The shift towards renewable energy sources (RES) is accelerating, with increasing investment in large-scale industrial installations and a growing number of energy prosumers. Prosumers typically contribute to distributed energy generation, characterized by smaller production units. Renewable energy production, particularly from wind turbines and solar panels, is inherently dependent on weather conditions.
This variability introduces challenges for grid stability. To date, the instability of RES generation has been counterbalanced using regulating reserves, also known as frequency-response reserves, from conventional energy sources. Sudden fluctuations in RES output due to changing weather conditions can pose significant risks to the energy system. As a result, accurate short-term and ultrashort-term predictions of RES production, along with real-time optimization, are critical for maintaining grid stability.
Forecasting horizons for RES energy production include:
Ultrashort-term
Up to a few hours ahead
Short-term
Several hours to a few days ahead
Medium-term
Several days ahead
Ultrashort-term forecasting focuses on immediate system balancing, while short-term predictions guide operational planning and market integration. Medium-term forecasts, on the other hand, are essential for strategic activities, including maintenance scheduling.
The accuracy of renewable energy production forecasts is shaped by the dynamic nature of converting primary energy sources into electrical energy. These forecasts depend heavily on the time horizon. This is because primary energy sources in RES, such as wind and solar, are subject to constant fluctuations, often shifting within minutes or tens of minutes. Numerical Weather Predictions (NWPs) provide the most reliable foundation for energy production forecasting. However, their acceptable accuracy is limited to short-term and ultrashort-term horizons, typically spanning several hours. This constraint makes them unsuitable for medium- or long-term planning needs.
Additionally, because predictions are estimated on a global scale, they often contain errors that become evident on a local level. These errors in final energy production forecasts are significantly larger than the inaccuracies in NWPs alone due to the nonlinear relationship between weather conditions and energy generation. To address this, it is necessary to develop advanced prediction methods that incorporate not only NWP data but also additional information, such as local environmental conditions and current production data. This approach allows for continuous adjustments to improve the accuracy of predictions.
Improving prediction accuracy can also be achieved through the application of advanced forecasting methods, tailored weather predictions, follow-up adjustments based on local measurement data, and the integration of various energy sources within virtual power plants.
4RES: Creating Accurate RES Generation Forecasts
The 4RES system incorporates all the advanced methods mentioned earlier and was developed by Globema’s Research & Development Center through two extensive research projects focused on predicting energy production. The first project concentrated on individual forecasts for specific renewable energy farms, particularly large wind installations. The second explored predictions for small, distributed energy systems, including prosumer energy, dispersed over wide areas such as an entire country.
Both projects were carried out in collaboration with companies managing renewable energy assets. For larger production units, the research led to the development of methods for locally correcting weather forecasts and identifying profitability thresholds for production predictions based on installed capacity.
The use of hybrid forecasting methods achieved normalized mean absolute errors (nMAE) of 9–12%, normalized by installed power. By aggregating multiple wind and solar farms into a single virtual power plant (VPP), these errors could be reduced even further, to as low as 6%.
For distributed energy systems, the research focused on identifying regions with similar weather conditions to streamline forecasting. Instead of relying on numerous localized forecasts, this approach uses a single, area-based weather prediction to schedule energy production across a region. This method is particularly effective for managing many small power sources collectively as a virtual power plant, representing one unified output on the energy market.
Analyses revealed that, despite the reduced precision of weather forecasts due to value averaging across the studied area, the errors in area-based production predictions remained at a satisfactory level of 5–6%. This demonstrates that the area-based approach, combined with the virtual power plant concept, enables accurate predictions of hourly energy production for several days in advance.
An additional aspect of the research focused on the impact of distributed RES production on energy networks, particularly the additional power flows in network nodes. These findings hold potential for future applications in local energy balancing. Currently, they are used to evaluate network performance and help prevent local overloading. These topics will be explored in a separate article, and we encourage you to check it out!
The 4RES system integrates all the insights from this research. It was designed to assist energy companies in creating accurate production schedules. It also optimizes maintenance windows for energy source operators. Additionally, it supports distribution system operators by predicting energy flows in the network’s main nodes from distributed sources.
How Does 4RES Work?
4RES is composed of two key components:
Computational engines
A business user application
The business application enables users to manage renewable energy sources, edit production schedules, and analyze historical data.
The computational engine modules draw on a variety of data sources, including historical production data, real-time output from specific energy sources, and weather data from multiple meteorological models such as ECMWF, UM, and GFS.
These inputs are used to fine-tune initial prediction models powered by artificial intelligence (AI) and machine learning (ML) algorithms. Over time, the models are further refined with updated data to improve their accuracy. In cases where historical data is unavailable, the computational engines rely on physical models of RES operations.
To optimize the cost and quality of predictions, the selection of weather forecast models is tailored to factors such as the type of energy source, installed capacity, and territorial distribution. This process ensures accurate forecasts covering time horizons ranging from several hours to up to 10 days ahead.
4RES leverages cutting-edge prediction trends by integrating parallel models that account for the influence of various weather variables on energy production.
Another critical component of the 4RES system is its ability to collect various types of historical data, including energy production records, technological process details, and maintenance logs. This information is utilized to refine predictions for subsequent periods. For solar energy forecasts, the system takes into account specific panel types—such as bifacial, tracker, or tilting panels—as well as external factors like snow accumulation and melting.
Building virtual power plants (VPPs) and managing scheduling units that encompass multiple energy sources are also essential features. By linking diverse sources across different locations, operators can enhance prediction accuracy and reduce fluctuations in total energy production. Incorporating regulated sources or energy storage into VPPs creates an optimal energy mix. It improves efficiency and mitigates the impact of distributed generation on the network.
The system’s architecture separates the computational engines from the user-facing application, ensuring continuous development without disrupting user operations. This modularity also enables the sharing of scheduling services through a dedicated client API, allowing users to manage farm lists, introduce detachments, and apply reductions.
The functionalities of the business user application cater to the needs of energy producers, energy sellers, and network operators. Energy producers can use the application to plan maintenance windows strategically, minimizing losses caused by unit outages. Energy sellers benefit from ready-to-use data for creating hourly energy sales schedules, whether for the current day with 15-minute forecasts or the next 15-minute interval. These schedules can be adjusted dynamically, incorporating expert insights not accounted for by the system’s model, such as market trends or unexpected outages.
Historical data analysis within the application helps identify deviations between actual production and forecasts, quantify prediction errors, and assess their financial impact. This analysis, combined with business agreements, can inform the development of future market strategies.
Through its area-based approach and virtual power plant capabilities, the system provides Distribution Network Operators (DNOs) and Operator Service Providers (OSPs) with reliable renewable energy production forecasts for up to 10 days. This enables better planning and management of local medium- and high-voltage network operations and offers a clear view of the country’s anticipated renewable energy output.
As the share of energy generated from renewable sources continues to grow rapidly, driven by Polish and European green energy goals, tools like 4RES that forecast renewable production and its impact on energy networks will only become more critical.
Let’s talk about 4RES!
If you are interested in the solution described in the article, write to us! Let’s talk about how we can help you develop your business.
