The traditional density peak clustering (DPC) algorithm requires subjective selection of the cutoff distance \({{d}_{c}}\) based on experimental experience, which compromises the objectivity of clustering results. Based on the dopamine regulatory system and reward learning mechanisms, we propose Dynamic Threshold Numerical Neural P Systems (DTNNP), which integrate dynamic threshold mechanisms with numerical spiking variables to overcome the static spiking constraints of traditional neural P systems. Inspired by the dynamic threshold and numerical mechanisms of this neural P system, we developed a Peeling Off Density Peak Clustering algorithm (PODPC) that overcomes the limitations of traditional DPC. The peeling off strategy we adopted, inspired by P systems, eliminates the non-objective influence of manual cutoff distance parameter and clustering center selection, while the new algorithm reduces the impact of global density on clustering results through the new density calculation method. Operating within the DTNNP system framework, the PODPC algorithm leverages dynamic thresholds, numerical mechanisms, and the inherent parallelism of neural P systems to theoretically enhance computational efficiency. Through the verification of 8 synthetic datasets, 7 UCI real-world datasets and face datasets, it has been proved that PODPC-DTNNP has a good performance effect when performing clustering tasks, especially in complex-shaped datasets.
