In the post-epidemic era, dynamic monitoring of expressway road freight volume is an important task. To accurately predict the daily freight volume of urban expressway, meteorological and other information are considered. Four commonly used algorithms, a random forest (RF), extreme gradient boosting (XGBoost), long short-term memory (LSTM) and K-nearest neighbour (KNN), are employed to predict freight volume based on expressway toll data sets, and a ridge regression method is used to fuse each algorithm. Nanjing and Suzhou in China are taken as a case study, using the meteorological data and freight volume data of the past week to predict the freight volume of the next day, next two days and three days. The performance of each algorithm is compared in terms of prediction accuracy and training time. The results show that in the forecast of freight volume in Nanjing, the overall prediction accuracies of the RF and XGBoost models are better; in the forecast of freight volume in Suzhou, the LSTM model has higher accuracy. The fusion forecasting method combines the advantages of each forecasting algorithm and presents the best results of forecasting the freight volumes in two cities.