AAU Student Projects is unavailable between June 15th 1.30pm and 17th 1.30pm due to planned system maintenance. The projects cannot be downloaded during this period.
AAU Student Projects - visit Aalborg University's student projects portal
An executive master's programme thesis from Aalborg University
Book cover


Short-Term Active Power Forecasting Under Uncertainty for Smart Distribution Grid Operation: Design of a Power Prediction model

Translated title

Short-Term Active Power Forecasting Under Uncertainty for Smart Distribution Grid Operation

Author

Term

4. semester

Publication year

2026

Submitted on

Abstract

Rooftop solar is changing local low-voltage grids: power now flows in both directions, and traditional load forecasts are no longer enough to prevent overloads. This thesis introduces a short-term active power forecasting approach that combines LSTM neural networks (a deep learning model for time series) with stochastic state estimation (a statistical method that quantifies uncertainty) to provide calibrated, uncertainty-aware support for grid operations. Sixteen LSTM models are trained on one year of substation measurements enriched with ERA5 weather data, achieving MAE = 18.95 kW. The predicted 95% confidence intervals contain the true values 99.5% of the time on a December 2023 test set, and 96.1% on a January 2024 cross-year validation. Forecasts are propagated through a physically consistent nine-node grid model that obeys Kirchhoff’s laws and then into a constrained maximum likelihood state estimator. This produces 95% confidence regions with coverage close to target (HRV = 95.0% and HRI = 95.1%). Finally, a three-zone congestion risk classification (low/medium/high) turns the results into actionable guidance for distribution system operators managing low-voltage grids with high levels of distributed energy resources.

Solceller på hustage ændrer lavspændingsnet i boligområder: strømmen kan nu løbe i begge retninger, og traditionelle belastningsprognoser rækker ikke længere til at undgå overbelastning. Denne afhandling præsenterer en kortsigtet prognosemetode for aktiv effekt, der kombinerer LSTM-neurale netværk (en dybdelæringsmodel til tidsserier) med stokastisk tilstandsestimering (en statistisk metode, der kvantificerer usikkerhed). Målet er kalibrerede, usikkerhedsbevidste beslutningsgrundlag for drift af elnettet. Vi træner 16 LSTM-modeller på ét års målinger fra en transformerstation suppleret med ERA5-vejrdata og opnår MAE = 18.95 kW. De forudsagte 95%-konfidensintervaller indeholder den sande værdi i 99.5% af tilfældene på et testsæt fra december 2023 og i 96.1% ved validering på tværs af år i januar 2024. Prognoserne køres igennem en fysik-konsistent model af et ni-noders net, der overholder Kirchhoffs love, og sendes videre til en begrænset maksimum likelihood-tilstandsestimator. Dette giver 95%-konfidensregioner med dækning tæt på målet (HRV = 95.0% og HRI = 95.1%). Til sidst omsættes resultaterne til en tre-niveaus klassificering af overbelastningsrisiko (lav/mellem/høj), som giver handlingsrettet beslutningsstøtte til netselskaber, der driver lavspændingsnet med høj andel af distribuerede energikilder.

[This abstract has been rewritten with the help of AI based on the project's original abstract]