Digital-twin platforms are increasingly adopted in energy infrastructure management for smart grids. Novel opportunities arise from emerging artificial intelligence technologies to increase user trust by enhancing predictive and prescriptive analytics capabilities and by improving user interaction paradigms. This paper presents a novel data-driven and knowledge-based energy digital-twin framework and architecture. Data integration and mining based on machine learning are integrated into a knowledge graph annotating asset status data, prediction outcomes, and background domain knowledge in order to support a retrieval-augmented generation approach, which enhances a conversational virtual assistant based on a large language model to provide user decision support in asset management and maintenance. Components of the proposed architecture have been mapped to commercial-off-the-shelf tools to implement a prototype framework, exploited in a case study on the management of a section of the high-voltage energy infrastructure in central Italy.

Artificial Intelligence (AI) in Cyber-Physical Systems allows machine learning inference on acquired data with ever greater accuracy, thanks to models trained with massive amounts of information generated by Internet of Things devices. Edge Intelligence is increasingly adopted to execute inference on data at the border of local networks, exploiting models trained in the Cloud. However, the training tasks on Edge nodes are not supported yet with flexible dynamic migration between Edge and Cloud. This paper proposes a Cloud-Edge AI microservice architecture, based on Osmotic Computing principles. Notable features include: (i) containerized architecture enabling training and inference on the Edge, Cloud, or both, exploiting computational resources opportunistically to reach the best prediction accuracy; and (ii) microservice encapsulation of each architectural module, allowing a direct mapping with Commercial-Off-The-Shelf (COTS) components. Grounding on the proposed architecture: (i) a prototype has been realized with commodity hardware leveraging open-source software technologies; and (ii) it has been then used in a small-scale intelligent manufacturing case study, carrying out experiments. The obtained results validate the feasibility and key benefits of the approach.

Machine Learning (ML) is increasingly exploited in a wide range of application areas to analyze data streams from large-scale sensor networks, train predictive models and perform inference. The Cloud-Edge Intelligence (CEI) computing paradigm integrates cloud infrastructures for resource-intensive ML tasks with devices at the border of a local network for distributed data preprocessing, small-scale model training and prediction tasks. This can achieve a tunable trade-off of ML accuracy with improved data privacy, response latency, and bandwidth usage. Prevalent CEI architectures are based on microservices encapsulated in containers, but serverless computing is emerging as an alternative model. It is based on stateless event-driven functions to facilitate development and provisioning of application components, increase infrastructure elasticity and reduce management effort. This paper proposes a novel CEI framework for sensor-based applications, exploiting serverless computing for data management and ML tasks. Small-scale model training occurs at the edge with local data for quick prediction response, while large-scale models are trained in the cloud with the full sensor network data and then they are fed back to edge nodes for a progressive accuracy improvement. A fully functional prototype has been built by leveraging open source software tools, selected devices for field sensing and edge computing, and a commercial cloud platform. Experiments validate the feasibility and sustainability of the proposal, compared to an existing container-oriented microservice architecture.