Forecasting Model for Railway Freight Volume Based on Dynamic Optimization and Deep Learning Considering Multiple Factors
Railway freight volume is a key index of regional economic vitality and logistics efficiency,whose accurate prediction is very important to the optimal allocation of railway transportation resources.However,railway freight volume is affected by a variety of social factors,and the traditional forecasting mod-el has some problems such as low prediction accuracy and limited generalization ability when dealing with complex situations.In view of these problems,we proposed the GRA-MABiLSTM-HHO forecasting model for railway freight volume based on dynamic optimization and deep learning considering multiple fac-tors.First of all,considering the quantitative indicators related to railway freight volume comprehensively,we selected several closely related factors in social activities from the direct and indirect dimensions,and adopted the improved grey relational analysis method,the entropy weight method and analytic hierarchy process method to determine the objective and subjective weights and strengthen the features.Next,we calculated the comprehensive grey correlation degree,screened out the strongly related factors and assigned them rea-sonable weights.Then,we constructed the BiLSTM to train the railway freight volume data sequence bi-directionally and recurrently to find the complex patterns and trends.Meanwhile,we introduced a multi-head attention mechanism to process the multiple attention layers in parallel,obtained the input matrix through linear trans-formation,calculated the attention output of each head before merging and transforming them to more accu-rately analyze the importance of each factor,and improve the focusing and generalization ability of the mod-el over key information.In order to optimize the initial parameters of the model,we proposed an improved HHO algorithm based on multi-strategy combination to deal with the local optimization and unbalanced local development ten-dency of the traditional HHO algorithm.Also,we carried out nonlinear improvement for the escape energy to make it more suitable for the actual prey energy consumption in the hunting process,and avoid local opti-mization.In addition,we introduced tent chaotic mapping to initialize the population,improve the popula-tion diversity and the global search ability,and further enhance the generalization ability and prediction accu-racy of the model.Finally,we used the railway freight volume data of Guangdong Province from 2013 to 2023 for empirical verification,the result of which shows that compared with BiLSTM,BilSTM-HHO,GRA-MABiLSTM,the GRA-MABiLSTM model significantly reduces the evaluation indicators such as mean absolute error,mean relative error and root mean square error;and compared with MI-PSO-RBF,EMD-APSO-SVR and other combined prediction models,it also has better performance,which fully proved that the model can effectively improve the forecasting accuracy of railway freight volume and provide strong support for the op-timal allocation of railway transportation resources.
railway freight volumegrey relational analysisBiLSTMmultiple head attentionimproved HHO algorithm
NETL
NSTL
万方数据
