IED Decision Logic Training for Faster Power Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for generating decision logic in electric power systems face challenges in achieving rapid, reliable, and accurate fault detection and mitigation, requiring human expert knowledge and lacking robustness in distinguishing between normal and fault states.
Innovation Solution
A method and system that automatically generates decision logic using machine learning models, adjusting training kernels independently for different cases, employing LSTM or GRU networks, and optimizing parameters to reduce delay in fault detection, enabling quick and accurate decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If human expert knowledge is used to classify training cases, then the decision logic can be tailored to specific fault scenarios, but the system requires human expertise and cannot operate independently
Solution Approach 1:
The system automatically generates training kernels without human expert intervention. The training kernel generation unit autonomously creates the necessary training data structures from raw measurements and labeled fault scenarios, enabling the decision logic to be trained and deployed independently of human expertise while maintaining high detection accuracy
Solution Approach 2:
The system performs preliminary automated processing of measurements into training kernels before the actual training phase. This pre-processing step automatically structures the data in a format ready for machine learning, eliminating the need for human experts to manually prepare training cases while ensuring the data is properly formatted for accurate model training
2Productivity
If the training kernel is adjusted independently for different cases, then the system can optimize performance for each specific scenario, but the complexity of the training process increases
Solution Approach 1:
The training process is segmented into distinct automated phases: measurement collection, training kernel generation, model training with independent kernel adjustment, and validation. Each phase handles specific tasks independently, allowing the system to optimize training kernels for different fault cases without overwhelming complexity, as each segmentation manages a specific aspect of the process
Solution Approach 2:
The training kernel parameters are dynamically adjusted during the training process based on the specific fault scenario being addressed. The system automatically modifies kernel characteristics for different cases rather than using fixed parameters, enabling scenario-specific optimization while the automated adjustment mechanisms keep the overall process manageable
3Loss of time
If the decision logic is trained to detect faults quickly, then the response time is reduced, but the risk of incorrect trips increases
Solution Approach 1:
The system incorporates feedback mechanisms where the decision logic output is continuously evaluated against actual system state and fault conditions. The feedback loop allows the model to learn from both correct detections and false trips, adjusting its decision boundaries to reduce incorrect trips while maintaining fast response times through iterative refinement of the training kernels
Solution Approach 2:
The system dynamically adjusts decision thresholds and kernel parameters based on the specific fault scenario and system conditions. By changing parameters adaptively rather than using fixed values, the system can optimize the balance between detection speed and accuracy for different operating conditions, reducing false trips while maintaining rapid fault detection
Data Source
Figure 1
Figure 2
Figure 3~6
AI summary
To generate a decision logic (34) for an IED (30), at least one machine learning model is trained in an iterative machine learning model training. Weighting functions are used to weight samples in the iterative machine learning model training. Weighting function(s) associated with one or several training cases are automatically modified in the iterative machine learning model training.