Parallel And Distributed Computation In Process Optimization

Process optimization has emerged as one of the most valuable techniques for system design, analysis and operation. Currently, the rapid trends toward increased model detail and rigor, dynamic optimization, on-line optimization, and scheduling accelerate the need to optimize very large systems of equations with many degrees of freedom. The large-scale optimization of process systems, however, continues to present a major challenge both in academia and in process industries.Even with the high performance computers nowadays, there still exists many difficulties for a single computer to solve large-scale chemical process optimization problems. Parallel computing with cluster of workstations, is a high performance/price solution to solve large-scale process optimization problems.This dissertation details the investigation, development and implementation of efficient algorithms and techniques for parallel optimization of process systems. The main contributions are as follows:1) The theory and object to parallel computing technology for process optimization problems are reviewed. Many parallel algorithms and parallel models are discussed. Details of cluster of workstations, a relatively recently developed technology, are given to highlight its advantages compared with other approaches in parallel computing. A new solution with cluster of workstations to solve large-scale process optimization problems is proposed. The plant can use cluster of workstations to gain strong computation power at low cost. Observations indicate that the degree of granularity plays a major role in this approach. It should be carefully schemed to balance the load of communication and the distributed calculation steps. How to improve computation efficiency in cluster of workstations is also discussed.2) Several parallel optimization algorithms are discussed and evaluated, and they can be divided into two classes: algorithms based on gradient and algorithms not based on gradient. Then some parallel solutions using these algorithms based on cluster of workstations are proposed.3) As SQP (Sequential Quadratic Programming) has emerged as the algorithm of choice for solving large-scale process optimization problems, several parallel strategies for SQP are presented. A parallel strategy utilizing cluster of workstations is proposed to solve process optimization problems efficiently. The most expensive computing parts in SQP are function and gradients evaluation. Using variable partition, a large-scale gradients evaluation problem can be divided into several sub-problems with smaller dimensions. The sub-problems are independent and can be calculated on different processors. Because communication overhead among workstations is very high on local area network, coarse granularity must be selected to reduce communication ratio. This method assures the need for coarse grain and low communication, and it's suitable for cluster of workstations. Computing results on a distillation column optimization problem demonstrate the efficiency of this approach.4) Decomposition-coordination method can be used to solve very large-scale problems. The two tier architecture of this method makes it time expensive. A modified decomposition-coordination method is presented for large-scale process optimization and parallel implementation. SQP is used at lower level and quasi Newton method is used at upper level in this method. In order to improve efficiency of decomposition-coordination algorithm, a parallel approach using cluster of workstations is developed. Computing results on a heat exchange optimization problem demonstrate the efficiency of this approach.5) A distributed parallel optimization computation environment based on Matlab is proposed. After details of parallel computation environment are discussed, the architecture of distributed parallel optimization computation environment is analyzed. Four kinds of parallel functions including initialization, communication, computation and integratio...