TwinOpt-Pro - Platform for forecasting and optimizing the operation of drinking water supply systems

Project goals

The main objective of the project is to develop a platform for robust real-time drinking water operation optimization based on a digital twin of the drinking water network, a forecasting toolbox and optimization tools. In the project, the innovative platform will be applied and evaluated using two use cases as examples:

• Use Case 1: Energy optimization as a contribution to climate neutrality: maximized use of regeneratively generated electricity
• Use Case 2: Optimization of drinking water production and treatment (water softening): On-demand water treatment and thus minimizing the use of chemicals.

These use cases are realized prototypically using the drinking water infrastructure of Bühl (Baden). Using the Thuringia district water supply system, a simulation will be used to investigate the potential of the innovative platform for Use Case 1 (optimization of pump use using renewable electricity).

Project results of the IOSB

Development of a digital platform for processing data from drinking water networks:

• Connectors for the integration of diverse data sources relevant to drinking water networks, including live operation, weather, electricity market, house stations, spring discharges and level measurements
• Processing of data streams in real time through event-based data processing
• Implementation of the platform as a distributed service-oriented architecture in a Kubernetes cluster
• High-performance availability of historical data for data analysis, monitoring and reports

Flexible forecasting toolbox simplifies the process from training to deployment of ML models

• Workflows support the training, evaluation and roll-out of new ML models for forecasting time series data
• Helm chart that summarizes frequently used ML tools across the entire workflow process (e.g. Jupyter Notebooks, MLflow and InfluxDB) and makes them accessible via a central entry website and can be executed in a Kubernetes cluster in just a few steps
• Easy access to GPU resources in the Kubernetes cluster
• Simple deployment of the trained ML model with MageAI
• Monitoring and evaluation of the predictions with Grafana