Welcome to the Energy Twin (ET) extension for SkySpark. Energy Twin is a machine learning-based SkySpark extension for energy data analysis and prediction.
There are three ET libraries available on StackHub:
- ET Analytics
- advanced ML library using detailed models with up to 15min sampling
- ideal for building electrical energy consumption analysis and prediction
- anomaly detection - predefined Sparks
- tailored KPIs
- based on the TOWT model by LBNL
- free to download, licensed by ET Core
- ET Interactive
- simple to use regression analysis using degree-days models and ASHRAE change point models
- interactive model tuning and outlier removal
- perfect for heat consumption analysis
- free to download, licensed by ET Core
- ET Core
- core mathematical modeling algorithms referenced by both ET Analytics and ET Interactive
- ET licensing mechanism for both ET Analytics and ET Interactive
See the ET website for case studies and more information: Energy Twin webpage.
- ET license is valid for both ET Analytics and ET Interactive
- ET license is defined by two parameters:
- expiration date
- the maximum number of models that can be defined at one time (see the explanation of a model below)
- the maximal number of models is calculated across all projects on the SkySpark server
- the sum of the models that are defined in ET Analytics and ET Interactive is evaluated
- What is a model?
- ET counts models in a similar way to SkySpark license points
- each model is linked to one point
- typically a kW point of a meter
- one model usually equals one meter
- model can be re-linked to any other point and re-trained anytime
- Ask ET sales (-----------------------------) for an individual offer or start with one of the two predefined starter licenses
- Starter 6 x 12
- use up to 6 models for 12 months
- Starter 72 x 1
- use up to 72 models for 1 month
Regression models such as change point and degree days models are widely used for predicting energy consumption based on outdoor temperature. These methods are used in M&V projects and in general for any energy consumption evaluation where one needs to suppress varying outdoor temperature.
Energy Twin (ET) Interactive is a SkySpark extension for energy evaluation using regression models without need for exporting data to other programs to avoid tedious and potentially erroneous data transfer. With interactive GUI, you can focus on data and its interpretation instead of lengthy manual data processing. ET Interactive will make our energy analysis work more efficient.
Get results in real-time as well as identify outliers easily with the ET Interactive graphical data representation. Exclude any data points and modify the model for better results based on metrics such as R2, CV(RMSE), or model shape validation. You can change the aggregation period to monthly, weekly, daily, and most importantly, daily with the weekend setback. Change points are also adjustable – tune your model and instantly observe the changes in statistical properties of the model.
Sometimes, you do not want to tune all parameters on your own, and for such a case, ET Interactive provides automated model parameters selection tools. Selecting the optimal change point or automatically determining the model type and the change point is the option. ET Interactive will try hundreds of possibilities, including all model type possibilities and their change point variations, and select the best model using multiple criteria.
- Change point models
- Two Parameter Model
- Three Parameter Heating Model
- Three Parameter Cooling Model
- Four Parameter Model
- Five Parameter Model
- Degree days models
- Heating Degree Day Model
- Cooling Degree Day Model
- Heating & Cooling Degree Day Model
- including a separate model for weekend setback
- M&V adherence
- Evaluation of model statistical properties according to M&V requirements
- Open API
- Integration of prediction into your applications
- Smooth SkySpark integration
- Prediction point provides prediction in all standard SkySpark applications
- Easy to use user interface
- On the edge
- ET Interactive uses just standard SkySpark libraries - no external packages are used
- All computations are done on the SkySpark server - no data sharing nor sending to the cloud
Energy Twin Interactive
Welcome to the Energy Twin Interactive manual. Energy Twin Interactive is a versatile SkySpark extension for building energy baseline modeling. The baseline model can be used to predict energy use and determine energy savings, and the identification of operational problems.
There are Administration, Model Definition, Manual, and Report tabs in the Energy Twin Interactive toolbar.
The administration tab contains a complete list of created models with their aggregation period, model type, span, and value of statistical properties like CV(RMSE).
You can edit any existing model or generate a prediction point. For creating a new model, click on New, and you will be automatically redirected to the Model Definition tab.
In the Model Definition tab, you can create a new model or edit previous models in the following steps:
- Select your point, time span, aggregation, and aggregation period. Possible aggregation is either sum (e.g., for kW) or last-first diff (e.g., for kWh, GJ, or m3). The aggregation period can be daily with no setback, daily with a weekend setback, weekly, or monthly.
- Set up the models - choose from Manual Setup, Auto Change point and Complete Auto Setup, for more details see section Model & model parameters selection.
- Visualization - efficiently identify outliers with real-time graphical data visualization. Check the overview of the model, measurement, excluded data, etc. See the data fit clearly for weekdays and the weekend immediately after applying changes to the model.
- Outliers - exclude (Exclude selected) or include (Include selected) various points. For easier outlier detection sort by measurement. Using ctrl select multiple outliers and exclude them at once. Visualisation can be adjusted to either show or hide excluded points.
- Statistical properties - various metrics, such as R2, RMSE, NDB, can be found in the overview chart for the combined, workday and weekend model. Green color indicates acceptable values according to M&V guidelines, whereas orange is borderline value and red color indicates bad model results.
- Equations characterizing the currently displayed model. Also helpful recommendations can be find here, e.g., use Sum aggregation instead of last-first diff.
A model can be edited without the need to create a new one each time. Also, after tuning the model, you can rewrite it or save it as a new model instead.
- Manual Setup = a pop-up window will appear where you can select the model type to the following: Cooling Degree Day, Heating Degree Day, Heating & Cooling Degree Day, Five Parameter, Three Parameter Heating, Three Parameter Cooling, Four Parameter, or Two Parameter according to your requirements. You can also manually set the change heating point and cooling point.
- Auto Change Point = find the most appropriate change point automatically, which is determined by the value of R2 metrics. The higher, the better. As you can see the change point can vary between the weekend and workday model.
- Complete Auto Setup = a pop-up window with a review of all possible model types and their result will appear. Models are sorted by adjusted R2. The model with acceptable shape and highest adjusted R2 is selected. Note that optimal change-points are also calculated in this process.
Modeled vs. Measured
In this section, you can generate an overview for models of your choice for any time span. Consequently, you can choose to either show or hide the Weather and Cumulative difference in the taskbar. The upper table shows a review of selected models, their measured energy consumption, predicted consumption, and the difference and relative difference. Underneath there is a graph showing the measured and predicted energy consumption and graphs showing the weather and cumulative difference depending on your settings.
If any errors occur check common mistakes:
- Select model(s) - the view will not work unless you select a model.
- In case of selecting aggregation to daily with a weekend setback, there will be three models: Combined, Weekend, and Workday. For the Combined model, use Complete Auto Setup only.
- Visualization is not possible if one model is a parameter model and the other is a degree day model since the resulting graphs have different quantities represented by the x-axis.
- Check if aggregation is set properly - when set incorrectly a recommendation will appear above the model's equation.
Model doesn't correspond to real data
If your model doesn't correspond to your data try the following:
- Check time span, aggregation type and whether you use valid data.
- Exclude outliers in the GUI.
- Check model type and change points - try using automatic tuning feature.
- Disclaimer: In case of HDD/CDD, it depends on the settings of SkySpark's function degreeDays.
API Functions are Axon functions that are available for users. Only a brief description will be provided, for more information see StackHub Axon Funcs tab
This function calculates a prediction of the given model.
etiPointIdentifyOptimal(point, span, aggregation, aggregationPeriod, subModelName, modelType)
This function returns R2 table for all changepoints.
- options: etiFoldDiff, etiFoldIntegral
- etiFoldDiff: difference between first and last value in rollup period
- etiFoldIntegral: integrated sum in rollup period
- used to rollup data
- options: 1day, 1wk, 1mo
- used to filter data
- options: "base", "weekend", "workday"
- "base": no filtering
- "weekend": only weekend data
- "workday": only workday data
- options: "twoParameter", "threeParameterCool", "threeParameterHeat", "fourParameter", "fiveParameter", "hdd", "cdd", "hddCdd"
etiPointIdentifyOptimal(read(point), 2019, etiFoldDiff, 1day, "weekend", "hddCdd")