Harnessing the advantage of digital twin (DT) technology, smart grid provides tempting prospects for efficient management of energy manufacturing, conservation, demand forecasting, pricing, and scheduling. However, the development of smart grid is challenged by volatility of individual users, cyber attacks, and hesitation about data sharing. In this study, a lightweight blockchain-enhanced federated learning (FL) framework is proposed to address the above issues and promote efficient collaborative DT modeling. Through encrypted blockchain retrieval, data aggregation and outlier detection, a Proof of Distribution consensus algorithm is developed to periodically update a conformal score for each edge user over time. The conformal score measures conformity of one user with patterns of the overall data distribution at current stage and is further utilized as basis for stochastic participant selection and model averaging, which improves stability and pertinence of FL in an uncertain environment with heterogeneity and concept drift. In the meantime, based on average learning loss of users with top-ranked conformal scores and engagement of users in the learning process, a reputation score is updated in real time for each user, which offers timely detection of malicious behaviors in the cooperation. The transmission of models and data are both implemented with blockchain, which guarantees privacy preservation and makes the process auditable. Through simulation of residential load forecasting based on real-world scenarios, double robustness of our framework against data volatility and adversarial attacks is verified.

The rapid development of Internet of Things (IoT) offers unprecedented opportunity for federated learning (FL). However, the increasing scale of IoT is accompanied by limitation of communication resource and endogenous heterogeneity of edge devices, which hinder FL’s popularization. Based on features of the edge network, this article proposes a novel one-shot screening scheme to identify edge users with homogeneous data distribution and orchestrate adaptive personalized FL tailored for any task initiator. To cope with the stipulation that raw data cannot be communicated, the learning loss calculated with a locally pretrained model is leveraged to measure the similarity between data distributions. In addition, a multiple hypothesis testing instrument, the Adaptive $p$ -value Thresholding (AdaPT) is utilized to automatically control false discovery rate (FDR) of the screening below any specific level. To demonstrate the effectiveness of the proposed collaborator selection scheme, a numerical study is designed on the real-world non-i.i.d. data set FEMNIST. Evaluations confirm that stable improvement in convergence rate and learning accuracy can be gained compared to vanilla global FL, iterative personalized reweighted FL as well as iterative clustered FL, along with additional welfare of economical wireless resource occupation. Finally, the experiment is carried out under differential privacy to verify robustness further.

Prevailing interest and advancement of fifth-generation (5G) related technology has inspired the study of the Internet of Vehicles (IoV). As per the request for privacy-preserving and collaborative learning, the federated learning paradigm has emerged as a significant supporting component, but it also requires facilitating the explainability of the incentive mechanism. Inspired by this, an asynchronous federated learning paradigm with graph-based reputation incentives has been proposed in the present paper for trajectory prediction tasks in IoV cases. According to prominent characteristics of IoV scenarios, where inter-relations of clients impose great effects on trajectory prediction, a feature extraction strategy based on Graph Neural Network(GNN) is proposed to extract latent features of IoV data and establish a reputation-based incentive mechanism. Integrated with this incentive scheme, an asynchronous federated learning framework is proposed to carry out collaborative trajectory forecasting tasks. As illustrated by the testing results on real-world datasets, the proposed scheme is capable of promoting overall federated learning performance greatly. Additionally, more experiments on proposed reputation incentives have been conducted and show an advance over loss function-based reputation incentives.

The design of 3D scenes is of great significance, and one of the crucial areas is interior scene design. This study not only pertains to the living environment of individuals but also has applications in the design and development of virtual environments. Previous work on indoor scenes has focused on understanding and editing existing indoor scenes, such as scene reconstruction, segmentation tasks, texture, object localization, and rendering. In this study, we propose a novel task in the realm of indoor scene comprehension, amalgamating interior design principles with professional evaluation criteria: 3D indoor scene design assessment. Furthermore, we propose an approach using a transformer encoder–decoder architecture and a dual-graph convolutional network. Our approach facilitates users in posing text-based inquiries; accepts input in two modalities, point cloud representations of indoor scenes and textual queries; and ultimately generates a probability distribution indicating positive, neutral, and negative assessments of interior design. The proposed method uses separately pre-trained modules, including a 3D visual question-answering module and a dual-graph convolutional network for identifying emotional tendencies of text.

With the advancement of the large language model (LLM), the demand for data labeling services has increased dramatically. Big models are inseparable from high-quality, specialized scene data, from training to deploying application iterations to landing generation. However, how to achieve intelligent labeling consistency and accuracy and improve labeling efficiency in distributed data middleware scenarios is the main difficulty in enhancing the quality of labeled data at present. In this paper, we proposed an asynchronous federated learning optimization method based on the combination of LLM and digital twin technology. By analysising and comparing and with other existing asynchronous federated learning algorithms, the experimental results show that our proposed method outperforms other algorithms in terms of performance, such as model accuracy and running time. The experimental validation results show that our proposed method has good performance compared with other algorithms in the process of intelligent labeling both in terms of accuracy and running solves the consistency and accuracy problems of intelligent labeling in a distributed data center.