Cognitive Digital Twin Framework: Modeling and Real-Time Decision Making
Abstract
Digital twins can support modeling and optimization of complex physical systems. However, many existing frameworks remain passive and cannot learn or decide online under non-stationary conditions. We present a Cognitive Digital Twin (CDT) framework that couples real-time perception, reasoning, and adaptation. CDT builds on deep reinforcement learning, temporal knowledge graphs, and federated learning. It makes three design choices. First, it uses a quality-aware multi-modal fusion module to weight heterogeneous inputs. Second, it uses a hierarchical reasoning engine with reactive (0–10 ms), deliberative (10 ms–1 s), and reflective (asynchronous) layers. Third, it uses a privacy-preserving federated protocol to coordinate multiple twins. Experiments on three industrial IoT testbeds show that CDT reduces prediction error by 25.5%, reduces average response latency by 30.0%, and improves decision quality by 25.7% over the best baseline. Ablations show that each module contributes, and the full system gives the best overall performance.