Neural Network Grid Protection for Adaptive Fault Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing protection devices in electrical energy supply grids struggle with inflexible protection functions that cannot be easily adapted to changing conditions, such as grid expansion and decentralized power feed-in, leading to inefficiencies in decision-making and reliability.
Innovation Solution
Implementing a protection device with an evaluation device designed as a single neural network that performs all protection functions, trained through deep and reinforcement learning methods, allowing for rapid adaptation to grid changes and independent operation without deterministic algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If protection functions are implemented using traditional deterministic algorithms, then the device structure is clear and deterministic, but the adaptability to changing grid conditions deteriorates
Solution Approach 1:
The patent replaces traditional deterministic protection algorithms (mechanical/systematic approach) with a neural network-based evaluation device (biological/adaptive approach). The neural network learns optimal protection strategies through training data, automatically adapting to grid changes without requiring manual reconfiguration of protection logic, thus resolving the contradiction between adaptability and structural complexity.
Solution Approach 2:
The patent changes the fundamental parameter of the evaluation device from fixed deterministic algorithms to trainable neural network parameters (weights and biases). These parameters are adjusted through learning from historical data and grid conditions, enabling the system to adapt to changing grid topologies and decentralized power feed-in scenarios while maintaining a unified device structure.
2Productivity
If multiple protection functions are implemented separately, then each function can be optimized independently, but the decision-making time increases
Solution Approach 1:
The patent merges multiple separate protection functions into a single unified evaluation device based on neural networks. Instead of executing multiple independent protection algorithms sequentially, the neural network processes all protection requirements simultaneously through parallel computation, significantly reducing decision-making time while maintaining comprehensive protection coverage.
Solution Approach 2:
The neural network-based evaluation device is designed as a universal system that can perform multiple protection functions (overcurrent protection, distance protection, differential protection, etc.) simultaneously. This multi-functional approach eliminates the need for separate dedicated algorithms for each protection type, improving overall decision-making efficiency.
3Reliability
If protection functions are made flexible and adaptive, then the reliability under changing conditions improves, but the ease of operation deteriorates
Solution Approach 1:
The neural network evaluation device is designed to self-adjust and self-optimize through automatic learning from training data. The system performs self-service by automatically adapting its parameters to changing grid conditions without requiring manual intervention or complex configuration, thus maintaining high reliability while preserving ease of operation. The training process is performed offline, and the trained model is deployed for automatic operation.
Data Source
AI summary
A protection device for monitoring an energy supply grid has a measured value acquisition device for acquiring measured values which indicate an electrical state of the energy supply grid, and an evaluation device which is connected to the measured value acquisition device and is configured to carry out a plurality of protection functions and, as a result, to make a decision as to whether the energy supply grid is in a permissible or impermissible operating state. Here, the evaluation device be designed entirely in the form of a single neural network with an input neuron layer that receives the measured values and/or values derived therefrom, intermediate neuron layers, and an output neuron layer which outputs a classification of operating states of the energy supply grid. The neural network for classifying the operating states is trained to carry out all protection functions together.