A sequential quadratic programming-based algorithm for optimization of gas networks

British Gas uses a complex, heavily looped network of pipes and controllable units (compressors and regulators) to transmit gas from coastal supply terminals to regional demand points. Computer algorithms are required for efficient management of the system. This paper describes an algorithm for optimal control over periods of up to a day. The problem is large scale and highly nonlinear in both objective function and constraints. The method is based on Sequential Quadratic Programming and takes account of the structure of the pipeflow equations by means of a reduced gradient technique which eliminates most of the variables from the quadratic subproblems. The latter involve only simple bound constraints, which are handled efficiently by a conjugate gradient-active set algorithm. Trust region techniques permit use of the exact Hessian, preserving sparsity. More general constraints are handled at an outer level by a truncated augmented Lagrangian method. Results are included for some realistic problems. The algorithm is generally applicable to problems with a control structure.