(444c) CAPE-OPEN Compliant Better Optimization of Non-Linear Uncertain Systems (BONUS) Capability for Large Scale Systems

Process optimization involves large scale nonlinear system. Optimization in the face of uncertainty for such large systems is a computationally intensive task. This paper describes a new CAPE-OPEN compliant capability for efficient optimization of large scale process systems under uncertainty. A case study of PC power plant optimization for water minimization illustrates this capability. This case study involves realistic model of the process using ASPEN Plus simulator. This approach requires the employment of several design specifications with their corresponding nested convergences yielding a highly non-linear non-convex problem. Additionally, water consumption of the PC plant is drastically affected by atmospheric conditions represented as air temperature and humidity. The presence of these uncertainties requires the employment of stochastic models to obtain a realistic representation of the process. Therefore, the water consumption management is a complex stochastic programming problem. To reduce the computational intensity of the stochastic programming, the CAPE-OPEN compliant capability employs BONUS algorithm. BONUS uses a sequential quadratic programming (SQP) for the solution of the non-linear programming problem (outer loop) and for the stochastic simulation (inner loop) the algorithm makes use of a re-weighting method with Kernel density estimation to approximate the objective function and its derivatives. Thus, the evaluation of all the sampled scenarios for each policy determined by the SQP is avoided. This reduction in computational effort becomes extremely valuable when several initial policies are required to be evaluated due to the non-convex nature of the problem. A 548 MW PC plant was studied under variable air temperature and humidity. Stochastic simulation and partially ranked correlation coefficients (PRCC) were employed to determine the most influential variables as decision parameters. Uncertainty in air conditions was characterized based on the weather data available for eight US Midwestern urban centers. BONUS results estimated that water consumption can be reduced by 12% and the computational time was reduced by 99.7%. BONUS estimation of the objective function at the optimum contains only 2% error.