Who’s not interested in knowing exactly how the weather will be – when you’re planning on seeing your favorite football team play, going for a hike in the woods, booking a beach holiday, or trying to teach your kid how to fly a kite?
The same curiosity, combined with the need of risk mitigation and profit optimization applies for energy traders, renewable energy developers, energy companies and utilities.
Let’s focus now on the ‘HOW’, i.e. ways of fine-tuning and constantly refining the inputs and the outputs of the wind or solar power energy generation formula.
A typical short-term(*) power production forecasting approach covers the following input information:
- Wind or solar farm location
- Installed capacity of the farm
- Historical data from the farm:
- production data – at least one year, but preferably as much as available (unless, of course the farm is brand new)
- local weather measurements from a met-mast or a measurement tower (if ever installed)
(*) Short-term is typically defined as intra-day (today), day-ahead (tomorrow), and up to 7 or sometimes 10 days.
To sum up, we need as much information as possible about the wind or solar farm itself. As a minimum, a typical forecasting company will need to know the installed capacity and the geographical location (latitude and longitude) of the farm.
The next step is adding the weather element to the equation. The de facto industry standard is to use weather data input from a Numerical Weather Prediction (NWP) model. Now, each model on the market behaves slightly differently. One may be better during summer, one may be better at predicting weather in difficult topography terrains, while yet another one. Well, would be completely useless.
For the specific site(s), the same model can also be very good at one location but at the same time bad at a neighbouring location
Weather models also come in different resolutions. The higher the resolution, the smaller the grid cell and the easier it is to get precise weather information about the location of your specific farm.
So, as you can imagine, if you pick the wrong weather model, you will get very inaccurate forecasts. What a good forecasting company then does is that they use multiple NWP models and decide on a relevant mix. What this means is that they are taking the best weather data and the physics of the model and apply it to your farm in a weighted manner. For example 40% from the main and most reliable model, 30% from the secondary model, and 20% from the 3rd model.
The ranking of the models does not necessarily depend only on absolute accuracy. But also on what they bring to the table. That is whether they add new information and hence value to the model mix – there is no benefit in adding more of the same.
An experienced wind and solar power forecasting company will have access to multiple weather models. Typical global NWP models that are in use in the energy market are:
- The European ECMWF model
- The American GFS model
- The UK UKMO model
Some, like ConWX, will also have their own meso scale (i.e. high resolution) models available.
So, data input number 2 is the weather and multiple weather data inputs result in 15-20% more accurate forecasts.
Only the most experienced ones, with huge customer bases, will have enough big data to work on and implement additional ways of improving the very short-term forecasts. Typically defined as the intra-hour (within the next hour and up to 6 hours ahead).
Also, only the most knowledgeable forecasting companies will apply AI and machine learning techniques to constantly learn from the big data and teach their forecasting algorithms.
It’s important to emphasize here that you should only trust a forecasting supplier that can guarantee a ‘Chinese wall’/a sealed no-access border between your data and the data of all their other customers. After all, you wouldn’t like your competitor to benefit from your own data.
How to improve short-term forecasts
Let’s have a look at what other improvements of a short-term forecast can be expected from a forecasting company like ConWX:
The most common add-on is using real-time production data from the wind or solar park to further fine-tune the next 6 hours of the intra-day forecast. A typical improvement that you can expect is 30-50% (YES! That much!). These corrections are often referred to as PERSISTENCE. If the client has access to high-frequency production data that they can share with their forecasting provider every 5 or 10 minutes. Then an additional improvement of 5-10% can be achieved.
Short Term Network System
What if you could use farms in your own portfolio to detect weather changes that will affect the rest of the portfolio downstream? Depending on how the climatic front moves. That is depending on how the wind blows. Changes observed in one or a group of farms can be used to further fine-tune the very short-term forecasts on other neighbouring farms. If an improvement of 25% on 8% of the worst events is not impressive, then I don’t know what is!
It’s obvious given that all of the farms must belong to the same customer of ours (alternatively a consortium of energy companies). As mentioned above, at ConWX, we would never share data between our customers.
Offshore wind is increasingly gaining momentum in most parts of the world. Finding targeted solutions that can help increase the quality of short-term power production forecasts in the offshore environment is therefore crucial. At ConWX, we have been able to reach an improvement of 13% over the first 20 minutes of the intra-day forecasts utilizing readings from 1 to 2 LiDARs located on the perimeter of a typically large offshore park. LiDARs can, as a matter of fact, be utilized onshore as well.
In areas with an extremely high concentration of wind energy sources, such as e.g. Germany or the Netherlands. Curtailment (or Einsmann in German) is what everybody fears. Curtailment can however be forecasted as well. And by curtailment, we mean here the actions taken by the grid operator to balance the grid by cutting off chosen wind parks. At ConWX, we have developed a method that results in 5-8% improvement of the intra-day forecast.
Yet another important parameter that a good forecasting company will need to take into account is icing. It comes in different forms and not all the icing can be forecasted based on the data available in the NWP models. Do yourself a favor and help your forecasting company with data input to the following questions:
- Are there any real-time observations (ice, temperature, humidity)?
- Is there heating in the blades, and how is it being operated?
- What is the response time?
- Can heating be started automatically?
A very crucial element of a reliable forecasting service is constant performance monitoring of the quality and accuracy of the forecasts; especially in difficult ramp situations.
The common denominator for all of the above-mentioned possibilities of improving and fine-tuning a short-term wind and solar power production forecast is data. You can never go wrong with sharing the data. And you can never provide your forecasting company with too much data. As you can probably guess, close and trust-based cooperation between an energy company and the forecasting company is the core of being able to generate the highest possible production forecasts.
Remark: the part that we anticipate a good forecasting company is taking into account while generating wind and solar power production forecasts is farm availability. Meaning the times when the farm is actually available to produce energy and there is no service or maintenance being done. Plus various types of shut-downs, such as e.g. shut-downs due to too strong winds to prevent turbine overload and noise. As well as visual and environmental curtailments.
Wind and solar energy sources are entirely dependent on weather. Full stop. Mastering the ability to forecast the weather and hence the energy produced by wind and solar energy sources is the most important task that energy traders, renewable energy developers, energy companies, and utilities are facing today. Their raison d’etre, financial results, and everyday operational tasks highly rely on accurate power production forecasts.