Abstract:

Deep fractured-vuggy carbonate reservoirs， usually a result of multi-stage tectonic movements and paleo-karst modification， are complex in structure and strong in heterogeneity. Conventional connectivity identification methods， primarily based on clastic rock reservoirs， are not suitable for carbonate reservoirs. This research aims to realize an automatic evaluation of inter-well connectivity by studying static and dynamic data and incorporating techniques such as multifractal and curve similarity analyses into the evaluation. These techniques can be employed to extract characteristic dynamic parameters and detect response level of neighboring wells under different mechanisms. Deep residual networks are used to integrate multi-attribute seismic data for characterizing the spatial structure of reservoirs. Additionally， reinforcement learning and multi-objective algorithms are also employed to automatically search for three-dimensional connectivity paths. Extracted dynamic response features and the distribution patterns of three-dimensional connectivity paths in typical fracture-cavity units in the Tahe oilfield with different karst backgrounds demonstrate that the fracture network serves as the primary communication channel among wells in weathered crust karst， exhibiting good multi-directional connectivity. The main and secondary faults are the primary channels connecting between wells within fault-controlled karst oil reservoirs， forming a strip-like connectivity pattern along the faults. The connectivity of paleo-underground river karst along multiple layers of underground river networks with local filling and collapse exhibiting segmental characteristics. The research results are of guiding significance for exploring remaining oil and enhancing recovery of deep fractured-vuggy reservoirs.

Key words: reinforcement learning, multi-objective algorithm, three-dimensional connectivity path, fusion of seismic multi-attribute, karst system, fractured-vuggy reservoir, Tahe oilfield