郑州大学学报（理学版）

极限学习机(ELM)是一种新型的前馈神经网络,相比于传统的单隐含层前馈神经网络(SLFN),ELM具有速度快、误差小的优点.由于随机给定输入权值和偏差,ELM通常需要较多隐含层节点才能达到理想精度.粒子群极限学习机算法为使用粒子群算法(particle swarm optimization,PSO)选择最优的输入权值矩阵和隐含层偏差,从而计算出输出权值矩阵.一维Sinc函数拟合实验表明,相比于ELM算法和传统神经网络算法,粒子群极限学习机算法依靠较少的隐含层节点能够获得较高精度.

Extreme learning machine(ELM) was a new type of feedforward neural network.Compared with traditional single hidden layer feedforward neural networks,ELM possessed higher training speed and smaller error.Due to random input weights and hidden biases,ELM might need numerous hidden neurons to achieve a reasonable accuracy.A new ELM learning algorithm,which was optimized by the particle swarm optimization(PSO),was proposed.PSO algorithm was used to select the input weights and bias of hidden layer,then the output weights could be calculated.To test the validity of proposed method,two simulation experiments were drawn on the approximation curves of the Sinc function.Experimental results showed that the proposed algorithm achieved better performance with less hidden neurons than other similar methods.