Cargando…

Intelligent Network Applications Monitoring and Diagnosis Employing Software Sensing and Machine Learning Solutions

The article presents a research in the field of complex sensing, detection, and recovery of communications networks applications and hardware, in case of failures, maloperations, or unauthorized intrusions. A case study, based on Davis AI engine operation versus human maintenance operation is perfor...

Descripción completa

Detalles Bibliográficos
Autores principales: Minea, Marius, Dumitrescu, Cătălin Marian, Minea, Viviana Laetitia
Formato: Online Artículo Texto
Lenguaje:English
Publicado: MDPI 2021
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8348484/
https://www.ncbi.nlm.nih.gov/pubmed/34372272
http://dx.doi.org/10.3390/s21155036
Descripción
Sumario:The article presents a research in the field of complex sensing, detection, and recovery of communications networks applications and hardware, in case of failures, maloperations, or unauthorized intrusions. A case study, based on Davis AI engine operation versus human maintenance operation is performed on the efficiency of artificial intelligence agents in detecting faulty operation, in the context of growing complexity of communications networks, and the perspective of future development of internet of things, big data, smart cities, and connected vehicles. (*). In the second part of the article, a new solution is proposed for the detection of applications faults or unauthorized intrusions in traffic of communications networks. The first objective of the proposed method is to propose an approach for predicting time series. This approach is based on a multi-resolution decomposition of the signals employing the undecimate wavelet transform (UWT). The second approach for assessing traffic flow is based on the analysis of long-range dependence (LRD) (for this case, a long-term dependence). Estimating the degree of long-range dependence is performed by estimating the Hurst parameter of the analyzed time series. This is a relatively new statistical concept in communications traffic analysis and can be implemented using UWT. This property has important implications for network performance, design, and sizing. The presence of long-range dependency in network traffic is assumed to have a significant impact on network performance, and the occurrence of LRD can be the result of faults that occur during certain periods. The strategy chosen for this purpose is based on long-term dependence on traffic, and for the prediction of faults occurrence, a predictive control model (MPC) is proposed, combined with a neural network with radial function (RBF). It is demonstrated via simulations that, in the case of communications traffic, time location is the most important feature of the proposed algorithm.