Research On Evolutionary Algorithms Inspired By Prey-Predator Model

Many complex problems in human practice can be converted into optimization problems. Therefore, an efficient and effective approach for solving optimization problems has great practical significance. The conventional mathematical optimization method is effective for optimizing problems with continuously differentiable objective functions. However, it may not be suitable for handling the increasingly complicated problems in science and engineering. In recent years, evolutionary algorithms, due to the characteristics of its intrinsic parallelism, self-organization, as well as self-adaptive, and self-learning, have attracted considerable attentions from both academic and industrial communities, and been successfully utilized in many fields in science and engineering. However, since evolutionary algorithms are essentially stochastic techniques with heuristic information, they tend to suffer from premature convergence and slow convergence rate when solving complex problems that are nonlinear, multimodal, and high-dimensional. The main reason is that the stochastic feature of evolutionary algorithms may lead to a lack of diversity in the population during the evolution process, which may result in trapping into local optima. In addition, promoting population diversity can increase the probability of finding the global optimum, but this may lead to week selective pressure and thus slows down the convergence speed. Hence, there is a conflict between population diversity and convergence speed in evolutionary algorithms. Moreover, it is a difficult issue in evolutionary algorithms to achieve a good trade-off between population diversity and convergence speed. Most research on designing a robust evolutionary algorithm focuses on how to keep a good balance between population diversity and convergence speed. In order to overcome these drawbacks of evolutionary algorithms to a certain degree, learning from nature as a starting point, prey-predator model (PPM) in Ecology is employed to mimic the relationship between population diversity and convergence speed so that evolutionary algorithms can obtain a trade-off between population diversity and convergence speed to some extent. The main research work and innovations of the dissertation are summarized as follows.(1) An evolutionary algorithm framework inspired by prey-predator model is proposed. Aiming at the difficulty of keeping a balance between population diversity and convergence speed, we presented a new evolutionary algorithm framework inspired by prey-predator model by analyzing the existing algorithms based on prey-predator model. The core idea of the proposed framework is learning from nature and mimicking the prey-predator model in Ecology. In the proposed framework, the population is dynamically divided into prey subpopulation and predator subpopulation, and then the sizes of these two subpopulations are determined by the prey-predator model according to the evolutionary state to obtain a trade-off between population diversity and convergence speed to some extent.(2) A particle swarm optimization algorithm based on prey-predator model (PPMPSO) is proposed. The conventional particle swarm optimization (PSO) algorithm tends to suffer from premature convergence and slow convergence rate when solving complex problems. In order to improve the performance of the traditional PSO algorithm, we present an enhanced PSO algorithm based on the proposed evolutionary algorithm framework inspired by prey-predator model. The performance of different operators of predator in PPMPSO algorithm is evaluated on13benchmark test functions, and then the validity of PPM for PPMPSO is verified. In addition, PPMPSO algorithm is also compared with other PSO algorithms. The experimental results illustrate that PPMPSO can obtain better performance in terms of solution accuracy, convergence rate and stability.(3) A differential evolution algorithm based on the prey-predator model (PPMDE) is proposed. The traditional differential evolution (DE) algorithm is prone to suffer from slow convergence and trapping into local optima when solving complicated problems in science and engineering. For the sake of improving the performance of the traditional DE algorithm, we present an enhanced DE algorithm based on the proposed evolutionary algorithm framework inspired by prey-predator model. The performance of PPMDE algorithm is compared with several DE algorithms on CEC2005contest instances. The experimental results indicate that PPMDE algorithm performs better in terms of the quality of the final solutions and the convergence rate. Moreover, the framework of PPMDE is combined with two improved DE variants, i.e., jDE and JADE. The experimental results demonstrate that the framework of PPMDE can also be used for improving the performances of jDE and JADE.(4) A prey-predator model based discrete differential evolution (PPMDDE) algorithm for solving combinatorial optimization problem is proposed. The0-1knapsack problem is a classical combinatorial optimization problem. However, the conventional discrete differential evolution (DDE) algorithm tends to suffer from premature convergence and slow convergence rate when solving0-1knapsack problem. In order to enhance the performance of the traditional DDE algorithm for solving0-1knapsack problem, we present an improved DDE algorithm based on the proposed evolutionary algorithm framework inspired by prey-predator model. The performance of PPMDE is evaluated on a suite of0-1knapsack test problems and compares with several combinatorial optimization algorithms. The comparisons show that PPMDE algorithm can obtain better performance.