Forecasting regional short-term freight volume using QPSO-LSTM algorithm from the perspective of the importance of spatial information.

It is of great significance to accurately and efficiently predict expressway freight volume to improving the supervision level of the transportation industry and reflect the performance of transportation. Using expressway toll system records to predict regional freight volume plays an important role in the development of expressway freight organization work; especially, the short-term (hour, daily or monthly) freight volume is directly related to the compilation of regional transportation plans. Artificial neural networks have been widely used in forecasting in various fields because of their unique structural characteristics and strong learning ability, among which the long short-term memory (LSTM) network is suitable for processing and predicting series with time interval attributes such as expressway freight volume data. Considering the factors affecting regional freight volume, the data set was reconstructed from the perspective of spatial importance; we then use a quantum particle swarm optimization (QPSO) algorithm to tune parameters for a conventional LSTM model. In order to verify the efficiency and practicability, we first selected the expressway toll collection system data of Jilin Province from January 2018 to June 2021, and then used database and statistical knowledge to construct the LSTM data set. In the end, we used a QPSO-LSTM algorithm to predict the freight volume at the future times (hour, daily or monthly). Compared with the conventional LSTM model without tuning, the results of four randomly selected grids naming Changchun City, Jilin city, Siping City and Nong'an County show that the QPSO-LSTM network model based on spatial importance has a better effect.