论文部分内容阅读
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。
Aiming at the problem that the integrated model prediction accuracy is low due to the inadequate updating of the integrated weights and the impact of the human factors on the traditional adaptive integrated limit learning machine prediction algorithm, a Variance Adaptive Ensemble of Extreme Learning Machine, VAE-ELM). In order to minimize the prediction error, this algorithm adaptively updates its integrated weights repeatedly according to the prediction errors of each weak learning machine, integrates each weak learning machine according to the final integrated weight vector to obtain the final output . The simulation results of time series and the prediction of hydraulic pump state parameters show that the proposed algorithm is more robust and predictive than the E-ELM and AE-ELM algorithms.