摘要
针对传统自适应集成极限学习机预测算法中集成权值更新不充分,受人为因素影响较大所导致的集成模型预测精度较低的问题,提出一种基于方差自适应集成极限学习机(Variance Adaptive Ensemble of Extreme Learning Machine,VAE-ELM)的时间序列预测算法。该算法以最小化预测误差为目标,根据各个弱学习机的预测误差,通过反复迭代自适应地对其集成权值进行多次更新,按照最终的集成权值向量集成各个弱学习机得到最终输出。时间序列的仿真结果及液压泵状态参数预测实例表明,与E-ELM和AE-ELM算法相比,该算法鲁棒性强,预测精度更高。
The ensemble weights of conventional adaptively ensemble prediction algorithm of Extreme Learning Machine are updated in- adequately and tended to be affected by the human factor. This leads to the prediction accuracy been reduced. An algorithm named Vafiance Adaptive Ensemble of Extreme Learning Machine-VAE-ELM is proposed to solve this problem. To minimize the prediction error, the ensemble weights are calculated adaptively based on the prediction error of the weak learning machines for many times. The final result is obtained by the weighted integration of the output of the weak learning machines according to the ensemble weights vector. The results of the time series simulation and the hydraulic pump prediction show that, comparing with the E-ELM and AE-ELM algorithm, the VAE-ELM algorithm shows stronger robustness property and higher prediction accuracy and is suitable to time series prediction.
出处
《控制工程》
CSCD
北大核心
2014年第4期529-532,共4页
Control Engineering of China
基金
军内科研项目
关键词
自适应
集成
极限学习机
时间序列预测
液压泵
adaptive
ensemble
extreme learning machine
time series prediction
hydraulic pump