Memory dendritic neural P systems for multi-ground type fault line selection in distribution networks

Rapid and accurate single-line-to-ground (SLG) fault line selection in distribution networks is crucial to power system safety. Although existing deep learning methods have achieved good performance, there is still room for improvement in the accuracy of fault line selection when faced with multiple neutral grounding types and varying fault conditions. The neural-like P systems use the "point neurons" computing mode and fails to fully utilize the multi-dendritic branching characteristics of biological neurons, resulting in limited computing efficiency. Therefore, this paper proposes a novel memory dendritic neural P (MDNP) system, which effectively improves the system's fault feature learning ability under different grounding types by introducing the triggering rules of memory dendritic factors, dynamically adjustable dendritic structure and adaptive learning mechanism, significantly reduces computing resource consumption and accelerates convergence speed. Furthermore, the design of online application scheme enables the MDNP system to achieve online optimization of fault line selection in dynamic power grid. The experimental results show that compared to seven baseline models, the MDNP system achieves fault line selection accuracy of 98.2% and 97.3% under non-grounding and arc suppression coil grounding types, respectively. The average fault line selection time is only 1.78s. Additionally, the computational resource consumption is reduced by 56.6% compared to traditional neural-like P systems, demonstrating its exceptional performance and potential for fault line selection tasks.