The detection of highlights in broadcast streams is essential for enhancing User Experience (UX) through automated summaries and efficient content retrieval. This is particularly relevant for live streaming environments common in sports and eSports, where audiences demand near real-time analysis. This paper presents a benchmark of models for highlight detection in broadcast audio, validated on the SoccerNet dataset but applicable to general competitive gaming streams. We propose a novel multi-modal architecture combining high-level semantic audio features (YAMNet) with Natural Language Processing (NLP) of transcribed commentary (analogous to eSports shoutcasting). Results show that fusing audio event detection with semantic text analysis significantly outperforms uni-modal baselines. The proposed framework offers a computationally efficient solution for AI-based broadcasting technologies, enabling scalable automation for content creators and improved viewer experiences.

Efficient scheduling of Virtual Power Plants (VPPs) is critical for integrating distributed energy resources into modern power systems. This paper introduces a CUDA-accelerated simulated annealing algorithm designed to optimize VPP scheduling by leveraging GPUs’ massive parallel processing capabilities. The proposed method reformulates the traditional simulated annealing process to exploit GPU parallelism, significantly reducing computational runtime and enhancing scalability even as the system dimensionality increases. Experimental results indicate that the GPU-based implementation maintains a consistent solution time irrespective of the number of prosumers, demonstrating robust performance in large-scale, real-world energy optimization scenarios. The findings suggest that CUDA-accelerated simulated annealing can serve as a highly effective tool for optimal VPP scheduling, offering substantial improvements in both efficiency and scalability.

This paper presents a high-performance architecture for the New Generation Storage (NGS) project, designed to tackle operational and scalability challenges in managing distributed energy resources within Virtual Power Plants (VPPs). The architecture enables real-time data processing and control, facilitating efficient VPP asset management in both on-premise and cloud environments. By incorporating digital twin technologies, it delivers a dynamic model of physical assets for accurate monitoring and control of distributed energy resources. Advanced data analytics and high-performance computing (HPC) support optimization algorithms targeting cost minimization, battery lifespan extension, and enhanced system efficiency. Its scalability accommodates energy network expansions while maintaining performance. The architecture’s integration with external services enhances predictive analytics for energy forecasting and resource management. Secure authentication and authorization mechanisms based on Keycloak ensure fine-grained access control, safeguarding sensitive data. This flexible solution is suitable for various VPP applications, from small microgrids to large, distributed energy systems, providing a comprehensive and future-proof framework for modern energy infrastructures, and significantly advancing the management of distributed energy resources.

This work provides a comprehensive analysis of the optimization of a Virtual Power Plant (VPP), that consider the presence of energy storage systems and controllable loads, through the benchmarking of various solvers.It delves into the development of a Mixed Integer Linear Programming (MILP) algorithm aiming at optimizing energy management and exchange within a VPP, that takes into account the operation of shift electric appliances and battery storage systems among different houses.The proposed model aims to minimize the overall electricity cost while ensuring that the energy demand of the system is met, the battery state of charge is maintained within safe operating limits, and the shift electrical appliance is scheduled.Furthermore, the experimental comparisons, the study evaluates the performance of commercial and open-source solvers in handling the complex dynamics of energy demand and supply.The findings highlight the importance of solver selection in enhancing the management, scalability, and reliability of VPP optimization strategies, offering insights into the optimal combination of programming interfaces and solvers for efficient VPP operation.

In the nowadays highly unstable manufacturing market, companies are faced, on a daily basis, with important strategic decisions, such as “does the company has the necessary capacity to accept a high volume order?” or “what measures need to be implemented if the product demand increases x% a year?”. Decision-makers, i.e. company's managers, rely on their experience and insights supported by classical tools to take such decisions. Classical mathematical solvers or agent-based systems are typical architectural solutions to implement strategic planning tools to support decision-makers on this important task. Within the ARUM (Adaptive Production Management) project, a hybrid strategic planning tool was specified and developed, combining the optimization features of classical solvers with the flexibility and agility of agent systems. This paper briefly presents such architecture and focuses on the generation of the “what-if game” mechanism to support the generation of more intelligent and dynamic planning solutions.

The continuous change in the manufacturing world is demanding more flexible, responsive and accurate planning tools, which are able to assist the decision-makers to take tactical and strategic decisions on short notice with a high level of confidence. For this purpose, these tools should dynamically explore different operative scenarios in the planning procedure and produce information about key performance indicators. This paper describes the development of an agent-based strategic planner, combining the flexibility of multi-agent systems principles with the optimization capability of a Mixed Integral Programming technique. The tool is integrated in an ecosystem of heterogeneous decision-making systems through an Enterprise Service Bus that also provides access to legacy data.

Manufacturing companies are currently forced to reconsider their production processes by adopting more flexible, robust, and adaptive systems, aiming to improve their competitiveness. Multiagent systems (MASs) technology is suitable to address this challenge by providing an alternative way to design these complex systems based on the decentralization of the control functions over distributed entities. This paper describes the installation of a MAS solution in an industrial factory plant producing laundry washing machines. The installed solution focuses on the integration of quality and process control, and contributes to the maximization of the factory profitability facing changing conditions by applying self-adaptation procedures at the local and global levels. The preliminary results show improvements in the production efficiency and product quality, as well as a reduction of the scrap costs.

Multi-agent system (MAS) is being pointed out as a suitable technology to develop systems that demand flexibility, robustness and re-configurability. Consequently, a significant effort has been noticed to apply MAS to industrial domains exhibiting these characteristics, like manufacturing and smart grids. In spite of the adequacy of the MAS principles to solve the industrial requirements, the truly deployment of MAS for industrial applications is far to be solved. This paper discusses the current challenges for the deployment of MAS in the context of industrial applications, mainly focusing the integration of agents with physical equipment and the ability to run agents directly in industrial or low cost controllers. An experimental MAS solution for a smart grid case study was deployed aiming to support the discussion.

Multi-agent systems paradigm is a suitable approach to implement distributed manufacturing systems addressing the emergent requirements of flexibility, robustness and responsiveness. In such systems, an ontology is a crucial piece to provide a common understanding on the vocabulary used by the intelligent, distributed agents during the exchange of shared knowledge. This paper describes the design of an ontology to define the structure of the knowledge that is used within a multi-agent system integrating process and quality control in production lines for home appliances, which is being developed within the EU FP7 GRACE (inteGration of pRocess and quAlity Control using multi-agEnt technology) project. The ontology schema is validated by instantiating for a case study derived from a washing machines production line.

In the last decade multi-agent systems (MAS) have been thoroughly investigated as a suitable paradigm for process control. Despite the difficulties in designing and maintaining a MAS architecture for real production scenarios, several EU research projects have been financed and many companies are looking at the improvement in the production efficiency of such systems. The EU FP7 GRACE project aims at the integration of process and quality control in a multi-agent environment. This paper discusses the integration of quality control stations into the GRACE MAS. The stations themselves will become autonomous agents, capable of self-reconfiguration according to the needs of the production to improve shop floor efficiency while maintaining the same (and possibly higher) quality level for the manufactured products. Details about the quality control agent behaviour, its integration with the physical hardware and communication with the other agents will be given. All the concepts have been tested in an experimental environment where a vision inspection station behaves as one of the agents of the MAS platform, communicating and exchanging data with the other agents and optimizing its operations over time.