0
Research Papers

Online Sequential Learning of Neural Networks in Solar Radiation Modeling Using Hybrid Bayesian Hierarchical Approach

[+] Author and Article Information
Sajid Hussain

Department of Mechanical Engineering,
The Petroleum Institute,
P.O. Box 2533,
Abu Dhabi, The United Arab Emirates
e-mail: sahussain@pi.ac.ae

Ali AlAlili

Mem. ASME
Department of Mechanical Engineering,
The Petroleum Institute,
P.O. Box 2533,
Abu Dhabi, The United Arab Emirates
e-mail: alialalili@pi.ac.ae

1Corresponding author.

Contributed by the Solar Energy Division of ASME for publication in the JOURNAL OF SOLAR ENERGY ENGINEERING: INCLUDING WIND ENERGY AND BUILDING ENERGY CONSERVATION. Manuscript received January 28, 2016; final manuscript received September 6, 2016; published online October 18, 2016. Assoc. Editor: Philippe Blanc.

J. Sol. Energy Eng 138(6), 061012 (Oct 18, 2016) (10 pages) Paper No: SOL-16-1057; doi: 10.1115/1.4034907 History: Received January 28, 2016; Revised September 06, 2016

Traditional learning process in solar radiation modeling usually requires historical data to perform regularization using training and cross-validation approaches. However, in applications where no historical data are available, regularization cannot be performed using traditional techniques. This paper presents a hierarchical Bayesian framework with the extended Kalman filter (Bayesian-EKF) to perform regularization in sequential learning of the artificial neural network (ANN) for solar radiation modeling. A highly stochastic time series for daily solar radiation, the global horizontal irradiance (GHI), is modeled based on different meteorological variables including temperature (T), relative humidity (RH), wind speed (WS), and sunshine duration (SSD). A comparison is made with well-known methods including the ANN-based nonlinear autoregressive with exogenous inputs neural network (NARX-NN) and Wiener filter-based multivariate linear regression (MLR). The method is validated on test data using coefficient of determination (R2) and root mean squared error (RMSE). The proposed technique effectively estimates the noise components in the data and achieves superior performance as compared to the traditional learning processes of NARX-NN and MLR. Moreover, it is more robust to statistical outliers in the data and does not require prior history for training and cross-validation. In the presence of the outliers, the performance of the NARX-NN degrades from R2 = 94.73% to R2 = 85.85% but there is virtually no difference in the case of Bayesian-EKF. Over and above, MLR performs better than NARX-NN but worse than Bayesian-EKF.

FIGURES IN THIS ARTICLE
<>
Copyright © 2016 by ASME
Your Session has timed out. Please sign back in to continue.

References

Figures

Grahic Jump Location
Fig. 1

Block diagram of the proposed method

Grahic Jump Location
Fig. 3

Daily evolution of T, RH, WS, SSD, and GHI over Abu Dhabi, UAE

Grahic Jump Location
Fig. 4

R2 boxplot versus hidden layer units for MLP (30 simulations each)

Grahic Jump Location
Fig. 5

NARX-NN simulations without outliers: (a) GHI measured versus simulated and (b) regression analysis

Grahic Jump Location
Fig. 6

NARX-NN simulations with outliers: (a) GHI measured versus simulated and (b) regression analysis

Grahic Jump Location
Fig. 7

Bayesian-EKF without outliers: (a) GHI measured versus simulated, (b) regression analysis, and (c) q-parameter

Grahic Jump Location
Fig. 8

Bayesian-EKF with outliers: (a) GHI measured versus simulated, (b) regression analysis, and (c) q-parameter

Grahic Jump Location
Fig. 9

Boxplots for NARX and EKF comparisons

Tables

Errata

Discussions

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In