基于改进烟花算法的ELM分类模型

The Extreme Learning Machine Classification Model Based on Improved Fireworks Algorithm

针对极限学习机(Extreme Learning Machine,ELM)的性能必须依赖于大量隐层节点的问题,提出了基于改进烟花算法(Improved Fireworks Algorithm,IFWA)的ELM分类模型。用改进的烟花算法进行迭代搜索,求得N个最优的烟花;选择ELM测试数据集的RMSE作为改进烟花算法的适应度值函数,来优化ELM每个隐层节点的输入权值和偏置,使得节点的决策水平提高,从而使ELM的决策性能显著提高;采用KDD99数据集验证表明:改进烟花算法的极限学习机(IFWAELM)能够以较少的隐层节点得到更高的测试平均正确率,提高了极限学习机的泛化性能。5种同类算法性能对比实验也表明IFWAELM是效果最优的。

The performance of extreme learning machine must depend on a large number of hidden layer nodes,the extreme learning machine classification model of improved fireworks algorithm is proposed. The improved fireworks algorithm obtainsn optimal fireworks by iteration,and selects the root mean square error of the elm test data set as the fitness function to improve the fireworks algorithm,and to optimize the input weight and bias of each hidden layer node of elm and to improve the decision level of the node,so that the performance of elm can be significantly improved.Experimental results on KDD CUP 99 datasets show that IFWAELM can get higher test mean accuracy with fewer hidden nodes,the generalization performance of extreme learning machine is improved. The performance comparison experiment of 5 kinds of similar algorithms also shows that IFWAELM is the most optimal.