Abstract: Mapping logical quantum circuits onto quantum devices with constrained connectivity and noise is a critical bottleneck for quantum computing. Executing multiple circuits in parallel on quantum hardware can increase qubit utilization, but it will also exacerbate mapping complexity. To reduce SWAP gate overhead and improve fidelity, an adaptive multi-programming mapping approach based on dynamic partitioning is proposed. A preprocessing mechanism filters out qubits with high error rates from physical connections prior to mapping. A dynamic partitioning strategy guided by a comprehensive quality metric is employed, and adaptive mapping is applied to generate high-quality mappings. Experiments on IBM Quantum Toronto achieved an average fidelity of 52.39%, representing a 2.17% improvement over existing methods. Regarding mapping overhead, an average of 26.25 additional CNOT (controlled NOT) gates were inserted, while SWAP gate overhead was reduced by 1.5 CNOT gates compared to other methods. The proposed method provides an optimized solution for multi-programming mapping, demonstrating the merits of both high fidelity and low mapping overhead.