Electricity load forecasting using genetic algorithm optimized sigmoid regression

Purpose This paper aims to determine an accurate model for forecasting electricity load demand by estimating sigmoid regression parameters using a genetic algorithm. Design/methodology/approach Meteorological data are used to prepare the city’s daily electricity load dataset from February 1, 2014 to November 30, 2019. A genetic algorithm optimizes five parameters of the sigmoid regression. Findings The study evaluates the usability of the genetic algorithm in sigmoid regression for forecasting electricity load demand. It finds that one of the two seasonal behaviors presented in load consumption is not a good fit for the electricity load models. Winter and summer seasonal models are estimated to increase accuracy. Furthermore, minimum, maximum and average temperatures are used to determine the optimal temperature type for minimizing errors. Test dataset errors are lower than training dataset errors, indicating that the models are sufficiently accurate for forecasting. Research limitations/implications The findings will be helpful to demand forecasting and energy marketplace experts who want accurate forecasts of their assets. Practical implications The findings will be helpful to demand forecasting and energy marketplace experts who want accurate forecasts of their assets. The proposed model’s relative mean absolute error is 71.8% and 18.1% better in the training and test datasets, respectively, compared with the naïve model, and the proposed model can lead to increased grid stability and cost savings. Originality/value Previous studies used four parameters, and the authors added one more parameter to the Sigmoid regression. In addition, prior studies employed the Levenberg–Marquardt method to estimate parameters, whereas the authors utilized genetic algorithms with extensive search capabilities.