Objective To develop a multivariate modeling and prediction approach using the data of influenza incidence, meteorological factors and PM_2.5 variables in Hefei City, and to provide new insights into multivariate modeling and prediction methods. Methods PM_2.5 data were transformed into fractal dimension data and, along with meteorological data, were incorporated into a ConvLSTM model. The performance of this model was compared with traditional ARIMAX and multivariate LSTM models. Results The ARIMAX model's testing set Mean Absolute Error (MAE) was 95.75, Root Mean Square Error (RMSE) was 176.72, and Index of Agreement (IA) was 0.396642. The multivariate LSTM model's testing set MAE was 22.18, RMSE was 43.06, and IA was 0.974611. For the fractal dimension-based ConvLSTM model, the testing set MAE was 17.37, RMSE was 32.25, and IA was 0.988149. Conclusion The fractal dimension effectively captures the complexity and self-similarity of PM_2.5 concentration, providing the model with richer feature information. The fractal dimension-based ConvLSTM model significantly outperforms the traditional ARIMAX model and the multivariate LSTM model in prediction accuracy and can be used for predicting the number of influenza cases.