Neural Network Protection Relay for Changing Grid Operating States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing protective devices for electrical energy supply networks are inflexible and require significant changes or new hardware/software to adapt to changing conditions, such as network expansions and decentralized power feeds, which slows down decision-making and adaptation to impermissible operating states.
Innovation Solution
A protective device utilizing a single neural network that integrates all protective functions, eliminating the need for deterministic algorithms and allowing independent adaptation to changing circumstances, with intermediate output neuron layers providing specific results for individual protective functions, enabling quick and adaptable decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional protective devices use deterministic algorithms and separate protective functions, then device complexity is reduced and ease of manufacture is improved, but adaptability to changing network conditions deteriorates and decision-making speed slows down
Solution Approach 1:
The patent combines multiple separate protective functions (earth fault protection, short-circuit protection, inrush detection, oscillation detection) into a single integrated neural network evaluation device. This unified neural network receives measured values from the network and simultaneously performs all protective functions, eliminating the need for separate deterministic algorithms for each function. The neural network is trained to recognize various fault patterns and operating states, providing adaptive protection that can handle changing network conditions without requiring manual reconfiguration or additional hardware.
Solution Approach 2:
The neural network evaluation device is designed as a universal protective device that can perform multiple protective functions simultaneously. A single neural network model is trained to handle earth fault protection, short-circuit protection, inrush detection, and oscillation detection, making the device multi-functional. This universal approach allows the protective device to adapt to different network configurations and fault types without requiring separate specialized devices or algorithms for each function.
2Reliability
If protective devices require significant changes or new hardware/software to adapt to changing conditions, then reliability is improved, but productivity and response time deteriorate
Solution Approach 1:
The patent implements a dynamic protective device using a neural network that can adapt its behavior based on changing network conditions. The neural network is trained offline to recognize various fault patterns and operating states, but once deployed, it dynamically adjusts its decision-making based on the actual measured values from the network. This dynamic approach allows the device to maintain high reliability by accurately recognizing faults while responding quickly without requiring manual reconfiguration or hardware changes when network conditions change.
Solution Approach 2:
The neural network is trained in advance (offline) using simulated and real data to learn the characteristics of various faults and operating states. This preliminary training phase allows the network to acquire knowledge about different fault patterns, inrush scenarios, and oscillation characteristics before being deployed in the actual protective device. Once trained, the neural network can immediately apply this knowledge to real-time protection without requiring additional training or reconfiguration, thus maintaining both reliability and fast response time.
3Manufacturing precision
If multiple separate protective functions are implemented, then manufacturing precision and ease of operation are improved, but loss of time in decision-making increases
Solution Approach 1:
The patent merges multiple separate protective functions into a single neural network evaluation device that processes all protective functions simultaneously. Instead of executing separate deterministic algorithms for earth fault protection, short-circuit protection, inrush detection, and oscillation detection in sequence, the unified neural network evaluates all these functions in parallel based on the same input measured values. This simultaneous evaluation maintains the accuracy of each protective function while eliminating the time delays associated with sequential processing.
Solution Approach 2:
The patent replaces traditional deterministic mechanical/algorithms-based protective functions with a neural network-based system. The neural network uses learned patterns and relationships from training data to make protection decisions, substituting the rigid deterministic algorithms with a more flexible intelligent system. This substitution allows the device to maintain high accuracy in fault detection while significantly reducing decision-making time by processing all protective functions simultaneously rather than sequentially.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a protective device (20) for monitoring a power supply network (30, 70, 80), comprising a measurement acquisition device (21) for acquiring measured values that indicate an electrical state of the power supply network (30, 70, 80), and an evaluation device (22) which is connected to the measurement acquisition device (21) and is configured to perform several protective functions and, as a result, to make a decision as to whether the power supply network (30, 70, 80) is in a permissible or an impermissible operating state.To enable even faster decisions, adaptable to different conditions of the power supply network, regarding whether an operating state is permissible or impermissible, it is proposed that the evaluation device (22) be designed entirely as a single neural network (40) and comprise an input neuron layer (42), to which the measured values and/or values derived therefrom are fed, at least one intermediate neuron layer (43a-n), and an output neuron layer (45) that outputs a classification of operating states of the power supply network. The neural network (40) is trained to perform all protective functions simultaneously for classifying the operating states. The invention also relates to a method for monitoring a power supply network implemented with such a protective device.