Probabilistic forecasting of power system imbalance using neural network-based ensembles

Overview

• The paper addresses the challenge of maintaining the balance between electricity generation and consumption, which is becoming increasingly difficult due to the growing use of renewables, electric vehicles, heat pumps, and industrial electrification.
• Accurate imbalance forecasts, along with reliable uncertainty estimations, can help transmission system operators (TSOs) manage balancing costs and enable market parties to exploit asset flexibility to support grid balancing and generate revenue.
• The authors propose an ensemble of C-VSNs, their adaptation of variable selection networks (VSNs), to forecast system imbalance (SI) and its uncertainty, with a focus on high imbalance magnitude situations.

Plain English Explanation

Keeping the electricity grid balanced, where the amount of electricity being generated matches the amount being used, is getting more and more challenging. This is mainly because of the increasing use of renewable energy sources, electric vehicles, heat pumps, and the electrification of industrial processes.

To help manage this challenge, forecasting system imbalance and the uncertainty around those forecasts is crucial. Transmission system operators (TSOs) can use these forecasts to decide how much backup electricity they need to have ready, which helps reduce the costs of balancing the grid. Market parties, like power companies, can also use the forecasts to find ways to help balance the grid and make money in the process.

The researchers in this paper developed a new forecasting model called an ensemble of C-VSNs, which is their version of a type of neural network called a variable selection network (VSN). Their model can predict the imbalance on the grid, not just for the current time but also for the next two 15-minute periods. It also provides estimates of how uncertain those predictions are.

The researchers tested their model on data from Belgium, where they defined "high imbalance magnitude" as when the imbalance is greater than 500 megawatts (MW) in either direction. This happens about 1.3% of the time in Belgium. For these high imbalance situations, the researchers' model outperformed the previous state-of-the-art models by 23.4% in terms of a metric called continuous ranked probability score (CRPS), which evaluates how good probabilistic forecasts are. Overall, the model achieved a 6.5% improvement in CRPS compared to other methods.

The researchers also developed a way to effectively incorporate new input data into their model, even if that data doesn't have a long history. This work was done in collaboration with the Belgian TSO, Elia, showing the real-world relevance of this research.

Technical Explanation

The authors propose an ensemble of C-VSNs, which are their adaptation of variable selection networks (VSNs), to forecast system imbalance (SI) and its uncertainty. VSNs are a type of neural network that can automatically select the most relevant input features for a given prediction task.

The authors' C-VSN ensemble model makes SI forecasts for the current time and the upcoming two 15-minute periods, along with uncertainty estimates for these forecasts. The model was evaluated on data from Belgium, where "high imbalance magnitude" was defined as |SI| > 500 MW, occurring 1.3% of the time.

For these high imbalance magnitude situations, the C-VSN ensemble outperformed the state-of-the-art by 23.4% in terms of continuous ranked probability score (CRPS), a metric for evaluating probabilistic forecasts. Overall, the model achieved a 6.5% improvement in CRPS compared to other methods. Similar improvements were seen in terms of root-mean-squared error.

Additionally, the authors developed a fine-tuning methodology to effectively incorporate new input features with limited historical data into their model. This work was conducted in collaboration with the Belgian transmission system operator, Elia, demonstrating the practical relevance of this research.

Critical Analysis

The paper addresses an important and timely challenge in the energy sector, namely the need for accurate and reliable forecasts of system imbalance as the grid becomes increasingly complex due to the rise of renewables, electric vehicles, and industrial electrification.

The authors' proposed C-VSN ensemble model appears to be a promising approach, as it outperforms the state-of-the-art methods, particularly in high imbalance magnitude situations, which are crucial for both TSOs and market parties to forecast accurately.

However, the paper does not provide much detail on the specific architecture and training of the C-VSN ensemble, making it difficult to fully assess the technical merits of the approach. Additionally, the evaluation was limited to a single case study in Belgium, and it would be valuable to see how the model performs in other power systems with different characteristics.

The authors also acknowledge that their fine-tuning methodology for incorporating new input features could be further improved, and they suggest exploring other techniques, such as data-efficient strategies for probabilistic voltage envelopes or deep learning-based weather-related power outage forecasting, to address this challenge.

Overall, the research presented in this paper represents a valuable contribution to the field of system imbalance forecasting, but there is still room for further development and testing of the proposed approach to fully assess its potential and limitations.

Conclusion

This paper addresses the increasingly challenging task of maintaining the balance between electricity generation and consumption, driven by the rise of renewables, electric vehicles, heat pumps, and industrial electrification. The authors propose an ensemble of C-VSNs, a novel adaptation of variable selection networks, to forecast system imbalance and its uncertainty, with a focus on high imbalance magnitude situations.

The researchers' model outperformed the state-of-the-art methods, particularly in forecasting high imbalance events, which are crucial for both transmission system operators and market parties. This work was conducted in collaboration with the Belgian TSO, Elia, demonstrating the real-world relevance and potential impact of this research.

While the paper presents a promising approach, further development and testing are needed to fully assess its capabilities and limitations. Exploring additional techniques for efficiently incorporating new input data and evaluating the model's performance in diverse power system contexts could help strengthen the research and its potential applications in the energy sector's ongoing transformation.