PV Inverter Time-Series Classification for Clipping and Power Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data classification methods for photovoltaic systems, based on controlled environments, fail to accurately classify real-world data, leading to false positives and negatives, necessitating an improved method for multiple classifications.
Innovation Solution
A multi-filter approach involving smoothing, max range, daily max, triangular waves, range, and PMax filters, along with break point identification and event type classification, to accurately label data as 'No Limitation', 'Clipping', 'PL', 'SPL', or 'LPL', using sensors and a computer program to process raw data from photovoltaic inverters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a single-label classification method based on controlled environment data is used, then the method is simple to implement, but the classification accuracy deteriorates when applied to real production data
Solution Approach 1:
The patent segments the classification task into multiple independent filters, each targeting specific clipping patterns. Instead of using a single classification method that attempts to handle all cases, the invention divides the problem into: a) a first filter for detecting hard clipping (saturated values), and b) a second filter for detecting soft clipping (gradual limitations). This segmentation allows each filter to be optimized for its specific detection task, improving overall accuracy while maintaining implementation simplicity.
Solution Approach 2:
The patent applies partial action by using multiple specialized filters rather than one comprehensive classification system. Each filter performs a specific function (detecting different types of clipping), and their results are combined. This approach of using multiple partial solutions exceeds the capability of a single filter, thereby improving classification accuracy on real production data while keeping each individual filter simple to implement.
2Adaptability or versatility
If multiple classification labels are implemented to capture different clipping types, then the classification completeness improves, but the system complexity increases
Solution Approach 1:
The patent segments the complex multi-label classification problem into simpler sub-tasks handled by separate filters. The first filter identifies hard clipping events, while the second filter detects soft clipping events. By segmenting the classification into distinct detection stages, the system achieves comprehensive multi-label classification without requiring a single complex algorithm to handle all cases simultaneously.
Solution Approach 2:
The patent introduces an intermediary mechanism in the form of a combination rule that integrates results from multiple filters. Rather than having filters directly compete or conflict, the combination rule acts as an intermediary that systematically merges their outputs into final classifications. This intermediary layer simplifies the overall system architecture while maintaining the ability to produce multiple classification labels.
3Measurement precision
If off-line filtering is applied to improve classification accuracy, then the measurement precision improves, but the processing time increases
Solution Approach 1:
The patent applies preliminary action by performing off-line filtering on historical production data before final analysis. The filters are applied to previously collected data, allowing computationally intensive processing to occur when system load is lower. This preliminary processing improves classification accuracy without impacting real-time operational response, as the filtering results can be stored and referenced later.
Solution Approach 2:
The patent implements periodic action by applying the off-line filtering process at scheduled intervals rather than continuously. The filters process data batches periodically, which reduces overall processing time compared to continuous real-time filtering. This periodic approach maintains classification accuracy while significantly reducing the time loss associated with constant processing.
Data Source
Figure 1
Figure 2
Figure 3A~3E
AI summary
Method (100) for classifying raw data of a time series of a physical quantity related to an inverter of a photovoltaic power plant, the physical quantity being selected among an intensity, a voltage, a power and an energy measured on DC terminals of the inverter and an intensity, a voltage, a power and an energy measured on AC terminals of the inverter, characterized in that the method comprises the step of: acquisition (105) of the time series of raw data over a plurality of successive days; applying (120) a maximum range filter to select each raw data as a candidate data when a time variation thereof is below a first cut off; and, applying (130) a daily maximum filter based on a difference between a value of the candidate data and a maximum value on a day of acquisition of the candidate data, the candidate data being labelled as "Clipping" data when the difference is smaller than a threshold and as "Power Limitation" data when the difference is higher than the threshold.