Differential Evolution Alogrithm With Adaptive Mutaion Strategy And Its Application

With the development of science and technology, optimization has become an important research field in control theory. How to find the optimal solution from all possible solutions is the problem which the modern industrial production process must solve. With the raise of the production process and automatic level, the optimization objects have been changed from the simple linear optimization problem into the complex, high-dimensional nonlinear optimization problems. Traditional optimization methods are difficult to obtain the global optimal solutions for this kind of problems. However, with the good global search capability, modern intelligent optimization algorithms give a new way for solving these problems.Differential evolution (DE) algorithm which is one of intelligent optimization algorithms is famous for its parallel, direct random search and has the strong optimizition ability. Control parameters, the mutation strategies and other factors greatly affect the performance and search efficiency of the DE algorithm. In this paper, it is studied a class of DE algorithms with adaptive strategies, by learning the experience of historical evolution to implement the adaptive mutation strategies, and the adaptive control parameters method is further enhanced the global search ability and convergence speed. In addition, this article attempts to do some works on the adaptive search space. According to the simulation results, it shows that the proposed three algorithms are highly robust and practical.Firstly, by learning from the current estimation of distribution algorithm which is based on the best individuals in the search space, building a feasible solution of the probability distribution model to guide the global search for the best individual, this paper proposes a population-based distribution of adaptive variable space DE algorithm (AVSDE). AVSDE algorithm uses the variable space to guide some individuals in search space, randomly selects mutation strategy and adaptive control parameters improve the global search capability for the algorithm. The results of 19 benchmak functions with different dimensions show that AVSDE algorithm has strong robustness.Secondly, according to the statistical learning of transferring different mutation strategies in the evolutionary process, updating the transition probability between different strategies to achieve the adaptive transfer strategies during a learning cycle, this paper presents an adaptive transfer DE algorithm (ATSDE). The results of 14 benchmark functions show that ATSDE algorithm has strong capability for optimizating, especially for the low-dimensional function optimization problems.Finally, inspiring by the scheduling algorithm of the computer system employs different process, this paper presents an adaptive scheduling strategy of DE algorithm (ASDE) to employ different mutation strategy without priority scheduling during a period, and different strategies use their own success rates to adjust their holding time. Using the 14 benckmark functions to verify its performance and comparing with the state-of-art adaptive DE algorithms (i.e. SaDE algorithm), ASDE algorithm still keeps the strong competitive ability.Moreover, these three DE algorithms have been used for solving two chemical process dynamic optimization problems, experimental results show that these three algorithms also has a strong ability to solve this kind of problems.