Regresar

Prediction of electrical energy demand by artificial neuronal networks

Abstract:

A prediction model of electricity demand based on Artificial neural networks is proposed to improve the planning, operation and maintenance of power plants, starting with the field observation of the electrical substation where 70,128 records were obtained corresponding to 10 years; By means of preprocessing, 726 lost and atypical data were processed, the network architecture was determined by the technique of dynamic and forced search of the best local minimums, to train the network with descending gradient. The percentage of average absolute error of the model was 2.63%, while that based on multiple linear regression 4.56%. The artificial neural network based prediction model has better performance than the multiple linear regression model. It is recommended before designing a neural model to perform pre-processing to correct outliers, lost data and others smooth the time series in order to obtain satisfactory results.