Digital twin state-realignment-based cloud infrastructure for resource allocation and fault prediction using HCPF and LST-TICaLUM

To monitor and analyze a large-scale cloud environment continuously, an intelligent cloud management system that influences Digital Twin (DT) technology is necessary. Nevertheless, the prevailing works overlooked the continuous real-time synchronization of the DT with the physical system. Hence, this work proposes a DT reliability-triggered update mechanism using the Fuzzy Neumann Polynomial Inference System (FNePIS) together with Twin State Realignment (TSR) based on the Hankel Contour Particle Filter (HCPF). Initially, the sensor and log details are collected from the physical edge-cloud layer. After that, by using Long Short-Term Trigonometric Integral Cauchy Linear Unit Memory (LST-TICaLUM), the workload is predicted in the DT layer. The explainability is offered using SHapley Logistic Additive exPlanation (SHLogAP). Furthermore, using the Negative Binomial Distribution-based Rat Optimization Algorithm (NBDROA), the Virtual Machines (VMs) are allocated. Later, LST-TICaLUM-based fault detection with explainability is done. Later, regarding the Dirichlet Integral - based Multi-Agent Reinforcement Learning (DIMARL)-based alert-driven self-healing process, agent feedback for the detected fault is updated. Then, regarding deviation measurement and FNePIS-based trigger updates, the twin reliability is validated. Finally, using HCPF, the continuous synchronization is performed. Thus, the proposed system allocated VM with a response time of 1283 ms for 100 nodes and continuously synchronized the DT with a Mean Absolute Error (MAE) of 0.0942.