基于状态转移算法的极限学习机

Extreme Learning Machine Based on State Transition Algorithm

目前极限学习机在训练模型时存在占用计算资源多和模型精度低等问题.为了解决上述问题,提出了一种基于状态转移算法的极限学习机,可提升算法计算效率和模型精度.利用状态转移算法的全局搜索特性求解线性方程组,得到极限学习机的输出权重矩阵,进而完成建模.在分类和回归数据集上与极限学习机和其他主流算法进行对比,所提方法可以利用较少的隐藏层节点得到高精度的模型,同时具有更好的学习准确率.这种高性能的建模方式弥补了极限学习机的不足.

In order to solve the problems of extreme learning machine(ELM), occupying more computing resources and low accuracy during model training, an extreme learning machine based on the state transition algorithm(STA) was proposed to improve the calculation efficiency of the algorithm and the accuracy of the model. Taking advantage of the global search feature of the state transition algorithm, the algorithm was arranged to solve the linear equations, obtain the output weight matrix of the extreme learning machine and complete the modeling. Compared with extreme learning machine and other mainstream algorithms on classification and regression data sets, the proposed algorithm can realize high model accuracy with fewer hidden layer nodes and achieve better learning accuracy. The high-performance modeling method can make up for the deficiencies of the extreme learning machine.