction methods frequently fall short of practical application needs, particularly in addressing uncertainty. To overcome these limitations, we proposed a freight volume trend prediction model that integrated Fuzzy Information Granulation (FIG) with evolutionary optimization. The three-phase methodology establishes: (1)A FIG method was utilized to transform raw time-series into tri-granular representations (Low, R, Up) through fuzzy c-means clustering with temporal constraints, extracting feature information from the raw time-series data and encapsulating it into information granules (2) For complex predictions with small samples, we applied a Support Vector Machine (SVM) for granular modeling, combined with an Improved Particle Swarm Optimization (IPSO) algorithm featuring dynamic inertia weights and mutation operators to prevent premature convergence during training. (3) A hybrid FIG-IPSO-SVM architecture implementing granular-level regression with uncertainty quantification. Validation using 9-year operational records (2013–2022) from the Lanzhou Freight Center (n = 114 monthly observations) in China reveals statistically significant enhancements: compared to the FIG-GS (grid search)-SVM and FIG-PSO (Particle Swarm Optimization)-SVM algorithms, the proposed IPSO-SVM algorithm achieved the smallest prediction error for each granulated set (Low, R, Up) and the smallest mean maxima of absolute percentage error ( A P E M ) for the prediction interval of freight volume, at 5.03%. Moreover,it yielded the tightest prediction interval, characterized by a relative width ( R w ) of just 8.53% and a corresponding interval width ( W ) of only 516,209 tons, surpassing all benchmark models.These findings validate that the FIG-IPSO-SVM framework substantially improves interval prediction precision and trend detection reliability, providing actionable intelligence for railway infrastructure planning and operational optimization.
Gao et al. (Thu,) studied this question.