The Models For Estimating The Performance Of Intelligence Optimization Algorithms

As an alternative to the traditional optimization algorithm, great progresses have been obtained in the field of theory and application of intelligence optimization algorithms. But the foundation of the theory is always been blamed for intelligence algorithms. Dividing the optimization theory into the theory of optimization problems and the theory of the optimization algorithms, can promote the developement of these two fields.There are many foundmental questions about the intelligence optimization algorithms, what make kinds of optimization problems solved? How can these algorithms run effectively? In addition, how do establish unified measurements and methods for the intelligence optimization algorithms? To solve these problems, the studies in this paper are shown as follows:1. The principles of the intelligence optimization algorithms are analyzed. In spite of all the theoretic differences, the traditional optimization algorithms and the intelligence optimization algorithms both depend on the same foundation, which is "hill climbing" model. The difference lies in the ways and means, the traditional optimization algorithms are determined, while the intelligence optimization algorithms are probabilistic. According to these conclusions, we find out that the key of the optimization algorithms are the search strategies of the algorithms, and present the unified model of the strategies. Taking Sampling model as the core, assisted by information collection model, the search model can completely summarize the search processes of the traditional optimization algorithms and the intelligence optimization algorithms. By the search model, this paper discusses the sampling models of the intelligence optimization algorithms, and concludes that the sampling models are parameterized probabilistic models. Through analyzing the typical the probability model of intelligent optimization algorithm, the distribution of the sampling model are obtained, which laid the foundation of the evaluation of the algorithm.2. The models and methods of estimation of the intelligence optimization algorithms are studied. As the probabilistic models are the core of the intelligence optimization algorithms, the methods to estimate the probabilistic models of intelligence algorithms are established. The measurements of the methods are the distribution density function and the probability of the optimal solutions. According to the splitting technique, an accurate method for estimating the probability of the optimal solutions set is introduced. The availability of the intelligence optimization operators is presented and the methods for measure the vailability are proposed also. Meanwhile the backwards of the measures of the traditional optimization algorithms are presented. The concepts of the convergence in objective function values, accuracy, the ratio between accuracy and running time are proposed. By studying the performance of the pure random search algorithm, the method based random search algorithm for estimating performance of intelligence optimization algorithms is proposed. The methods to estimate the behaviors of the intelligence optimization algorithms are suggested.3. The principles of the design of the algorithms of traditional optimization and intelligence optimization algorithms are presented. The recognition of the features of the optimization problems is the premise of the design of these algorithms. The estimation of the distribution can get the features of these problems.4. Considering the intelligence optimization algorithms as the method for recognizing the feature of the optimization problems, then the fusion of informations the algorithms obtained can generate new effective algorithms. The experiments show that these novel algorithms have improved the performance greatly.