CAJ | 학술논문

极限学习机在实际应用中具有学习速度快、训练误差小的优点，但其稳定性与泛化能力却较差。针对其缺点，将自适应差分进化算法引入极限学习机对其改进，利用自适应差分进化算法的全局寻优能力，求取训练误差较小时极限学习机的输入权值矩阵以及隐含层偏置矩阵，从而优化极限学习机。将优化后的极限学习机应用于球磨机料位测量，实验结果表明，优化后的极限学习机与传统极限学习机相比具有较高的测量精度和较好的稳定性。
겁한학습궤재실제응용중구유학습속도쾌、훈련오차소적우점，단기은정성여범화능력각교차。침대기결점，장자괄응차분진화산법인입겁한학습궤대기개진，이용자괄응차분진화산법적전국심우능력，구취훈련오차교소시겁한학습궤적수입권치구진이급은함층편치구진，종이우화겁한학습궤。장우화후적겁한학습궤응용우구마궤료위측량，실험결과표명，우화후적겁한학습궤여전통겁한학습궤상비구유교고적측량정도화교호적은정성。
The advantages of extreme learning machine has strong learning capacity and smaller training deviation. To further improve the reliability and decreasing the test deviation of extreme learning machine,self-adaptive differential algorithm was intro-duced to extreme learning machine.Then smaller test deviation of sample sets was acquired in this way.Lastly,optimized extreme learning machine was applied to measure the ball mill material level.The experiment result shows that the test deviation and training deviation of this method are largely smaller than that of extreme learning machine.At the same time,learning capacity and generali-zation performance of this method are also better than that of original extreme learning machine.