Case Study: Administrador del Mercado Mayorista achieves 50% reduction in short-term electricity demand forecast error with MathWorks (MATLAB)

Administrador del Mercado Mayorista Develops AI-Based Models for Predicting Electricity Demand

Administrador del Mercado Mayorista (AMM), which operates Guatemala’s national grid and wholesale electricity market, needed accurate hour-by-hour demand forecasts to maximize hydroelectric resources, optimize reserves, and lower energy costs. Their previous regression-based approach was labor-intensive and didn’t account for ambient temperature effects, so AMM turned to MathWorks and used MATLAB with Deep Learning Toolbox, Statistics and Machine Learning Toolbox, and MATLAB Compiler to build a modern forecasting solution.

Using MathWorks’ MATLAB, AMM cleaned 17 years of hourly demand and temperature data, evaluated multiple models, and implemented a hybrid solution combining a bagged decision tree and an artificial neural network packaged as a standalone application with a graphical interface. The MathWorks-based tool halved prediction error (MAPE improved from 3.7% to 1.7%), was developed and deployed in six months, automatically retrains to adapt to pandemic-driven changes, and now provides day-ahead and 10-day forecasts with extensions underway for medium- and long-term predictions.

Administrador del Mercado Mayorista