Well Flow Simulation Using Segmented Sub-Networks and Learning Logic

The patent segments the flow network into multiple sub-networks that can be processed independently and in parallel. Each sub-network represents a portion of the overall system, allowing the computational problem to be divided into smaller, more manageable units that reduce the size of matrices required for each calculation while maintaining solution accuracy through coordinated processing of all sub-networks.

The patent applies preliminary action by using learning logic to predict pressure bands and memorize pressure versus flow relations before performing detailed calculations. This pre-processing step establishes initial conditions and constraints that guide the subsequent computational process, reducing the complexity of iterative calculations and improving solution stability by avoiding exploration of unstable solution spaces.

The patent applies dynamics by implementing adaptive learning logic that adjusts its processing based on the detected characteristics of the flow network. The system dynamically identifies regions with strong inhomogeneities and discontinuities and applies appropriate handling strategies to these specific areas, rather than uniformly processing the entire network. This dynamic adaptation maintains high simulation accuracy for complex regions while reducing algorithmic complexity in simpler regions.

The patent uses copying by memorizing pressure versus flow relations from previous calculations and similar operational conditions. This creates a knowledge base of pre-computed relationships that can be referenced during simulation, avoiding the need to re-solve complex non-linear equations from scratch for each scenario. The copied knowledge accelerates computation while maintaining accuracy through the learning logic's ability to generalize from stored examples.

The patent segments the large flow network into multiple sub-networks that can be processed independently. This segmentation allows the system to handle networks with unlimited branches by dividing them into manageable portions that fit within available memory constraints. Each sub-network is processed separately, and results are integrated to provide the complete solution, enabling scalability without requiring proportional increases in memory resources.

The patent applies preliminary action by using learning logic to predict pressure bands and memorize pressure versus flow relations before performing detailed calculations. This pre-processing step reduces the computational burden during the main simulation phase, allowing the system to handle larger networks with limited memory resources. The learned relationships are stored in compact form and used to guide subsequent processing, reducing the need to maintain large data structures in memory during computation.