Research Of Continuous Ant Colony Optimization Algorithm And Its Application In Chemical Engineering

Optimization is an effective technique for improving the performance of complex process system. It can remarkably increase the efficiency, reduce the use of energy, utilize the resource reasonably, and boost the economic benefit. Chemical engineering system is a typical complex system. As the scale of object problem becomes more and more large, of which the model structure becomes more and more complicated too, thus makes it more and more difficult to be optimized, and has been a great challenge to the existing optimizing method. So the require for high efficiency intelligent optimization techniques becoming more and more necessary.Ant Colony Optimization (ACO) was put forward in early 1990s, which is a class of intelligent optimization methods. Its predominant distributed pattern of problem solving achieves great success in combinational problems, and brings extensively attentions of related research area. But to many practical engineering problems, they always are expressed as continuous optimization problems. It is an imperative challenge on how to apply the basic ant colony optimization idea to the problems solving in continuous space, which is the major work of this thesis.In this thesis, aimed at the discreet nature of basic ant colony optimization algorithm, we put forward two kinds of hybrid continuous optimization algorithms based on ACO, which combined the original ideal of ACO with other optimization techniques according the common principles of algorithm hybrid. The optimization experimentations of standard test functions validate that these algorithms have very good global optimization performance. Finally we apply them to the optimization of actual chemical engineering systems, the result is very well, which shows that ACO based continuous optimization algorithms have a goodish potential in chemical engineering optimization area. The major contributions of this work are summarized as follows.1) Aimed at the discrete nature of basic ant colony optimization algorithm, a hybrid continuous ant colony optimization (HCACO) algorithm was constructed. It combined the ACO strategy with Float Genetic Algorithm and introduced Powell method as eugenic strategy. In HCACO, ants were divided as global ants and local ants, and led the individuals of GA's population to do global exploration search and local exploitation search separately, and deposit pheromone on them to increase their attraction of other ants. The algorithm scheduled the behaviors of these two kinds of ants, and coordinated their actions via pheromone, which well balanced the exploration and exploitation of algorithm. All these made HCACO have good efficiency and stability for solving continuous optimization problem and have powerful global optimization ability, which was well validated by the result of optimization of test functions. And then, the parameters of HCACO were discussed in detail based on the optimization tests, and referenced values of them were presented. Finally, HCACO was successfully applied to training of Neural Networks (NN), and established a Neural Networks model for quantitative structure-activity relationships (QSAR) of herbicidal N-(l -methyl-1-phenylethyl) phenylacetamides, and achieved good predicting accuracy.2) A common framework of continuous ACO based on division of ant colony was designed on In-depth analysis of HCACO, which was induced and generalized from the optimization mechanism of HCACO. Under this framework, a new algorithm named Evolutionary Programming—Ant Colony Optimization (EP-ACO) was put forward. It introduced Evolutionary Programming in global exploration and Pattern Search Method in local search. The greatest characteristic of EP-ACO is its concision. The results of optimization experiments showed that EP-ACO exhibited high optimization efficiency and excellent global capability. Comparing with HCACO, the concision of EP-ACO improved the optimization efficiency evidently, while the optimization results was still very good, and stability of results was some enhanced. Finally, EP-ACO was successfully applied to the operation optimization of RBF-MCSR model of the equipment of xylene isomerization.3) Inspired by the hunting behavior of biological ants in continuous space. A hypothesis was proposed that the distribution of pheromone on continuous space was normal distribution. With this hypothesis, ants could directly move in the continuous space according to the pheromone, which lay a foundation for applying ACO idea to continuous optimization problems. The normal distribution of pheromone simulated the congregating of ants around the current optimal ant, and reflected the basic principle of ACO idea that "the concentration of the search around the best solutions found during the search."4) Based on the normal distribution hypothesis of pheromone, a model based continuous ant colony optimization algorithm - Hybrid Continuous Ant Colony System (HCACS) was put forward, in which the goal of ant colony is not to find the shortest path but the optimal food source. In HCACS, ants could use a random generator with pheromone's normal distribution function as the state transition rule to choose the next point to move to, and pheromone was updated through adjusting parameters of the distribution according to transitions of ant colony. Ants congregate gradually around the optimal food source under the direction of pheromone, and the congregation of ants also changes the distribution of pheromone, thus comes to the positive feedback mechanism, which is the major characteristic of ACO. Further more, HCACS integrates the eugenic strategy and mutation strategy, which enhance the exploitation and exploration of ant colony separately. HCACS has fewer control parameters, which can be set easily. Finally, HCACS was successfully applied to the nonlinear parameter estimation of a mathematical kinetic model for Heavy Oil Thermal Cracking.5) With in-depth analysis of creation mechanic of new candidates, the continuous ant colony optimization algorithms of this thesis were classified to two categories according to Quinlan's classifying rules of Meta-Heuristic methods: HCACO and EP-ACO belonged to instance-based continuous ant colony optimization algorithms, while HCACS belonged to model-based continuous ant colony optimization algorithms. However, for the reason of algorithm hybrid, in HCACO andEP-ACO, it partly exhibited the characters of model-based method, and in HCACS, it partly exhibited the characters of instance-based method.