<p>Neighborhood schemes are commonly embedded in evolutionary algorithms (EAs) to address multimodal optimization problems (MMOPs) because they explicitly or implicitly separate populations, enabling each subpopulation to evolve in parallel, thereby tracking different peaks. However, constructing neighborhoods that can aggregate similar individuals and efficiently guide coevolution remains challenging. To address this, this paper proposes a potential-optima guided adaptive neighborhood differential evolution algorithm, termed POGAN-DE. In POGAN-DE, an adaptive neighborhood mechanism (ANM) is designed to regulate the search behaviors of individuals with distinct characteristics across evolutionary stages, balancing exploration and exploitation capabilities. Furthermore, considering that multiple optima may coexist within a single neighborhood, a potential optimal identification (POI) scheme is developed using a flexible clustering method to detect undiscovered optima located on different peaks, thereby enhancing peak identification capability. A population crowding relieving technique is integrated to avoid the waste of fitness evaluations and enhance population diversity. Additionally, species-level historical success records are utilized to enable self-adaptation of F and CR parameters. Finally, POGAN-DE is validated on a benchmark suite of 20 multimodal problems and real-world applications. Comparative experiments with state-of-the-art algorithms confirm its robust and comprehensive performance.</p>