DLNN Power Converter Synchronization Under Grid Disturbances

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing power converters face challenges in synchronizing with the electrical utility grid due to difficulties in accurately determining parameters like amplitude, frequency, and phase, especially in single-phase systems, leading to instability and oscillations during grid disturbances.

Innovation Solution

Utilizing deep learning neural networks (DLNNs) to monitor and adjust electrical parameters in real-time, enabling precise synchronization by training on synthetic datasets that simulate various grid conditions, including harmonics and noise, and allowing for online updates of neural network weights.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional PLL schemes are used for grid synchronization, then the system can maintain basic synchronization, but the system becomes unstable and oscillatory during grid disturbances such as voltage sags, swells, and frequency variations

Engineering Contradiction:

Improvesynchronization stabilityVSAvoidsystem stability during disturbances

Core Design Contradiction:

ReliabilityVSStability of the object's composition

Solution Approach 1:

The patent replaces the conventional mechanical PLL feedback control system with a deep learning neural network-based synchronization system. The DLNN learns optimal synchronization strategies from training data containing various grid disturbance scenarios, substituting the traditional feedback loop with an intelligent system that can adapt to different disturbance patterns without causing oscillations or instability.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental parameter of the synchronization system from fixed feedback gains in conventional PLL to dynamic, learned parameters in the DLNN model. The neural network adjusts synchronization parameters based on real-time grid conditions by processing voltage and frequency measurements through learned weight matrices, enabling stable operation across varying disturbance scenarios.

Inventive Principle:

Principle #35Parameter changes

2Speed

If the PLL loop filter is designed for faster response, then the synchronization response time improves, but the system becomes more prone to instability and oscillations

Engineering Contradiction:

Improvesynchronization response speedVSAvoidsynchronization stability

Core Design Contradiction:

SpeedVSReliability

Solution Approach 1:

The patent introduces dynamics into the synchronization system by using a neural network that can adapt its response characteristics based on real-time grid conditions. The DLNN processes current grid state measurements and dynamically adjusts synchronization commands, enabling fast response during normal conditions while maintaining stability during disturbances through learned control strategies rather than fixed dynamic parameters.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent implements an enhanced feedback mechanism where the DLNN continuously monitors grid voltage and frequency measurements and adjusts synchronization commands based on learned patterns from training data. This intelligent feedback system replaces conventional fixed-gain feedback, enabling the system to achieve both fast response and stability by learning optimal feedback adjustments for different operating conditions.

Inventive Principle:

Principle #23Feedback

3Loss of information

If single-phase systems use Quadrature Signal Generator or all-pass filter to generate fictitious quadrature signal, then phase information can be obtained, but the estimation deteriorates during fast transients causing extra oscillations or instability

Engineering Contradiction:

Improvephase information availabilityVSAvoidphase estimation accuracy during transients

Core Design Contradiction:

Loss of informationVSReliability

Solution Approach 1:

The patent replaces the conventional QSG or all-pass filter based quadrature signal generation with a DLNN-based phase estimation system. The neural network directly processes single-phase voltage measurements and outputs accurate phase information by learning the relationship between voltage patterns and phase angles from training data containing various transient scenarios, eliminating the need for fictitious quadrature signal generation.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the approach to phase estimation from mathematical signal manipulation (QSG/APF) to pattern recognition through neural networks. The DLNN learns to extract phase information directly from single-phase voltage measurements by processing sequences of voltage samples through learned weight matrices, achieving accurate phase estimation during fast transients without generating oscillations.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12567746B2Power converters utilizing deep learning neural networks and methods of use

Publication Date: 2026.03.03 GEORGIA TECH RES CORP

AI summary

An exemplary embodiment of the present disclosure provides a power converter system comprising a power converter. The power converter system can comprise a power converter electrically connected to a local power supply and an electrical utility grid. The power converter can comprise an output configured to exchange electrical power with the electrical utility grid. The power converter can be further configured to monitor one or more electrical parameters of the electrical utility grid over a period of time and alter one or more electrical parameters of the output of the power converter based on the monitored one or more electrical parameters of the electrical utility grid in real time using a deep learning neural network.