Interpretable Controller Decision Logic for Adaptive Power Protection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in modern industrial automation control systems, particularly in power transmission and distribution systems, is the complexity of designing protection logic that adapts to changing grid environments due to converter-interfaced generation and e-mobility, which introduces stochasticity and requires frequent reevaluation, while existing machine learning approaches lack interpretability for human engineers.
Innovation Solution
A machine learning-based decision logic generation method that decomposes the learning process into interpretable sub-processes, using checkpoints to visualize and verify the decision-making process, allowing human engineers to trace and adjust the logic effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning is used to generate decision logic, then the system can adapt to changing grid environments, but the decision logic becomes uninterpretable to human engineers
Solution Approach 1:
The patent segments the machine learning decision-making process into multiple interpretable sub-processes with defined checkpoints. Each sub-process handles a specific aspect of the decision logic, allowing human engineers to trace and understand the reasoning path while maintaining the adaptability benefits of machine learning for changing grid environments.
2Loss of information
If conventional protection logic is designed by human engineers, then the decision logic is interpretable, but the system cannot adapt frequently to changing grid conditions
Solution Approach 1:
The patent introduces an intermediary layer that bridges human-engineered interpretability requirements and machine learning adaptability. This intermediary structure provides checkpoints where human engineers can verify and understand the decision logic while the underlying machine learning models continue to adapt to changing grid conditions through automated retraining.
3Adaptability or versatility
If protection logic is reevaluated frequently due to grid changes, then the system adapts to new conditions, but the design complexity and time increase
Solution Approach 1:
The patent implements a dynamic protection logic system where the decision logic can be automatically retrained and updated in response to changing grid conditions. The segmented architecture with checkpoints allows for efficient reevaluation without requiring complete redesign, reducing the complexity burden of frequent adaptations.
4Adaptability or versatility
If protection logic is reevaluated frequently due to grid changes, then the system adapts to new conditions, but the design time increases
Solution Approach 1:
The patent performs preliminary segmentation of the decision logic into modular sub-processes with defined checkpoints during the initial design phase. This preliminary structuring enables faster subsequent reevaluations and retrainings when grid conditions change, as the modular architecture can be updated incrementally rather than requiring complete redesign.
Data Source
AI summary
Machine learning, ML, using labeled data is performed to thereby train a ML model to generate decision logic for a controller of an IACS, in particular a power system controller. The ML model has ML model inputs and ML model outputs. The ML model is decomposed into a plurality of computational sub-processes, wherein intermediate signals output by one of the computational sub-processes are input into a consecutive one of the computational sub-processes. Information on the trained ML model is generated based on the intermediate signals and is output for interpretation, verification, visualization, and/or inspection of at least one of the computational sub-processes.