Research On Optimization Methods Of Cell-like Membrane Computing And Their Applications

Cell-like membrane computing is a molecular model, with powerful computing capability, which can provide a new solution for solving complex problems. It has been studied and de-veloped in various disciplines, such as control, mathematic and biology, etc. In recent years, scholars have paid attention to the research of cell-like membrane computing in the field of opti-mization. Moreover, how to apply cell-like membrane computing for effectively improving the performance of optimization methods has became a problem deserved to be studied deeply.This research focusses on making a systematic attempt for solving optimization problems by using cell-like membrane computing. Three heuristic optimization methods, which are based on cell-like membrane computing, are proposed to solve single-objective, multi-objective and dynamic optimization problems, respectively. These methods are validated by using benchmark test functions and the practical problems in basic oxygen furnace. The thesis is divided into three parts:(1) An evolutionary membrane algorithm is proposed, which is based on cell-like membrane computing. Three elements of cell-like membrane computing, including objects, reaction rules and a membrane structure, are improved according to the characteristics of the optimization problems. In the skin membrane, a mechanism based on the chaotic system is employed to initiate objects, which can improve the distribution of objects in the search space. In the region of the elementary membrane, an operation is proposed to evolve objects by cellular automata invoking the rewriting rules, which can generate excellent candidate objects. After all reaction rules have been executed, the evolved objects are sent back to the skin membrane. In the region of the skin membrane, objects from different elementary membranes are regrouped, and new object sets are formed. The object sets are useful to share information among different elementary membranes. Finally, the above-mentioned process is repeated until the termination condition is met. The optimal object in the skin membrane is the global optimal solution of the optimization problems.(2) A multi-objective evolutionary membrane algorithm is proposed to solve multi-objective optimization problems. Unlike with single-objective optimization problems, multi-objective op-timization problems have some different characteristics such as multiple objective functions, and mutual restraint relationship among these objective functions. When the performance of an objective function is improved, the performance of the other objective functions will be de-creased. To solve the above-mentioned multi-objective problems, multi-objective evolutionary membrane algorithm is proposed based on evolutionary membrane algorithm. But, the initial-ization of objects and the design of rewriting rules in the proposed multi-objective algorithm are different from evolutionary membrane algorithm’s. In addition, non-dominated sorting and crowding distance are introduced into the proposed algorithm to maintain diversity and space distribution of candidate objects. An archive object set is designed in the region of the skin membrane, which can avoid the loss of the best object by recoding the object with the best fit-ness throughout the optimization process. The archive object set can accelerate the convergence speed of the proposed algorithm.(3) The thesis also proposes a dynamic evolutionary membrane algorithm to solve dynamic optimization problems. Three elements of cell-like membrane computing are improved accord-ing to the characteristics of dynamic optimization problems. More specifically, four special objects are proposed, and some of them can improve the convergence of the proposed algorith-m, and the others can enhance the diversity of the candidate objects. To calculate the number of elementary membranes, a single chain hierarchical clustering is introduced into the proposed algorithm. In addition, after objects are evolved by using rewriting rules, some of them could be in the overcrowded area, or they jump out of the region of the current elementary membrane. To solve the problem of overcrowding objects, a crowding rule is designed to delete the crowd-ed objects, which can be used to enhance the diversity of candidate objects. An overlap rule is introduced to solve the problem of the objects moving from the current membrane to the other membranes. The overlap rule can improve the search efficiency by merging elementary mem-branes, because these membranes exist the same searching region. Moreover, an environmental detection rule is designed to judge whether a new environment is reached or not. The rule can improve the ability of the proposed algorithm for adapting to the new environment.Finally, benchmark test functions are employed to verify and analyze effectiveness and fea-sibility of the proposed optimization algorithms. In addition, two data-driven models, which is based on the extreme learning machine network, are established to solve two problems. These problems include the prediction of endpoint carbon temperature and the calculation of adding al-loy amount in the steelmaking process of basic oxygen furnace. One of two models is established for a prediction of endpoint carbon and temperature in which evolutionary membrane algorith-m is employed to optimize the parameters of the extreme learning machine network. Another model is established for the calculation of adding alloy amount. According to both amount and cost of adding alloy, a two-objective optimization problem is constructed on the premise that the steel composition of adding alloy is satisfied with the requirements of steel procedure. In addition, the proposed multi-objective evolutionary membrane algorithm is employed to solve the two-objective optimization problem. At last, the amount of adding alloy is obtained from the solution of the two-objective optimization problem. In the simulation, the actual smelting data in the steel factory are utilized to validate the effectiveness of two proposed models.