The rapid evolution of cyber-physical smart cities necessitates intelligent, adaptive systems capable of handling vast urban complexities. Therefore, a Cognitive Digital Twin (CDT) framework built on real time sensor data, deep learning, anomaly detection and blockchain based security to enhance urban operation like traffic control, energy management and public safety has been presented in this research. The target is to construct a scalable self-learning AI model that assists the prediction analytics and thereby the decision making for the smart cities. In contrast to methodical digital twins, the proposed CDT combines both the advanced AI and distributed cloud-edge computing to both speed up data processing and to create more accurate forecasting. The contribution of this work is in its holistic approach by integrating time series forecasting, anomaly detection, and energy consumption prediction in a single AI Driven system and has been verified for both simulation environments as well as real world setup. RMSE and MAPE metrics are used to assess the system performance in terms of processing latency values of around 50 milliseconds for one thousand IoT devices and MAPE values below industry thresholds. Robustness, scalability and precision of urban data management is demonstrated with CDT through optimization of resource allocation, improvement of public safety responsiveness and sustainability of urban development. These results confirm the proposed model’s ability to improve the urban management efficiency to a great extent while ensuring the stringent data security and system resilience. © 2025 IEEE.