Neural Network Protective Device for Power Grid Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional protective devices for electrical energy supply networks face challenges in detecting impermissible operating states, particularly high-resistance earth faults, and require rigid algorithms that cannot adapt to changing network conditions, limiting their reliability and selectivity.
Innovation Solution
A protective device utilizing a multi-stage neural network architecture with a convolutional neural network (CNN) for graphical pattern recognition, capable of adapting to changing circumstances through continued training, to evaluate measured values and determine the operating state of the energy supply network, including fault type and location.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional protective devices use rigid deterministic protection algorithms, then the device structure remains simple, but the device cannot adapt to changing network conditions and new error scenarios
Solution Approach 1:
The patent applies the dynamics principle by implementing a neural network that can dynamically adapt its evaluation criteria through continuous training with new data. The protective device transitions from static deterministic algorithms to a dynamic system that learns and adjusts to changing network conditions, new error scenarios, and evolving operating patterns without requiring structural changes to the device itself.
Solution Approach 2:
The patent utilizes parameter changes by modifying the evaluation parameters through neural network training. Instead of changing the device hardware or fundamental structure, the system adapts by changing the parameters (weights and biases) of the neural network based on learned patterns from training data, enabling flexibility while maintaining device simplicity.
2Measurement precision
If multiple specialized protective functions are combined to detect fault type, direction, and location, then the detection accuracy improves, but the device complexity increases
Solution Approach 1:
The patent applies the merging principle by integrating multiple specialized protective functions (fault type detection, direction detection, location determination) into a single unified neural network evaluation device. This consolidation maintains comprehensive detection capabilities while simplifying the overall device structure by replacing multiple separate algorithms with one integrated neural network that processes all aspects simultaneously.
Solution Approach 2:
The neural network evaluation device embodies universality by being designed to perform multiple protective functions through a single system. The same neural network structure can detect fault types, determine fault directions, and locate faults by processing different aspects of the input data, eliminating the need for multiple specialized devices or algorithms.
3Reliability
If protective device functionality is permanently implemented using protection algorithms, then the device operates reliably, but the functionality cannot be changed or adapted subsequently
Solution Approach 1:
The patent applies self-service by enabling the protective device to automatically adapt its own functionality through continuous training with new operating data. The neural network performs self-learning and self-adjustment by processing new data patterns and updating its internal parameters, allowing the device to maintain reliability while simultaneously adapting to new conditions without external intervention or firmware updates.
4Ease of manufacture
If conventional protection algorithms are used, then the device is easy to manufacture and configure, but certain error scenarios like high-resistance earth faults cannot be detected
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network with comprehensive data that includes rare and difficult-to-detect error scenarios like high-resistance earth faults. This preliminary training equips the device with the ability to recognize these subtle patterns before they occur in operation, maintaining ease of configuration while significantly improving detection capability for previously undetectable fault types.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The invention relates to a protective device (22) for monitoring an electrical power supply network (10), comprising an evaluation unit (40) configured to determine, using measured values indicating the electrical state of the power supply network (10) at at least one measuring point, whether the power supply network (10) is in a permissible or impermissible operating state, wherein the evaluation unit (40) includes an artificial neural network. To provide a particularly flexible and reliable method for detecting impermissible operating states in power supply networks, it is proposed that the evaluation unit (40) be configured to use pattern recognition of a graphical representation (G) based on the measured values to determine the operating state of the power supply network.The invention also relates to a method for monitoring a power supply network (10) carried out with such a protective device (22) and to a corresponding computer program product.