Computationally effective optimization methods for complex process control and scheduling problems.

摘要

Motivated by the soaring production cost, intensive competitions and public attentions on environmental issues, how to reduce the operational cost, raise the profit and enhance the operational safety attracts tremendous interests in the chemical and petroleum industry. Since the regulatory control strategy may not achieve such rigorous requirements, higher level process control activities, such as production planning, real time optimization (RTO) and multi-variable control are more frequently taken into account. Moreover, to attain the better performance, process control engineers often consider plant-wide operations rather than unit-based actions. As a result, both dynamic and discrete optimization techniques for the large scale problem nowadays play a more important role in the industry than before. Even the classical optimization based techniques, such as model predictive control (MPC), have seen considerable successes in many practical applications. However, they are still suffering from computational issues in the circumstances of a large-scale plant, complex dynamic system or the short sampling time period. Furthermore, these traditional optimization techniques usually employ the deterministic formulations, but often become unsuitable for uncertain dynamics. Hence, this thesis is mainly concerned with developing computationally effective algorithms to solve practical problems arising from those high level process control activities and highly affected by the disturbances.;Approximate dynamic programming (ADP) is one of the most efficient computational frameworks to handle large-scale, stochastic dynamic optimization problems. While a large number of successful cases based on ADP have been reported, several critical issues, including risk management, continuous state space representation and the stability of the control policy, prohibit its application in process control. To overcome these shortcomings, • We developed a systematic approach to extract the probabilistic model from the operational data of a plant-wide system and proposed a risk-sensitive RTO approach based on ADP. • An innovative procedure for designing control Lyapunov function (CLF) and robust control Lyapunov function (RCLF) is presented for a nonlinear control affine system under the input and state constraints. • Based on the well-designed RCLF, a mixed control strategy, combining the advantages of MPC and ADP, is proposed to handle the stability issue of the ADP control scheme.;In addition to dynamic optimization, another focus of this research is the discrete optimization. Considering mixed integer linear programming (MILP) becomes increasingly common in the planning and scheduling of the chemical production, it is worthwhile to explore a more efficient algorithm for solving this NP hard problem. A modified Benders decomposition approach, featured by its tighter cutting plane, is presented to accelerate the solution procedure.;All the proposed approaches are demonstrated and evaluated by several bench-mark examples. The comparisons with previous works also show the superiority of the suggested methods.