演化策略(英文版).pptVIP

演化策略ES Evolutionary Strategies 程 睿 2015年12月12日 * outline Background Algorithm principle Strategy parameter Strategy parameter types Mutation of strategy parameter Adaptive strategy Evolutionary strategy operator selection operator crossover?operator mutation?operator Application Function and Circle packing problem Background Background of ES * Evolutionary strategy was proposed by I. Rechenbery and H. P. Schweful at the time they investigated the problem of fluid force question in wind tunnel when they studied in Berlin Technology University in Germany 1964. This method produces an optimal solution by using the idea of biological variation to alter randomly the parameter values. The evolution strategy is specially designed for solving parameter optimization problems, where the auto-adapted mechanism is adopted. The concealed parallelism and the community’s globe searching are its two prominent characteristic. In addition, it has a strong robustness and also has the unique superior performance when solving some complex nonlinear systems. （1+1）策略算法 * 第一个进化策略是由于动力学原理问题而被发明的，这个（1+1）-进化策略是最早讲一个个体视为一个被优化的决策向量x和策略向量δ构成的元祖进化算法。 个体表示为如下元组： 子代： 通过如下的高斯扰动产生： Development of ES * μ表示亲代个数，λ表示子代个数 （μ+λ）进化策略：也被称为加法策略，进化策略从μ个亲代和其产生的λ个子代，下一代包括从μ个亲代和其产生的λ个子代中选出最好的μ个个体作为新的亲代。这是一种精英主义，最合适的亲代存活到下一代 （ μ，λ ）进化策略：逗号策略，下一代包括λ个子代产生的最优秀的μ个个体。这不是一种经营策略，相对于加法策略表现出较低的选择压力 Algorithm principle General evolutionary strategy algorithm Evolutionary strategy algorithm proceeds as follows. Confirm the expression of the question. (X, δ) = ((x1, x2, · · xi· · · , xn), (δ 1, δ 2, · · δ i,· · , δ n)) (2) Produce randomly the initial community (3) Calculate the fitness value of each initial individuals (4) A new colony is generated as follows: (a) Recombine: Exchange the information based on the two father generation individuals to produce the new individual. (b) Mutation: δ i’= δ i· exp(l’ · N(0, 1) + l · Ni(0,