Enhanced RES Infeed Forecasting (wind only)

The massive penetration of weather-dependent Renewable Energy Sources (RES) generation brings several challenges to power system operation. Short-term forecasting of RES generation, a few minutes up to days ahead is a cornerstone prerequisite for the secure and economic operation of power systems with high RES penetration.

In the past two decades, considerable research has been carried out leading to several forecasting tools, most of them originating from the specific field of a particular RES. Wind power forecasting techniques were developed through several EC funded R&D projects (FP5 Anemos, FP6 Anemos.plus, FP7 SafeWind), while solar forecasting being addressed through FP7 DNICast and FP7 Performance Plus projects. Spinoffs emerged providing forecasting services and operational tools have been adopted by electricity value chain stakeholders (TSOs, DSOs, RES plant operators, aggregators and forecast service providers) in applications including power system scheduling, estimation of reserves, congestion management, coordination of renewables with storage or trading in electricity market once the RES plants are commissioned and operational.

Originating from meteorology, and not initially oriented to the energy sector, Numerical Weather Predictions (NWP) tools have been developed resulting from collaboration from the energy and meteorology with the aim to align NWP models to the expectations of power and energy industry. These models are used to forecast wind together with algorithms that provide the non-linear transformation of wind speed into power under consideration of the other relevant meteorological and orographic effects, as well as the wind turbine type and/or wind farm layout, including shadow effects. Depending on the forecast time horizon, such forecasts are used by grid operators for intraday and near to real time grid operations (few hours ahead; 15 to 30 min), day-ahead market clearing check (24h) and operational planning (hundreds hours-ahead).


Technology Types

RES forecasting involves a model and value chain encompassing models for weather forecasting and for RES forecasting as well as models related to the applications the forecasts are used for (i.e. decision making for trading). Three types of methodologies exist to forecast wind output:

  • Physical method / deterministic method that uses meteorological data to obtain wind speed forecast and convert it into wind power. Forecasts are based on numerical weather predictions (NWP) using weather data like temperature, pressure, surface roughness and obstacles. NWP, satellite and sky image based methods generate forecasts of weather variables for horizons up to several days ahead with updates usually every 6 hours, and spatial resolution that can go down to 1 km and 1-3 hours temporal resolution
  • Statistical method is based on vast amount of historical data without considering present meteorological conditions. It usually involves advanced data science and time series analysis approaches (machine learning)
  • Hybrid method combines physical methods and statistical methods using weather forecasts and time series analysis approaches (machine learning).

Components & enablers

  • Meteorological data (temperature, pressure, surface roughness and obstacles, wind speed and direction) and related satellite and sky image based methods
  • Numerical weather prediction tools
  • Drones and Lidar in data-spare areas (e.g. offshore for wind energy forecasting)
  • Wind turbine parameters and characteristics
  • Wind turbines maintenance schedules and grid maintenance schedules
  • Historical forecast and generation data.

Another classification of wind forecasting approach is also used by some authors and is based on the time horizon, namely immediate-short-term (8 hours-ahead) forecasting, short-term (day-ahead) forecasting, long-term (multiple-days-ahead) forecasting.


Advantages & field of application

Over 20 years of experience with different forecasting systems is capitalised in early wind-adopting utilities in Denmark, Germany, Spain, the Netherlands and Ireland. Gaps and bottlenecks limiting the performance of the forecasting model chain include the intrinsic limitations of weather forecasting models, the dependability of RES/wind forecasting models on the type of input data, the constraints regarding confidentiality or privacy of data, the lack of standardisation of forecasting products and the under-usage of the collected data for improving wind forecasts. One more structural barrier is the necessity to connect research carried out in the respective fields of meteorology and wind forecasting. Thus, room for improvement concerns simultaneously the weather prediction side, the usage of the forecast as well as their interaction.

At the forecasting model chain end, an important room for improvement remains: the accuracy at levels which appear as adequate for current RES penetration, but inadequate for systems with high penetration of renewable. Such levels result to over-dimensioning of costly remedies, to potential financial penalties by grid operators for wind power plants participating in electricity markets, or to the difficulty to provide ancillary services with a high reliability.


Technology Readiness Level

A commercial offer is available (TRL 9 – System incorporated in commercial design, which do not preclude current development at lower maturity levels on next generation multi-source RES forecasting).


Research & Development

Current fields of research: A steady trend in research is confirmed bridging statistical and meteorological models and focusing on predicting impact of weather events on the forecast.

In Europe advanced modelling tools for modelling and forecasting energy production from variable renewables are expected in a short-term time horizon. The ETIP SNET R&I Implementation Plan 2021-2024 for transmission foresees the following R&I priority in relation to RES Forecasting: “Advanced RES forecasting considering weather forecasts, local ad-hoc models, historical data and on-line measurements”.

The 4-year H2020 Smart4RES collaborative R&D (started in 2019) aims to substantially improve the entire model and value chain in renewable energy prediction by proposing the next generation of RES forecasting models, enabling an increase of at least 15% in RES forecasting performance. It is built on the vision of a holistic approach, covering the whole model and value chain related to RES forecasting from weather forecasting up to end-use applications.\ Some of the ‘advanced modelling tools’ outputs of the projects will include:

improved forecasting of RES-oriented variables with NWP, satellite and ground based all-sky images for Weather modelling and forecasting

  • data science tools for blending information from multiple sources and creating a seamless view of RES forecasting at various temporal and spatial granularity levels for RES power forecasts
  • new business models for RES forecasting ensuring distributed and privacy-preserving forecasting and based on data marketplace for RES forecasting.
  • risk-based decision-aid tools under risk for use cases including storage and RES joint optimisation; predictive dispatch of inertia and frequency containment reserve in isolated power systems with massive integration of non-synchronous generation; distributed voltage control and congestion management in distribution grids; data-driven RES trading in multiple markets.

Machine learning algorithms constitute a key domain of research including the application of statistical learning, of machine learning (to locally approximate the atmospheric equations of motion and extrapolate the system’s behaviour into the near future) and deep learning (convolutional neural networks, long short-term memory recurrent neural networks) models to combine multi-source data and improve forecasting skill up to several days-ahead.

Variable for improvement: Focus should be on improvements in forecasting of high impact (meteorological) events, probabilistic methods for spatiotemporal forecasting, numerical weather prediction models that are specific to energy in general and to variable energy resources, and joint forecasting of wind and solar power generation. The IEA Wind Task 36 aims to coordinate international efforts to improve methods and adoption in wind power forecasting. A survey results in 2016 from IEA Wind Task 36 shows a large untapped potential in systems operators and market actors in the use of forecasting uncertainty in their business practice.

Other: A crucial effort is to be made on data management to address the further mentioned gaps: dependability of forecasting models on the type of input data, data privacy, harmonisation of data structure and data processing protocols for quality and efficiency purposes.


Best practice performance

It is proved that the error of the prediction result will enlarge with larger time horizon. Several benchmarking exercises on forecasting skill compared the performance of weather and power forecasting models for wind power. They showed that performance depends on several factors: weather conditions, time of the year, the terrain typology, the time horizon, the type of models, as well as the level of aggregation of RES plants due to the smoothing effect that permits errors to compensate for geographical distributed RES plants.

For a typical single wind farm with lead time of 24 hours ahead, the normalised mean absolute error (NMAE) can be up to 9-12% of installed capacity, which levels are considered as too high by most users and urge for improvements. For shorter horizons (between a few minutes and up to 6-8 hours ahead), potential for improvement with spatial-temporal approaches and multi-sources combination can reach up to 15-20% for wind energy.


Best practice application

France

2014

Description
Researchers compared different forecasting methods. The data set has been provided by the wind energy company operating the wind farm. Each wind turbine provides 10 min. interval measurements of electrical power, wind speed, wind direction, temperature, as well as an indicator of the working state of the turbine.

Design
The electrical power output of the whole farm is also provided on a 10 min. interval basis. All measures are recorded simultaneously. Data is available for 3 different farms made up of 4 to 6 turbines, in the North and East of France, from 2011 to 2014.

Results
The CART-Bagging algorithm (machine learning) outperforms other forecast methods for the case study evaluated. The RMSE of the forecast was as low as 1.65% with CART-Bagging compared to 2.4% which was at that time state of the art record recorded by French industries according to the renewable energy union.

Norway

2018

Description
A comparative study on assessing the potential benefit of using a deterministic NWP model with 1-hour generation time compared to an NWP ensemble with 2.5 hours generation time.

Design
Nine months of data for the Norwegian wind farms Bessakerfjellet and Hitra were organized to evaluate several forecast models and based on various use of the predictive information sources hourly quantile wind power forecasts are made.

Results
In terms of mean absolute errors the results were neutral, but in situations where moderate to large changes in wind speed were forecast, the scores were in favour of the deterministic NWP model with one hour generation time.

Germany (ParkCast project)

expected Oct. 2021 (project started Nov. 2018)

Description
ParkCast project deals with optimisation of minute-scale power forecasts of offshore wind farms using long-range lidar measurements and data assimilation. It aims to develop, optimise and evaluate new methods for short-term forecasts of the performance of offshore wind farms. The power forecasts focus on the time range up to 60 minutes with high temporal resolution. The aim is to significantly improve the temporal resolution and forecasting quality of the parking performance in the above-mentioned time period and thus make a contribution to grid stability and supply security.

Design
To this end, long-range lidar measurement data are assimilated into a high-resolution, local weather model using new methods based on machine learning. Physical and advanced machine learning-based prediction models are then used for the power prediction and validated in real time for the alpha ventus offshore wind farm as part of an online test phase.

Results
A dedicated work package will focus on ‘Wind Farm Power Forecast’. Assessment of different forecast methods for windfarm power will contribute to the minute-scale power forecast of an offshore wind farm.

Europe

expected 2023 (project started in 2019)

Description
The Smart4RES project aims at the research, development, and validation of a next generation of tools for modelling and forecasting energy production from variable renewables and decision aid for a number of use cases.

Design
Smart4RES proposes disruptive research ideas for developing and validating the next generation tools ensuring jointly an increase of at least 15% in RES forecasting performance and a leverage of the economic value of RES forecasting by considering the whole value chain from weather forecasting to end-use applications.

Results
Results are expected based on the proposed project objectives which include (i) the definition of requirements for forecasting technologies to enable near 100% RES penetration by 2030 and beyond, (ii) the development of a RES-dedicated view of weather forecasting, leading to improvements in forecasting of the relevant weather variables in the order of 10-15% using various sources of data and the development of very high-resolution forecasting approaches, (iii) the development of a new generation of RES forecasting tools that are able to improve RES power production forecasting by at least 15%, (iv) the development of new forecasting products, data marketplaces, and novel business models to get optimal value from data and forecasts, (v) last but not least, new data-driven optimisation and decision-aid tools will enable the large-scale penetration of RES into the electricity market and the provision of system services towards TSOs and DSOs.


References

[1] Wang X, Guo P, Huang X. A Review of Wind Power Forecasting Models. [Link]

[2] IEA Wind. Wind power prediction project list. [Link]

[3] HAL. Statistical learning for wind power: a modeling and stability study towards forecasting. [Link]

[4] Kariniotakis, G., et al. (2004). What performance can be expected by short-term wind power prediction models depending on site characteristics. In Proceedings of the European Wind Energy Conference EWEC [Link]

[5] Kariniotakis, George & Mayer, Didier & Team, Anemos. The Anemos Project: Next Generation forecasting of Wind power. [Link]

[6] Giebel G, Cline J, Frank H, Shaw W, Pinson P, Hodge B-M, Kariniotakis G, Madsen J, Möhrlen C. Wind power forecasting: IEA Wind Task 36 & future research issues. [Link]

[7] The Parkcast project [Link]

[8] The Forewer project (ANR funded): Forecasting and Risk Evaluation of Wind Energy Production [Link]

[9] The Smart4RES project website. Next Generation Modelling and Forecasting of Variable Renewable Generation for Large-scale Integration in Energy Systems and Markets [Link]

[10] Norwegian Meteorological Institute. MET report N° 6/2018 J. B. Bremnes. On the use of NWP forecasts in wind power forecasts for the next few hours [Link]

[11] Sperati, S., et al. (2015). The “weather intelligence for renewable energies” benchmarking exercise on short-term forecasting of wind and solar power generation. Energies, 8(9), 9594-9619 [Link]

[12] B. Alonzo, H.-K.Ringkjob, B. Jourdier, P. Drobinski, R. Plougonven, et al.. Modelling the Variability of the Wind Energy Resource on Monthly and Seasonal Timescales. 2016. hal-01344869 [Link]

[13] ETIP SNET R&I Roadmap 2030 [Link]

[14] ETIP SNET R&I Implementation Plan 2021-2024 [Link]