Power Asset Control Using Multi-Variable Production Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for forecasting power production in power generating assets, such as wind turbines, often result in large variations due to their reliance on linear relationships and small sample sizes, leading to inaccurate performance predictions.
Innovation Solution
The use of machine-learning, model-based analytics that generate performance predictions based on various model-variable combinations, incorporating multiple environmental and operational variables, such as wind speed, direction, temperature, and turbulence intensity, to create more accurate and confident power generation forecasts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional linear modeling methods are used for power production forecasting, then the forecasting process is simple and easy to implement, but the prediction accuracy deteriorates due to large variations and small sample sizes
Solution Approach 1:
The patent transforms the forecasting approach by changing the parameters used in modeling - transitioning from simple linear relationships to non-linear relationships, and from monthly-scale variables to multi-scale variables including sub-hourly intervals. This parameter transformation enables the system to capture complex wind turbine performance patterns while maintaining computational feasibility through automated model generation and selection.
Solution Approach 2:
The patent implements dynamic modeling by allowing the forecasting system to adaptively select different model types (linear or non-linear) and different variable combinations based on the specific conditions and data characteristics. The system dynamically adjusts the complexity of the model to match the underlying patterns in the data, improving accuracy without requiring a fixed complex structure.
2Measurement precision
If multiple model-variable combinations are generated and trained, then the prediction accuracy improves with smaller uncertainty, but the computational complexity and system complexity increases
Solution Approach 1:
The patent segments the forecasting task into multiple independent model-variable combinations, each trained on specific subsets of variables at different scales. By dividing the complex forecasting problem into smaller, manageable segments (different model types, different variable combinations), the system can process each segment separately and then aggregate results, reducing the computational burden of any single model while improving overall accuracy.
Solution Approach 2:
The patent generates and evaluates multiple model-variable combinations beyond what a single conventional model would provide. By creating more models than traditionally necessary (excessive action), the system ensures that at least some models will capture the underlying patterns accurately, then selects the best performing models. This approach trades increased initial computational effort for significantly improved prediction reliability.
3Ease of operation
If monthly-scale variables are used for modeling, then the data processing is straightforward, but the forecasting accuracy deteriorates due to inability to capture short-term variations
Solution Approach 1:
The patent adds temporal dimensionality to the forecasting model by incorporating variables at multiple time scales - not just monthly averages but also sub-hourly, hourly, and daily variations. This multi-scale approach captures both short-term fluctuations and long-term trends, providing a more complete picture of wind turbine performance while maintaining systematic data processing through automated variable generation at different scales.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system and method are provided for operating a power generating asset. Accordingly, at least one external data set indicative of a plurality of variables affecting the performance of the power generating asset is received by the controller. The controller also receives at least one operational data set indicative of the performance of the power generating asset. A plurality of production-assessment models for the power generating asset are generated and trained based on the data sets. A performance prediction is then generated for each of a plurality of model-variable combinations and a control action is implemented based on one of the performance predictions.