Aiolos Forecast Studio 11.3

11.3 is a major release with a lot of amazing new features. Main aim with the release has been Machine learning enhancements, automatic model weighting, price dependent forecasts, Indata improvements with automatic availability handling for windpower, GUI improvements and introduction of Regimes. Below follow the news in abbreviated form, but see also our Educandi-session on the topic. 382 improvements have been made since last version.

Indata Control

  • Automatically detect partial or complete stops in the History data for Wind power plants with the new function Optimize Availability for Wind power. 
  • Optimize Availability handling for Wind power. Automatically handle detected stops by adding new Scheduled items.
  • New visualizations with the diagrams Power Curve and Scatter plot for all types of Energy series.
  • Zoom box for the new diagrams Power curve and Scatter plot.
  • Batch name from autopar-file now displayed at available Indata control imports to easier find desired one.
  • New Detection to find Last values in import or from Now. This can be used to detect and prevent wrongfully sent future values. It can also in combination with Multiple Breach with another rule be used to have stricter rules for the last set of data. 

More Indata Control

  • Sort and Filter function added to the Series tree to work with the biggest series first.
  • Scheduler reachable from Indata Ribbon.
  • Stop button to abort a wrongfully started Indata run.
  • New Detection algorithm to find Constants in the Historical data
  • New Detection algorithm to find zeros in the Historical data
  • New function Update Max, which can automatically detect an increased Max of a series and handle it with by automatically adding the new Scheduler type of Update Max.

Scheduler

  • New Scheduled item type called Update Max, which can both increase and decrease a Forecast series Max. The advantage with using the scheduled item rather than changing directly in Config, is that the forecasting model can use data from both before and after the change and explicitly understand that a level change has occurred.
  • Multiple parallel use of the Scheduler. Instantly see new schedules created in another instance of Aiolos or automatically created from Indata when opening the Scheduler.
  • Scheduler now remembers last chosen Scheduled item type, so it’s faster to add new items of the type you want

Home and View

  • History moved to quick buttons in Diagram for faster access
  • Calendar added in Diagram to freely set a history period
  • Daytypes from chosen Forecast series’ Calendar visual in Home with coloring in Diagram, and as tool tip in Diagram and in Values grid. Coloring steered in Config for the Daytype.
  • Availability (Max output) now displayed for Forecast series with Data Type Wind or Solar. Earlier it wasn’t displayed for Solar and needed to be a specific Statistical model type for it to be displayed, which didn’t fit the rise of new AI models.

More Home and View

  • Resolution added to visualize forecast and history in other resolutions, such as Hourly, Daily, Weekly, Monthly values. This gives an overview of the forecast level in both Values table and Diagram. Previous function Totals that could only show values table in one step, deprecated with this and thus removed.
  • Sort and Filter function moved to icons near the Series tree from the Ribbon to be closer where it is needed.
  • Model weights GUI added to visualize and weight between different Models.
  • Thinner lines for validation curves, to enhance the forecast
  • Less background lines to give a cleaner look. Previous line setting possible to get back by right-clicking in Diagram if needed
  • Copy Forecast series name by right clicking in Series tree or by using short cut key commands Ctrl+C.
  • GUI improvements. Diagram surface increased by rearranging surrounding texts and functions. See separate document “11.3 Version map.png” for further details.
  • Stay zoomed in setting via right click in Diagram. When changing between different Forecast series the setting keeps current zoom if you are zoomed in. Default behavior is otherwise to zoom out when changing series.

Forecasting

  • AI multivariate model with multiple outputs allows for simultaneous forecasts for a number of series, which improves calculation performance considerably.
  • Significantly improved calculation performance for Aiolos AI I models.
  • New weather independent AI I Industry model
  • Price dependent forecasts added to Aiolos. Spot price can be added as a model predictor and to steer a wanted behavior. Example can be wind power that shuts down during negative prices, or energy consumption decreasing at high prices. 
  • Regimes added to Aiolos. Steer model weighting and forecast model behavior according to different scenarios (Regimes), such as high/low wind speeds, temperatures, time of day, day of year.
  • New SerieType Aggregated Forecast & Has History, that handles the situation where child nodes and parent both has historical data. Example can be that the child nodes are wind turbines with real time measurements, and the parent is the whole wind park of all turbines that has summed Settled measurements. The new serie type can use both types of data to make a forecast.

Follow Up

  • Automatically weight between different Models to optimize usage across multiple models. Can also be run as a batch job to continuously optimize the model usage.
  • Sort and Filter function added to find the biggest series first
  • Model weights GUI added to visualize and weight between different Models
  • Resolution improved to give an overview of values in Table and in Diagram
  • Rarely used function Aggregations moved from Ribbon to Extracts menu
  • Follow up Ribbon box name changed from Selection to Weather to be clearer of what it contains.

Config

  • MasterDataSync, steer Modelsettings when syncing in new Forecast series. Example can be to set a specific DaytypeCalendar when adding a new Forecast series, but possible to steer any model parameter
  • MasterDataSync, autoconnect new Forecasts series to desired templates to include them in the export automatically
  • More Model places, now 18 to give room for a continuously growing amount of AI and other models
  • Relative file paths (<>\) now available for MasterDataSync and SQL import
  • Automatic backup of aiodef performed every time before MasterDataSync to backup folder in <>\Data.
  • Filter out “not empty” in Grid added as new function
  • StdTextEx added as new Export protocol
  • Wildcard introduced for LoadImport, in autopar file to pick out which LoadImportProtocolName that should be used, so multiple can be runned from the same autopar-file
  • Calculate Distance from Forecast series to connected weather points. Can be used to find weather points that have a longer distance than desired, to go further and order new points.

Miscellaneous functions

  • New Graphical look across the whole software that is more uniform and stylish. 
  • General bug fixes and UX improvements
  • New cleaning function in System settings to Merge Scheduled items to aiofile to free up space in aiodef, which makes everything in Aiolos faster. Applies to customers with a lot or old scheduled items
  • New log level categories added to steer amount of information more in detail from System settings.
  • A version map between 11.1 and 11.3 to ease up the navigation changes between the versions added to the Help section (found at the question mark “?”) in the menu bar.
  • Configlog moved to datasbase (aiodef.db) from textfiles and expanded from a few weeks to include last 2 years of changes as default. The configlog is a log of changes made in the system to see what, when and by whom a change in Config is implemented.

Weather

Weather can now be fetched from Bazefield for weather measured directly at the production park.
Increased use of actual weather forecasts by prolonging a short weather forecast with an older longer from the same weather provider if available before switching to Normal weather. Some weather providers send short weather forecasts (12 hours) at occasions, where it is preferred to use older weather forecasts instead of normal weather.
Increased use of actual weather forecasts by prolonging a short weather forecast with a forecast from another weather provider if a longer one is available from another provider, before switching to Normal weather. Some weather providers have 10 days long forecasts, which reduces the need of Normal weather.