AVACON: AI-supported data analysis as the key to resilient energy grids

• Sector(s): Energy equipment and solutions
• Customer: Avacon Natur GmbH
• Project title: Avacon Data-Quality
• Year: 2023
• Period: July - September 2023

Optimized efficiency and performance analyses through the use of neural networks.

Avacon Natur GmbH uses intelligent grid management to ensure a secure energy supply for municipalities, industry and households. Combined heat and power (CHP) plants generate electricity and heat simultaneously, thereby increasing their efficiency. Sensors record extensive data on temperature, flow and output during operation. It is important to standardize this data to enable efficiency and performance analyses. Regular measurements optimize CHP efficiency and meet legal requirements, ensuring transparent and sustainable operation.

Technical faults in the CHP units can affect data collection and transmission. Hardware and software problems as well as communication problems in the network lead to data gaps that make analysis difficult. External factors such as regulatory changes or grid fluctuations also influence performance and operation. This makes it more difficult to identify and analyze anomalies and leads to cost-intensive manual rework, such as direct queries to the system operator.

Unclear causes jeopardize data plausibility, make troubleshooting complex and lead to longer downtimes and higher maintenance costs. Inaccurate data also harbors compliance risks and can jeopardize adherence to legal requirements. 

Efficient data quality strategies with advanced diagnostic tools are therefore essential to minimize negative impacts and ensure operational efficiency.

As part of our data and AI analysis, we examined the operating data of a CHP plant near Magdeburg. We analyzed the data collection, the data quality requirements,
the data sources used and the existing internal data quality methods. The aim was to develop effective optimization strategies and a prototype to increase data plausibility and supplement missing information.

We developed a self-learning system based on artificial neural networks that recognizes external effects and technical faults in the operational data. It identifies and fills in gaps in data sets and missing data points while retaining the context of the original data. The system recognizes patterns and anomalies that indicate unusual operating conditions
or malfunctions.

The neural networks are trained with historical time series data, manually corrected and aggregated values in various resolutions down to the minute range of the individual sensors. They analyze multivariate causal relationships, measure effect sizes and create forecasts that are evaluated internally. This teaches the system to differentiate between normal and abnormal conditions, enabling precise and rapid problem detection even in the event of external faults.

Strategy:

• Data and AI potential analysis, architecture consulting

Develop:

• Development of a prototype

Lifecycle Services

• DevOps, maintenance & change management

The system was developed by analyzing the existing data and AI potential. We focused on overcoming jointly identified challenges such as data gaps and operational anomalies. This methodical approach led to concrete recommendations for action and a proposal for an expandable system architecture.

The aim is to develop a robust solution that offers significant added value for monitoring and fine-tuning operational processes in smart energy grid management. This solution allows AVACON Natur employees a high degree of flexibility in its application.

Known external events and influences can be managed efficiently, supported by machine learning, artificial neural networks and advanced statistical methods as modular filter options. An intuitive user interface enables employees to optimize and expand the core functionalities. In addition, regulatory requirements for AI applications are taken into account in the functional scope in order to meet legal requirements.

• Blob Storage
• Event Grid
• Message Queues
• Azure Function
• Azure DataBricks
• Azure AI Studio
• Data Analysis
  • Python
  • R