Goose Queue Approaches With Applications In Process Real-Time Optimization

To solve large-scale process optimization problems, traditional decomposition and coordination methods are easily flung into dilemmas in the presence of increase in calculation, no practical meaning of the iniatial factors, as well as dependency on objective separability.Inspired by biologic nature of flying geese, so-called goose queue (GQ) approaches towards a kind of large-scale system decomposition and coordination optimization are explicitly introduced in this thesis. In the context of analysis of upwash and Ⅴ-formation seeking mechanism, an ad-hoc GQ structure is introduced, along with the operators of following, exchange and transmission. Based on this philosophy, a large-scale optimization problem can be firstly decomposed into hierarchically connected multi-layer GQs according to the relationship among variables of objective and constraints. Subsequently, optimization solutions are converted into coordination issues among GQs in a novel way, where the GQ-Objective is optimized firstly, presenting the result as the objective of individual GQ before the operators of following, exchange and transmission are invoked to coordinate the stepwise optimization. KKT optimality analysis of GQ approaches is carried out, which reveals that the architecture and the optimum points of GQ achievement are consistent with lagrangian multiplier methods. Further advantages over the lagrangian multiplier methods are proven based on whether the parameters of lagrangian multiplier and separable objectives are required.PGQ (process goose queue) approaches are responsible for GQ applications in process industries. Similarly, PGQ strategies can decompose a process optimization problem into multi-layer PGQs based on process operational units and then coordinate them by PGQ operators. The disturbances stemmed from process supervision are corresponding to two PGQ deviation situations, which refer to leading goose driven deviation and state goose driven deviation, respectively. Accordingly, two adjustment rules named leading goose driven adjustment rule and formation driven adjustment rule are proposed for adjustment the formation into optimum, which is suitable for steady state optimization and real-time optimization respectively. This thesis particularly addresses a supervision-driven RTO issue concerning the economic performance deterioration caused by process supervision. Therein, the process operational unit whose process variables shifted by human operators can be regarded as an Ill-PGQ, which would trigger the autonomous adjustments and optimization. To implement the process RTO caused by Ⅲ-PGQ emergence, leading goose driven adjustment rule is firstly employed to find the optimums in steady state optimization, which act as the objectives of PGQs in RTO. The formation driven adjustment rule is then employed to find the optimums of RTO. In order to overcome the long-waiting deficiencies for steady state in process RTO, a real-time evolution (RTE) approach is adopted, which divides the process waiting time for steady state into small time intervals, in which the corresponding process state is deemed as pseudo-steady state. Within time partitions, formation driven adjustment rule is employed to find the optimum from the Ill-PGQ to PGQ-Objective.It is worth noting that GQ approaches adopt the heuristic mechanism of information diffusion as well as the gradient operator so that fully utilize the real-time measurements and achieve more accurate optimums by fine mathematical search direction simultaneously. To solve the large-scale system optimization problems, GQ approaches take advantage of the state variables in information diffusion to decompose the nonlinear objective and then model the complex constraints by multi-layer GQ. The optimal operators of following, exchange and transmission are employed to follow up and coordinate the optimums.Furthermore, the supervision-driven RTO issues are first put forward for the process optimization. Two adjusting rules involved in PGQ approaches, named leading goose driven adjustment rule and formation driven adjustment rule are proposed to coordinate the steady-state and RTO problems with the RTE strategy problems respectively. The PGQ approaches are consistent with the concept ot process operation, which provide the decentralized optimization structure for coordination and optimization and it can overcome the algorithmic deficiencies associated with conventional optimization approaches such as huge process models, enormous manipulated variables and long waiting time for detecting steady states.To demonstrate the feasibility and validity of the contribution, TE benchmark process is employed as an extensive case study through PGQ modeling, PGQ steady state optimization and RTO, showing that the proposed approaches could enjoy considerable computational simplicity in contrast with global strategies.