Transformer Station Reactive Power Forecasting Under Abnormal Grid Events
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing forecasting tools for reactive power in electrical distribution networks fail to accurately predict reactive power flows due to changes in consumer behavior, emergence of renewable energies, and spatio-temporal variability, especially during events like load transfers and reactive power regulation, which disrupt normal operating patterns.
Innovation Solution
A method for processing electrical power data at transformer stations that involves identifying and correcting power data outside normal operating patterns by detecting jumps and deviations, using Group Lasso linear regression to establish a reactive power forecast model, accounting for events like load transfers and reactive power regulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If constant power factor is used to model reactive power in forecasting tools, then the forecasting process is simple, but the prediction accuracy deteriorates due to changes in consumer behavior, renewable energies, and network configuration
Solution Approach 1:
The patent changes the modeling approach from using a constant power factor to using a dynamic power factor that varies according to different network operating conditions, consumer behavior patterns, and renewable energy generation levels. This allows the forecasting model to adapt to changing parameters in the electrical network, significantly improving prediction accuracy while maintaining computational feasibility through pattern recognition algorithms.
Solution Approach 2:
The patent introduces dynamic elements into the forecasting model by making the power factor a variable parameter that changes with network conditions, consumer behavior, and renewable energy production. This dynamic approach replaces the static constant power factor assumption, enabling the model to capture temporal variations and improve reactive power prediction accuracy in modern electrical networks.
2Quantity of substance
If power data during abnormal operating patterns (load transfers, reactive power regulation) is included in forecasting, then more data is available for model training, but prediction accuracy deteriorates due to spatio-temporal variability and disruptive events
Solution Approach 1:
The patent extracts and separates abnormal operating pattern data (load transfers, reactive power regulation events) from the normal operating data used for forecasting model training. By identifying and removing these disruptive events, the model trains exclusively on representative normal operating conditions, eliminating the spatio-temporal variability introduced by abnormal events while preserving sufficient training data for accurate pattern recognition.
Solution Approach 2:
The patent converts the potentially harmful effect of abnormal operating data into a benefit by using these events as indicators to define and exclude abnormal periods. The detection of load transfers and reactive power regulation events, which would otherwise degrade forecast accuracy, is used to create a filtering mechanism that improves overall model performance by ensuring training data represents normal operating conditions.
3Reliability
If forecasting is performed at local grid level, then local network constraints can be identified, but prediction difficulty increases due to spatio-temporal variability and diminished swarming effect
Solution Approach 1:
The patent segments the electrical network into local grid areas and develops forecasting models tailored to each segment's specific characteristics. This segmentation allows local network constraints to be identified and addressed while using localized training data that captures regional patterns, reducing the spatio-temporal variability problem that plagues aggregate-level forecasting of distributed networks.
Data Source
Figure 1~2a
Figure 2b~3
Figure 4
AI summary
The invention relates to a method for processing data Dm concerning electrical power consumed by an electrical transformation station, comprising the following steps: a. identifying power data outside of the normal operating pattern Dhsne; b. identifying a type of situation Shsne at the origin of the data Dhsne; c. processing the data Dhsne according to the type of situation Shsne identified, and obtaining corrected power data Dcc; d. establishing, by learning from the data Dcc, a model Mp for predicting reactive power outside of a situation Shsne of the type identified in step c). The invention also relates to a device and a computer program for implementing such a method.