ML Grid Criticality Assessment for Voltage Limit Violations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for evaluating the criticality of electrical power distribution grids are hindered by the lack of accurate power-flow models, which are difficult to create due to varying data formats and limited accuracy, leading to incomplete grid impact assessments and inability to predict future voltage states.
Innovation Solution
A computer-implemented method using a machine learning model fitted to smart meter data and voltage data from the power distribution grid, allowing for the calculation of a grid impact score that evaluates criticality based on power and voltage limit violations, eliminating the need for a conventional power-flow model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a conventional power-flow model is used to evaluate grid criticality, then voltage and power calculations can be performed, but the model creation requires high digitalization and manual engineering efforts, and accurate data is typically unavailable
Solution Approach 1:
The patent replaces the conventional mechanical/power-flow modeling approach with a machine learning-based system. Instead of manually creating detailed power-flow models with accurate topology and line parameters, the system uses neural networks trained on smart meter data to directly predict voltage magnitudes and grid criticality scores, eliminating the need for manual model construction while maintaining calculation accuracy
Solution Approach 2:
The patent creates a virtual copy of the power distribution grid behavior through machine learning models. The neural network learns the relationship between power consumption patterns and voltage states from historical smart meter data, creating a digital surrogate that replicates grid behavior without requiring actual detailed electrical parameters or manual modeling efforts
2Reliability
If manual engineering efforts are used to create accurate grid models, then voltage predictions can be made, but the process is time-consuming and requires substantial manual work
Solution Approach 1:
The patent performs preliminary action by training the machine learning model on historical smart meter data in advance. Once trained, the model can rapidly predict voltage states and grid criticality for future scenarios without requiring manual model creation each time. The model is fitted to represent the specific power distribution grid, enabling fast and reliable predictions for planning and operation purposes
Solution Approach 2:
The patent substitutes manual model creation and updating processes with an automated machine learning system. The neural network automatically learns from historical data and can be retrained on new data without requiring manual engineering intervention, significantly reducing the time required to generate reliable grid state predictions
3Loss of information
If GIS data is used for grid modeling, then location information is available, but accurate electrical connectivity and phase information are missing
Solution Approach 1:
The patent uses smart meter data as an intermediary to bridge the gap between available GIS location data and the needed electrical connectivity information. The machine learning model learns the relationship between power consumption patterns (which reflect electrical connectivity and topology) and voltage states, effectively inferring the missing electrical network structure from operational data without requiring manual topology input
4Adaptability or versatility
If the grid impact score considers hypothetical power changes, then load and generation margins can be evaluated, but a data model describing voltage changes in response to power changes is lacking
Solution Approach 1:
The patent replaces the missing voltage response model with a machine learning approach. The neural network is trained on historical data containing both power consumption and voltage measurements, enabling it to predict how voltages will change in response to hypothetical or actual power changes. This allows the system to calculate load and generation margins and perform root-cause analysis for voltage limit violations without requiring a conventional power-flow model
Data Source
Figure 1~2
Figure 3~4
AI summary
A computer implemented method and apparatus for determining a criticality of an electrical power distribution grid comprising the steps of: providing (S1) a machine learning model (ML) of the power distribution grid fitted to smart meter data (smd) representing a power consumption of customers connected via a distribution transformer (DT) to conducting equipment within a substation of the power distribution grid and fitted to data representing a voltage at selected conducting equipment of the power distribution grid; using (S2) the provided machine learning model (ML) for calculating a grid impact score (GISC) representing the grid state of the power distribution grid based on power limit violations and/or based on voltage limit violations; and evaluating (S3) the calculated grid impact score (GISC) to determine the criticality of the electrical power distribution grid.