How can I verify your day-ahead accuracy?
The Historical Load Forecast feature in our TESLAWeb interface allows you to display our day-ahead forecast against the observed data. Accuracy statistics for that forecast will be displayed in the Statistics view. Alternatively, you can contact your local TESLA office. We maintain accuracy statistics for all of our forecasts as well as for many of those provided by regional or national transmission operators or wholesale market operators. Ultimately, we offer free trials of any of our forecasts to allow you accumulate the data you need to be sure that TESLA forecasts are the most accurate available.
Why has the forecast changed dramatically since yesterday?
Dramatic changes to the weather data supporting a given forecast can cause dramatic shifts in our demand or generation forecasts. Additionally, our analysts are constantly updating and evaluating the models used to generate the forecasts. These updates will not often cause dramatic changes in the forecast, but will certainly contribute to minor changes and may coincidentally combine with small weather data changes to result in dramatic demand or generation forecast changes.
What is Stage 2?
Stage 2 is our proprietary forecast error correction filter. It works by evaluating our forecasts against incoming observed data, analyzing the error patterns, and forecasting those error patterns into the forecasts. The result will be lower forecasts where we’ve been running high for a few hours, and higher forecasts where we’ve been running low.
Why does the load shape look so jagged?
This is a possible side effect of our Stage 2 error correction filter. Often toggling Stage 2 off will result in smoother forecasts. Alternatively, forecast jaggedness may be a sign that our model is over-specified and needs to be more finely tuned by our analysts in order to smooth it out. In general, the jagged forecast will be statistically superior, but obviously not what is likely to be observed.
Why is there a forecast of zero?
While this is not uncommon when forecasting renewable generation, it is highly irregular in a demand forecast. It is likely caused by a missing data input, such as weather or customer count data, or by a deficiency in the underlying model. Please contact your local TESLA office, and we will be happy to resolve the issue as quickly as possible.
When are your wholesale market forecasts published?
All of our wholesale market forecasts are published no less than hourly using the most up-to-date data available. In addition, we make every effort to tune the underlying models each day prior to the close of bidding to ensure that our last forecast before the market closes is our best possible.
Do you offer file delivery via both FTP and SFTP?
Yes. We certainly do.
How can I learn to use your tools?
We are happy to provide printed documentation as well as live demonstrations of our products and services via the internet at your convenience. Alternatively, we can also send members of our team to train members of yours.
How much data do you need to build a TESLA model?
In an ideal world, eleven years is the optimal amount of historical demand data as it allows for observation of moving holidays on every day of the week. However, we’ve built quality models with far far less. Let us take a look at what you have available, and we’ll let you know what we can do with it.
Can I view the forecast by trading period instead of time of day?
Yes. Please visit the Date Time Format page from the Settings menu in TESLAWeb and tick the Display Period box.
I already have an interface with which I am comfortable. Do I have to use yours?
Absolutely not. Our forecasts can be accessed as flat files via FTP or via a web service for integration into your interface.
Can I use TESLA software to build my own forecasting model?
Not at this time. We are working on making this option available. Please contact your local TESLA office to see how that project is progressing.
Does the y-axis have to begin at zero?
No. Please visit the Graph Settings page from the Settings menu in TESLAWeb and tick the Variable Y-Axis box.
Do you offer energy price forecasting?
We do not. However, depending on the market, we may be able to direct you to someone who does. We’re happy to help wherever we can.
Can you customize the format of my forecast output files?
Of course. We have experience with a wide variety of formats. Just let us know what you need.
Can I view sub-hourly data aggregated hourly?
Yes. Please visit the Miscellaneous Settings page from the Settings menu in TESLAWeb and tick the Allow Aggregation box.
I disagree with your forecast. What should I do?
Please do let us know. A TESLA forecast is a service, not a number, so we are happy to double and triple check our forecasts upon request. There are also options within our interfaces to allow you to edit our forecasts. Please contact your local TESLA office for more details.
How reliable is access to your forecasts?
We make every effort to ensure that your forecasts are there when you need them. Our forecasts are supported by two parallel systems on two different continents. If our primary systems are taken offline for any reason, you will be seamlessly redirected to our secondary systems. If you are currently having trouble accessing your forecasts. Please contact your local TESLA office for immediate assistance.
MAPE or RMSE?
Mean Absolute Percent Error, MAPE, measures forecast errors as a percentage of the demand being forecast. By definition, this approach puts greater weight on periods of lowest load. Overnight loads are more influential than peak loads, weekend loads more than weekday loads, and holiday loads more than non-holiday loads. Using MAPE places the greatest emphasis on times when power prices are low.
If the TESLA forecast is off by 100 MW when the load is 10,000 MW at the peak hour for the day. This represents a 1% Absolute Percentage Error, APE. That same 100 MW error in the middle of the night when the load may be 5,000 MW represents an APE of 2.0%. When you average the APEs to get a MAPE of 1.5%, the night-time error impacts the MAPE by twice as much as the same size error at the daytime peak. This offers a distorted view of what is really at stake when power prices are high.
By contrast those two errors contribute the same amount of influence to the Root Mean Squared Error, RMSE, which measures forecast error in terms of megawatts. The RMSE between those two observations is 100 MW. This offers a more meaningful view of what’s really at stake at any given moment, including those times when power prices are high.
What kind of support can I expect during my free trial period?
The same level of support that you will receive should you decide to go to contract and the same level of support on which we’ve built our reputation.