Mid Sweden University

Accurate mobile network traffic prediction is crucial for transit infrastructure service optimization in industrial informatization. Traditional linear models fail to capture complex non-linear dynamics, while existing deep learning methods struggle with rapid temporal changes, signal fluctuations, and diverse network conditions, limiting real-time applicability. To address these challenges, this paper proposes a hybrid model integrating Convolutional Neural Networks (CNNs) and Transformers, tailored for High-Speed Railway (HSR) environments. The proposed hybrid model is evaluated using both public datasets and a real-world HSR dataset collected through empirical field measurements, it not only achieves state-of-the-art (SOTA) predictive accuracy, reducing root mean square error by 4.7% over strong baselines in the challenging HSR environment, but also delivers this performance with superior computational efficiency, achieving over 3.6 times lower inference latency than leading SOTA models. This establishes an optimal performance-to-cost ratio, demonstrating its practical value for real-time HSR systems.

Accurate traffic forecasting in high-speed railway (HSR) systems is hindered by abrupt signal fluctuations and varied mobility scenarios. Conventional approaches that rely on fixed weighted combinations of local and global features are unable to adjust rapidly to real-time changes, resulting in suboptimal performance. To address this limitation, we propose a novel gated guided serial CNN and Transformer network (GsCT) that employs a dynamic combination mechanism implemented via a multilayer perceptron (MLP). In GsCT, CNNs capture fine-grained local variations while Transformers model long-range dependencies, and the adaptive MLP-based gating module adjusts the contribution of each branch based on time-window statistics. This dynamic fusion improves prediction quality by 6.5% compared to conventional fixed weighting mechanisms. Evaluations on both public and real-world HSR datasets demonstrate that GsCT achieves a 2.4% reduction in RMSE relative to LSTM-based methods, and the learned gating coefficients offer transparent interpretability of the feature fusion process. Overall, GsCT provides an effective solution for real-time railway traffic forecasting, paving the way for next-generation HSR services.

Integrating cyber-physical systems and the Internet of Things into industrial operations has significantly improved automation, efficiency, and data-driven decision making. However, these advances have also made industrial environments more vulnerable to cybersecurity risks. Our previous work explored lightweight deep learning models for real-time intrusion detection systems on edge devices, yet these models often operate as black boxes, limiting their trustworthiness. This issue is especially critical in the European Union, where the AI Act mandates transparency, accountability, and human oversight for AI solutions to be interpretable. In this paper, we integrate explainable AI solutions into lightweight real-time intrusion detection systems on edge devices to enhance the transparency and interpretability of black-box models. The study demonstrates that integrating SHapley Additive exPlanations significantly enhances the interpretability of intrusion detection systems, providing more transparent insights into model decision-making processes while maintaining accuracy and computational efficiency. This work contributes to the development of more secure and trustworthy industrial ecosystems by improving the effectiveness and reliability of intrusion detection.

The fast expansion of AI within distributed computing environments emphasizes questions about trustworthiness, particularly in contexts involving sensitive data and resource- constrained edge devices. To address this, we implement and evaluate a lightweight federated learning intrusion detection system in a realistic smart home scenario that combines TensorFlow Lite inference on edge devices, MQTT-based publishing, and co- ordinated training via the Flower framework. By operationalizing a previously proposed taxonomy and ontology of trustworthy AI in distributed systems, the implementation in this paper demonstrates how key trust dimensions, such as data integrity, model reliability, and process transparency, can be realized in edge environments. Our implementation utilizes local inference with TensorFlow Lite on IoT devices and coordinated federated evaluation via the Flower framework. We also introduce a trust score to quantify how the implementation aligns with the trust principles. The results indicate that the trust mechanisms are maintained without compromising accuracy or loss, contributing to practical insights into the application of theoretical trust frameworks within distributed AI systems.

The increasing need for robustness, reliability, and determinism in wireless networks for industrial and missioncritical applications is the driver for the growth of new innovative methods. The study presented in this work makes use of machine learning techniques to predict channel quality in a Wi-Fi network in terms of the frame delivery ratio. Predictions can be used proactively to adjust communication parameters at runtime and optimize network operations for industrial applications. Methods including convolutional neural networks and long short-term memory were analyzed on datasets acquired from a real Wi-Fi setup across multiple channels. The models were compared in terms of prediction accuracy and computational complexity. Results show that the frame delivery ratio can be reliably predicted, and convolutional neural networks, although slightly less effective than other models, are more efficient in terms of CPU usage and memory consumption. This enhances the model's usability on embedded and industrial systems.

Ensuring reliable and predictable communications is one of the main goals in modern industrial systems that rely on Wi-Fi networks, especially in scenarios where continuity of operation and low latency are required. In these contexts, the ability to predict changes in wireless channel quality can enable adaptive strategies and significantly improve system robustness. This contribution focuses on the prediction of the Frame Delivery Ratio (FDR), a key metric that represents the percentage of successful transmissions, starting from time sequences of binary outcomes (success/failure) collected in a real scenario. The analysis focuses on two models of deep learning: a Convolutional Neural Network (CNN) and a Long Short-Term Memory network (LSTM), both selected for their ability to predict the outcome of time sequences. Models are compared in terms of prediction accuracy and computational complexity, with the aim of evaluating their applicability to systems with limited resources. Preliminary results show that both models are able to predict the evolution of the FDR with good accuracy, even from minimal information (a single binary sequence). In particular, CNN shows a significantly lower inference latency, with a marginal loss in accuracy compared to LSTM.

Classifying channels into line of sight (LOS) and non-line of sight (NLOS) is essential for accurate ranging/positioning, which underpins all fundamental measurements of time, angle, and carrier phase. Although several statistical and data-driven LOS/NLOS methods are examined in the literature, the heterogeneous parallel ensemble learning (EL), which offers computational efficiency and effectively addresses overfitting issues, has not been investigated, especially for 5G New Radio (5G-NR) positioning signals in indoor factory (InF) channel conditions. In this paper, we address this gap by a) extracting the statistical features of the received positioning reference signal (PRS) in a raytracing-based indoor factory (InF) propagation environment and b) designing a stacking ensemble learning to predict LOS/NLOS channels using these features. Our results indicate that double-level heterogeneous parallel EL outperforms the single-level classification even when sequential or homogeneous EL (boosting) classifiers are compared, both in terms of model performance metrics and prediction accuracy since the worst true LOS and NLOS label prediction is more than 90% and 80%, respectively. Moreover, EL offers a straightforward solution to classifying imbalanced binary samples by employing an internal cross-validation (CV) algorithm for model evaluation. Thus, it can be considered a novel ML-assisted 5G-NR positioning accuracy enhancement (PAE) method.

With the development of intelligent edge computing (IEC) in industrial IoT (IIoT), there is a growing number of service providers trying to leverage computing resources in the cloud and at the edge to meet the users’ demand for low latency and high reliability in diversified applications. This evolving landscape necessitates innovative approaches to manage and process the vast amounts of data generated by IIoT devices. Among these approaches, distributed learning frameworks, such as federated learning (FL), have emerged as popular solutions. However, compared to computing, communication remains the primary bottleneck that constrains the speed of federated model training. Most of the previous solutions have focused on reducing communication overhead. Differently, we propose a transmission-centric approach by designing an efficient communication architecture for FL with cloud-edge collaboration, specifically aimed at enhancing communication capabilities through multi-path transmission. We deploy this FL system in a real environment and conduct extensive testing. The results demonstrate that the new approach can significantly reduce communication time in FL setting, thereby enhancing model aggregation efficiency and shortening the overall training duration. Compared to conventional single-path transmission, the proposed solution improves training efficiency by up to 26.4%. 

In the rapidly evolving cybersecurity landscape, detecting and preventing network attacks has become crucial within the industrial sector. This study aims to explore the potential of intrusion detection by employing deep learning within edge computing, especially for the Industrial Internet of Things. Specifically, TinyML converted CNN, LSTM, Transformer-LSTM, and GCN models on the UNSW-NB15 dataset. A comprehensive dataset analysis gained insights into the nature of attack behavior data. Subsequently, a comparative analysis in an edge computing setup using Raspberry Pi units revealed that the GCN model, with its accuracy of 97.5%, was the best suited of the compared models for this application. However, the study also explored variables like time consumption, where the CNN model was the fastest out of the compared models. This research also highlights the need for continued exploration, especially in addressing dataset imbalances and enhancing model generalizability. By recognizing each model's strengths and areas of improvement, this research serves as a step toward bolstering digital safety and security in an increasingly interconnected industrial world.

The increasing number of IoT devices in the network brings new challenges to the network carrying capacity of intelligent edge computing, and the complicated network services make the demand for network resources in industrial production scenarios or ordinary network users often exceed the carrying capacity of the edge computing network. To alleviate this problem, this paper proposes an intelligent edge computing architecture that introduces network service identification, extracts and analyses the data characteristics of network traffic, and designs appropriate algorithms to classify network traffic into six different service types. This enables real-time and computing-requiring tasks to be prioritised in the network. Using two machine learning algorithms, KNN and MLP, a model validation is carried out on the constructed dataset, and the results show the effectiveness of the method, with the correct rate of data validation reaching 85%, which is more than 5% higher than the correct rate of direct classification of the specified applications, and the accuracy can be as high as 97% in certain scenarios.