Ant Colony Optimization with Stepwise Localization of the Discrete Search Space to Solve Function Optimization
Ant Colony Optimization with Stepwise Localization of the Discrete Search Space to Solve Function Optimization
复制标题
通过离散搜索空间逐步定位的蚁群优化来求解函数优化
DOI：
10.1109/icmla.2017.00-78
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
Ryouei Takahashi and Yukihiro Nakamura
中科院分区：
文献类型：
--
作者：
Yuya Kaneda;Yan Pei;Qiangfu Zhao;Yong Liu;Asaki Saito and Akihiro Yamaguchi;Yukiko Yamamoto; Setsuo Tsuruta; Takayuki Muranushi; Yuko Hada-Muranushi; Syoji Kobashi; Yoshiyuki Mizuno; Rainer Knauf;Mikawa Masahiko;Takuya Yoshimoto and Hiroyuki Torikai;笹山 友裕，伊藤 秀昭，福本 尚生，和久屋 寛，古川 達也;Ryouei Takahashi and Yukihiro Nakamura
提出了蚁群算法在求解函数优化问题中的一种新应用——改进蚁群算法（i-EAS）。i-EAS是针对旅行商问题（TSP）而设计的精英蚁群系统（EAS）的改进。它能够逐步定位搜索空间。在这里，我们研究了通过使用离散值（二进制数据）逼近在实空间Rn中搜索解的方法。我们提出的蚁群算法通过逐步定位搜索空间来提高解的准确性。为了定位当前步骤的搜索空间，我们的蚁群算法在前一步找到的最佳解β的邻居中递归地搜索解。我们假设α是搜索空间局部化的次数，然后我们缩减搜索空间使R（α）满足方程R(α) = RANGE × （1 / 2）（α × ln α），其中RANGE作为初始值提供。虽然减小了每个自变量的定义域，但可观测点的个数为2l，且不随α变化，这意味着可以提高解的精度，使得d(α) = RANGE × （1 / 2）（α × ln α）+l-1，其中d（α）为可观测数据的区间。为了进一步提高解的准确性，我们在每只蚂蚁每次完成搜索解的旅程时对它找到的解执行变异操作。此外，为了保持种群多样性，我们动态地循环改变精英蚁信息素的权重。采用多峰标准测试函数验证了i-EAS的有效性。
A new application of Ant Colony Optimization (ACO) called improved-EAS (i-EAS) is proposed for solving the function optimization problem. i-EAS is an improvement on the elitist ant system (EAS) which was devised to solve the Travelling Salesman Problem (TSP). It is capable of stepwise localization of the search space. Here, we examined methods that search for solutions in the real space Rn by approximating them using discrete values (binary data). Our proposed ACO uses stepwise localization of the search space to improve the accuracy of solutions. To localize the search space on the current step, our ACO searches for solutions recursively in neighbors of the best solution β found on the previous step. We assume that α is the number of times that the search space is localized, then we reduce the search space so that R(α) satisfies the equation R(α) = RANGE × (1 / 2)(α × ln α), where RANGE is provided as an initial value. Although the domain of each independent variable is reduced, the number of observable points is 2l and does not change according to α, which means that solution accuracy can be improved so that d(α) = RANGE × (1 / 2)(α × ln α)+l-1, where d(α) is the interval of the observable data. To improve solution accuracy further, we perform mutation operations on the solution found by each ant every time it finishes a tour in search of solutions. Furthermore, in order to maintain population diversity, we dynamically altered the weight of elitist ant pheromone cyclically. The validity of i-EAS is verified by using well-known standard test functions with multiple peaks.