Enhanced multivariate singular spectrum analysis-based network traffic forecasting for real time industrial IoT applications

Abstract

Industrial IoT (IIoT) applications are widely used in multiple use cases to automate the industrial environment. Industry 4.0 presents challenges in numerous areas, including heterogeneous data, efficient data sensing and collection, real-time data processing, and higher request arrival rates, due to the massive amount of industrial data. Building a time-sensitive network that supports the voluminous and dynamic IoT traffic from heterogeneous applications is complex. Therefore, the authors provide insights into the challenges of industrial networks and propose a strategy for enhanced traffic management. An efficient multivariate forecasting model that adapts the Multivariate Singular Spectrum Analysis is employed for an SDN-based IIoT network. The proposed method considers multiple traffic flow parameters, such as packet sent and received, flow bytes sent and received, source rate, round trip time, jitter, packet arrival rate and flow duration to predict future flows. The experimental results show that the proposed method can effectively predict by contemplating every possible variation in the observed samples and predict average load, delay, inter-packet arrival rate and source sending rate with improved accuracy. The forecast results shows reduced error estimation when compared with existing methods with Mean Absolute Percentage Error of 1.64%, Mean Squared Error of 11.99, Root Mean Squared Error of 3.46 and Mean Absolute Error of 2.63.