Dailly Forecasting of Photovoltaic Power Using Non-Linear Auto-Regressive Exogenous Method

Accurate and realistic models of forecasting are very relevant as methods for the successful implementation of renewable energy sources into smart grids. This work, present a short-term forecasting of the power output of building photovoltaic system installed in the Heliopolis university, Cairo, Egypt. The proposed method forecast one day using the autoregressive exogenous (NARX) method, based in neural network times series and based in the exogenous input data of the neural networks. The neural network (NN) is trained with Bayesian algorithm to attain the best performance of the forecasted results. However, the NARX neural networks are efficient to overcome a variety of problems and are common for nonlinear control applications as well as for the forecasting. The NARX model opts for accelerated training and level of convergence and good representation and is characterized by attractive dynamics and sensitivity to interference. This model differs in the use of the exogenous neural network input or exogenous variables. The performance of the NARX approach was analyzed as a function of training data sets, error descriptions based on experimental results of the photovoltaic system. The NARX results is compared with experimental data and neural network method. In addition, a statistical criterion is used to evaluate the effectiveness and performance of NARX model. NARX method performed the neural network in term of the statistical criterion which has a low value. Generally, the efficiency and precision of the results depend heavily on the environmental conditions and the data input.