Work

Electricity demand forecasting for Tetouan, Morocco using Darts in machine-learning: A comparative study of forecasting models

Year: 2026

Type: article

Abstract: Accurate electricity demand forecasting is crucial for maintaining grid stability, reducing costs, and ensuring reliable energy delivery in rapidly growing urban areas. This study presents a unified comparative analysis of forecasting models for Tetouan City, Morocco, using the Darts time series framework with SCADA data collected at 10 minute intervals across three distribution zones during 2017. Meteorological variables including temperature, humidity, wind speed, and solar radiation, along with engineered temporal features such as hour, weekday, and month, are incorporated to enhance predictive capability. Six models, Linear Regression, Random Forest, XGBoost, TiDE, TSMixer, and Prophet, are evaluated under three preprocessing techniques (Min-Max, Max Absolute, and Standard scaling) at both 10 minute and hourly intervals. Model performance is assessed using MAPE, RMSE, and MAE metrics. Results show that TSMixer excels in fine grained, short term forecasting, achieving a MAPE of 5.66% for 10 minute Boussafou demand and the lowest MAE of 623.7 for citywide mean consumption. TiDE performs best for aggregated consumption patterns with a MAPE of 1.92%for combined 10 minute forecasts, while Random Forest achieves the highest accuracy in hourly forecasting with a minimum MAPE of 1.04% and MAE of 224.2 for mean consumption. Compared with previous Tetouan forecasting studies, our Darts based approach lowers RMSE and MAE by up to 60%, surpassing Random Forest, Prophet, and Bayesian Fine Tree baselines. These findings establish a scalable forecasting pipeline that integrates classical and deep learning models, providing a transferable methodology for utilities and planners seeking improved operational efficiency and energy management.

Source: Results in Engineering

DOI: 10.1016/j.rineng.2026.110289

Authors: Nripendro Biswas, Munaem Ahmed Mahdi, Mohona Das Gupta

Institution: Shahjalal University of Science and Technology

Language: English

Topic: Energy Load and Power Forecasting

Subfield: Electrical and Electronic Engineering

Field: Engineering

Domain: Physical Sciences

SDGs: Affordable and clean energy, Industry, innovation and infrastructure, Climate action, Sustainable cities and communities

Open Access status: gold

Estimated APC paid: $1,940