Knowledge-Based Multi-Agent Mind Evolutionary Algorithm And Its Applications In Engineering

As a new generation of intelligent computing methods, evolution computing, rough set and granular computing have not only made the remarkable progress in the fields of their own, but also the fusion of them will powerfully improve the development and application in the soft intelligent information processing technique. Mind evolutionary algorithm(MEA) belongs to evolution computing and it imitates the thought mechanism of mankind. However MEA does not form the explicit knowledge carrier, knowledge processing system and thinking model. It does not embody the essential trait of people's thinking activities. From intelligent computing point of view this thesis introduces the intelligent agent and knowledge discovery technique into the evolution computing and forms a hybrid intelligent system (HIS). The generated data in the evolution process can be regarded as the knowledge database. Knowledge in the unsolved problem can be acquired through the knowledge discovery based on rough set theory and granular computing theory. It forms the knowledge-based multi-agent mind evolutionary algorithm (KMMEA). The main work and the innovative achievements of this thesis can be concluded as follows:On the basis of the basic mind evolutionary algorithm, the knowledge-based multi-agent mind evolutionary algorithm was put forward. The KMMEA takes the agent as the carrier of the thinking activity. As agents rough set and granular computing are used as the tools to discover the knowledge, to inference and to make the decision. Furthermore, the specific evolution operators are applied to realize the belief, the desire and the intention of the agent, which form the multi-agent evolution system with the granular-hierarchy structure.Through the concepts in rough set the solution space and the objective function spaces can be divided into the subspace granule and objective function granule in the form of equivalence classes. This thesis had put forward the concepts as the relationship between the individual variable and the objective function, granular fitness landscape and individual eigenvector which help one to understand and master the internal law of the problem. The internal law can be taken as the knowledge to describe the characteristic of the unsolved problem. Through the binary granular computing, the relationship between individual variable and the objective function can be obtained fast, and then the characteristic of the individual variable and the granular fitness landscape can be calculated to judge the type of the unsolved problem. In terms of the achieved knowledge to acquire the optimizing subspace, it can reduce the searching scope and enhance the optimization efficiency and precision.Aiming at the complication of multi-objective problem, the application of KMMEA has been studied and discussed in the multi-objective optimization by the dominance relation and dominance classes in the information system. The solution space can be divided by the dominance relation of the information system and forms the evolution strategy which mines in the Pareto dominance space and searches in the rest space. The characteristic of subspace granule and objective function granule can be applied to maintain the population diversity in order to find the global and uniform-distributed Pareto optimum.The thesis put forward dominance granule multi-objective sorting algorithm(DGSA). The dominance granule can be obtained by the dominance relation in the information system and granulation computing. It is the basis of multi-objective sorting and fitness assignment. Therefore, the dominance granule multi-objective sorting algorithm is designed and reduces the computational complexity highly.Two important operators 'similar-taxis' and 'dissimilation' in the MEA are reconstructed based on knowledge discovered in evolution process. It enhances the intelligence of operator which can be used in macro-population evolution. This thesis introduced the mutation operator on the basis of knowledge guidance which is the specific application of granular fitness landscape and individual eigenvector. The mutation operator can enhance the algorithm convergence effectively and the optimization precision. Meantime, new individual generation method is researched according to the dominance-class model.Furthermore, KMMEA's application in engineering is researched. KMMEA was used in neural network optimization in the multi-sensor fusion system and nonlinear updating of voltage-controlled oscillator in the linear frequency modulation radar level meter. It can remarkably improve the frequency-modulation linearity of voltage-controlled oscillator and depress the influence of spectrum spread and phrase noise to distance resolution. KMMEA enhances the signal-to-noise ratio of measurement signals. The application has been granted by the national invention patent(patent No. 200410092447.2) .The main idea of the thesis is to research the application of knowledge discovery technique in the evolution process, guide the evolution process by the discovered knowledge and form the hybrid intelligent computing system.