ALTEN Group Case Studies Sharing
Today’s energy providers are looking for ways to improve the performance and productivity of alternative energy sources. LINCOLN, an ALTEN consultancy that empowers companies to make the most of their data, helped design solutions for managing the equipment in wind and photovoltaic farms, including statistical and prediction-based models to detect anomalies and facilitate maintenance.
Traditional methods of identifying issues with equipment and performance are typically reactive and inadequate. One of Europe’s leading energy providers called on LINCOLN to harness the power of AI by developing advanced statistical methods that could be used to detect anomalies and enable predictive maintenance in renewable energy production.
Challenge: Design statistical and AI methods, leveraging various heterogenous data, for anomaly detection and predictive maintenance in wind and photovoltaic farm management
Solutions: Robust statistical models and local prediction systems to pinpoint anomalies and predict potential losses in energy production, enhancing productivity and efficiency
Benefits:
- Enhanced understanding of energy assets
- Deeper insights into the performance of wind and photovoltaic farms
- Proactive maintenance
- Optimized energy performance
- Improved efficiency and reduced downtime
Productivity-enhancing data
The project focused on leveraging data analytics, such as data cleansing and indicator construction, to aggregate and preprocess data from various sources. The project team drew on contextual data (e.g., weather data, information on the model and age of equipment) to identify the root causes of the anomalies, as well as data from IoT sensors to detect changes in the environment. The data from weather devices and photovoltaic panels were collected and stored in a centralized data lake. The team then constructed indicators and “cleansed” the measurements provided. The next step was the construction of advanced statistical models, a process that included the application of Markov chains, sequence analysis and pattern recognition. Local prediction-based models were developed for specific areas and then compared to actual reported data to identify discrepancies and estimate the percentage of loss, allowing for timely detection of underperformance. To test the performance of the system, the team created operational models using cold data from previously detected anomalies. Comprehensive data integration from multiple sources enhanced the quality of the insights.
Tools and technologies
Python and R programming languages were used for data analysis and statistical modeling, as well as for statistical computations and analysis; Docker facilitated containerized deployment of the models; and GitHub supported version control and collaboration. Data visualization tools were also used for monitoring and reporting.
From models to maintenance
LINCOLN’s client plans to apply these advanced statistical techniques to ensure accurate detection of anomalies and losses. Using real-time monitoring and operational measures, they will be able to anticipate breakdowns (for example, blade breakage, rotor and generator bearing degradation), reducing the frequency and cost of interventions, minimizing energy losses and enabling proactive maintenance actions.
In this way, LINCOLN’s expertise in data analytics will contribute to improving overall productivity, boosting efficiency and reducing downtime.
For more information, please visit AI and Machine Learning or contact us via marketing@cienet.com.