基于Lognormal-GMM模型的船舶电力监测网络流量建模方法

A Traffic Modeling Approach for Ship Electrical Power Monitoring Network Based on Lognormal-Gaussian Mixture Model

为解决船舶电力监测网络中冲击性、周期性流量负荷给传统的基于Hurst参数的流量建模方法带来的模型误差问题,给出了基于Lognormal-GMM模型的船舶电力监测网络流量模型。为辨识该模型参数,提出了基于变分贝叶斯近似传播聚类的辨识方法。将流量波形等时分段;使用具有分布估计能力的变分贝叶斯理论辨识出每个时间段上的Lognormal流量参数分布函数;使用具有自适应聚类中心识别能力的近似传播聚类算法将分布函数聚类,得到高斯混合模型的混合参数。实验结果表明,基于上述方法的流量模型的拟合性比基于Hurst参数的流量模型更优。

To solve the model error problem in the Hurst parameter based traditional traffic modeling method brought about by impactive and periodical traffic load, a traffic model for ship electrical power monitoring network based on Lognormal-Gaussian mixture model （GMM） is given. To identify the parameters of this model, an identification method based on variational Bayesian affinity propagation clustering is proposed. In the proposed method, the waveform of the traffic is divided into equitime intervals; then the distribution function of Lognormal traffic parameters in each time interval is identified by variational Bayesian theory that possesses the ability of distributed estimation; finally, using affinity propagation clustering algorithm that can adaptively identify the clustering center, the distribution function is clustered to achieve the mixture parameters for GMM. Experimental results show that the fitting performance of the traffic model based on the proposed method is better than that of the traffic model based on Hurst parameter.