论文部分内容阅读
为了实现极端学习机(ELM)的在线训练,提出一种限定记忆极端学习机(FM-ELM).FM-ELM以逐次增加新训练样本与删除旧训练样本的方式,提高其对于系统动态变化特性的自适应性,并根据矩阵求逆引理实现了网络输出权值的递推求解,减小了在线训练过程的计算代价.应用于具有动态变化特性的非线性系统在线状态预测表明,FM-ELM是一种有效的ELM在线训练模式,相比于在线贯序极端学习机,FM-ELM具有更快的调节速度和更高的预测精度.
In order to realize the online training of Extreme Learning Machine (ELM), a Limited Memory Extreme Learning Machine (FM-ELM) is proposed.FM-ELM improves the dynamic characteristics of the system by adding new training samples successively and deleting old training samples , And according to matrix inversion lemma, the recursive solution of network output weight is realized, which reduces the computational cost of online training process.The on-line state prediction of nonlinear system with dynamic characteristics shows that the FM- The ELM is an effective ELM online training mode, which offers faster tuning speed and higher prediction accuracy than online sequential learning machine.