为提升A类通用机场终端区在动态环境下的冲突探测与解脱能力，提出一种生成对抗网络（GAN）与长短期记忆网络（LSTM）深度融合的端到端冲突探测与智能解脱方法。该方法的核心创新包括：①构建双任务判别器架构，通过共享特征实现轨迹真伪判别与冲突概率预测；②设计物理约束引导的生成器，在满足飞行约束条件下生成多样化解脱轨迹，并通过多准则筛选最优方案；③提出自适应损失权重调整策略，动态平衡轨迹重建精度、对抗训练与冲突规避等多个目标。基于TrajAir数据集的综合实验表明，所提方法的冲突检测准确率达93.4%，解脱成功率达88%，显著优于所对比的传统几何规则方法；所生成轨迹误差小、符合飞行性能约束，体现出良好的实时性、准确性及决策灵活性。研究可为通用航空空中交通管理智能化提供技术参考，有望促进低空空域的安全高效运行。

To enhance the conflict detection and resolution capability in dynamic environments of Category A general aviation terminal areas, this paper proposes an end-to-end conflict detection and intelligent resolution method based on the deep integration of Generative Adversarial Networks (GAN) and Long Short-Term Memory (LSTM). The core innovations of the method are: ①constructing a dual-task discriminator architecture that performs both trajectory authenticity discrimination and conflict probability prediction through shared feature representation; ②designing a physics-constrained generator that produces diverse resolution trajectories under flight performance constraints and selects the optimal solution via a multi-criteria screening mechanism; ③proposing an adaptive loss weight adjustment strategy to dynamically balance multiple objectives such as trajectory reconstruction accuracy, adversarial training stability, and conflict avoidance. Comprehensive experiments based on the TrajAir dataset show that the proposed method achieves a conflict detection accuracy of 93.4% and a resolution success rate of 88%, significantly outperforming conventional geometric rule-based methods. The generated trajectories exhibit small errors, comply with flight performance constraints, demonstrating the real-time capability, accuracy, and decision-making flexibility of the method. This study provides a feasible technical solution for intelligent air traffic management in general aviation, contributing to safer and more efficient low-altitude airspace operations.