Due to its ability to effectively capture the evolution and spatial dynamic structure of convective weather systems, dual-polarization radar data has emerged as a critical component in achieving accurate predictions of heavy precipitation, thus garnering significant attention in weather forecasting research. This study aims to address the challenge of low prediction accuracy associated with convective weather using dual-polarization radar data within deep learning models by proposing a novel approach. Specifically, we propose a multi-branch feature extraction model that integrates horizontal reflectivity factor, differential reflectivity, and specific differential phase-shift multimodal data fusion strategies. This approach incorporates a convolutional neural network architecture for predicting convective weather. The model employs data fusion strategies both at the input stage and across modalities, resulting in a four-branch feature extraction network. Additionally, the robustness of the extracted features is enhanced using the squeeze and excitation algorithm. Finally, a convolutional neural network is utilized to generate long-term predictions of horizontal reflectivity factor, which are subsequently employed for heavy precipitation prediction. Through comprehensive ablation and comparative experiments, the proposed algorithm demonstrates promising results, achieving prediction accuracies of 46.28% and 28.68% for the probability of detection (POD) and critical success index (CSI) parameters, respectively, in 1-hour forecasts.