A non-linear auto-regressive exogenous method to forecast the photovoltaic power output. (April 2020)

• Record Type: Journal Article
• Authors: Louzazni, Mohamed; Mosalam, Heba; Khouya, Ahmed; Amechnoue, Khalid
• Is Part Of: Sustainable energy technologies and assessments. Volume 38(2020)
• Publication Date: 2020-04
• Subjects: PV Photovoltaic -- PVG Photovoltaic generator -- PVP Photovoltaic plant -- NARX Non-linear auto-regressive exogenous -- LM Levenberg-Marquardt -- NWP Numerical weather prediction -- SVM Support vector machine -- AR Autorepression model -- ARX Autoregressive with exogenous input -- MA Moving average model -- ARMA Autoregressive moving average model -- ARMAX Autoregressive moving average with exogenous inputs -- ARIMA Autoregressive integrated moving average model -- ANN Artificial neural networks -- BJ Box-Jenkins -- H Hessian matrix -- SW Solar World -- DC-DC Direct-current -- GNA Gauss-Newton algorithm -- R-NARX Recurring non-linear auto-regressive exogenous -- STC Standard test conditions -- η Efficiency of photovoltaic plant -- A Plant area -- G Solar irradiance -- Tambt Ambient temperature -- Tc Temperature of solar cell -- μ(t) Input -- γ(t) Output -- na System exponent number -- nb Input exponent number -- aij, bi, and ci Fixed real valued weights -- J Jacobian matrix -- x Weight vector -- µ Variable scalar -- I Identity matrix -- ε Residual error vector -- Pexp, I Observed values -- Pfor, I Forecasted values -- IAE Individual absolute error -- RMSE Root means square error -- R2 Coefficient of determination -- Pexp, i, Pfor, i Observed and forecasted values -- HN Different hidden neurons -- ND number of delayers; Forecasting method -- Photovoltaic plant -- Power output -- NARX -- Times Series Analysis; Renewable energy sources -- Periodicals; Energy development -- Technological innovations -- Periodicals; Electric power production -- Periodicals; Energy storage -- Periodicals; 333.79
• DOI: 10.1016/j.seta.2020.100670
• Languages: English
• ISSNs: 2213-1388

Abstract

Highlights: Forecasting of photovoltaic power production installed in Belbis, Egypt. Using the nonlinear autoregressive with exogenous model (NARX) based in neural network times series and Levenberg-Marquardt training algorithm. The performance of the used technique has been studied as a function of training data sets, error definitions and based on photovoltaic plant experimental data. The obtained results shown that the forecasted power obtained by NARX method give a high correlation with the experimental data. Abstract: This paper deal about the prediction of SunModule SW 175 monocrystalline photovoltaic (PV) module power output installed in Belbis, Egypt. The proposes prediction model forecast one month using a non-linear auto-regressive exogenous method, based in neural network times series and Levenberg-Marquardt training algorithm. NARX neural network are powerful to solve several problems and popular in nonlinear control applications. The NARX model is choosing for rapid training and convergence speed and strong representativeness and is characterized by favourable dynamics and resistance to interference. Besides, the exactitude of NARX method has examined as a function of training data sets, error definitions relying on experimental data of a PV framework. The predicted power acquired by the NARX method gives a high correlativity with the experimental data and comparatively low errors. The forecast of output power obtained with the NARX method are compared with neural network and experimentally measured data. The obtained result is very accurate in R 2 coefficient 99.47% and MSE = 20.5753% compared to NARX-Bayesian R 2 = 99.47 and RMSE = 21.71%. Generally, the execution and exactness of the results are exceedingly relying upon the climate condition, and the R 2 took a low value if the user data in series analysis are not very accurate.