Energy production forecasts for distributed photovoltaic and wind energy sources
About the project
Learn how Globema helped to reduce Tradea’s balancing market participation costs thanks to reliable weather forecasts provided for 250 distributed solar and wind energy sources, using hybrid models utilizing artificial intelligence/ machine learning (AI/ML) and analytical computing.
Table of contents:
Challenge
Trading in energy from Renewable Energy Sources (RES) involves the risk of deviation of the actual production from the forecast and, consequently, incurring additional costs of participation in the balancing market. With the growing volume of energy generated from renewable sources, it has become a challenge to increase the accuracy of forecasts and reduce errors every hour of the day.
Solution
Start of collaboration – initial stage
In late 2019, Globema and Tradea partnered to develop optimal methods for forecasting energy production in solar and wind farms. In 2020 we carried out work that included:
- completing the energy production measurement data and correcting this data, e.g. in terms of identification of measurement points (PPE) and values of installed capacity and technical data of the equipment – on Tradea’s side,
- verifying this data in terms of its correctness, completeness and possible anomalies – on Globema’s side,
- building preliminary predictive models and verifying their accuracy – on Globema’s side.
The cooperation was based on meetings, during which we presented research results to the customer and decided together on the further course of work. After developing a satisfactory prediction model, we started a six-month test phase of delivering and verifying the quality of forecasts.
Further cooperation
The six-month trial period resulted in a positive verification of both the service itself and the production forecast error rate. Tradea’s evaluation of the results of the cooperation was positive and they decided to launch the service, based on 4RES, from the start of 2021.
The forecasting service covers approximately 250 facilities dispersed across Poland, with a total installed capacity of about 240 MW.
It is based on area weather forecasts, which are created with the use of UM and GFS meteorological models and represents a compromise between the precision of spot weather forecasts and their cost – given the large number of objects for which forecasts have to be created.
The conversion of weather forecasts into production forecasts is done with the use of hybrid models that combine analytical calculations with artificial intelligence/machine learning (AI/ML).
The forecasts are delivered daily, before 9:00 a.m., via two channels: e-mail and FTP, in the form of text files. They cover the hours from 1:00 a.m. until midnight on the next day, for each individual energy farm. The farms are grouped into balancing units according to the market clearing method: by fixed price, balancing market, day-ahead market, etc.
The assignment of farms to groups changes from month to month, there are also new installations, which, at the request of the customer, are included in the forecasting system from day to day.
The form and method of file transfer agreed upon with the customer allows the files to be handled automatically by Tradea and the production schedules to be prepared for entities responsible for trade balancing. An intervention in the files prepared is necessary only in very rare cases, e.g. when one of the sources fails.
Improving the forecasts
Globema monitors the service quality on a monthly basis. Forecast errors in a given month are compared with errors in the previous months and the corresponding month of the previous year. We also check for the possibility of reducing the error by spiking the model with the latest measurement data.
If such an attempt, confirmed by a 3-month test period, brings an improvement, it is applied to the target model. The reported nMAE error of the service, defined as the mean absolute error normalized by the maximum power of the farm, ranges from 1.8% to 7.3%, depending on the month (errors are larger in the summer due to higher insolation values and length of the solar day).
Importantly, these values drop to 0.9%-3.2% if, instead of considering the average error of individual farms, we take into account the error of the whole settlement group (VPP). This can be explained by the mutual compensation of errors of different installations in different locations. This effect also confirms the validity of the area-based approach in the case of a large number of dispersed RES installations with small unit power.
Service reliability
We work incessantly to make our forecast delivery mechanism even more reliable. The delivery channel redundancy mentioned earlier (FTP and email) increases reliability in the event that one of them fails. We also prepare backup forecasts more in advance in case the weather forecast service itself fails – then, even in the absence of the latest weather forecast, our service does not fail, although the output forecast sent is statistically slightly less accurate.
Did you know that…
As a part of Globema R&D Center’s activity, we also conduct continuous research on improving the quality of our forecasts, in cooperation with our customers and weather forecast provider – the Interdisciplinary Centre for Mathematical and Computational Modelling UW
An interesting example of such cooperation is the issue of the impact of snowfall and snow accumulation on photovoltaic panels on the forecasting error, reported by Tradea. Although, with the warming climate, heavy snowfall, especially in the lowlands, is slowly becoming a thing of the past, it had a noticeable impact on forecasting errors in the 2020/21 season.
Based on this experience, we have developed a snow coverage model for the panels, which has been tested in the snowiest period in January and February of 2021, and is ready to be implemented in the upcoming winter season.
Effects
The launched and systematically developed service of forecasting and generating hourly schedules of energy production allowed our Customer to reduce its share in the balancing market and thus limit the risk of additional, often unpredictable, costs of this share.
The flexibility of the tool additionally allows for an easy and current way of taking into account changes in the installed power and assigning generating units to relevant balancing groups, which saves the employees’ time and guarantees correctness of the settlements.
The daily generation of production forecasts for the next day is the foundation of our business, as it translates into the financial effects of our market game.
We chose Globema after a pilot project that ran in 2019 and 2020, because we were convinced that the quality of these forecasts developed with a professional method based on artificial intelligence, significantly exceeds our previous attempts to prepare forecast on our own.
Radoslaw Bartnicki, Electricity Market Analyst, Tradea
Discover other success stories
Polenergia
Network inventory supports Polenergia’s growth based on M&A
Learn how we standardized the network management at one of the largest energy distributors in Poland and organized data about 30 areas covering 11 thousand customers.
PERN
Developing Network Inventory Management system in the Pay-As-You-Go model
SeZaM system based on the GE Smallworld platform supports pipelines as well as telecommunications and energy networks at PERN S.A.
GE Global Transmission Office/Smallworld | Network Inventory Management | Oil and Gas
