Investigating the impact of feature selection on the prediction of solar radiation in different locations in Saudi Arabia

doi:10.1016/j.asoc.2018.02.029

Applied Soft Computing

Volume 66, May 2018, Pages 250-263

https://doi.org/10.1016/j.asoc.2018.02.029 Get rights and content

Highlights

•
A study of different feature selection methods is carried out to predict the daily amounts of solar radiation in different locations in Saudi Arabia.
•
Four feature selection algorithms are applied: ReliefF algorithm, Monte Carlo uninformative variable elimination algorithm (MCUVE), random-frog algorithm, and Laplacian score algorithm (LS).
•
A computational intelligence model of a multi-layer neural network is used as the predictor.
•
The results showed the importance of using feature selection methods in order to obtain a reliable prediction of the amount of solar radiation compared with using all the features available.

Abstract

Predictions about the future amount of solar radiation is an important factor that affects the planning and operating of solar energy projects. However, it is a difficult task due to the existence of a high level of uncertainty that is associated with unknown future weather conditions. Accordingly, the prediction process, especially for the short term, requires more attention to unveil hidden relationships and interactions between related variables. Since a large number of parameters affect the estimation and prediction processes, there is a need to apply an effective and efficient feature selection of the input feature space. In this paper, an investigation of different feature selection methods is carried out in order to predict the daily amounts of solar radiation in different locations in Saudi Arabia using a neural networks (NN) predictor. First, the selection of the most important variables is carried out using four different algorithms: ReliefF algorithm, Monte Carlo uninformative variable elimination algorithm (MCUVE), random-frog algorithm, and Laplacian score algorithm (LS). Then, a computational intelligence model of a multi-layer neural network is used as the predictor. The predictor aims to predict the next-day global horizontal irradiance using selected meteorological and solar radiation observations. The experimentation and results of the feature selection methods and the prediction process are described. The results showed the importance of using feature selection methods in order to obtain a reliable prediction of the amount of solar radiation compared with using all the features available.

Introduction

Solar energy is linked with the variability of different types of weather, topography, and geography. Such variability is considered to be one of the main obstacles that impacts energy availability and, thus, prevents the optimal exploitation of solar energy, especially in the short term. To best exploit solar energy, a reliable prediction model is crucial. For any location, a reliable source of solar radiation data is a requirement in many associated applications [1]. The prediction process is affected by the uncertainty about future weather conditions, which are linked to a large number of meteorological and solar radiation variables in different time horizons. Historical solar radiation and meteorological records can be used to build automated prediction models that do not require experts knowledge. Therefore, the prediction process, especially for short-term, is tough task for researchers and operators. Accordingly, many models of solar radiation have been presented in the literature. Two of the main tracks taken to develop prediction models are numerical weather prediction (NWP) and artificial intelligence (AI) models (see [[2], [3], [4]]). In order to obtain reliable prediction models, there is a clear need to reduce the large number of meteorological and solar radiation variables into a representative set that enhance the accuracy of the prediction model and reduce the data redundancy before applying the prediction model.

This paper focuses on the investigation of the impact of the feature selection process on the prediction of solar radiation. The proposed methodology uses four different feature selection algorithms to determine the most important variables. A neural network predictor is then used to predict the next-day solar radiation for eight locations in Saudi Arabia. The background for this study is discussed in Section 2. The methodology, the results, and the analysis are described in Sections 3 and 4. Finally, the conclusions and future work are highlighted in Section 5.

Section snippets

Background

The objective of this study is to investigate the impacts of using feature selection methods prior to predicting next-day solar radiation. In this section, we review the problem considerations and the methods applied in this study.

Methodology

The experiment is made up of four main stages: preparing the data sets, selecting the features, building the neural networks predictor, and testing the new data set using the predictor model. Testing process uses new data sets that are different spatially and temporally from the training data. The experiment stages are illustrated in Fig. 2 where neural network models are built with and without the four feature selection algorithms. The experiment is carried out in the MATLAB environment using

Results and discussion

The experiment was run 50 times using a Matlab environment. The average, maximum, minimum, and standard deviations of the training and testing errors have been calculated using different error metrics, including root mean squared error (RMSE), mean absolute error (MAE), and mean bias error (MBE). The errors metrics are defined as follows: $RMSE = \sqrt{\frac{1}{n} \sum_{i = 1}^{n} {(f (x) - \hat{f} (x))}^{2}}$ $MAE = \frac{1}{n} \sum_{i = 1}^{n} |(f (x) - \hat{f} (x)|)$ $MBE = \frac{1}{n} \sum_{i = 1}^{n} (f (x) - \hat{f} (x)),$

where f(x) is the predicted value, and $\hat{f} (x)$ is the actual value. Estimation error

Conclusion and future works

In this article, the use of different feature selection algorithms and neural networks to predict daily solar radiation in eight locations in Saudi Arabia has been presented. According to the experimental results for feature selection, the use of four feature selection algorithms produced the best prediction behavior for all locations compared with algorithms that used all the features. The improvements in RMSE were found to be between 19.9% and 52.1% compared with the prediction values when

References (48)

K. Ulgen et al.
Diffuse solar radiation estimation models for Turkey's big cities
Energy Convers. Manag.
(2009)
Z. Sen
Solar energy in progress and future research trends
Prog. Energy Combust. Sci.
(2004)
T. Muneer et al.
Discourses on solar radiation modeling
Renew. Sustain. Energy Rev.
(2007)
A.K. Yadav et al.
Solar radiation prediction using artificial neural network techniques: a review
Renew. Sustain. Energy Rev.
(2014)
Z. Şen
Fuzzy algorithm for estimation of solar irradiation from sunshine duration
Sol. Energy
(1998)
A. Mellit et al.
A 24-h forecast of solar irradiance using artificial neural network: application for performance prediction of a grid-connected PV plant at Trieste, Italy
Sol. Energy
(2010)
C. Voyant et al.
Numerical weather prediction (NWP) and hybrid ARMA/ANN model to predict global radiation
Energy
(2012)
S. Bhardwaj et al.
Estimation of solar radiation using a combination of hidden Markov model and generalized fuzzy model
Sol. Energy
(2013)
J. Zeng et al.
Short-term solar power prediction using a support vector machine
Renew. Energy
(2013)
J.C. Lam et al.
Solar radiation modelling using ANNs for different climates in China
Energy Convers. Manag.
(2008)

A. Hepbasli et al.

A key review on present status and future directions of solar energy studies and applications in Saudi Arabia

Renew. Sustain. Energy Rev.

(2011)

A.A. El-Sebaii et al.

Estimation of global solar radiation on horizontal surfaces in Jeddah, Saudi Arabia

Energy Policy

(2009)

H. Bulut et al.

Simple model for the generation of daily global solar-radiation data in Turkey

Appl. Energy

(2007)

S. Rehman et al.

Spatial estimation of global solar radiation using geostatistics

Renew. Energy

(2000)

S.H. Alawaji

Evaluation of solar energy research and its applications in Saudi Arabia: 20 years of experience

Renew. Sustain. Energy Rev.

(2001)

M. Benghanem et al.

ANN-based modeling and estimation of daily global solar radiation data: a case study

Energy Convers. Manag.

(2009)

M. Benghanem et al.

A multiple correlation between different solar parameters in Medina, Saudi Arabia

Renew. Energy

(2007)

M.A. Ramli et al.

Investigating the performance of support vector machine and artificial neural networks in predicting solar radiation on a tilted surface: Saudi Arabia case study

Energy Convers. Manag.

(2015)

S. Salcedo-Sanz et al.

A CRO-species optimization scheme for robust global solar radiation statistical downscaling

Renew. Energy

(2017)

A.K. Yadav et al.

Application of rapid miner in ANN based prediction of solar radiation for assessment of solar energy resource potential of 76 sites in Northwestern India

Renew. Sustain. Energy Rev.

(2015)

H.D. Li et al.

Random frog: an efficient reversible jump Markov chain Monte Carlo-like approach for variable selection with applications to gene selection and disease classification

Anal. Chim. Acta

(2012)

K. Hornik et al.

Multilayer feedforward networks are universal approximators

Neural Netw.

(1989)

M. Mohandes et al.

Estimation of global solar radiation using artificial neural networks

Renew. Energy

(1998)

T. Khatib et al.

A review of solar energy modeling techniques

Renew. Sustain. Energy Rev.

(2012)

Cited by (42)

Semi-real-time decision tree ensemble algorithms for very short-term solar irradiance forecasting
2024, International Journal of Electrical Power and Energy Systems
Industrial activities are transitioning towards decarbonization, focusing on renewable energy sources, particularly photovoltaic solar energy. However, the inherent high variability of photovoltaic energy poses challenges. Some of them can be partially addressed by predicting electricity production, which in the case of photovoltaic solar energy is heavily based on solar irradiance prediction. Although extensive research has been conducted in this field, there is a noticeable gap in research regarding very short-term (intra-minute) forecasting under high-variability scenarios. In this proposal, real data from a photovoltaic solar plant in Alderville (Canada) were used to predict irradiance with a horizon of 15 and 30 s. The objective is to make this prediction in near-real time. To achieve this, we propose the use of machine learning algorithms based on decision tree ensembles, due to their low computational training cost and known effectiveness. On the other hand, we propose pre-processing the data through a temporal and spatial correlation analysis between measurements from different sensors. Feature selection analysis allows us to determine the direction of the wind and consequently identify the most relevant panels for model training. This preprocessing enhances the model retraining without the need for external information such as sky images or wind speed and direction on days with highly variable cloud cover. The presented methodology offers promising results with significantly reduced training times, demonstrating the suitability of this semi-online training approach for highly variable time series forecasting.
Composition of feature selection techniques for improving the global horizontal irradiance estimation via machine learning models
2024, Thermal Science and Engineering Progress
The rise in photovoltaic capacity installed globally and the sporadic nature of solar resources underline the significance of solar radiation forecasting for grid integration. The choice of relevant climatic variables for estimating global horizontal irradiance (GHI) is crucial because it influences estimation accuracy. To keep this in mind, different feature selection techniques such as filter, wrapper, and embedded are computed and compared to identify a pertinent input variable for the prediction of GHI. In order to reduce the dimensionality of the input features, it is pertinent to remove unnecessary variables using feature selection techniques. This comprehensive study explores the use of tree-based ensemble machine learning (ML) models in the modeling of global solar radiation. Eighteen statistical and ML regressor algorithms, along with ten feature selection methods are evaluated based on statistical comparisons of their computational effectiveness and estimation errors. The lowest mean absolute error (MAE) and root mean squared error (RMSE) of 0.9938 and 2.5327 achieved by the extra-tree regressor algorithm reveals that it outperforms the other models on the test dataset. The integration of feature selection techniques improves predictive capabilities, reducing MAE and RMSE by 13.15% and 5.24%, respectively, for the Mutual Information method. This study also identifies influential features critical to accurate GHI estimation, such as solar zenith angle, diffuse horizontal irradiance, and direct normal irradiance, which provide valuable insights into the dynamics governing solar radiation estimation.
Ensemble robust local mean decomposition integrated with random forest for short-term significant wave height forecasting
2023, Renewable Energy
A robust short-term significant wave height (Hs) modelling framework based on an ensemble local mean decomposition method integrated with random forest (i.e., En-RLMD-RF) is developed. The robust local mean decomposition (RLMD) decomposed the Hs data series into three subseries; amplitude modulation, frequency modulation and the low-frequency product function (PFs). The partial autocorrelation function was employed to determine the correlation-based significant predictor signals between the PFs at t₀ and t₁. Then the statistically significant PFs were incorporated into the random forest (RF) to construct the RLMD-RF model. The RLMD-RF based forecasted PFs were used again in the RF model as input predictors resulting in an ensemble-based RLMD-RF (i.e., En-RLMD-RF) model for forecasting short-term Hs. The En-RLMD-RF model is validated and compared with RF, extreme learning machine (ELM) and multiple linear regression (MLR) models and their hybrids RLMD-RF, RLMD-ELM, RLMD-MLR, En-RLMD-ELM and En-RLMD-MLR counterparts using a set of performance metrics. The results demonstrated that the En-RLMD-RF model generates better forecasting accuracy against the benchmarking models. This study is beneficial for the application and optimization of more clean energy resources worldwide for sustained energy generation.
The role of input selection and climate pre-classification on the performance of neural networks irradiance models
2022, Applied Soft Computing
Citation Excerpt :
In [31] an Automatic Weather Station (AWS) located in Abu Dhabi International Airport and its data from 2004 to 2013 was used to reveal that the variables most affecting model precision are sunshine duration and temperature. Next-day solar radiation prediction using neural networks using three algorithms for feature selection in three locations in Saudi Arabia (Al-Uyaynah, Hafar Al-Batin, and Al-Qunfuthah) is developed in [5]. The authors found that the predictions based on feature selection algorithms demonstrated better prediction behavior than using all the data.
Neural Network (NN) models are widely accepted in the derivation of solar radiation models due to their ability to adapt to various geographic and environmental conditions. Despite this, it is unclear what training variables are required to enhance the precision of the NN model or which of them could be considered redundant; therefore, this paper intends to clarify this issue by investigating if the Köppen climate classification could be used to substitute climatic measurements.
To this end, We analyzed a variety of NN architectures using 20 years of data from 1629 weather stations belonging to three different climate types (Climate A, B, and C). We found that Köppen climate sub-classification had a limited effect on the models’ performance when the information of all data types was processed together, resulting in barely noticeable improvements from 1.2% to 2.5%. However, if data were pre-classified according to climate type, the climate sub-classification input induced significant differences. Improvements up to 14% in the precision of the models were found for Climates B and A; moreover, temperature and relative humidity daily measurements could be replaced by Köppen climate information. Cross-validation analysis, using the same amount of data for all climate types, allowed us to confirm our findings for Climates A and B and revealed that data pre-classification according to climate type for Climate C, systematically increased errors from 10% to 24%, so replacing actual climatological measurements was not possible for this climate type.
Revealing such patterns would facilitate the derivation of models for scenarios of limited information on temperature and relative humidity in some locations and reveals the usefulness of soft computing to go beyond understanding climate complexity.
Deep learning CNN-LSTM-MLP hybrid fusion model for feature optimizations and daily solar radiation prediction
2022, Measurement: Journal of the International Measurement Confederation
Citation Excerpt :
It first describes the SMA approach for feature selection and then describe the foundations of the CNN algorithm and the LSTM approach. In this study, a wrapper feature selection method [85–87] based upon a meta-heuristic algorithm called Slime Mould Algorithm (SMA) is firstly used to select the optimal features for GSR prediction. We have selected SMA based on its recent performance as a metaheuristic algorithm derived from the diffusion and foraging behaviour of slime mould [88].
Global solar radiation (GSR) prediction plays an essential role in planning, controlling and monitoring solar power systems. However, its stochastic behaviour is a significant challenge in achieving satisfactory prediction results. This study aims to design an innovative hybrid prediction model that integrates a feature selection mechanism using a Slime-Mould algorithm, a Convolutional-Neural-Network (CNN), a Long–Short-Term-Memory Neural Network (LSTM) and a final CNN with Multilayer-Perceptron output (SCLC algorithm hereafter). The proposed model was applied to six solar farms in Queensland (Australia) at daily temporal horizons in six different time steps. The comprehensive benchmarking of the obtained results with those from two Deep-Learning (CNN-LSTM, Deep-Neural-Network) and three Machine-Learning (Artificial-Neural-Network, Random-Forest, Self-Adaptive Differential-Evolutionary Extreme-Learning-Machines) models highlighted a higher performance of the proposed prediction model in all the six selected solar farms. From the results obtained, this work establishes that the designed SCLC algorithm could have a practical utility for applications in renewable and sustainable energy resource management.
Solar energy modelling and forecasting using artificial neural networks: a review, a case study, and applications
2022, Artificial Neural Networks for Renewable Energy Systems and Real-World Applications
In this chapter, for forecasting with high possible accuracy, solar radiation intensity, an approach for identifying the optimum set of input data from large sets of input parameters, is elaborated, assessed, and proposed. This approach is an input variable selection (IVS) technique and it is established using artificial neural networks (ANNs) as an effective means for approximating numerically the computed objective function. Based on the approach presented, we can determine the best combinations of inputs giving a great possible correlation and approximation with the considered output parameter. Recorded data from about 35 stations in different climatic zones were used for both training and testing purposes. Several new linear formulas between the global solar radiation and other climatological and meteorological parameters were developed and evaluated. Based on these relationships, we can forecast other climatological and meteorological parameters such as temperature, wind speed, and humidity. The statistical analysis was done, and the best performance of the proposed approach was checked and duly validated. Finally, a review of the most possible applications of solar energy for both human health and engineering is briefly summarized and presented.

View all citing articles on Scopus

^☆: This study is part of a funded research project by King Abdulaziz City for Science and Technology (KACST) (Grant Number: 13-ENES2373-10). In addition, the author thanks King Abdullah City for Atomic and Renewable Energy (KACARE) for providing data.

View full text

Investigating the impact of feature selection on the prediction of solar radiation in different locations in Saudi Arabia☆

Highlights

Abstract

Introduction

Section snippets

Background

Methodology

Results and discussion

Conclusion and future works

Energy Convers. Manag.

Prog. Energy Combust. Sci.

Renew. Sustain. Energy Rev.

Renew. Sustain. Energy Rev.

Sol. Energy

Sol. Energy

Energy

Sol. Energy

Renew. Energy

Energy Convers. Manag.

Renew. Sustain. Energy Rev.

Energy Policy

Appl. Energy

Renew. Energy

Renew. Sustain. Energy Rev.

Energy Convers. Manag.

Renew. Energy

Energy Convers. Manag.

Renew. Energy

Renew. Sustain. Energy Rev.

Anal. Chim. Acta

Neural Netw.

Renew. Energy

Renew. Sustain. Energy Rev.