OCTOBER 19-20, 2012 - YMCA University of Science & Technology

Proceedings of the National Conference on
Trends and Advances in Mechanical Engineering,
YMCA University of Science & Technology, Faridabad, Haryana, Oct 19-20, 2012
Kalogirou et al. [11] applied Artificial Neural Network (ANN) to estimate the useful energy extracted and the
temperature rise in the stored water of Solar Domestic Heating Water System (SDHWS). The data from 33 sets
were randomly selected. 30 sets were used for training and testing and the remaining 3 were randomly selected
for validation of model. The input data included was collector area (from 1.81 m 2 to 4.38 m 2 ), storage tank heat
loss coefficient (U-value), tank type, total daily solar radiation, mean ambient air temperature, storage volume,
water temperature in storage tank and type of system. A multilayer feed forward neural network consists of an
input layer (2 neurons), some hidden layers (8 neurons) and an output layer (2 neurons). The results for the
useful energy extracted from the system and temperature rise in stored water was 0.9722 and 0.9751 respectively.
The ANN method can use at different weather conditions and for completely unknown systems. The results
obtained within 7.1 % and 9.7 %. Its performance can be improved by knowing collector performance
characteristics.
Kalogirou et al. [12] estimated useful energy extracted from the thermosyphon solar water system and the stored
water temperature rise. For this an ANN network has been trained to handle a number of unusual cases using
performance data for four types of systems, with same collector panel and varying weather conditions. The result
obtained maximum deviations of 1 MJ and 2.2 o C using random data for both with performance equations
developed from the experimental measurements and with the artificial neural network. The predicted values
enlightened the effectiveness of the proposed ANN method for the estimation of the performance of the
particular thermosyphon solar water system in any of the configurations considered in this study. 30
thermosyphon solar domestic water-heating systems have been tested and modeled as per the procedures
outlined in the standard ISO 9459-2, for three different locations in Greece [13].
Out of these, data of 27 systems were used for training and testing the network and data of remaining 3 have
been used for validation. Two networks were trained for solar energy output estimation first for storage-tank
capacity and another for the system and the average quantity of hot water required per month at desired
temperatures of 35 and 40 o C. The R 2 -value set was: for first network was equal to 0.9993 and for second were
equal to 0.9848 and 0.9926 for the two output parameters for the training data [13].
A similar type of approach has been adopted for predicting the long-term performance of three forcedcirculation-type
solar domestic water-heating systems [14]. The maximum percentage differences of 1.9 and
5.5% have been obtained for the two networks, when unknown data have been used, respectively.
B. Solar Radiation Estimation
ANNs applied to predict global solar radiation in the areas that were not covered by direct measurement
instrumentation [15]. In this study, input data considered for the network are such as: the location, month, mean
temperature, mean relative humidity, mean pressure, mean vapor pressure, mean wind speed and sunshine
duration. An ANN model proposed by Mubriu et al. [16] efficiently used to estimate monthly average daily
global solar radiation on a horizontal surface. They used data obtained from three different sites for training a
neural network and formulating an empirical model and one site for checking the ANN and Empirical models. In
this study a feed-forward back propagation neural network with on hidden layer consisting of 15 neurons with
tangent sigmoid as the transfer function was used. The input data included was sunshine hours, cloud cover,
maximum temperature together with latitude, altitude and longitude of location. The maximum temperature and
cloud cover data was obtained from Uganda Meteorological Department from 1993 to 2005. The result shows
that the mean bias error (MBE) of 0.059 MJ/m 2 and root mean square error (RMSE) of 0.385 MJ/m 2 . Due to the
ability of reliably capturing non linearity nature of solar radiation, the developed ANN model proved to be
superior over empirical model. Global solar irradiation can be estimated from different ANN models [15-23] and
the results are compared for the correlation coefficients and the absolute percentage error for these different
models are shown in Table II.
Table 2:Comparison of global solar radiation model for correlation coefficients and the absolute percentage
error
Root
S.
Model by
Mean Correlation
No
Year
author
Absolute coefficient
.
Error
1
Alwai and
Hinai [15]
1998 7.3 0.989
2
Mihalakakou
et al. [17]
2000 0.2 0.96
3 Sozen et al. 2004 6.7 0.99
190