Power Grid State Estimation Training Data for Critical Error Scenarios

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for state estimation in power grids face challenges in achieving real-time accuracy and reliability, particularly due to measurement deviations and communication delays, which affect the performance of higher-level grid applications.

Innovation Solution

A method for creating a second training data set for artificial neural networks by identifying and modifying training pairs with high errors, using load flow calculations and measurement models to generate new data pairs, thereby enhancing the network's training data for critical scenarios.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If an artificial neural network is used for state estimation in real-time, then the speed of state estimation is improved, but the reliability and accuracy of the estimation deteriorates

Engineering Contradiction:
Improvestate estimation speedVSAvoidestimation reliability
Core Design Contradiction:
SpeedVSReliability

Solution Approach 1:

The patent applies preliminary action by pre-training the artificial neural network offline using a comprehensive training dataset that includes various grid conditions and measurement scenarios. This pre-training prepares the network in advance for real-time operation, allowing it to provide reliable estimates quickly when deployed online without requiring extensive real-time computation or data collection.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If a large amount of training data is collected to improve neural network accuracy, then the measurement precision improves, but the loss of time and complexity of data collection increases

Engineering Contradiction:
Improvestate estimation accuracyVSAvoiddata collection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent uses copying by generating synthetic training data that replicates real measurement scenarios and grid conditions. Instead of collecting extensive real-world measurement data over long periods, the system creates artificial training pairs that mimic actual operating conditions, including various fault scenarios and measurement configurations. This approach provides sufficient training data for accurate state estimation without the time-consuming process of collecting and storing large volumes of real measurement data.

Inventive Principle:
Principle #26Copying

3Adaptability or versatility

If the training data set is expanded to cover more scenarios, then the adaptability of the neural network improves, but the device complexity and data processing requirements increase

Engineering Contradiction:
Improveneural network adaptabilityVSAvoiddata processing complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent applies local quality by creating targeted training data that focuses on specific, critical grid scenarios and conditions rather than attempting to cover all possible scenarios uniformly. The training dataset is designed with particular emphasis on locally important aspects such as specific fault types, measurement configurations, or grid topologies that are most relevant to the application, providing high adaptability for critical scenarios without the complexity of comprehensively modeling every possible grid condition.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4060559B1Training data set, training and artificial neural network for estimating the condition of a power network
Publication Date: 2024.01.10 SIEMENS AG
  • EP4060559B1 patent drawingFigure 1
  • EP4060559B1 patent drawingFigure 2
  • EP4060559B1 patent drawing

AI summary

A method for creating a second training dataset for training an artificial neural network trained for state estimation of a power grid (1) from a provided first training dataset is proposed, wherein the second training dataset comprises several training pairs, each formed by a measurement dataset and an associated state of the power grid (1), and the respective measurement dataset includes at least complex apparent powers associated with the power grid.The method is characterized by the following steps: - (S1) Determining at least one first training pair and an associated first measurement data set and first state, the first state of which exhibits an error greater than or equal to a defined error threshold with respect to training the artificial neural network using the first training data set; - (S2) Calculating a second state of the power grid using a load flow calculation based on at least one complex apparent power modified with respect to the complex apparent power of the first measurement data set; - (S3) Calculating a second measurement data set using a measurement model of the power grid from the calculated second state; and - (S4) Creating the second training data set from the first training data set by adding a second training pair formed from the second measurement data set and the associated second state.Furthermore, the invention relates to a method for training an artificial neural network and a method for estimating the state of a power grid (1).