摘要
针对专家制定作战方案训练样本时容易受主观性影响的问题,提出了基于ELM的样本验证及评估方法。首先根据ELM建立作战方案样本的预测模型,然后更正错误标记的样本。仿真实验表明,利用ELM模型训练更正后的样本集能有效降低均方根误差值和提高预测的准确率。与RBF神经网络相比,训练ELM模型的时间缩短了98.8%,而且无需调节激活函数的参数就可以得到足够好的泛化性能。
In order to overcome the deficiency of subjectivity in deciding the label of battle scheme training samples by experts, a verification and evaluation method based on extreme learning machine （ELM） is proposed. The prediction model of battle scheme samples is constructed based on ELM and the wrong samples are corrected. The simulation experimental results show that lower root mean square error and better testing rate can be obtained by training ELM model on correct samples. In contrast with RBF neural network, the training time of ELM is reduced by 98.8% , and a good generalization ability can be obtained without parameters of activation function needing to be adjusted.
出处
《现代防御技术》
北大核心
2015年第4期204-209,共6页
Modern Defence Technology
关键词
超限学习机
径向基函数
作战方案
评估
泛化性能
extreme learning machine （ ELM ）
radial basis function （RBF）
battle scheme
evaluation
generalization ability
