Strengthening Power Grid Reliability Amid Renewable Energy Challenges

The reliability and safety of power grids are increasingly important as we use more renewable energy sources like solar and wind. These energy sources are not always consistent, which makes it tricky to match the amount of energy produced with the demand. To make this work, we are suggesting a new method that combines advanced technologies to improve the reliability of power grids that have many small energy sources, known as Distributed Energy Resources (DER). Our approach mixes a system that can detect attacks against the grid and a local market setup that helps manage these challenges.

#Challenges in Power Grids

As more renewable energy sources come into play, we face new challenges in keeping the power grid stable. Weather-related changes in energy production can create problems in balancing energy supply with demand. Adding technologies like batteries and electric vehicles increases these challenges. A power grid that is rich in DER is also more open to Cyber-attacks, which can disrupt operations. Studies have shown various types of attacks that can target these kinds of grids. When uncertainties arise, they can affect many important functions in power systems, like planning, security, and market operations. This can lead to problems like exceeding safety limits or failing to detect attacks. Hence, accurate Forecasting is key to managing these concerns effectively.

#The ALAMO Project

The Accurate Federated Learning with Uncertainty Quantification for DER Forecasting Applied to Smart Grids (ALAMO) project focuses on building techniques to handle power grids that have a lot of DER while ensuring privacy for users. Its goal is to create accurate forecasting methods that can quantify uncertainties in energy production and consumption. The project will look into different methods for selecting clients and combining models in order to improve forecasting accuracy, which currently falls short when compared to traditional methods. In addition, it aims to create and evaluate methods to measure uncertainties, ensuring we have reliable estimates for planning.

#A New Approach to Enhance Grid Reliability

In this discussion, we propose to utilize the ALAMO project framework to show how accurate forecasts can work together with a local market structure, ultimately improving the reliability of the grid during normal operations and also during potential attacks. We will outline the proposed framework, which integrates federated learning techniques into forecasting tasks and market mechanisms.

#Understanding Federated Learning

Federated learning (FL) is a method that allows multiple devices to work together to improve machine learning models while keeping data stored locally. This is important for protecting privacy since sensitive data does not leave its original location. FL also makes it easier to manage resources and scaling, as it reduces the need for significant data transfers. In the energy sector, this approach allows for the use of data from various locations, leading to more accurate predictions.

Several studies have shown how FL can be applied in the power sector. For instance, it has been used for forecasting household energy needs and solar energy production. These studies demonstrate that FL can save time and resources, all while prioritizing privacy. By applying FL for various forecasting tasks, including predicting energy demand and solar output, we can improve overall grid management.

#Detecting Cyber Attacks Using FL

Cyber attacks on power grids generally fall into three categories: deception, disclosure, and disruption. We focus specifically on disruption attacks, which aim to disconnect resources from the grid. Our approach involves using FL to obtain daily forecasts for household energy demand and solar energy production from individual prosumers, or energy consumers who also generate power. These forecasts help us identify attacks through a method based on threshold values. By comparing forecast errors and actual energy import data, we can detect anomalies and determine whether an attack has taken place.

While some studies have used FL for detecting unusual patterns and identifying other cyber threats, our work uniquely combines the distributed FL method with a local market structure for both detecting and mitigating attacks on DER-heavy grids.

#Local Electricity Markets

We build on an existing framework of a local electricity market (LEM) that we developed in previous studies. This market involves multiple agents, each representing a prosumer in the grid. The market allows for the coordination of energy supply and demand, even in the event of cyber attacks. The local market offers a chance to better manage energy resources while also ensuring that the grid remains stable.

In our current study, we focus on how FL and LEM can work together. Each agent in the grid submits flexibility bids, which are used to determine how to best allocate energy resources during normal operations and emergencies. This method allows us to optimize energy flow and control losses while still responding to market conditions.

#Mitigating Attacks Through Load Flexibility

Once an attack is detected, we take steps to mitigate its impact. The market operator compares actual energy readings to forecasted values to identify discrepancies caused by an attack. This information helps us adjust the allocation of energy resources effectively. By leveraging flexibility options available throughout the grid, we can reduce the overall energy import required from external sources during an attack. This ensures that the grid remains as stable as possible, even amid disruptions.

#Simulation Case Study

We tested our approach using a real-world low voltage (LV) distribution network in Madeira Island. The network consists of 88 nodes connected by 87 lines, all powered by a transformer that changes high voltage from the grid down to usable levels. Since we lacked actual measurements for each node, we used data from 12 prosumers to generate consumption and production patterns for the simulation.

During the testing phase, we simulated the local electricity market for a 24-hour period and introduced a cyber attack targeting the solar panels. The attack lowered solar energy production to zero, increasing the demand on the main grid. Our system responded by reallocating resources and utilizing load flexibility to reduce energy imports, although it was unable to fully restore the grid to prior conditions.

#Results

The simulation revealed a significant drop in solar energy production due to the attack. This led to a considerable increase in power drawn from the main grid. However, our mitigation measures helped reduce the overall energy import considerably, showcasing the effectiveness of our approach. While we could not completely counteract the attack's effects, our system ensured that the impact on the larger grid was minimized.

#Conclusions and Future Work

We have proposed a method to combine federated learning with local energy markets to effectively confront cyber attacks targeting power grids rich in distributed energy resources. Nonetheless, we face several challenges that require further exploration, particularly in the training of these federated learning models.

Efficient communication is essential since federated systems, often involving many local devices, can be slower compared to centralized systems. Variability in the data from different devices adds to this complexity, making it harder to analyze and model.

Future work will focus on improving the predictive capabilities of our models and addressing their limitations. We also intend to look into more advanced detection methods that go beyond our initial threshold approach. Lastly, incorporating uncertainty in our energy forecasts will enhance the accuracy of our models and help us better address challenges in grid operations.