Artificial neural network model with different backpropagation algorithms and meteorological data for solar radiation prediction

Abstract

Solar energy serves as a great alternative to fossil fuels as they are clean and renewable energy. Accurate solar radiation (SR) prediction can substantially lower down the impact cost pertaining to the development of solar energy. Lately, many SR forecasting system has been developed such as support vector machine, autoregressive moving average and artificial neural network (ANN). This paper presents a comprehensive study on the meteorological data and types of backpropagation (BP) algorithms used to train and develop the best SR predicting ANN model. The meteorological data, which includes temperature, relative humidity and wind speed are collected from a meteorological station from Kuala Terrenganu, Malaysia. Three different BP algorithms are employed into training the model i.e., Levenberg–Marquardt, Scaled Conjugate Gradient and Bayesian Regularization (BR). This paper presents a comparison study to select the best combination of meteorological data and BP algorithm which can develop the ANN model with the best predictive ability. The findings from this study shows that temperature and relative humidity both have high correlation with SR whereas wind temperature has little influence over SR. The results also showed that BR algorithm trained ANN models with maximum R of 0.8113 and minimum RMSE of 0.2581, outperform other algorithm trained models, as indicated by the performance score of the respective models.

A memetic dynamic coral reef optimisation algorithm for simultaneous training, design, and optimisation of artificial neural networks

“…Learning algorithms are divided into two main groups: local search (LS) and global search algorithms. The Backpropagation algorithms (BPs) 3 or Extreme Learning Machines (ELM) 4 belong to the first group, commonly used for weight optimisation, while Evolutionary Algorithms (EAs) 5 belong to the second group. This second group is usually referred to Neuroevolution 6 or application of metaheuristics such as EAs to the evolution of ANNs, also known in the literature as Evolutionary Artificial Neural Networks (EANNs) 7 – 9 , so that both the weights and the ANN architecture are optimised.…”

A novel framework for the evaluation of coastal protection schemes through integration of numerical modelling and artificial intelligence into the Sand Engine App

“…The function of ANN is to provide prediction values based on the historical data on which it is trained. Training allows ANN models to learn the relationship between input and output variables 39 , 44 , 45 and these relationships are then used for future predictions. ANN models can also learn non-linear relationships between different variables 46 , 47 .…”

Deep learning approach analysis model prediction and classification poverty status

“…Based on the concept that has been explained that ANN can do learning by adopting a mathematical calculation process [40]. Learning is capable of a model that is applied in the form of an algorithm to produce decisions [41]. The model is presented in an architectural pattern based on the input layer, hidden layer, and output layer [42].…”