ADVANCED FORECASTING

OF VARIABLE
RENEWABLE POWER
GENERATION
INNOVATION LANDSCAPE BRIEF
© IRENA 2020

Unless otherwise stated, material in this publication may be freely used, shared, copied, reproduced, printed
and/or stored, provided that appropriate acknowledgement is given of IRENA as the source and copyright
holder. Material in this publication that is attributed to third parties may be subject to separate terms of use
and restrictions, and appropriate permissions from these third parties may need to be secured before any use
of such material.

ISBN 978-92-9260-179-9

Citation: IRENA (2020), Innovation landscape brief: Advanced forecasting of variable renewable power generation,
International Renewable Energy Agency, Abu Dhabi.

ABOUT IRENA
The International Renewable Energy Agency (IRENA) is an intergovernmental organisation that supports
countries in their transition to a sustainable energy future and serves as the principal platform for
international co-operation, a centre of excellence, and a repository of policy, technology, resource and
financial knowledge on renewable energy. IRENA promotes the widespread adoption and sustainable
use of all forms of renewable energy, including bioenergy, geothermal, hydropower, ocean, solar and
wind energy, in the pursuit of sustainable development, energy access, energy security and low-carbon
economic growth and prosperity. www.irena.org

ACKNOWLEDGEMENTS
This report was prepared by the Innovation team at IRENA’s Innovation and Technology Centre (IITC) and was authored
by Francisco Boshell and Arina Anisie, with additional contributions and support from Santosh Kamath, Harsh Kanani and
Shikhin Mehrotra (KPMG India).

Valuable review was provided by David Moser (Eurac), Keith Parks (Xcel Energy), Ilya Chernyakhovskiy (National Renewable
Energy Laboratory) and Kui Luo (State Grid Corporation of China), along with Harold Anuta, Gayathri Nair, Laura Casado,
Elena Ocenic, Nina Litman-Roventa, Martina Lyons and Paul Komor (IRENA).

DISCLAIMER
This publication and the material herein are provided “as is”. All reasonable precautions have been taken by IRENA to verify
the reliability of the material in this publication. However, neither IRENA nor any of its officials, agents, data or other third-
party content providers provides a warranty of any kind, either expressed or implied, and they accept no responsibility or
liability for any consequence of use of the publication or material herein.

The information contained herein does not necessarily represent the views of all Members of IRENA. The mention of
specific companies or certain projects or products does not imply that they are endorsed or recommended by IRENA
in preference to others of a similar nature that are not mentioned. The designations employed and the presentation of
material herein do not imply the expression of any opinion on the part of IRENA concerning the legal status of any region,
country, territory, city or area or of its authorities, or concerning the delimitation of frontiers or boundaries.

Photographs are from Shutterstock unless otherwise indicated.

This document does not represent the official position of IRENA on any particular topic. Rather, it is intended as a contribution
to technical discussions on the promotion of renewable energy.
www.irena.org

1 BENEFITS
Accurate generation forecasts for solar and wind power – short term and long term, centralised and decentralised – are
valuable to system operators and renewable generators.

GENERATORS
Electricity trading in spot markets Renewables plant placement
BENEFIT

SYSTEM OPERATORS
System stability System planning
BENEFIT

SHORT-TERM FORECASTING LONG-TERM FORECASTING

2 KEY ENABLING FACTORS 3 SNAPSHOT

Regulatory incentives for accurate variable ➜ Australia invests USD 5.6 million in advanced wind
renewable energy (VRE) forecasting and solar forecasting to improve decisions made
on spot markets.
Open source systems for weather data
collection and sharing ➜ The UK system operator uses artificial intelligence
to better predict renewable generation.
Advanced meteorological devices
➜ A study from the California Energy Commission
indicates potential savings of USD 2 million yearly
with improved solar and load forecasting.

WHAT IS ADVANCED GENERATION FORECASTING?

Meteorological technology captures real-time, site-specific weather data.

Algorithms produce advanced forecasts for solar and wind output.

ADVANCED FORECASTING OF VARIABLE

RENEWABLE POWER GENERATION
Improved weather forecasts allow accurate estimates of the amounts of solar
and wind electricity likely to be available in specific time frames.
 I N N OVAT I O N L A N DS C A P E B R I E F

ABOUT THIS BRIEF

T his brief forms part of the IRENA project

“Innovation landscape for a renewable-
powered future”, which maps the relevant
innovations to create actual solutions. Solutions
to drive the uptake of solar and wind power span
four broad dimensions of innovation: enabling
innovations, identifies the synergies and formulates technologies, business models, market design
solutions for integrating high shares of variable and system operation.
renewable energy (VRE) into power systems.
Along with the synthesis report, the project
The synthesis report, “Innovation landscape for a includes a series of briefs, each covering one of
renewable-powered future: Solutions to integrate 30 key innovations identified across those four
variable renewables” (IRENA, 2019a), illustrates dimensions. The 30 innovations are listed in the
the need for synergies between different figure below.

INNOVATION DIMENSIONS

ENABLING TECHNOLOGIES BUSINESS MODELS MARKET DESIGN SYSTEM OPERATION

1 Utility scale batteries 12 Aggregators 17 Increasing time 25 Future role of distribution

2 Behind-the-meter 13 Peer-to-peer electricity granularity in electricity system operators
batteries trading markets 26 Co-operation between
14 Energy-as-a-service 18 Increasing space transmission and
3 Electric-vehicle
granularity in electricity distribution system
smart charging 15 Community-ownership
markets operators
4 Renewable models
19 Innovative ancillary
power-to-heat 16 Pay-as-you-go models
services
27 Advanced forecasting
5 Renewable 20 Re-designing capacity
of variable renewable
power-to-hydrogen power generation
markets
21 Regional markets 28 Innovative operation
6 Internet of Things
of pumped hydropower
7 Artificial intelligence
22 storage
and big data 23 Market integration
8 Blockchain of distributed energy 29 Virtual power lines
resources 30 Dynamic line rating
9 Renewable mini-grids
10 Supergrids 24 Net billing schemes

11 Flexibility in conventional
power plants

4
 A DVA N C E D FO R EC A S T I N G O F VA R I A B L E R E N E WA B L E P OW E R G E N E R AT I O N


This brief provides an overview of the concept
The brief is structured as follows:
of advanced weather forecasting and its
importance for VRE integration into power I Description
systems. Emphasis is placed on the contribution
of both short- and long-term weather II Contribution to power sector transformation
forecasting for renewable generators and power
system operators. Key enablers required for its III Key factors to enable deployment
development and implementation are presented,
together with examples of ongoing initiatives. IV Current status and examples of ongoing
initiatives

V Implementation requirements: Checklist



5
 I N N OVAT I O N L A N DS C A P E B R I E F

I. DESCRIPTION

A ccurate weather forecasting is crucial for

integrating wind and solar power generating
resources into the grid, especially at high
For weather-dependent renewable generators,
like solar and wind power plants, the most critical
scheduling input comes from weather forecasting
penetration levels. It is a crucial, cost-effective tool data. A power generation forecast is a combination
available to both renewable energy generators of plant availability and weather forecasts for the
and system operators (NREL, 2016). location, as illustrated in Figure 1.

Figure 1 Weather and power generation forecast

WEATHER FORECAST
Accurate weather prodictions enable
renewable generators to forecast their
power generation

POWER

+ GENERATION
FORECAST

PLANT AVAILABILITY
The availability of a plant accounts for
both planned and unplanned outages

6
 A DVA N C E D FO R EC A S T I N G O F VA R I A B L E R E N E WA B L E P OW E R G E N E R AT I O N

Advanced weather forecasting methods take These simulated weather predictions can then be
advantage of advances in digital technologies, converted to corresponding energy production
such as artificial intelligence (AI) and big data, from wind and solar resources. NWPs are
to analyse live and historical weather data and normally used for 15-days-ahead forecasting.
make predictions. In fact, advanced weather However, these models are not accurate over
forecasting is one of the main applications of short timescales (less than a few hours).
AI in facilitating and improving VRE integration Statistical approaches are also commonly used
(for more information see Innovation landscape and are based purely on historical learnings. Solar
brief: Artificial Intelligence and Big Data [IRENA, irradiation forecasts also employ sky imagers
2019b]). Driven by an increase in computing (digital cameras that produce high-quality sky
power and improvement in algorithms, power images) and satellite imaging (data from networks
generation forecasts have become more accurate. of geostationary satellites) to track and predict
In a similar vein, thanks to the increasing use of cloud formations at different timescales. Hybrid
AI fuelled by big data, time granularity for short- models use two or more techniques in conjunction
term predictions has increased as well. These to minimise the forecasting error. These hybrid
factors can greatly contribute to the integration methods have produced the best forecasting
of renewable power into the grid (Bullis, 2014). results when compared with individual statistical
and machine learning techniques for all types of
Improving VRE generation forecasts on short- time horizon (Akhter et al., 2019).
term and long-term timescales engenders a
diverse set of benefits for various stakeholders in Compared with large-scale dispatchable plants,
the power sector. At short timescales, accurate forecasts for distributed solar photovoltaic (PV)
VRE generation forecasting can help asset generation are more difficult to produce because
owners and market players to better bid in the of the relatively small size and large number of
electricity markets, where applicable. Bids based solar PV sites. Such forecasts are most accurate
on more accurate forecasts would reduce the when near-real-time power generation data and
risk of incurring penalties for imbalances (i.e. detailed static data (e.g. location, hardware
for not complying with the generation offered in information, panel orientation) are available for
the bid). For power system operators, accurate all connected systems (NREL, 2016).
short-term VRE generation forecasting can
improve unit commitment (operation scheduling For example, 11 Renewable Energy Management
of the generating units) and operational planning, Centres (REMCs) are being set up in India. The
increase dispatch efficiency, reduce reliability REMCs are equipped with AI-based renewable
issues and, therefore, minimise the amount of energy forecasting and scheduling tools at the
operating reserves needed in the system. regional level and provide greater visualisation
and enhanced situational awareness to the grid
Over longer timescales (e.g. over days or seasons), operators. In total, 55 gigawatts (GW) of renewable
improved VRE generation forecasting based on power (solar and wind) is being monitored through
accurate weather forecasting brings significant the 11 REMCs (Asian Power, 2020). In Germany,
benefits to system operators, especially when the power generation forecasts for transmission
planning for extreme weather events. By system operators (TSOs) and distribution system
contributing to the allocation of appropriate operators (DSOs) are calibrated and evaluated
balancing reserves, long-term weather forecasting against estimations of the solar power production
assists in ensuring safe and reliable system on a postal code level. Forecasts of aggregated
operations. It can also help in better planning distributed solar PV production are developed
the long-term expansion of the system, both and validated by upscaling the output from
generation and network transmission capacity, a subset of representative solar PV sites. The
needed to efficiently meet future demand. process is similar in California, where information
about all solar PV systems in the state is recorded
Every energy forecast invariably starts with and then combined with high-resolution solar
numerical weather prediction (NWP) models, irradiance values and weather predictions to
which are the accepted baseline predictions being forecast power output for the entire state. This
tuned and run by large and mostly government- bottom-up approach is employed by the California
funded organisations. NWP methods take weather Independent System Operator (CAISO) to predict
data, such as temperature, pressure, humidity, the total contribution of behind-the-meter solar
as inputs to simulate weather conditions in the plants to its grid.
future using physical and mathematical laws.

7
 I N N OVAT I O N L A N DS C A P E B R I E F

Table 1 lists the methods used for power horizon, even for the same forecasting technique.
generation forecasting at different time horizons, Thus, the selection of a proper time horizon
as well as the key applications of these forecasts before designing a forecasting model is key to
in the power sector. The forecasting accuracy maintaining the accuracy of forecasting at an
decreases with the increase of forecast time acceptable level (Akhter et al., 2019).

Table 1 Generation forecast methods and applications

Time Horizon Methods Key Applications

5–60 min ahead of Regulation, real-time dispatch,

Statisticala, persistenceb
real time trading, market clearing
1–6 hours ahead of Blend of statistical and Scheduling, load following,
real time NWP models congestion management
Power
Generation NWP with corrections for Scheduling, reserve requirement,
Day(s) ahead of real time
Forecast systematic biasesc trading, congestion management
Resource investment planning
Week(s), seasonal, 1 year or Climatological forecasts, (generation, network), contingency
more ahead of real time NWP analysis, maintenance planning,
operation management

Source: Based on NREL (2016).

Note: min = minutes; NWP = numerical weather prediction.
a
Use historic and real-time generation data to provide statistical corrections to predictions
of the NWP models.
b
Simple statistical methods that assume that the current generation will remain unchanged in the near future.
They are used as benchmarks to evaluate other advanced forecasting models.
C
Errors inherent in the measurement process.

Very short-term forecasting (from seconds Long-term forecasting (one month to one year)
to 1 hour) is useful for real-time electricity helps transmission and distribution authorities in
dispatch, optimal reserves, and smoothing electricity generation planning, in addition to energy
power production of solar and wind. Short-term bidding and security operations (Akhter et al., 2019).
forecasting (from 1 hour to 24 hours) is useful to
increase the stability of the grid. Medium-term Complex modelling is required to account for all
forecasting (one week to one month) maintains the variables that affect local weather. Figure 2
the power system planning and maintenance illustrates, in a simplified way, the data needed
schedule by predicting the available electric and the methodology applied for VRE power
power in the near future. plants’ generation forecasts.

Figure 2 From weather to power generation forecast

NUMERICAL WEATHER FORECAST MODELS

TEMPERATURE SOLAR
WEATHER ENERGY
IRRADIANCE WIND GENERATION
FORECAST FORECAST
VELOCITY HUMIDITY FORECAST
MODELS MODELS
RAIN, ETC.

PUBLIC WEATHER DATA PLANT CHARACTERISTICS

PRIVATE WEATHER DATA PLANT AVAILABILITY

PUBLIC SENSORS DATA

8
 A DVA N C E D FO R EC A S T I N G O F VA R I A B L E R E N E WA B L E P OW E R G E N E R AT I O N

II. CONTRIBUTION TO POWER

SECTOR TRANSFORMATION

A ccurate weather forecasts and very short-term

to long-term forecasting are key for effectively
integrating VRE generation into the grid and
Centralised forecasts provide systemwide
forecasts for all VRE generators within a
balancing area. Decentralised forecasts,
bring valuable contributions for both renewable administered by individual VRE asset operators,
generators and system operators. Figure 3 provide plant-level information to help inform
summarises the benefits of short-term forecasting system operators of potential transmission
(defined here as minutes to one day ahead) and congestion due to a single plant’s output,
long-term forecasting (weeks to one year ahead). as well as help position the plant’s bids in
the forward or short-term markets (NREL,
System operators usually use centralised 2016). System operators can also benefit from
forecasting of renewable generation, widely decentralised forecasting by aggregating these
considered a best-practice approach for a cost- data and using them for both short- and long-
effective dispatch. term operational procedures.

Figure 3 Benefits of weather forecasting to system operators and renewable generators

CENTRALISED
BENEFITS OF DECENTRALISED FORECASTING
FORECASTING
WEATHER
FORECASTING For system operators For renewable generators

Improved network management Advantages for intraday and

Short-term forecasting
and system balancing day-ahead electricity market trading

Reserve planning and

operation management Efficient placement of
Long-term forecasting
Planning for extreme renewable plants
weather events

9
 I N N OVAT I O N L A N DS C A P E B R I E F

Contributions of short-term Accurate weather forecasts can enable

weather forecasting renewable power generators to better estimate
the generation and bid accordingly in intraday
Improved network management and or day-ahead markets, reducing penalties
system balancing imposed for deviations between actual and
scheduled power generation. By using a more
Short-term centralised forecasting is useful advanced forecasting solution with NWP
for applications related to system operations, models, renewable energy forecasts can be
such as real-time dispatch, market clearing and predicted closer to actual generation time.
load following. Furthermore, accurate weather Statistical methods using AI models consider
forecasting can also provide advantages in short- project-specific data along with nearby
term electricity trading and system balancing. weather conditions for more accurate short-
This leads to improved grid reliability and enables term forecasts. System operators in the United
the efficient use of renewable energy resources. States, such as the Midcontinent Independent
System Operator and Electric Reliability Council
Better weather forecasting using AI of Texas, combine short-term dispatch with very
supercomputers that synthesise data from various short-term forecasts (within 10 minutes of the
sources, including historical weather data, real- actual flow of power), which allows wind power
time measurement of local weather conditions, plants to be fully integrated into real-time
satellites and sensors, is expected to lead to intraday markets (Orwig et al., 2015).
more accurate estimations of generation ramping
requirements. More accurate estimations will be Contributions of long-term
useful in improving grid reliability. weather forecasting

Day-ahead forecasts provide hourly power values Reserve planning and operation management
used in the scheduling process to help avoid costs
and inefficiencies due to unnecessary starts and Long-term weather forecasts are valuable
stops of thermal generators. Intraday forecasts for system operators in applications such as
typically provide power values with frequent time reserve planning and operation management.
steps (i.e., every 10 minutes) up to a few hours Such forecasts may further assist generators
ahead of real time. They are used in real-time and system operators in investment planning
dispatch and market-clearing decisions. for, respectively, power plant construction and
system expansion.
VRE generation forecasts can be, for example,
integrated with load forecasting to produce Planning for extreme weather events
net load forecasts, which improve the visibility
of demand-side variations. VRE forecasts can Long-term weather forecasts can be used by
be also integrated by system operators into a system operators to predict extreme weather
power flow module, which is a part of energy events and better plan and prepare for such
management systems, to detect voltage and occasions. For instance, the North Atlantic
congestion problems with a certain probability Oscillations, which are seasonal weather
threshold. phenomena over the North Atlantic and Europe
caused by pressure differentials, can cause as
Advantages for intraday and day-ahead much as a 1 020% variation in wind and solar
electricity market trading generation (Jerez et al., 2013). Long-term
weather forecasts for such weather events can
Short-term decentralised forecasts engender thus improve the resilience of the system, helping
benefits for generators, especially if there is system operators in planning for alternative
a functioning electricity market in place, and resources to ensure the security of supply.
help renewable generators define their bidding
strategies.

10
 A DVA N C E D FO R EC A S T I N G O F VA R I A B L E R E N E WA B L E P OW E R G E N E R AT I O N

Efficient placement of renewable plants Other important parameters are the zenith angle
and the orientation of the PV panel. Advanced
Long-term decentralised weather forecasts can weather forecasting tools can be used for
help in the generation expansion of the system by the identification of optimal sites for wind
identifying the best locations for the construction turbine and solar PV installations to improve
of new renewable power plants. For example, generation outputs and reduce maintenance
turbine placement in the wind industry is an costs. Supercomputers can be used to analyse
important parameter affecting power generation. petabytes of structured and unstructured data –
Major factors for appropriate turbine placement such as weather reports, tidal phases, geospatial
include wind speed, wind turbulence, wind and sensor data, satellite images, deforestation
direction, space and ecological considerations. maps, and weather modelling research – to
Solar irradiation, also location specific, is crucial identify the optimal location of wind turbines
for solar PV projects. (IBM, 2011).

Potential impact on power sector transformation

Several studies have explored the impact and implications of advanced forecasting for utilities,
system operators and VRE generators:

• The impact of improvement in solar power Coal, gas combined-cycle, and gas and oil
forecasting was analysed in one study by steam turbine generators are committed in the
evaluating the operations of the entire power day-ahead run, while gas and oil turbines and
system for four scenarios that represented internal combustion generators are committed
four forecast improvement levels: 25%, in the real-time run. Table 2 shows, for different
50%, 75% and 100% (perfect forecast). solar penetration levels, the impact of forecast
The study showed that electricity improvement on the overall system generation
generation from conventional sources (gas mix. Day-ahead operation can be better
and oil generators) decreases with solar planned, and faster generators committed in
power forecasting improvement (Martinez- real time can be used less, when solar power
Anido et al., 2016). uncertainty decreases.

Table 2 Impact of solar power forecasting improvement on power generation

Solar Penetration (%) 4.5 9.0 13.5 18.0

Forecast Improvement (%) 25 50 75 100 25 50 75 100 25 50 75 100 25 50 75 100

Coal (%change) 0.0 0.0 0.0 0.1 0.1 0.2 0.2 0.3 0.1 0.3 0.4 0.5 0.2 0.4 0.4 0.4

Gas CC (%change) 0.0 0.2 0.1 0.1 0.1 0.1 0.2 0.3 0.3 0.7 0.9 1.1 0.5 0.9 1.4 1.8

Gas and Oil ST (%change) -0.1 -1.4 -0.8 -0.8 0.6 1.2 0.5 0.0 -1.3 -2.8 -3.9 -4.8 -1.9 -3.5 -5.3 -5.5

Gas and Oil GT & IC (%change) -0.8 -1.8 -1.9 -1.7 -4.5 -7.5 -8.6 -10.2 -3.9 -8.8 -12.1 -13.7 -5.1 -10.5 -14.3 -17.5

Source: Martinez-Anido et al. (2016).

Note: CC = combined-cycle; ST = steam turbine ; GT = gas and oil turbine; IC = internal combustion;.

11
 I N N OVAT I O N L A N DS C A P E B R I E F

The study shows also a considerable decrease forecast. The results of the study are shown
in solar power curtailment with the improved in Table 3.

Table 3 Impact of solar power forecasting improvement on solar power curtailment

Impact (% change) Solar penetration (%)

Forecast Improvement (%) 9.0 13.5 18.0

25 -2.3 -1.2 -1.5

50 -4.0 -2.3 -2.7

75 -5.2 -3.5 -3.5

100 -5.4 -4.2 -3.8

Source: Martinez-Anido et al. (2016).

The same study found that the impact of solar starts and shutdowns on fossil fuel generators,
forecasting improvements on solar curtailment results in lower operational costs for the
and electricity generation, including ramping, system, as shown in Table 4.

Table 4 Cost savings from solar power forecasting improvement per unit of solar power generation

Forecasting improvement
Solar penetration (%)
cost savings ($/MWh)
Forecast Improvement (%) 4.5 9.0 13.5 18.0

25 0.11 0.33 0.39 0.50

50 0.29 0.62 0.77 0.95

75 0.30 0.74 1.03 1.25

100 0.32 0.82 1.13 1.42

Source: Martinez-Anido et al. (2016).

Note: MWh = megawatt-hours.

• National Grid, the TSO in the United Kingdom, information, including temperature data, solar
is using AI and machine learning to help irradiation data, and historic weather data,
predict solar and wind generation. National to reach an output generation figure, which
Grid announced that its new AI prediction is then tested against 80 weather forecasts
models have improved solar forecasting to give an energy generation forecast
by one-third. The new system combines (Cuff, 2019).

12
 A DVA N C E D FO R EC A S T I N G O F VA R I A B L E R E N E WA B L E P OW E R G E N E R AT I O N

• A study by the California Energy within California. The results indicate

Commission evaluates the impact of that potential savings of approximately
improvement in solar and load forecasting USD 2 million per year can be made, based
on system operation in California (California on an average annual CAISO load of 26 GW
Energy Commission, 2019). Solar forecasting and an average regulation cost of USD 9 per
includes forecasting for individual utility-scale megawatt-hour. In addition to financial savings
resources and aggregated behind-the-meter from operation, emission savings should result
resources, which total nearly 6 GW. Possible from the reduction in the need for spinning
improvements to the solar forecasts include reserves (California Energy Commission, 2019).
incorporating age-related degradation;
improving inverter modelling, incorporating • A study conducted for the California
ever-changing amounts of solar capacity; and Independent System Operator (CASIO) shows
handling real-world performance issues, such that improved short-term wind forecasting
as soiling, system outages and shading. in the CAISO market can result in annual
total cost savings1 between USD 5 million
However, the impact of behind-the-meter and USD 146 million, depending on the
solar PV is not considered in forecasting scenario, as shown in Table 5 (Hodge et al.,
loads in California. The study extended the 2015). In the low wind scenario, available
existing load forecasting models to capture wind capacity amounts to 7 299 megawatts
the influence of behind-the-meter solar PV (MW), and in the high wind scenario,
and predict an increasingly volatile load. 11 109 MW of available wind capacity is
The results showed that improvements in expected. The time-based variability in the
solar and net load forecasting methods wind speed determines the instantaneous
can provide positive financial impacts penetration level and the degree to which
in the scheduling and procurement of the forecasting accuracy influences the
electricity in the wholesale electric market actual dispatch of generation.

Table 5 Total cost savings from improved wind power forecasting

Wind Scenario Forecast improvement Annual Savings (USD)

Low 5 050 000

10% uniform improvement
High 25 100 000

Low 14 800 000

25% uniform improvement
High 62 900 000

Low 34 700 000

50% uniform improvement
High 146 000 000

Source: Based on Hodge, et al. (2015).

1 Capacity reserve, frequency regulation reserve and production cost savings.

13
 I N N OVAT I O N L A N DS C A P E B R I E F

III. K
 EY FACTORS TO
ENABLE DEPLOYMENT

A regulatory environment incentivising and decentralised forecasting for VRE generators,

accurate VRE power generation enabling them to revise their schedules and
forecasting dispatch instructions. System operators can also
benefit from the decentralised forecasting of
To enhance the operation of a power system VRE generators if communication between these
with significant VRE shares, the regulatory stakeholders is established and automatised.
and electricity market arrangements need to Penalties for significant deviations in generation
increase time granularity; in other words, the forecasts, compared with actual generation, can
dispatch and scheduling time interval, the be implemented to incentivise improving the
pricing of market time units, financial settlement accuracy of decentralised forecasts.
periods, and the time span between gate closure
and the real-time delivery of power should be Open source systems for weather data
reduced (for more infomration see Innovation collection and sharing
landscape brief: Increasing time granularity in
electricity markets [IRENA, 2019c]). Advances in digital technologies and data science,
such as AI and big data – including predictive
In this way, advanced weather forecasting analytics and machine learning algorithms – can
methods, by delivering granular results and improve weather forecasting. However, these
data closer to real time, become very important algorithms require large numbers of datasets
for VRE power generation. For instance, the to make accurate predictions. Therefore, open
Electric Reliability Council of Texas has reduced sourcing weather data collected by the weather
the dispatch time intervals from 15 minutes monitoring stations of VRE generators, national
to 5 minutes, allowing updates in generation meteorological institutes, and information and
schedules until 10 minutes before the actual power communication technology developers can foster
dispatch. This change in rule has incentivised the rapid advances in data analytical techniques and
use of better weather and generation forecasting consequently in weather forecasting.
methods and has resulted in reduction of wind
curtailment due to better accuracy in generation For weather data collection, a network of weather
forecasts 5 minutes before the actual generation, stations at national or regional level can be deployed
compared with the previous 15 minutes to collect and store long-term meteorological
(Bridge to India, 2017). data, which can be used to characterise renewable
energy resources. The Institute for Environmental
Simultaneously, VRE generators and system Research and Sustainable Development, for
operators must be incentivised to produce example, operates such a network of automated
accurate generation forecasts at different weather stations across Greece. As of December
timescales (i.e. week ahead, day ahead, intraday), 2016, 335 weather stations were operational
providing the same time granularity as that used and providing real-time data at 10-minute time
in the electricity markets. Accurate forecasts intervals. These weather data are used by power
engender different benefits to generators and system operators and private VRE generators,
system operators. For this reason, two types of along with other industries, to plan and forecast
forecast are needed: centralised forecasting for both the load and the generation from various
system operators to maintain overall reliability; power generating resources.

14
 A DVA N C E D FO R EC A S T I N G O F VA R I A B L E R E N E WA B L E P OW E R G E N E R AT I O N

For the last 30 years, the Joint Research Centre installed on turbines to monitor wind speed,
has produced a dataset (known as EMHIRES) temperature and direction. Using such advanced
of wind energy production by the hour at the meteorological devices, which are connected to
national, regional and local levels across the the Internet, may help in gathering information
European Union; these data can be fed into of real-time, site-specific weather conditions (for
advanced weather forecasting models for more more information on Internet-connected devices,
accurate wind power generation forecasts to help see Innovation Landscape Brief: Internet of Things
policy makers devise better energy frameworks [IRENA, 2019d]).
(European Commission, 2016).
However, the impact of these instruments is
At large power plants, meter data are typically still being investigated, and the technologies
available in near real time, and site metadata remain relatively expensive. Advanced
are generally known. For smaller distribution ultrasonic sensors using ultrasound are being
connected plants, dedicated meter data are trialled to measure horizontal wind speed
usually available and recorded but are often not and direction. Sky cameras are also being
telemetered in real time, nor systematically used tested to study cloud coverage, ultraviolet
for system operation procedures. index, cloud movement, cloud heights, sky
polarisation and wind speed at cloud heights.
Advanced meteorological devices Automated weather stations attached to solar
panels or wind turbines to report real-time
Other weather forecasting tools and models that information are also being developed. These
are being experimented with include the use of weather stations contain data loggers and
advanced cloud-imaging technology; sky-facing meteorological sensors that collect and save
cameras to track cloud movements; and sensors weather data for later applications.

15
 I N N OVAT I O N L A N DS C A P E B R I E F

IV. C
 URRENT CONTEXT AND
ONGOING INITIATIVES

Short-term VRE generation forecasting These are applied to high-resolution wind

solutions in Australia turbine data from the supervisory control and
data acquisition system, granular short-term
The Australian Renewable Energy Agency hyperlocal weather forecasts and meteorological
(ARENA) awarded funding of some data. The solution will couple physical and
USD 5.6 million (AUD 9.4 million) to 11 projects statistical models with an industry-best 1 square
to trial short-term forecasting at large-scale kilometre (km²) precision, compared with
wind and solar farms across Australia. The trial traditional global weather forecasts, which
covers at least 45% of the National Electricity operate at coarser temporal resolution and
Market’s registered wind and solar capacity, 16 km² spatial resolution (Utopus Insights, 2019).
which collectively provides a total of 3.5 GW of
renewable electricity generation (ARENA, 2019). Sun4Cast Solar Generation Forecasting
System in United States
The funding is part of ARENA’s Advancing
Renewables Program, and part of the study will The Sun4Cast solar generation forecasting system
focus on improving the Australian Wind Energy combines various forecasting technologies,
Forecasting System’s 5-minute forecast. This covering a variety of temporal and spatial scales,
solution will allow ARENA to more accurately to predict local solar irradiance. Forecasts from
forecast wind generation, reduce wind multiple NWP models are combined via the Dynamic
generators’ dispatch uncertainty and improve Integrated foreCast System2, used for deriving
system stability by balancing the energy supply forecasts beyond 6 hours, and the observation-
and demand in the market. based “nowcasting” technologies3, used for short-
term forecasts ranging from 0 to 6 hours. These
The solution involves applying advanced data technologies are integrated to derive irradiance
science techniques, including deep learning, to forecasts. These irradiance forecasts are converted
deliver greater accuracy in energy forecasts, into expected electricity generation values, which
both for specific sites and technologies and for are then provided to industry partners for real-time
the system as a whole. decision-making (Haupt et al., 2018).

2 The Dynamic Integrated foreCast System uses meteorological data (observations, numerical model output, statistical data,
climate data, etc.) and produces tuned meteorological forecasts at user-defined forecast sites and lead times
(NCAR Research Applications Laboratory, 2017).

3 Nowcasting combines the current state of the atmosphere with a short-term forecast of how the atmosphere will evolve over the
next several hours (Mass, 2012).

16
 A DVA N C E D FO R EC A S T I N G O F VA R I A B L E R E N E WA B L E P OW E R G E N E R AT I O N

EWeLiNE, ORKA, ORKA2 and Gridcast The new models also have better weather
projects improving VRE generation warnings, which are adapted to grid operation,
forecasts in Germany especially when faced with extreme weather
conditions, such as strong winds. Therefore, solar
Since 2012, the Deutscher Wetterdienst (German radiation data are calculated every 15 minutes,
Meteorological Service) has been working on enabling the system operators to better estimate
optimising its weather forecasts for renewable if they need additional resources to maintain grid
energy applications within the research projects stability. As a next step, the Gridcast project
EWeLiNE and ORKA, funded by the Federal aims to integrate satellite images for solar
Ministry for Economic Affairs and Energy. On forecasts in addition to the existing weather
the basis of the findings of the ORKA project in forecast data, thereby helping system operators
December 2015, this successful co-operation was better manage the system with high shares of
continued in a new project, ORKA2, implemented VRE (IRENA, 2020).
in 2016, followed by a new project called
Gridcast in 2017. In these projects, the German Hybrid Renewable Energy Forecasting
Meteorological Service and the Fraunhofer for advanced wind forecasting in China
Institute for Wind Energy and Energy System
Technology worked with the three German TSOs The Hybrid Renewable Energy Forecasting
(Amprion GmbH, TenneT TSO GmbH and 50Hertz (HyRef) project installed at a hybrid 670 MW
Transmission GmbH), one manufacturer of wind solar and wind generation unit, created by IBM
energy systems and two DSOs. for the Chinese State Grid’s Jibei Electricity
Power Company Limited, is using advanced
Their goal was to improve the weather and power data analytics to improve the forecasting of
forecasts for wind turbines and solar PV plants wind power output. By using advanced tools for
and to develop new forecast products focusing weather modelling, cloud-imaging technology
specifically on grid stability. These projects allowed and sky-facing cameras, paired with sensors on
them to use real-time data from solar panels and the wind turbine, HyRef can forecast the power
wind turbines around Germany and to feed those output months ahead, but also up to 15 minutes
data into an algorithm that uses machine learning before actual generation. As a consequence of
to calculate the renewable energy output. After using these technologies, wind curtailment has
testing, the researchers concluded that the newly been reduced by 10%, which is the equivalent of
developed forecast models have better forecast supplying some additional 14 000 homes with
accuracy, with higher temporal and spatial electricity (Cochran et al., 2013).
resolution, than conventional models.

17
 I N N OVAT I O N L A N DS C A P E B R I E F

V. IMPLEMENTATION
REQUIREMENTS: CHECKLIST

Hardware:
• Smart meters to monitor real-time power production
• Weather sensors to continuously monitor changes in weather
TECHNICAL
REQUIREMENTS • Wind turbine sensors to monitor the functioning of wind turbines
• Regional/national network of weather stations to continuously monitor weather patterns
• Sky imagers
• Satellite data

Software:
• Advanced weather forecasting tools based on a combination of input data
(historical data, real-time data, etc.)
• Advanced power generation forecasting tools based on weather forecasting and power plant
parameters (e.g. availability)
• Data analytics software (i.e. AI software, such as machine learning platforms)

Communication protocols:
• Common interoperable protocol co-ordination between VRE developers, asset owners,
system operators (and consumers)

Retail market:
• Establish common open databases of weather data that can be accessed for free or at a small cost,
REGULATORY which is necessary to bring down costs for smaller asset owners; these databases can have a large
REQUIREMENTS impact on the grid when aggregated
• Enable visibility of generation from distributed energy resources, like small-scale wind and solar PV
generators (typically sub-megawatt)

Transmission and distribution system operators:

• Allow DSOs to monitor power generation data from distributed solar and small-scale wind projects
• Allow TSOs to integrate real-time VRE forecasts into the dispatch schedule
• Grid codes that cover data requirements for efficient VRE power generation forecasting

Wholesale market:
• Reward accuracy and penalise large deviations in scheduling and dispatch from market participants,
where markets are in place

Policy makers:
• Support the development of advanced forecasting tools for weather and VRE power generation
• Deploy mechanisms to record weather data at various locations to accurately forecast VRE power
STAKEHOLDER generation
ROLES AND
RESPONSIBILITIES System operators:
• Enable transmission and/or system operators to play the role of “data custodians”
• In liberalised markets, allow authorised third-party vendors to use data for developing
new forecasting tools

VRE generators:
• Equip VRE plants with supervisory control and data acquisition systems, sensors and software
for accurate power generation forecasts
• Make use of the most advanced weather forecasting technology for optimised VRE generation
(to participate in markets or to inform system operators of the forecasted generation output)

Meteorological and research institutes:

• Collaborate with the renewable power industry, including system operators, to refine the weather
forecasting tools to improve VRE integration into the power system

18
 A DVA N C E D FO R EC A S T I N G O F VA R I A B L E R E N E WA B L E P OW E R G E N E R AT I O N

ABBREVIATIONS

AI artificial intelligence km² square kilometre

ARENA Australian Renewable Energy MW megawatt

Agency
NWP numerical weather prediction
CAISO California Independent System
Operator PV photovoltaic

DSO distribution system operator REMC Renewable Energy Management

Centre
GW gigawatt
TSO transmission system operator
HyRef Hybrid Renewable Energy
Forecasting VRE variable renewable energy

BIBLIOGRAPHY

Akhter, M.N. et al. (2019), “Review on forecasting California Energy Commission (2019),
of photovoltaic power generation based on Improving Solar and Load Forecasts by Reducing
machine learning and metaheuristic techniques”, Operational Uncertainty, Energy Research and
IET Renewable Power Generation, Vol. 13/7, Development Division, San Diego,
Institution of Engineering and Technology, https://ww2.energy.ca.gov/2019publications/
Stevenage, United Kingdom, pp. 1 009–1 023, CEC-500-2019-023/CEC-500-2019-023.pdf.
https://ieeexplore.ieee.org/document/8707118.
Cochran, J. et al. (2013), Market Evolution:
ARENA (2019), “$9 million funding to enhance Wholesale Electricity Market Design for 21st
short term forecasting of wind and solar farms”, Century Power Systems, National Renewable
Australian Renewable Energy Agency, Energy Laboratory, Golden, Colorado,
https://arena.gov.au/news/9-million-funding- www.nrel.gov/docs/fy14osti/57477.pdf.
to-enhance-short-term-forecasting-of-wind-and-
solar-farms. Cuff, M. (2019), “AI-powered weather forecasts
are improving predictions for smart grids’ energy
Asian Power (2020), “India inaugurates 11 outputs”, GreenBiz, www.greenbiz.com/article/
renewable energy management centres”, ai-powered-weather-forecasts-are-improving-
https://asian-power.com/power-utility/news/ predictions-smart-grids-energy-outputs.
india-inaugurates-11-renewable-energy-
management-centres. European Commission (2016), “How much wind
energy can be generated in your area? Check
Bridge to India (2017), “Southern region to lead EMHIRES”, EU Science Hub, 23 November,
India in grid integration of renewable energy – https://ec.europa.eu/jrc/en/news/how-much-
Bridge to India”, PV Tech, 26 May, www.pv-tech. wind-energy-can-be-generated-your-area-check-
org/guest-blog/southern-region-to-lead-india-in- emhires.
grid-integration-of-renewable-energy-bridg.
Haupt, S.E. et al. (2018), “Building the
Bullis, K. (2014), “Smart wind and solar power”, Sun4Cast system: Improvements in solar
MIT Technology Review, May/June, power forecasting”, Bulletin of the American
www.technologyreview.com/s/526541/smart- Meteorological Society, Vol. 99, American
wind-and-solar-power. Meteorological Society, Boston, pp. 121–136,
https://journals.ametsoc.org/doi/abs/10.1175/
BAMS-D-16-0221.1.

19
 I N N OVAT I O N L A N DS C A P E B R I E F

Hodge, B.-M. et al. (2015), The Value of Jerez, S. et al. (2013), “The impact of the
Improved Short-Term Wind Power Forecasting, North Atlantic Oscillation on renewable energy
National Renewable Energy Laboratory, resources in southwestern Europe”, Journal of
Golden, Colorado, www.nrel.gov/docs/ Applied Meteorology and Climatology, Vol. 52,
fy15osti/63175.pdf. American Meteorological Society, Boston,
pp. 2 204–2 225, https://journals.ametsoc.org/
IBM (2011), “Vestas Wind Systems turns to IBM doi/full/10.1175/JAMC-D-12-0257.1.
big data analytics for smarter wind energy”, IBM,
24 October, www-03.ibm.com/press/us/en/ Martinez-Anido, C.B. et al. (2016), “The
pressrelease/35737.wss. value of day-ahead solar power forecasting
improvement,” Solar Energy, Vol. 129, Elsevier,
IRENA (2020), Innovative solutions for 100% Amsterdam, pp. 192–203.
renewable power in Sweden, IRENA, Abu
Dhabi, www.irena.org/publications/2020/Jan/ Mass, C. (2012), “Nowcasting: The promise of
Innovative-solutions-for-100-percent-renewable- new technologies of communication, modeling,
power-in-Sweden. and observation”, Bulletin of the American
Meteorological Society, Vol. 93, American
IRENA (2019a), Innovation landscape for a Meteorological Society, Boston, pp. 797–809,
renewable-powered future: Solutions to integrate https://journals.ametsoc.org/doi/abs/10.1175/
variable renewables, International Renewable BAMS-D-11-00153.1.
Energy Agency, Abu Dhabi, www.irena.org/
publications/2019/Feb/Innovation-landscape-for- NCAR Research Applications Laboratory (2017),
a-renewable-powered-future. “Dynamic integrated forecast (dicast®) system”,
https://ral.ucar.edu/projects/dynamic-integrated-
IRENA (2019b), Innovation landscape brief: forecast-dicast%C2%AE-system.
Artificial intelligence and big data, International
Renewable Energy Agency, Abu Dhabi, NREL (2016), Forecasting wind and solar
www.irena.org/publications/2019/Sep/Artificial- generation: improving system operations,
Intelligence-and-Big-Data. National Renewable Energy Laboratory,
Golden, Colorado, www.greeningthegrid.
IRENA (2019c), Innovation landscape brief: org/resources/factsheets/copy_of_
Increasing time granularity in electricity markets, ForecastingWindandSolarGeneration.pdf.
International Renewable Energy Agency,
Abu Dhabi, https://irena.org/-/media/Files/ Orwig, K.D. et al. (2015), “Recent trends in
IRENA/Agency/Publication/2019/Feb/IRENA_ variable generation forecasting and its value
Increasing_time_granularity_2019.pdf. to the power system”, IEEE Transactions on
Sustainable Energy, Vol. 6/3, pp. 924–933,
IRENA (2019d), Innovation landscape brief: https://ieeexplore.ieee.org/document/6996049.
Internet of Things, International Renewable
Energy Agency, Abu Dhabi, www.irena.org/ Utopus Insights (2019), “Vestas and Utopus
publications/2019/Sep/Internet-of-Things. Insights to pilot energy forecasting solutions for
renewable power plants in Australia”,
www.utopusinsights.com/australiapilot.

20
ADVANCED FORECASTING OF VARIABLE
RENEWABLE POWER GENERATION
INNOVATION LANDSCAPE BRIEF

© IRENA 2020

IRENA HEADQUARTERS
P.O. Box 236, Abu Dhabi
United Arab Emirates

www.irena.org