Heuristic Knowledge Based Evolutionary Algorithms For Complex Constrained Optimization Problems

With the increasing of complexity and size of constrained optimization problems, constrained optimization problems have become challenging tasks in scientific research and engineering practice. Meanwhile, researchers have obtained more knowledge of the constrained optimization problems. Therefore, it is more necessary for researchers to use such knowledge to solve constrained optimization problems, especially the complex ones.The main contributions of the thesis are listed as follows:1. The historical and current research progress of constrained optimization problems, evolutionary algorithms, knowledge incorporation in evolutionary algorithms, and constrain handling approaches in evolutionary algorithms are briefly reviewed.2. A constraint handling approach based on the heuristic knowledge of the location of optimal solutions is proposed to solve constrained optimization problems where the optimal solutions are at the boundary of the feasible region. Active constraints are distinguished by the feasible solution and infeasible solution in the population of an evolutionary algorithm. According to these active constraints, a constrained optimization problem is transformed into a multi-objective constrained optimization problem by adding an active constraint helper-objective. This method incorporates the knowledge of active constraints into the selection operation of an evolutionary algorithm. This method can improve the efficiency of an evolutionary algorithm and guide its search away from local optima.3. A constraint handling approach based on the heuristic knowledge of feasible regions of constrained optimization problems and constraints is proposed to solve the constrained optimization problems with low feasibility. A new concept of the relative feasibility degree of a solution candidate is proposed to represent the amount by which the "feasibility" of the solution candidate exceeds that of another. Furthermore, relative feasibility degree based selection rules are also proposed. The rules can be used to make evolutionary computation techniques accelerate the search process of reaching feasible region.4. A set of interior penalty rules based on the heuristic knowledge of the superiority of feasible solutions is proposed to improve the efficiency of evolutionary algorithms in solving constrained optimization problems with lower feasibility. Interior penalty rules punished the feasible solutions to increase the probability of generation of feasible solutions. The theory validity of these rules is analyzed based on the relationship between the objective and the probability of a child generated by a feasible solution being feasible.5. A feasibility degree of the building block based on the heuristic knowledge of partial solutions is proposed to solve sparse constrained optimization problems. A new concept of the feasibility degree of the building block of feasible solutions is proposed to measure the probability of an individual containing the building block of feasible solutions being a feasible solution. In this way, the feasibility degree information can be used as heuristic knowledge to guide the search process in constructing an offspring and preserving the generated partial solutions which satisfy some constraints. The theoretical validity of the new crossover operators is proved by the schema theorem and the mixing ladder climbing model.6. The active constraint aided-objective method has been used to solve gasoline blending recipe optimization problems based on the characteristics of these problems. This method can effectively meet the quality control requirements and improve the economic benefits of the oil refinery. The application of the strategy is demonstrated by a case study.In the end of this dissertation, some suggestions about further researches in these fields are also provided.