<p>Biodiversity, which underpins the resilience of ecosystems, has declined sharply as the global species extinction rates have accelerated. Current biodiversity conservation cannot meet the needs of all species, and keystone species provide a key entry point for maximizing their conservation benefits. However, identifying specific keystone species remains challenging, and their roles in ecosystems need further analysis. In this study, the linear inverse model—Markov Chain Monte Carlo—was constructed by simulating the removal of candidate keystone species in Haizhou Bay ecosystem of China. Using the percentage changes of energy flows and ecological network analysis (ENA) in food web as evaluation indicators, specific keystone species were identified, and their roles in the ecosystem were comprehensively analyzed. The results showed that the keystone species identified based on energy flows and ENA were <i>Johnius belangerii</i>, <i>Alpheus distinguendu</i>s, and <i>Larimichthys polyactis</i>. Energy flow was more responsive to simulated keystone species removal than ENA analysis, with higher percentage changes in energy flow to detritus and respiration than those to predators. In addition, the five ENA indicators responded differently to simulated removal scenarios, with <i>HP</i> and <i>i/d</i> responding most extensively to these scenarios. These findings will contribute to a comprehensive understanding of the role of keystone species in marine ecosystems and improve the understanding of biodiversity conservation. This work will guide the selection of priority conservation species and provide essential information for ecosystem-based fisheries management.</p>