Optimization Of Uncertain Systems With Interval Parameters And Its Application In Gasoline Blending

The traditional optimization theory has been successfully applied into the various fields, but in fact the coefficients of the models are usually uncertainty and imprecision. The optimal decisions based on these imprecise data will result in poor solution quality. The optimization theory under uncertainty has attracted a lot of attention in recent years. Three major approaches have been proposed for decision making problems under uncertainty: stochastic programming, fuzzy programming and interval programming.To deal with the static and dynamic systems involving interval coefficients, the thesis proposes the deterministic interpretation of the uncertain optimization problems, and solves the transformed nonlinear programming, using the master-slave parallel genetic algorithm, and applies the proposed algorithm into the gasoline blending problem. The main research work is depicted as the follows.1. This thesis proposes the deterministic interpretation of interval nonlinear programming. This formulation shows the degree of the risk which the decision maker should undertake when pursuing the maximum objective function value, and the error size of the objective function value due to the uncertain parameters as the decision makers expect.2. This thesis proposes the deterministic interpretation of dynamic optimization involving the interval parameters, puts forward the path constraint satisfactory factors, analysises the feasibility problem of the path constraint relaxation, and gives the computation method of the path constraint satisfactory factors.3. After transforming the interval nonlinear programming into the minimax optimization problem, the thesis proposes the master-slave parallel genetic algorithm to solve the large scale nonlinear programming. Compared with the traditional master-slave parallel genetic algorithm, the presented algorithm can solve the problem of unbalanced distribution of the computational load among the slave computers.4. The thesis proposes the interval programming formulation of the gasoline blending problem, gives the solution of the gasoline blending problem of an oil refinery plant, and discusses the performance of the definite gasoline blending algorithm and the interval gasoline blending algorithm under the fluctuation of theoctane numbers of the feedstock streams. The results show that the interval gasoline blending algorithm is capable of decreasing the sensitivity of the optimal solution under uncertainty, and ensures the desired product quality and profit.5. To improve the performance of solving the nonlinear programming problem, the thesis proposes a new hybrid genetic algorithm which combines a genetic algorithm with sequential linear programming. During the iterative computation process, if the iterative points in the genetic algorithm do not obtain crossover or mutation operation, the objection function and constraints at these points will be linearized, and solved by sequential linear programming.