Compared to the widely distributed porous and fractured-porous carbonate reservoirs, the naturally fractured-vuggy carbonate reservoirs (NFVCRs) found in Tarim Basin, China, suffer from strong heterogeneity, multiple types of reservoir bodies, poor connectivity, complicated flow behavior and various water–oil interplay relationship in most cases. Due to lack of powerful and suitable EOR (enhanced oil recovery) strategies, depletion-drive recovery is commonly adopted for practical production, and the efficient oilfield development remains a great challenge to all of us. Understanding the relationship between reservoir bodies, configuration, water–oil interplay and depletion-drive performance, is essential to exchange of oilfield development scheme and further potential tapping of the remaining oil. By introducing an self-adaptive particle swarm optimization (DPSO) to tackle the shortcomings of the conventional fuzzy c-means (FCM) in terms of sensitivity to initial values, initialization with randomly generated clustering centers, and easy involvement in local minima, a novel integrated dynamic evaluation method for depletion-drive performance in NFVCRs based on DPSO–FCM clustering is developed. In this paper, different fractured-vuggy reservoir types are identified combing drilling & well logging response, numerical well test analysis and production test. The integrated diagnosis techniques including rate transient analysis, production decline analysis and reservoir energy evaluation, are employed to evaluate reservoir properties, furthermore, principal component analysis (PCA) is performed to reduce the dimensionality of input vector and establish the dynamic evaluation index system of depletion-drive performance. Taking a typical NFVCR in northern Tarim Basin as example, the proposed method is implemented to classify the production responses into four major patterns with eight sub-classes. Moreover, according to the insight of water–oil interplay relationship, depletion-drive production responses and identification of reservoir types, the water influx behaviors are divided into four general patterns. Eventually, the relevant dynamic clustering criterion and suitable potential-tapping strategies or measures are provided.