AI-Enabled Dynamic Edge-Cloud Resource Allocation for Smart Cities and Smart Buildings.

The rapid expansion of IoT devices represents significant progress in areas such as smart buildings and smart cities, but at the same time, the volume of data generated represents a challenge, which can lead to real bottlenecks in the data analysis process, thus resulting in increased waiting times for end users. The use of cloud-based solutions may prove inefficient in some cases, as the bandwidth required for transmitting data generated by IoT devices is limited. The integration with Edge computing mitigates this issue, bringing data processing closer to the resource that generates it. Edge computing plays a key role in improving cloud performance by offloading tasks closer to the data source, optimizing resource allocation. Achieving the desired performance requires a dynamic approach to resource management, where task execution can be prioritized based on current load conditions: either at the Edge node or the Cloud node. This paper proposes an approach based on the Seasonal Auto Regressive Integrated Moving Average (SARIMA) model for seamlessly switching between the Cloud and Edge nodes in the event of a loss of connection between the Cloud and Edge nodes. Thereby ensuring the command loop remains closed by transferring the task to the Edge node until the Cloud node becomes available. In this way, the prediction that could underlie a command is not jeopardized by the lack of connection to the cloud node. The method was evaluated using real-world resource utilization data and compared against a Simple Moving Average (SMA) baseline using standard metrics: RMSE, MAE, MAPE, and MSE. Experimental results demonstrate that SRIMA significantly improves prediction accuracy, achieving up to 64% improvement for CPU usage and 35% for RAM usage compared to SMA. These findings highlight the effectiveness of incorporating seasonality and autoregressive components in predictive models for edge computing, contributing to more efficient resource allocation and enhanced performance in smart city environments.