论文部分内容阅读
针对极限学习机(ELM)网络结构优化问题,提出一种改进的灵敏度剪枝ELM(Im SAP-ELM).Im SAP-ELM将2正则化因子引入SAP-ELM中,采用留一准则确定最优隐节点数.推导基于奇异值分解的输出权重计算公式,避免矩阵奇异导致求解无效的问题.将Im SAP-ELM用于故障预测,利用多组同类型故障数据建立多个Im SAP-ELM模型,基于加权思想融合不同Im SAP-ELM的预测值.某型无人机发射机实例表明,相比于ELM、OP-ELM(最优剪枝ELM)和SAP-ELM,Im SAP-ELM耗时最高,但是Im SAP-ELM的预测误差小于其他3种方法.
In order to solve the ELM network structure optimization problem, an improved sensitivity pruning ELM (Im SAP-ELM) is proposed. ImM SAP-ELM introduces 2 regularization factors into SAP-ELM, ELM is used to calculate the output weight of singular value decomposition and to avoid the inequality caused by matrix singularity. Im Immersion-ELM is used for fault prediction, and multiple Im SAP-ELM models are established by using multiple sets of fault data of the same type, Based on the weighted thought fusion of the predictions of different Im SAP-ELMs, an example of a UAV transmitter shows that Im SAP-ELM consumes the most time compared to ELM, OP-ELM and SAP-ELM , But Im SAP-ELM forecast error is less than the other three methods.