In recent years, with the improvement of data openness, database watermark has become increasingly important in database security. Database watermark can carry out copyright authentication and traceability of leaked data to ensure data security. However, existing watermark models have low watermark capacity and weak anti-attack robustness. This paper proposes a new database watermark method PADEW. PADEW used an improved particle swarm algorithm based on simulated annealing to avoid falling into the local optimal solution. This enhancement found better watermark embedding positions, thereby increased the watermark capacity and reducing distortion. In addition, this research proposed to use a weighted loss function based on attribute importance to improve the robustness against attribute dimension attacks. The experiments employed watermark capacity, average distortion, and watermark detection rate against multiple attacks to evaluate the performance of PADEW. Experiment results show that PADEW can reduce the distortion caused by watermark embedding while providing more watermark capacity. In addition, PADEW has stronger robustness against various attacks, including tuple deletion attacks, tuple addition attacks, bit flip attacks, and attribute deletion attacks. Especially in the face of 50% attribute deletion attacks, the watermark detection rate is still as high as 81%.
