¶ Aim and function
QASSANDRA is a desktop application that is able to make predictions for a target variable in a water treatment facility, based on historic datasets and weather forecast. A target, such as flow, is usually a variable that changes over time, due to operating conditions, rainfall, temperature, snow, and previous weather events.
QASSANDRA is designed to combine the different elements of the forecasting process into a single workflow. Instead of treating plant data, weather data, model training, and forecast as separate jobs, it connects them within one project. On top of that, QASSANDRA includes automatic retraining and evaluation every now and then (according to how the user sets up the project), so after the initial configuration, it is a basically hands-free service.
¶ Easy-to-follow workflow
With QASSANDRA, a project follows a clear sequence:
- Select the target variable to predict and connect to the plant for historic and actual data
- Connect to a weather service for historical records and actual forecast
- Configure the forecasting model
- Preview the training result
- Accept the model
- Start scheduled predictions
- Monitor the results continuously on the dashboard
¶ Flexibility with different data sources
QASSANDRA can use actual plant data from:
- Google Drive folder that holds CSV files extracted regularly from the plant's SCADA system
- OPC UA (directly mapping data from the plant)
Historical plant data can also be imported from a local file, so that the model can be set up to learn from past behavior right away from the start.
QASSANDRA can use weather data from the following online sources via APIs:
- Open-Meteo
- Visual Crossing
- Weatherbit
¶ Choose the modeling approach
QASSANDRA supports several forecasting ML models:
- Gradient Boosting
- XGBoost
- Random Forest
- LSTM
- automatic comparison of the supported models
One algorithm is not always the best choice for every plant or every dataset, so QASSANDRA is equipped with multiple options. The choice is in your hands - it is easy to experiment with the various models and evaluate them before deciding for the preferred one.