加速并行遗传算法及其在暴雨强度公式参数优化中的应用

Application of acceleration parallel genetic algorithm to estimating parameters of storm intensity formula

针对标准遗传算法中存在早熟收敛、后期收敛速度慢以及解精度低的问题,结合正交试验设计和元胞自动机模型,提出了一种改进的遗传算法———加速并行遗传算法(APGA)。APGA利用正交试验设计确定较好的初始种群,利用元胞自动机模型固有的并行计算能力设计并行遗传算法,借助元胞信息的动态性和多元性实现正交加速过程。该方法不仅能加快遗传算法的收敛速度,而且能提高遗传算法的搜索效率和解的精度,适用于求解复杂的非线性优化问题。APGA用于暴雨强度公式参数的优化实例结果表明,该算法是有效和可行的。

The present paper aims to introduce an improved algorithm known as the acceleration parallel genetic algorithm for global numerical optimization with continuous variables. It comes from the necessity to resolve the currently existing problems of simple genetic algorithm such as premature convergence, low speed of later convergence and its rough results. The paper＇ s goal is to apply the method of experimental design to enhance the genetic algorithm so as to make the resuiting algorithm more robust and statistically sound. In doing so, a quantization technique is suggested to make up for the empirical method called orthogonal design. The authors intend to use the method to generate an initial population points that are scattering uniformly over the feasible solution space so as for the algorithm to be able to scan the feasible solution space more evenly once the good points for further exploration are located in the subsequent iterations. In addition, the model of cellular automata has been inserted to the genetic algorithm, where the Moore model of 2D cellular automata is used. As a matter of fact, every cellular has eight neighbors, whose boundary rule can be defined as follows： topside cellular joins with nethermost cellular, leftmost cellular joins with rightmost cellular, and so on and so forth. Furthermore, each cellular and its neighbors confirm a new sub-feasible solution space. When the algorithm fulfils its parallel circulation number, it will confirm the new sub-feasible solution space in conformity with its best offspring. And now the orthogonal design can be used continuously to generate an initial population of points until the algorithm fulfils the orthogonal acceleration circulation number. It is clear that the intrinsic capacity of cellular automata can be used here to design the parallel genetic algorithm while the dynamic and multiple of cellular information can be applied to achieve orthogonal acceleration process. Thus, it can be concluded that not only can the APGA help to increase the convergence speed but also it contributes to heightening the search efficiency and improving the result precision of genetic algorithm so as to optimizing the parameters of the storm intensity formula by using the APGA, The application results show that the newly improved algorithm both practical and efficient.