Approach And Application Of Evolutionary Computation For Optimization And Modeling

Evolutionary Computation is a heuristic method which mimics the evolutionary process of nature biology according Darwin’s theory of natural selection,"the Survival of the Fittest". It uses group search strategy. It estimates the individuals in the population and renews the population with the selection iteratively until the optimal solution appears. This method has the features of self-organization, self-adaptive, and self-learning and these features make the method capable of settling a large number of difficult problems which cannot be approached by the traditional methods. This intelligent method was used widely in many domains.At first this thesis introduces the development of the evolutionary computation and some of its major branch. We focus on applying the evolutionary computation in the areas of auto programming and modeling, numeric optimization, multi-objective optimization and time series analysis. We mentioned some problems in these areas. We had studied these problems and designed the solutions to settle them.In the second chapter we proposed a new evolutionary multi-objective optimization algorithm with elite-subspace strategy. We also use rank-based fitness and niche strategy to keep convergence of algorithm and diversity of the solutions.In the third chapter we present a new gene expression programming algorithm which could use tree-based genetic operators. We noticed that the linear chromosome makes it is easy to design the genetic operators in gene expression programming algorithm and makes the algorithm speedy as well as the parse tree in genetic programming makes the operators in GP more purposive. So we discussed how to use tree-based genetic operators in GEP and designed a new GEP with these operators.We present a new evolution strategy algorithm which using Generalized Predictive Control(GPC) to adapt the global step size in chapter4. In our method, the evolution strategy algorithm is regarded as a controlled system and modeled as a CARIMA model. We calculate the model’s parameters and then the current global optimum step size is calculated by the GPC to feed back to evolution strategy algorithm. Simulation on various test functions reveals local and global search properties of the evolution strategy with GPC.In the fifth chapter we propose a new evolutionary threshold autoregressive modeling algorithm. We analyzed the arguments of the threshold autoregressive model and designed a simple evolutionary algorithm to search the optimal value of them. We tested this method with data of Canada lynx. The model which founded by this method can reveal the real scene of the lynx number’s fluctuation. It also can make a good prediction.