Transformers for Multi-Horizon Forecasting in an Industry 4.0 Use Case.

Recently, a novel approach in the field of Industry 4.0 factory operations was proposed for a new generation of automated guided vehicles (AGVs) that are connected to a virtualized programmable logic controller (PLC) via a 5G multi-access edge-computing (MEC) platform to enable remote control. However, this approach faces a critical challenge as the 5G network may encounter communication disruptions that can lead to AGV deviations and, with this, potential safety risks and workplace issues. To mitigate this problem, several works have proposed the use of fixed-horizon forecasting techniques based on deep-learning models that can anticipate AGV trajectory deviations and take corrective maneuvers accordingly. However, these methods have limited prediction flexibility for the AGV operator and are not robust against network instability. To address this limitation, this study proposes a novel approach based on multi-horizon forecasting techniques to predict the deviation of remotely controlled AGVs. As its primary contribution, the work presents two new versions of the state-of-the-art transformer architecture that are well-suited to the multi-horizon prediction problem. We conduct a comprehensive comparison between the proposed models and traditional deep-learning models, such as the long short-term memory (LSTM) neural network, to evaluate the performance and capabilities of the proposed models in relation to traditional deep-learning architectures. The results indicate that (i) the transformer-based models outperform LSTM in both multi-horizon and fixed-horizon scenarios, (ii) the prediction accuracy at a specific time-step of the best multi-horizon forecasting model is very close to that obtained by the best fixed-horizon forecasting model at the same step, (iii) models that use a time-sequence structure in their inputs tend to perform better in multi-horizon scenarios compared to their fixed horizon counterparts and other multi-horizon models that do not consider a time topology in their inputs, and (iv) our experiments showed that the proposed models can perform inference within the required time constraints for real-time decision making.

Chlorophyll soft-sensor based on machine learning models for algal bloom predictions.

Harmful algal blooms (HABs) are a growing concern to public health and aquatic ecosystems. Long-term water monitoring conducted by hand poses several limitations to the proper implementation of water safety plans. This work combines automatic high-frequency monitoring (AFHM) systems with machine learning (ML) techniques to build a data-driven chlorophyll-a (Chl-a) soft-sensor. Massive data for water temperature, pH, electrical conductivity (EC) and system battery were taken for three years at intervals of 15 min from two different areas of As Conchas freshwater reservoir (NW Spain). We designed a set of soft-sensors based on compact and energy efficient ML algorithms to infer Chl-a fluorescence by using low-cost input variables and to be deployed on buoys with limited battery and hardware resources. Input and output aggregations were applied in ML models to increase their inference performance. A component capable of triggering a 10 [Formula: see text]g/L Chl-a alert was also developed. The results showed that Chl-a soft-sensors could be a rapid and inexpensive tool to support manual sampling in water bodies at risk.

B5GEMINI: AI-Driven Network Digital Twin.

Network Digital Twin (NDT) is a new technology that builds on the concept of Digital Twins (DT) to create a virtual representation of the physical objects of a telecommunications network. NDT bridges physical and virtual spaces to enable coordination and synchronization of physical parts while eliminating the need to directly interact with them. There is broad consensus that Artificial Intelligence (AI) and Machine Learning (ML) are among the key enablers to this technology. In this work, we present B5GEMINI, which is an NDT for 5G and beyond networks that makes an extensive use of AI and ML. First, we present the infrastructural and architectural components that support B5GEMINI. Next, we explore four paradigmatic applications where AI/ML can leverage B5GEMINI for building new AI-powered applications. In addition, we identify the main components of the AI ecosystem of B5GEMINI, outlining emerging research trends and identifying the open challenges that must be solved along the way. Finally, we present two relevant use cases in the application of NDTs with an extensive use of ML. The first use case lays in the cybersecurity domain and proposes the use of B5GEMINI to facilitate the design of ML-based attack detectors and the second addresses the design of energy efficient ML components and introduces the modular development of NDTs adopting the Digital Map concept as a novelty.

Synthetic flow-based cryptomining attack generation through Generative Adversarial Networks.

Due to the growing rise of cyber attacks in the Internet, the demand of accurate intrusion detection systems (IDS) to prevent these vulnerabilities is increasing. To this aim, Machine Learning (ML) components have been proposed as an efficient and effective solution. However, its applicability scope is limited by two important issues: (i) the shortage of network traffic data datasets for attack analysis, and (ii) the data privacy constraints of the data to be used. To overcome these problems, Generative Adversarial Networks (GANs) have been proposed for synthetic flow-based network traffic generation. However, due to the ill-convergence of the GAN training, none of the existing solutions can generate high-quality fully synthetic data that can totally substitute real data in the training of ML components. In contrast, they mix real with synthetic data, which acts only as data augmentation components, leading to privacy breaches as real data is used. In sharp contrast, in this work we propose a novel and deterministic way to measure the quality of the synthetic data produced by a GAN both with respect to the real data and to its performance when used for ML tasks. As a by-product, we present a heuristic that uses these metrics for selecting the best performing generator during GAN training, leading to a novel stopping criterion, which can be applied even when different types of synthetic data are to be used in the same ML task. We demonstrate the adequacy of our proposal by generating synthetic cryptomining attacks and normal traffic flow-based data using an enhanced version of a Wasserstein GAN. The results evidence that the generated synthetic network traffic can completely replace real data when training a ML-based cryptomining detector, obtaining similar performance and avoiding privacy violations, since real data is not used in the training of the ML-based detector.

Forecasting short-term data center network traffic load with convolutional neural networks.

Efficient resource management in data centers is of central importance to content service providers as 90 percent of the network traffic is expected to go through them in the coming years. In this context we propose the use of convolutional neural networks (CNNs) to forecast short-term changes in the amount of traffic crossing a data center network. This value is an indicator of virtual machine activity and can be utilized to shape the data center infrastructure accordingly. The behaviour of network traffic at the seconds scale is highly chaotic and therefore traditional time-series-analysis approaches such as ARIMA fail to obtain accurate forecasts. We show that our convolutional neural network approach can exploit the non-linear regularities of network traffic, providing significant improvements with respect to the mean absolute and standard deviation of the data, and outperforming ARIMA by an increasingly significant margin as the forecasting granularity is above the 16-second resolution. In order to increase the accuracy of the forecasting model, we exploit the architecture of the CNNs using multiresolution input distributed among separate channels of the first convolutional layer. We validate our approach with an extensive set of experiments using a data set collected at the core network of an Internet Service Provider over a period of 5 months, totalling 70 days of traffic at the one-second resolution.