A TENSOR DECOMPOSITION–BASED METHOD FOR ADAPTIVE RESOURCE ALLOCATION IN REAL-TIME SYSTEMS

Abstract

The paper presents a tensor-decomposition–based method for dynamic resource allocation, aimed at substantially improving computational resource management and scheduling in real-time systems. The proposed approach enables a comprehensive analysis of the multidimensional characteristics of processes—including resource consumption profiles, priority weights, and execution-time parameters—thereby yielding a deeper understanding of the internal dynamics and interdependencies among system components. Unlike traditional models that rely on linear approximations or simplified heuristics, tensor decomposition allows for capturing complex correlations across multiple dimensions simultaneously, which results in more precise and context-aware allocation strategies.

By leveraging tensor operations, the method endows platforms with both flexibility and scalability: it can be applied equally to resource-constrained embedded systems, where optimization must occur under strict hardware limitations, and to large-scale cyber-physical environments featuring numerous nodes, heterogeneous architectures, and variable workloads. Integration of the tensor-decomposition algorithms occurs transparently over standard data-transfer channels and remains fully compatible with conventional memory-management schemes supported by modern processor architectures. This ensures that the method does not require disruptive infrastructure changes, making it suitable for practical deployment in existing technological ecosystems.

Results from experimental evaluations confirm that this approach reduces average task idle time, decreases the incidence of spurious scheduling events, and improves load balancing across computational cores. Furthermore, the method demonstrates resilience under fluctuating workloads and dynamically changing operational conditions, maintaining efficiency even in highly volatile environments. Thanks to its scalability, adaptability, and potential for integration with intelligent decision-making modules, the proposed solution also provides a solid foundation for future research in intelligent resource management, including predictive scheduling, anomaly detection, and adaptive optimization of distributed computing infrastructures.