Optimizing electricity trading decisions through weather-influenced demand and supply forecasting
Ivanova, Monika (2025)
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025120231609
https://urn.fi/URN:NBN:fi:amk-2025120231609
Tiivistelmä
The objective of this thesis was to improve the short-term electricity price forecasting to support trading decisions at BlueEnergy, a qualified supplier operating in Mexico’s Wholesale Electricity Market. The work focused on understanding how weather conditions influence electricity demand, supply and price formation, and on developing a forecasting framework capable of capturing these interactions in a volatile and weather-sensitive market environment.
The Study was carried out using historical data from 2021 to 2023, consisting of hourly electricity prices, demand and generation levels, natural gas prices and meteorological variables. A case study was first conducted to analyse the relationship between temperature, demand and price behaviour across Mexican market zones. The data were pre-processed into daily averages and examined to identify seasonal patterns, weather-driven demand responses and the influence of calendar effects on price variability.
Building on these findings, a practical forecasting framework was developed using Long Short-Term Memory neural network. Its performance was evaluated against a linear regression baseline using a rolling test period. In addition, Conformal Prediction was applied to generate calibrated 90% prediction intervals, providing probabilistic measures of forecast uncertainty.
The results showed that the LSTM model improved forecast accuracy significantly compared to the linear baseline, reducing error metrics across most zones and capturing both short-term fluctuations and broader seasonal trends.
The study concluded that advanced forecasting tools can strengthen operational decision making for electricity trading in Mexico. The findings also highlighted areas for further development, including the integration of renewable generation data, transmission constraints and more advanced modelling architectures to refine local price forecasts.
The Study was carried out using historical data from 2021 to 2023, consisting of hourly electricity prices, demand and generation levels, natural gas prices and meteorological variables. A case study was first conducted to analyse the relationship between temperature, demand and price behaviour across Mexican market zones. The data were pre-processed into daily averages and examined to identify seasonal patterns, weather-driven demand responses and the influence of calendar effects on price variability.
Building on these findings, a practical forecasting framework was developed using Long Short-Term Memory neural network. Its performance was evaluated against a linear regression baseline using a rolling test period. In addition, Conformal Prediction was applied to generate calibrated 90% prediction intervals, providing probabilistic measures of forecast uncertainty.
The results showed that the LSTM model improved forecast accuracy significantly compared to the linear baseline, reducing error metrics across most zones and capturing both short-term fluctuations and broader seasonal trends.
The study concluded that advanced forecasting tools can strengthen operational decision making for electricity trading in Mexico. The findings also highlighted areas for further development, including the integration of renewable generation data, transmission constraints and more advanced modelling architectures to refine local price forecasts.