Researchers are exploring the integration of IoT and the cloud continuum, together with AI to enhance the cost-effectiveness and efficiency of critical infrastructure (CI) systems. This integration, however, increases susceptibility of CI systems to cyberattacks, potentially leading to disruptions like power outages, oil spills, or even a nuclear mishap. CI systems are inherently complex and generate vast amounts of heterogeneous and high-dimensional data, which crosses many trust boundaries in their journey across the IoT, edge, and cloud domains over the communication network interconnecting them. As a result, they face expanded attack surfaces. To ensure the security of these dataflows, researchers have used deep neural network models with encouraging results. Nevertheless, two important challenges that remain are tackling the computational complexity of these models to reduce convergence times and preserving the accuracy of detection of integrity-violating intrusions. In this paper, we propose an innovative approach that utilizes trained edge cloud models to synthesize central cloud models, effectively overcoming these challenges. We empirically validate the effectiveness of the proposed method by comparing it with traditional centralized and distributed techniques, including a contemporary collaborative technique.