The widespread dissemination of misinformation poses significant challenges to public health, safety, and information integrity, particularly on social media platforms. To address this, we propose the HeteroGAT model, a novel heterogeneous graph attention network designed for multimodal Chinese fake news detection integrated with genetic algorithm-based hyperparameter optimization. The proposed model combines textual features extracted via BERT-base-chinese and visual features obtained through ResNet-50 in a unified heterogeneous graph structure. Hyperparameters are optimized with a genetic algorithm, resulting in a compact and efficient architecture. Experimental evaluation demonstrates that HeteroGAT achieves state-of-the-art performance with an accuracy of 99.41% and an F1-score of 99.23%, significantly outperforming strong baselines including deep learning and transformer-based models. These results highlight the critical role of integrating multimodal features and heterogeneous social graph structures for the robust identification of fake news.