Artificial intelligence applied to ENERGY

How can we ensure that electricity is available in the right quantities, at the right place and at the right time? With increasing uncertainty in energy supply, research projects in this field have become essential for companies and the population. Managing energy peaks, e.g. during specific times of the day or extreme weather conditions, is particularly important. By automatically adjusting the electricity consumption of specific infrastructures, reliance on energy imports can be reduced and local electricity production can be maximised. Smart energy management could make grids more reliable, cost-effective and flexible. 

AI and SCADA systems for enhanced energy management

The AI algorithms developed by researchers at the Institute of Informatics provide practical solutions for reliable consumption forecasts, reference consumption calculations, and managing complex energy data streams. Optimising the generation, distribution, conversion, usage and storage of electrical energy - particularly through the use of electric vehicles – relies on SCADA (Supervisory Control and Data Acquisition) systems. These systems enable real-time data collection and analysis, facilitating the visualisation of load curves, power factors and energy efficiency indicators, the identification of major energy consumers and the optimisation of sub-metering. These energy efficiency measures aim to reduce the carbon footprint of human activities and mitigate the risk of energy shortages or blackouts.

What is the difference between microgrids and smart grids? Microgrids are regional, small to medium-sized closed, intelligent electricity supply networks that directly link producers and consumers. In contrast, smart grids are decentralised intelligent grids designed to distribute electricity supplied by both industry and consumers. Unlike microgrids, smart grids enable all participants to communicate via gateways - devices connected to the internet bridging different grids - and distribute energy as required. 

The next generation of electricity grids

Smart grids represent the next generation of electricity networks capable of managing demand (consumption, generation, distribution) reliably and cost-effectively while integrating fluctuating renewable energy sources, such as solar and wind power. Controlled by artificial intelligence, these grids can simultaneously analyse consumption and production data. This optimises energy distribution while considering environmental conditions, market conditions and consumption patterns. 

Innovative solutions for intelligent energy management

The researchers at the School of Management want to offer practical solutions to meet the growing demand for intelligent energy management. They focus on modern metering infrastructures, automatic meter reading, automation and decentralised generation in microgrids to create an energy community of prosumers who consume and distribute locally generated renewable energy. Information technology is essential for connecting different networks and analysing data to support municipalities and companies in their energy management efforts.

The Institute of Informatics develops advanced machine learning and deep learning algorithms to collect, process and present energy data and to make it available in a smart grid. Artificial intelligence supports the integration of automation solutions or remote control systems for power plants and energy grids. Through these innovations, AI developed by researchers significantly supports the energy transition.

From intelligent networks to the automation of buildings and infrastructures, AI provides effective and cost-efficient tools to optimise the use of resources. These include: 

• managing smart meters or energy flow control systems
• processing smart meter data in energy network communication infrastructures
• analysing data for signal and status analysis, system monitoring, and consumer data evaluation
• automating load management and controlling storage, distribution or production. 

The collected and analysed data facilitates decision-making, ensuring a stable energy supply and efficient energy use.

The Internet of Things (IoT) is a network of physical objects equipped with sensors and software, capable of communicating with other systems or devices. These energy-efficient connected objects provide users with real time information and enable them to adapt their energy consumption behaviour.

Smart home solutions for better energy efficiency

Sensors can be installed in homes to record data on heating, lighting, the status of door and window, humidity, CO2 levels, noise and movement. Each power socket and the electricity meter can be monitored to create a detailed overview of electricity usage and a model of the home. Residents then receive advice on how to improve their energy consumption habits. 

Energy management systems for businesses

Energy management information systems can also prove interesting for companies wanting to control their energy consumption and expenses. The results of the data analysis can be integrated into an environmental management system to improve performance indicators. 

Industry integration

Networked objects are also useful for industry and can be seamlessly integrated into smart grid projects. Smart grids have sensors or nodes that enable remote monitoring of the infrastructure (power lines, transformers, meters), weather conditions and system temperatures. The data collected is processed by our research institute to assess transmission capacity, adjust energy flow, optimise distribution or propose energy-saving measures.  

The increasing densification of both population and transportation networks calls for improved management of transport flows. A vast amount of mobility data is available, including freely accessible sources such as those from the Federal Statistical Office. However, access to data alone is not enough; it must be thoroughly analysed and interpreted to be effectively used. Unlike other areas of sustainability, numerous stakeholders in the mobility sector are constantly impacted by user actions. Influencing individual behaviour is also challenging as it falls within the area of freedom of movement, which is viewed as an essential freedom. Public policy changes in this context can provoke strong reactions, as evidenced by the "yellow vest" protests in France or the surge in panic buying during the 2021 fuel shortage in the UK.

The importance of data science 

A private vehicle represents freedom but is often essential for people living in remote areas. Political decisions in this area are, therefore, delicate, as individual motivations vary widely—be they financial, practical, or electoral. Relying on data science is crucial for making decisions that align with public expectations and implementing effective and sustainable public policies. 

 

Decentralised federated learning

In traditional machine learning, individual computers send data to a central server, which analyses it to make predictions (e.g. diagnosing a disease). In contrast, Gossip learning is based on a decentralised approach, where data remains on the devices where it was generated. Each computer trains its own model and shares it with other computers, thereby preserving the privacy of the users. 

Why is this called gossip learning? 

The term “gossip learning” refers to how information is shared within the network. Similar to how gossip spreads in social settings, gossip learning relies on individual computers exchanging their locally trained models with others in the network. This distributed approach is scalable, decentralised and reduces the volume of communication, making it ideal for analysing large data sets, improving the flow of energy systems, or simulating complex mechanical or physical concepts to optimise products or devices virtually. 

The advantages of gossip learning

Gossip learning systems have numerous advantages. They are scalable, as new nodes (or computers) can be added to the network. These systems are also robust: If one node fails, the other computers continue without disruption. The distributed and optimised resources enhance performance and increase the calculation speed. Such systems are also cost-effective as they rely on several small nodes rather than a powerful central server. Privacy protection is another major benefit, as sensitive user data is not shared. 

Gossip learning can be applied to practical challenges in the field of mobility, such as making predictions about traffic volumes, preventing accidents involving autonomous vehicles, or studying the behaviour of pedestrians to improve urban planning. 

The models developed at the Institute of Informatics are partially based on digital twins. A digital twin is a virtual replica of a system that is updated in real-time. Digital twins can be used to simulate scenarios such as a city’s traffic volume, allowing researchers to estimate the potential impact of a change in legislation, a political decision, a traffic accident or a deterioration in road conditions on mobility. 

Real-time is not instantaneous and includes a certain level of latency. The digital twin receives real-world information every minute – for example, the number of vehicles and their speed at a particular point in traffic - and continuously adjusts the simulation to reflect reality as closely as possible.

Optimising urban logistics systems 

Various mobility projects are currently being carried out, including one that analyses a motorway model to predict traffic at different points in the road network and to test traffic management strategies. Other projects focus on analysing and reducing freight traffic in urban areas. The research teams use agent-based modelling to optimise urban logistics systems and reduce CO2 emissions. 

These projects are at the interface between the economy and agent-based modelling. Intelligent traffic and mobility management improve the quality of life in cities by minimising noise and pollution, reducing negative environmental impacts, and lowering energy consumption, particularly the use of fossil fuels, which contribute to greenhouse gas emissions and global warming.

Virtual Power Plants for a Flexible Energy Future

Our teams are also working on projects involving virtual power plants. A virtual power plant is an integrated network of decentralised units within the electricity grid, coordinated through a common control system. These units include electricity producers (using thermal or renewable energy sources such as wind, solar, hydropower, biomass and geothermal energy), consumers and energy storage facilities. The virtual power plant is used to collectively market electricity and increase system flexibility. 

All decentralised participants – whether involved in production, storage or consumption - can be part of a virtual power plant, the operation of which is quantified and coordinated by an AI algorithm. This algorithm takes into consideration various factors such as grid status, weather conditions, demand surges and production levels. The virtual power plant reacts quickly and efficiently to fluctuations in electricity market prices and can adjust its operations dynamically. It can use aggregated data to automatically request an increase in production when needed.