Novel Training Methods Based ANN for the Consumed Energy Forecasting

Abstract

Artificial Neural Networks have demonstrated best effectiveness and excellent scheduling capabilities in realizing many purposes like recognition, clustering, classification, management and even prediction. For this reason, we have used RBF based Artificial NN for the dynamic forecasting of load and Photovoltaic production using many operations like forecasting, training and validation of the data accuracy. For the validation, the Mean Absolute Percent Error is calculated in function of the most three relevant input parameters, which are previous load and Photovoltaic production measurements, seasonability and temperature or solar radiation data. This work has used real-time measurements of load and Photovoltaic production for their comparison with the predicted load data using RBFNN algorithms for the calculation of MAE and MAPE, to deduce the performance of forecasting algorithms including the accuracy of the forecasted data. This research paper has treated 2 goals. The first is the short-term energy and Photovoltaic production forecasting including training operations. The 2nd goal is the calculation of Mean Absolute Error and Mean Absolute Percent Error via the comparison between the forecasted data and real-time measurements to evaluate the reliability of forecasted data and the performance of the forecasting algorithms. By this way, the dynamic prediction algorithms were implemented, the predicted data were compared to the same time-series measurements and forecasted energy MAPE was calculated.