Study on the Physical Approach to Wind Power Prediction

 

Study on the physical approach to wind power prediction aims to reproduce the atmospheric processes related to the wind by solving the equations of conservation of mass, momentum and energy in the airflow.

One of the benefits of the physical forecasts is that, in opposition to statistical approaches, results can be obtain from initial and boundary conditions without considering historical series.

Furthermore, the calculations of physical models are perform in a given domain, generating gridded results, and thereby providing spatial wind descriptions which are difficult to obtain from statistical approaches study on the physical approach to wind power prediction.

However, these models are computationally expensive, and they involve high delays in the generation of results, particularly when high resolution is required.

Wind energy forecasts for wind farms would require a high resolution in order to consider local features, but the usability of predictions also require short computational time.

Thus, wind energy forecasts for wind farms are rarely based exclusively in physical approaches, especially for forecasting horizons shorter than a few hours.

However, forecasts obtained by physical models (as those provided by NWP models) are a cornerstone in wind energy management.

3.3.1 Numerical weather prediction(Study on the Physical Approach to Wind Power Prediction)

The term ‘NWP model’ is associated to dynamical models which perform meteorological predictions, not only related to wind power.

All NWP approaches are essentially similar as they solve the hydrodynamical and thermodynamical equations of the system atmosphere-surface.

They only differ on scales, parameterizations and resolution of the output grids.

The NWP models can be divide in global models and regional models, also called limited area models.

3.3.1.1 Global models

Global NWP models calculate the current state of the atmosphere and its evolution at planetary scale.

The level of data resources and computational power necessary to carry out these predictions is only accessible for international and state institutions.

The two most frequent global NWP data sources use in wind power predictions are the Global Forecasting System (GFS) manage by the National Centre of Environmental Prediction (NCEP) and the forecast products of the European Centre for Medium-Range Weather Forecasts (ECMWF).

These institutions typically disseminate the results via web server, allowing easy and computer-friendly access to their forecasts.

As a result of the involved resources and the accessibility, global NWP models are used as input in most of the physical and statistical wind power forecasting models study on the physical approach to wind power prediction.

The NWP models

The NWP models are not only focuse on providing wind information, but also on producing estimations of a wide range of meteorological variables.

Thus they implement different modules in order to deal with different geophysical elements: ocean, surface, ice, radiation, etc.

The evaluation of these elements to perform global meteorological forecasts also requires a geophysical description of the scenario, including topography and land surface properties.

In the case of global models

In the case of global models, the scenario is the Earth. Hence, in order to maintain appropriated computational times, the geophysical description of the Earth must be done with a coarse resolution.

Concretely, the domain is divide into cells with a resolution of dozens of kilometres with a basic topographical/geophysical description:

mean height elevation above sea level, percentage of land/sea and an approximate description of the surface coverage.

As a result of the resolution in the geophysical description, global NWP models cannot reproduce local atmospheric features study on the physical approach to wind power prediction.

An important element in the global NWP models

An important element in the global NWP models is the data assimilation system which allows updating feedback between real and calculated meteorological data.

The data assimilation system introduces the real information that is compose of remotely sense data acquire by space-borne instruments, and in situ measurements from surface weather stations, ships, buoys, radiosonde stations and aircraft.

The data assimilation system, after quality control, processes all the valid measurements to build a sequence of past situations which support the initialization of the model study on the physical approach to wind power prediction.

In the cases of GFS and ECMWF

In the cases of GFS and ECMWF, both deliver their results four times a day at 00, 06, 12 and 18 UTC.

Each delivery is compose of a sequence of forecasts which in the case of ECMWF cover 6 days in steps of 3 h, and 4 additional days in steps of 6 h.

GFS delivers 5 days of hourly forecasts, 5 additional days of 3-hourly forecasts and 6 additional days of 12-hourly forecasts.

The spatial resolution depends on the chosen product, but the highest resolution is comparable in both models ranging around 10 km.

There are other global NWP models as the Unified Model of the UK Meteorological Office, the Analysis and Prediction of the Australian Bureau of Meteorology or the GSM managed by the Japan Meteorology Agency, but their use is in wind power forecasting is exceptional respect GFS and ECMWF.

The results of global NWP data have two important drawbacks for a direct application to wind power forecasting.

Firstly, results are deliver with a delay of few hours as result of the computing time. Thus, forecasts of the shorter horizons are pass in the time of delivery.

Secondly, the coarse resolution and topographic simplification requires additional calculations for local applications, which is the case of wind farms.

For instance, in, ECMWF advise about the weakness of their near-surface wind forecasts in mountainous areas resulting from the ‘highly varying subgrid orography’ study on the physical approach to wind power prediction.

3.3.1.2 Limited area models

The limited area models (LAM) are basically similar to the global models comment in the previous subsection, but they are design to be apply in a concrete area.

The main purpose of these models is twofold: refining of the global models' output and including local features.

The applied equations

The applied equations are similar although they can include specific modules to deal with local phenomena, as turbulence, clouds or eddies.

LAMs are able to provide useful forecasts for wind power production but, as the resolution and local details increase, more computation and time is require to produce the results and, consequently, these approaches are not efficient for shortest horizons.

The most extended LAM is the Weather Research and Forecasting (WRF) model, develop by NCEP and able to work with GFS data.

The dissemination of the WRF model has been benefit by the previous diffusion of the fifth-generation mesoscale model (MM5) which can be consider its origin.

Furthermore, the code is freely downloadable in the WRF website, and enhance code can be also propose by the WRF community, being the model in permanent evolution.

As a result of this dissemination multitude of WRF configurations have appeare for the different modules and computational schemes, as the land surface model,

the planetary boundary layer or the microphysics scheme.

In the WRF model

In the WRF model, feed with GFS data, is run to test how these different configurations affect the WRF wind forecasts.

WRF is able to achieve resolutions of tens of meters but, as said, consuming important amounts of time and computation.

Thus, this level of resolution is only apply in long-term wind resource assessments in which computational times are not as critical as in wind forecasting.

There are numerous organizations that run WRF to produce forecasts for concrete areas.

Some of these organizations, as well as the downloadable code and additional information of the model, can be find in The High-Resolution Limited Area Model (HIRLAM) is other LAM with extended use, although its diffusion is not comparable to WRF.

HIRLAM

HIRLAM has been develop by a European consortium and it is use by different meteorological agencies of Europe.

The model is adapt to use ECMWF data as initial and boundary conditions, producing grids up to 5 km.

As a third step, the HIRLAM output can be also process by HARMONIE to obtain resolutions of 2.5 km.

As in the case of the global models, there are other LAMs with lower diffusion, as Consortium of Small-scale Modelling (COSMO) led by Federal Office of Meteorology and Climatology Meteoswiss, the Lokal-Modell managed by the German Weather Service or the North-American Mesoscale Forecast System (NAM) run by NCEP.

3.3.1.3 Competitive ensemble forecasting

As NWP results are solutions of deterministic equations, they produce single-value forecasts.

As comment in Section 2.3, this expression of wind predictions is not fully adapt to the wind energy integration problems which are benefit by an uncertainty evaluation of these results.

Competitive ensemble forecasting is the way to obtain uncertainty information from NWP.

Ensemble forecasts are produce by running NWP models under different initial conditions or different parameterizations, which leads to a set of different results.

This set of predictions is then process under the framework of wind power uncertainty analysis to adapt the results to the desired way of expression.

Regarding global NWP forecasts, both ECMWF and GFS implement ensemble products. Some works have analysed and compared their performance regarding wind speed and directions [36–38].

LAMs also can  run under different configurations to provide ensemble forecasts.

In [39], the WRF model is run to perform ensemble wind forecasts.

3.3.2 Physical approaches focused on wind forecasting

NWP data are the base of the wind power predictions, especially at a regional scale and horizons longer than a few hours.

However, in most cases, NWP forecasts are post-process to a better adaptation to the wind energy context by using statistical approaches, in order to not add additional delays.

Foley et al. compile 15 wind forecasting systems which process NWP data, showing than only two of them are purely based on physical considerations:

Predictor (developed by Riso, Denmark) and SOWIE (developed by GmbH, Germany).

The rest of them are based on statistical or hybrid approaches.

Other physical approaches for wind power forecasting consider computational fluid dynamics to translate the NWP data to a concrete local scenario.

In the complex numerical simulations needed to perform these simulations.

In order to reduce the computation time, Li and Liu propose a pre-calculation of flow fields at the local scale to conform to a database.

Then, each NWP situation is associate with a local pre-calculate flow field, thereby reducing the computational time require to generate the local prediction.