potential in Indonesiausing artificial neural networks (ANNs) approach. In this study, the meteorologicaldata during 2005 to 2009 from 3 cities (Jakarta, Manado, Bengkulu) are used for training the neural networks and the data from 1 city (Makasar) is used for testing the estimated values. The testing data are not used in the training of the network in order to give an indication of the performance of the system at unknown locations. Fifteen combinations of ANN models were developed and evaluated. The multi layer perceptron ANNs model, with 7 inputs variables (average temperature, average relative humidity, average sunshine duration, longitude, latitude, latitude, month of the year) are proposed to estimate the global solar irradiation as output. To evaluate the performance of ANN models, statistical error analyses in terms of mean absolute percentage error (MAPE) are conducted for testing data. The best result of MAPE are found to be 7.4% when 7 neurons were set up in the hidden layer. The result demonstrates the capability of ANN approach to generate the solarradiationestimation in Indonesiausingmeteorologicaldata.
Technique of ANN has been applied to solving meteorological global solarradiation problems by a number of researchers [6,9,8,13]. The potential use of ANN in this study is to validate the collected Global SolarRadiation (GSR) data with results obtained from ANN for prediction of solar global radiation for our stations and any other locations where solar energy is required.
The ANN models use different geographical parameters of a location as inputs for the prediction of solar radia- tion as discussed in . Al-Alawi and Al-Hinai  discussed multi-layer feed forward network, back propaga- tion (BP) training algorithm for global radiation prediction in Seeb, Oman. The inputs used in network were lo- cation, month, mean pressure, mean temperature, mean vapor pressure, mean relative humidity, mean wind speed and mean sunshine hours. Sözen et al.   used meteorological and geographical data as input va- riables in the ANN model for solarradiationestimation in Turkey. The transfer function for model is logistic sigmoid and learning algorithm is Scaled conjugate gradient, Pola-Ribiere conjugate gradient Levenberg-Mar- quardt.
The reanalysis data collected from the Era- Interim archive of the European Centre for Medium-Range Weather forecast were used to estimate the GSR over the Sahel, Guinea Savannah, Derived Savannah and Coastal regions in Nigeria using the NARX and MLR models for 2010 – 2015. The Surface data of minimum and maximum temperatures; relative humidity and wind speed for a period of 30 years (1980 – 2009) were used as input variables to develop the models. Multivariate linear regression analysis showed that minimum and maximum temperatures have a significant positive relationship while relative humidity and wind speed have a negative relationship with global solarradiation in all the four climatic regions in Nigeria. However, minimum and maximum temperatures have a significant negative relationship while relative humidity and wind speed have a positive relationship with GSR in Sahel region only.
As mentioned above, NNs has the appealing capability to recognize patterns in data. Indeed, NNs are able to approximate any continuous function at an arbitrary accuracy, provided the number of hidden neurons is sufficient. However, it is necessary to match the complexity of the NN to the problem being solved. The complexity determines the generalization capability (measured by the test error) of the model since a NN that is too complex will give poor predictions. In the NN community, this problem is called overfitting. Several techniques like pruning or Bayesian regularization  can be employed to control the NN complexity. In this work, we used the Bayesian Technique in order to control the NN complexity and therefore the generalization capability of the model .
In this study, nonlinear autoregressive recurrent neural networks with exogenous input (NARX) were used to predict global solarradiation across New Zealand. Data for nine hourly weather variables recorded across New Zealand from January 2006 to December 2012 were used to create, train and test Artificial NeuralNetwork (ANN) models using the Levenberg−Marquardt (LM) training algorithm, with global solarradiation as the objective function. In doing this, ANN models with different numbers of neurons (from 5 to 250) in the hidden layer as well as different numbers of delays were experimented with, and their effect on prediction accuracy was analyzed. Subsequently the most accurate ANN model was used for global solarradiation prediction in ten cities across New Zealand. The predicted values of hourly global solarradiation were compared with the measured values, and it was found that the mean squared error (MSE) and regression (R) values showed close correlation. As such, the study illustrates the capability of the model to forecast radiation values at a later time. These results demonstrate the generalization capability of this approach over unseen data and its ability to produce accurate estimates and forecasts.
All the input parameters feed their respective values to the input layer which processes that data towards hidden layer. All the functioning (training, validation and testing) of developed neuralnetwork takes place in the hidden layer of the network. The estimated solarradiation value for the respective location gets reflected at output layer. The measured data (1962-1978) of eleven parameters (mean value of year, duration, month, latitude, longitude, altitude, sun shine hours, temperature, humidity, wind speed and rain fall) given by Mani and Rangarajan for eighteen locations (Ahmedabad, Bangalore, Bhavnagar, Mumbai, Kolkatta, Goa, Jodhpur, Kadaikanal, Chennai, Mangalore, Nagpur, Nandi Hills, New Delhi, Poona, Port Blair, Shillong, Thiruvanathapuram and Vishakhapatnam) have been taken to train, validate and test the developed model (Mani and Rangarajan, 1980). In order to train the developed artificial neuralnetwork model for the estimation of global solarradiation 70 % of the measured data has been used while 15 % data has been used for the purpose of validation and the remaining 15 % data has been used for the testing purpose. A graphical user interface for solarradiationestimation with MATLAB programming has been framed and developed software has been checked for valid results. All the data has been linked with the developed graphical user interface to give the output for the desired inputs. The developed graphical user interface can be used to estimate global solarradiation at any location in India.
ABSTRACT- Solar energy is one of the basic elements for all renewable and fossil fuels. Solarradiationdata is always a necessary basis for the design of any solar energy conversion device and for a feasibility study of the possible use of solar energy in variety of applications. In the existing system the prediction of solarradiation is based on feed forward neural networks. The drawback behind that is they are less accurate. Hence the proposed system is used to predict solarradiation based on a non-linear autoregressive exogenous input model and find the suitability of crops for cultivation based on the atmospheric conditions. The daily average solarradiation for Tiruvallur region is predicted based on climatic factors such as minimum air temperature, maximum air temperature, air pressure and humidity that influence the crop yield for Tiruvallur region. A non-linear autoregressive exogenous input (NARX) is built to predict the solarradiation. The performance of the network is then analyzed by calculating the mean squared error and Regression analysis. The performance analysis is done to determine the accuracy between the actual data and predicted data. Based on the predicted solarradiation and along with other climatic parameters such as the amount of rainfall, soil, duration and suitable months the crops that are suitable for cultivation around Tiruvallur region are analyzed, identified and suggestions are provided to farmers. This work results in benefit of the farmer community around Tiruvallur region.
Photovoltaic (PV) energy is widely used over the world as the main step to solve the electricity shortage problems. Also, in remote and rural areas the use of PV systems can be considered the most important and more efficient source of electricity for different domestic applications and water supplying systems. Since the performance of any PV system is directly influenced by weather data such as ambient temperature or solarradiation, the accurate estimation of these parameters is essential for good design and operation. Different attempts have been carried out to determine the PV module surface temperatures using mathematical models of the PV module, empirical formula and by neural networks. Neuralnetwork (NN) doesn’t require any analysis of the system or scientific details; it only needs data from the system for training purposes. The present research describes the estimation of the PV module surface temperature using NN based on measured ambient temperatures and incident solarradiation. The NN is composed of input layer with two inputs (solarradiation and ambient temperature), hidden layer that has eight neurons and output layer to estimate the PV module surface temperature. Error back propagation algorithm was used to train the NN based on the measured data pairs at various working conditions. The result showed that, the estimation accuracy of the PV module surface temperature by the NN reached more than 96% of the measured value for clear and sunny days.
The period considered goes from the 01 of January to the 31 December 2015. Images of geostationary meteorological satellites are processed to produce esti- mates of solarradiation at the surface . The measurements taken by the satel- lite will probably offer the advantage of being more accurate and representative of the conditions prevailing between the ground stations than others, predicted by interpolation . Some authors use satellite data and data from the network of ground stations to build a map of the solar irradiation of a locality . The sites of Chad, for which we have had the data, are: Abeche, Aozou, Faya Largeau, Moundou and N’Djamena. That’s why it is on these sites that our study will be applied. Table 1 gives the geographical coordinates of each site.
The solarradiation has temporal and spatial variations. To collect this information, a network of solar monitoring stations equipped with pyranometers and data acquisition systems are generally established in the desired locations. However, the number of such stations in the network is usually not sufficient to provide solarradiationdata of the desired areas, especially in developing countries. This is mainly due to because of not being able to afford the measuring equipments and techniques involved. Therefore, it is necessary to develop methods to estimate the solarradiation on the basis of the more readily available meteorologicaldata.
Where, the subscripts ‘‘c’’ and ‘‘m’’ refer to the calculated and measured global solar irradiation values, respectively. [ P is the mean calculated global radiation and [ E is the mean measured global radiation. The low values of RMSE, MBE, MPE, MABE and MAPE are desirable but correlation coefficient (r) should approach to1 as closely as possible. Coefficient of determination (R 2 ) is a statistical measure of how well the regression line approximates the real data points. RMSE provides information on the short-term performance of the correlations by allowing a term by term comparison of the actual deviation between the estimated and measured values. It is also possible to have large RMSE values at the same time a small MBE or vice versa. The positive MBE and MPE shows overestimation while the negative MBE and MPE indicates under-estimation of the values [6, 23]. Error measures with absolute values (such as MABE and MAPE) provide valuable information because they do not have the problem of errors of opposite signs cancelling themselves out; a low mean error can at times be misleading but this is avoided with absolute error measures which provide the average total magnitude of the error. On the other hand, absolute error measures can also assume a symmetric loss function through the total magnitude of error without the true bias or direction of the error. Hence, when using the mean absolute values, it is also useful to compute a measure of bias [1, 6, 23].
Satellite (GOES) has a 4-km resolution in the termal infrared, while the NOAA-Advanced Very High Resolution Radiometer (AVHRR), Terra and Aqua- Moderate Resolution Imaging Spectroradiometer (MODIS) have 1-km spatial resolution. High resolution data from the Terra-Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) has a 90-m resolution and Landsat-7 Enhanced Thematic Mapper (ETM+) has a 60-m resolution in thermal region (Li, z.-l. et al., 2013). The AVHRR is among the most used for the agro- meteorological monitoring and climate studies by enabling data acquisition in daily global scans, for having calibrated thermal spectral bands as well as wide data availability because of simultaneous operation in several meteorological satellites. From the various products that can be obtained through these data, the LST is very important due to its great usefulness in agricultural monitoring, fire detection, sea surface state monitoring, and in the studies of climate change (Gusso & Fontana, 2007; Ferreira, 2004).
The main objective of this work is to predict average daily global solarradiation (GSR) without any measuring instruments , in future time domain for Madurai city, located in Tamilnadu (India) by using Standard multilayered feed-forward, back-propagation neuralnetwork with Levenberg- Marquardt (LM) training algorithm and Gradient descent back propagation (GD) algorithm. In order to train and test the neuralnetwork, three different artificial neuralnetwork models are developed, based on daily average meteorologicaldata like maximum ambient air temperature, minimum ambient air temperature and minimum relative humidity values for predicting global solarradiation. The measured data were randomly selected for training, validation and testing the neuralnetwork. The results from the three artificial neuralnetwork models shows that using the minimum air temperature and day of the year outperforms the other cases with absolute mean percentage error of 5.36% and mean square error of 0.006 when training was done by using LM back propagation learning algorithm. From the results it is very clear that neuralnetwork is well capable of estimating GSR from simple and available meteorologicaldata. It is expected that the models developed for daily global solarradiation will be useful to the designers of energy-related systems as well as to those who need to estimate the daily variation of global solarradiation for the specific location in Tamilnadu (India).
radiation on horizontal surfaces for India was done by Katiyar and Pandey (2010). Liu et al. (2015) found the changes in the relationship between solarradiation and sunshine duration in large cities of China. Robaa (2008) has done the evaluation of sunshine duration from cloud data in Egypt. Yang et al. (2012) have done the hourly solar irradiance time series forecasting using cloud cover index. Nimnuan and Janjai (2012) have found another approach for estimating average daily global solarradiation from cloud cover in Thailand. New types of simple nonlinear models to compute solar global irradiance from cloud cover amount were found by Badescu and Dumitrescu (2014). Ehnberg and Bollen (2005) have done a simulation of global solarradiation based on cloud observations. Reddy (1974) has devel- oped an empirical method for estimating sunshine from total cloud amount. Morf (2014) has done sunshine and cloud cover predictions based on Markov processes. Babatunde and Aro (1995) have established a relation- ship between “clearness index” and “cloudiness index” at a tropical station, for instance, Ilorin, Nigeria. Al- Mostafa et al. (2014b) have done a review of sunshine- based global radiation models. Manzano et al. (2015) established a method to estimate the daily global solarradiation from monthly data. Abraha and Savage (2008) have done a comparison of estimates of daily solar radi- ation from air temperature range. Temperature-related models were also derived by several researchers such as
The ability to accurately predict the incoming solarradiation is an important factor to improve the efficiency of a solar energy conversion system. One of the methods utilized to predict incoming solarradiation is the use of an empirical model. The empirical model is a technique which uses meteorological parame- How to cite this paper: Chan, C.K. and
Based on the result, the study proved that Levenberg-Marquadt back propagation algorithm for ANN can simulate well in predicting the missing stream flow daily data if the model customize with good configuration. The finding in this research will help set up a good platform for running the simulation of SWAT model in upperpart of Langat River Basin for climate change impact studies. The net will be basis on infilling missing river discharge data on Sungai Langat at Kajang station for 30 years that will be discussed on another paper.
As the FIS structure is now made available, ANFIS utilizes the hybrid learning algorithm to tune (optimize) the premise (nonlinear) and consequent (linear) parameters of the FIS via learning from the training data set and minimizing the error in order to realize the desired ANFIS model through 100 training epoch and error tolerance of zero. The training stage stopped, and the realized ANFIS model is supplied with the testing data to evaluate its generalization capability, since the testing error is the real performance measure of the model. The prediction performance of the ANFIS model is evaluated based on the RMSE, MAD, MAPD and R.
In Figure 7 the comparison of pre and post code effort estimations is given. The model developed has been applied to some student projects and the graph is plotted. In the figure the X axis represents the project number and Y-axis represents the effort. Series1 indicates the pre coding test effort for the proposed model and Series2 represents traditional method pre coding effort estimation, Series3 represents the post coding test effort for the pro- posed model and Series4 represents the traditional me- thod post coding test effort estimation. It is showing a variation of about 8% over large number of projects. Thus it confirms the fact the estimated efforts both in pre and post coding phase have higher accuracy than the conventional models which as shown earlier show large deviation.
Abstract: The non-renewable sources of energy are limited and will get exhausted eventually. Looking at the current need of electric power and its fulfilment, the non-conventional way of generating this energy has become essential. Climate change and energy crisis have motivated us to make use of renewable non-conventional source of energy. This paper discusses the theoretical assumptions and design aspects of developing a Model which will predict the solar power generation beforehand. The paper aims at promoting the use of renewable source of energy by developing a model which will accurately predict the solar power generation. The suggested model uses Long Short Term Memory Recurrent NeuralNetwork (LSTM RNN) Algorithms to predict the power generation which will be beneficial to both Industries and Residents.