Typification of the demand-generation relationship of Colombian electricity market and forecast of demand at an hourly-daily level based on consumption patterns

Abstract

This investigation establishes the relationship between demand and generation of the Colombian energy market by characterizing hourly and daily consumption patterns and later forecasting electricity energy demand at the hourly-daily level. The dataset used had variables demand and generation of electricity at hourly and daily levels, from 1 January 2019 to 30 September 2020. Ward’s method was applied with cosine similarity to establish the consumption patterns. Linear Regression, Support Vector Machine, and Random Forest were implemented to forecast consumption, and the model chosen was the one whose lowest Mean Absolute Percentage Error (MAPE) was selected. Daily energy consumption was classified into three groups and hourly energy consumption in six groups. The generation is in line with the demand, which indicates that the system is efficient. The algorithm that best forecasted hourly energy demand was linear regression, except for days with low demand peaks, such as October and December holidays.