Parallel Network Simulation for Sub-Surface Wells
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing reservoir simulators face a parallel scalability problem, where increasing the number of processors does not significantly reduce the overall CPU time for network simulations, leading to inefficient computation and simulation time.
Innovation Solution
The approach involves dividing the reservoir grid into sub-grids and assigning each sub-grid to multiple processors for parallel computation, using a distributed Jacobian matrix construction and partial factorization to solve network equations in parallel, thereby reducing CPU time and simulation time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sequential network simulation is used, then implementation simplicity is maintained, but parallel scalability is poor and CPU time does not decrease with more processors
Solution Approach 1:
The network simulation is divided into multiple independent sub-networks, each assigned to a separate processor. This segmentation allows parallel computation while maintaining simplicity within each sub-network solver, resolving the contradiction between parallel scalability and implementation complexity.
Solution Approach 2:
Multiple sequential sub-network solvers are merged into a parallel system where processors work simultaneously on different sub-networks. The overall network solution is obtained by combining the results from individual sub-network solutions, achieving parallel scalability without requiring complete redesign of the simulation architecture.
2Productivity
If all processors solve the same network redundantly, then load balancing is achieved, but computational efficiency is poor and CPU time is wasted
Solution Approach 1:
The network is segmented into distinct sub-networks, with each processor assigned to solve a unique portion. This eliminates redundant computation by ensuring each processor works on different data, while maintaining load balance through equitable distribution of sub-networks across processors.
3Productivity
If processors are assigned to reservoir grid blocks, then reservoir simulation parallelism is achieved, but network simulation remains sequential and creates a bottleneck
Solution Approach 1:
The network simulation is segmented into independent sub-networks that can be solved in parallel by the same processors used for reservoir simulation. This eliminates the sequential bottleneck while minimizing communication overhead since processors continue to work on their assigned domains with only minimal interface exchanges required.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In some embodiments, systems, methods, and articles may operate to compute, in parallel, to determine values of unknowns in network equations associated with a network of sub-surface wells and at least one surface facility, for intra-well subdivisions of the network, and then for inter-well subdivisions of the network, wherein the computing is based on default values of the unknowns, or prior determined values of the unknowns. Additional activities may include constructing a distributed Jacobian matrix having portions comprising coefficients of the unknowns distributed among a number of processors, wherein each of the portions is distributed to a particular one of the processors previously assigned to corresponding ones of the subdivisions. The Jacobian matrix may be factored to provide factors and eliminate some of the unknowns. Back-solving is used to determine remaining unsolved ones of the unknowns, using the factors. Additional apparatus, systems, and methods are described.