融合随机森林决策的海鸥优化算法

A Seagull Optimization Algorithm Incorporating Random Forest Decision Making

针对海鸥优化算法(SOA)在求解优化问题时容易早熟、算法性能过于依赖参数以及解的精确度较低等问题,提出了一种融合随机森林决策的海鸥优化算法(RFSOA)。首先,在海鸥前期迁徙阶段,引入非线性递减参数A对海鸥位置进行改变,调整算法的全局寻优能力,避免在迭代过程前期算法便会陷入局部最优的情况;其次,在海鸥攻击阶段,引入机器学习中随机森林思想,将每个海鸥个体位置看作任一决策树的分支,利用其构造决策树进行海鸥位置的改变,位置变化时采取螺旋状位置更新策略或最优高斯游走策略,增加海鸥位置改变的随机性,并利用贪心策略保留优质个体,从而提高寻优精度。选择Benchmark测试函数及0-1背包问题部分算例进行算法性能测试,结果显示RFSOA具有寻优能力强、精确度高的优点,更适合求解较高维度的连续函数优化问题及带有约束的实际应用问题。

A seagull optimization algorithm (RFSOA) that incorporates random forest decision making is pro-posed to address the problems that the seagull optimization algorithm (SOA) tends to be premature in solving optimization problems, the performance of the algorithm is too dependent on parame-ters, and the accuracy of the solution is low. Firstly, in the premigration stage of the seagull, a non-linear decreasing parameter A is introduced to change the position of the seagull to adjust the global optimization capability of the algorithm and avoid the situation that the algorithm will fall into the local optimum at the early stage of the iterative process;secondly, in the attack stage of the seagull, the random forest idea in machine learning is introduced to consider each individual position of the seagull as a branch of any decision tree and use it to construct a decision tree to change the position of the seagull. The spiral position update strategy or optimal Gaussian wandering strategy is adopted when the position changes to increase the randomness of the gull position change, and the greedy strategy is used to retain the high quality individuals, so as to improve the optimization seeking accuracy. Benchmark test function and some examples of 0-1 knapsack problem are selected to test the performance of the algorithm. The results show that RFSOA is more suitable for solving continuous function optimization problems with higher dimensionality and practical application problems with constraints.