Multimodel neural networks using socioeconomic variables for the prediction of residential electric consumption

This work intends to focus on the use of Multimodel Artificial Neural Networks (ANNs) for the projection of the residential electric demand, taking into account that this is the basis for an adequate planning of electricity distribution networks. The ANNs were developed in MATLAB®, trained according to the data recorded, and the final results of the different regions of study were compared with the official data provided by the UPME for the year 2017. Models of 2-layer ANNs capable of accurately predicting medium-term residential electrical consumption were designed, taking into account variables such as GDP per capita, population, residential electrical consumption and temperature. The ANNs were found and suggested as a model capable of incorporating the nonlinearities of the different study variables, in addition to having no complexity for the planner in their mathematical modeling. Thus, in addition to estimate the degree of precision of the forecast used, it is sought to achieve a high degree of accuracy in the decisions, taking into consideration that the increase of residential users and load are important topics for the energy supply companies in the next decade.