基于改进遗传算法的支持向量机参数优化方法

摘要

Over-study or under-study phenomenon sometimes happens, since nuclear parameters are chosen inappropriately in re-gression forecasting. The paper proposes a kind of support vector machine parameters optimization model based on improved ge-netic algorithm. By combining genetic algorithm with support vector machine algorithm, the model makes use of the principle of evolutionary of genetic algorithm to optimize penalty parameter, nuclear parameter and loss function at the same time, which are of great significance to support vector machine algorithm. Three sets of standard experiment data sets are selected as the test data set, and simulation test results are compared among the improved algorithm, genetic algorithm, particle swarm optimization algo-rithm and grid search algorithm. Experiment results show that the improved algorithm greatly improves the whole optimization abil-ity of support vector machine algorithm.%针对支持向量机算法在回归预测时由于参数选取不当导致过学习或欠学习的情况，提出一种基于改进遗传算法的支持向量机参数优化模型。该模型将遗传算法与支持向量机结合，利用遗传算法进化搜索的原理对支持向量机具有重要意义的惩罚参数、核参数和损失函数同时优化。实验选取3组标准数据集作为测试数据集，并将改进算法同时与遗传算法、网格寻址算法、粒子群算法进行仿真测试结果对比。实验结果表明改进的算法较大地提高了支持向量机算法整体的寻优能力。