Two cell architectures based on buried nano-electrodes are under investigation.

One electrode of the asymmetric set-up is realised as a comb-like structure of verti­cal electrodes and one planar counter elec­trode. The proof of principle was made and first experimental results have been pre­sented recently [9]. On the basis of the same acrylic substrate, an interdigital nano­electrode configuration can be realised.

The dimensions of the acrylic microstruc­ture are a period of the lamellas of 720nm, a depth of approximately 400nm and a width of the cavities of 400nm. The vertical metal electrodes are made by evaporation of the lamellae under a certain angle in such a

way that only the vertical regions and necessarily the tips of the structure are coated. Two electrodes can be realised by oblique evaporation of different metals. The proper separa­tion of the two electrodes is challenging from an technological point of view (figure 7).This
is done by a lift-off procedure. Once the separation is successful, there is no danger in creating shunts as the active polymer fills the cavities in the final step (figure 8).In con­trast to the planar solar cell architecture based on an ITO-substrate, both electrodes are deposited prior application of the photo active film.

This can be of an advantage for device preparation. The interface formation is an important issue. Suitable contacts with high selectivity are under investigation. As the dimensions of these structures are in the range of the wavelength of the light, near field optics play an important role. Strong interactions of the light in terms of high ab­sorption in TM-polarisation are observed.

Besides the application for solar cells, the concept of vertical interdigital nano­electrodes with distances down to 200nm offers a wide range of different applications like sensors or OLEDs.

Acknowledgements

We would like to thank Philippe Lalanne from Laboratoire Charles Fabry de L’Institut d’Optique, Orsay Cedex, France for the sim­ulation tool, J. C.Hummelen from University of Groningen for the delivery of the chemical components, Harald Hoppe from University of Linz for the determination of the optical constants and Thomas Schaller from Institut fQr Mikroverfahrenstechnik Forschungszen- trum Karlsruhe for the delivery of the microprism masterstructure.

The atmospheric radiation model proposed by Martin and Berdahl [12] is used in this study to estimate the cooling power. In this model the radiation is expressed by

F, (л, е) = 1 -(l — Ss Jt{X)/tav ]■ Exp[b(l.7 — 1/cos <?)]. (5)

Here, es is the averaged emissivity in clear sky, and is given by

= 0.711 + 0.5б(гф/100)+ 0.73(ГФ/100)2 . (6)

Here, 7dp is the dew point temperature [°C] (-13°C<Tdp<24°C) , b is a constant parameter, and t(X) is a shape function of the atmospheric window. tav is represented by

4 = £■ dXt {x)Eb (Л, Ta )/£ dAEb (Л, Ta) . (7)

Here, Ta is ambient temperature, and Eb(X, Ta) is the monochromatic thermal radiation power defined by Planck’s law. Using these formulations, the atmospheric radiation R is represented by

Rs (Л, в) = є, {x,0)Eb(Л, Ta)

Varying the zenithal angle the spectral emissive power shown in Fig.6 is calculated by using the above equations.

The emissive power absorbed by the sky radiator surface is expressed by

4 (Л, в) = Е, (Л, в)кs (Л,0). (9)

Here, es(49) Is the spectral emissivity of sky radiator surface. On the other hand, the emissive power from sky radiator surface is represented by

Re M = e, (Л,0)Еь (Л, Te). (10)

Here, Te is the surface temperature of sky radiator. Using eq’s (9) and (10), the cooling power is expressed by

Ce = 2я% dX fj1 sin в cos 0d0{Re (X,0)~ Ae (Л, в)). (11)

Hybrid system must include operation strategies that describe the energy flow between the generator and the load. The operation decision variables to be optimized represent routing and operation decisions that are based on the power flow modeled for the hybrid system. The main operation decision variables of the hybrid system model are presented as follows:

CSCPV, CSCW, CSIN, SOCmin, SOCmax, CSL, Vmax-off, Vmax-on, Vmin-off, Vmin-on

Some of these decision variables may be set before optimization of the hybrid system. To protect the battery against overcharging, PV array or wind generator is disconnected from the system, when the DC bus voltage increases above Vmax-off and when the current required by the load is less than the current generated by the PV array and wind generators. They are connected again when DC bus voltage decreases below Vmax-on. To protect the battery against excessive discharging, the load is disconnected when the DC bus voltage falls below Vmin-off and when the current required by the load is greater than the current generated by the PV array and wind generators. The load is connected again when DC bus voltage increase above Vmin-on. Control switch and charger controller operation strategies are described below.

Charger control switch of PV, CSCPV,

CSCPV = 0, If V> Vmax-off and ILOAD<IPV CSCPV = 1, If V< Vmax-on Charger control switch of wind generator, CSCW,

CSCW = 0, If V> Vmax-off and ILOAD<IW CSCW = 1, If V< Vmax-on Load control switch, CSL,

CSCW = 0, If V< Vmin-off and ILOAD>IW + IPV CSCW = 1, If V>Vmin-on

The hybrid system model has many constrains such as technological, socio-economic, legal or physical. The constraints in the presented approach are given by technical characteristics of battery operation and by matching demand and generated energy. Constraints can be formulated as follows;

SOCmax > SOC(t) > SOCmin

IPV (t) + IW (t) + IBD (t) > Iload(t) + IBC (t)

At all times, equation 13 can not be proved. According to the application, the load may not be served with desired amount of energy. This situation is described as loss of load probability. LLP of the energy system can be calculated using equation 14. Also, LLP is the size of system reliability.

T T n Energy _ Demand

LLi —

Energy _ Served (14)

The object of optimization procedure is to achieve hybrid system that generates energy with the lowest cost. The objective function of the hybrid system model considered LCC function that was defined with in costing model of hybrid system. The hybrid system model needs to be optimized with respect to the decision variables and operation strategies such that the minimum life-cycle cost is achieved.

The order in which the independent variables are assigned to the vertical column is essential. In the case of mixed variables and interaction between variables, the variables are to be assigned at right columns as stipulated by the orthogonal array. Hence, TPC, TEVA and Temperature are assigned to columns 1,2 and 3 respectively as shown below in table 2.

Conducting the experiment

Randomisation is the mixing or shuffling of the order in which events occur so that each have an equal chance of being selected. There is a substantial amount of uncontrolled or residual variation in any experiment, so no test can ever be repeated exactly. This residual variation is caused by factors, which have not been (or cannot be) controlled by the experimenter. These factors are called uncontrolled factors. Randomisation reduces the effect of these uncontrollable factors. Failure to randomise can cause successive error terms to be correlated, which can lead to misleading data acquisition (Ranjit, Roy, 1996).

Now that the orthogonal array has been selected the experiments are conducted as per the level combinations. It is essential that all the experiments be conducted. The experiments are run in random order.

3.1 General considerations

Up to now, the PV cells used in products have been mainly limited to cells with flat surfaces. The main reason for this is found in the early PV cells made from thick Silicon wavers which were rigid and fragile. Also only flat laminations were feasible. Nowadays however there are several new emerging technologies enabling curved PV surfaces resulting in new product design opportunities for integrating PV cells into or on products. Also PV cells can have various colours. It is rather strange that this freedom in shape and colour in photovoltaic cells is not yet widespread in use in industrially designed products, while in architectural design it has already been recognised and is in fact identified as one of the major factors that pushed the growing use of photovoltaic on buildings [Pellegrino et. al., 2002].

3.2 Curved PV surfaces

In general the PV technologies with potentials for curved surface can be divided into:

1. Thin wafer mono and multi crystalline Silicon PV cells

2. Thin Film amorphous Silicon PV cells on flexible substrates

3. Thin Film PV cells deposited on or between flexible substrates

4. Dye Sensitised PV cells on or between flexible substrates

5.
Other emerging technologies Ad. 1.

In the same category one could place thin micro-crystalline and multi-crystalline PV wavers.

Figure 11:

Curved PV panel by using multi-crystalline wafers pressed between curved glass plates

For safety and stability the thin wafers could be pressed and sandwiched in between two curved glass plates as can be seen in Figure 11 [Beers, 2003]. Colour if functional can be
added either by colouring the individual PV cells or by colouring the thick glass plates. The prototype panel in Figure 11 is made consisting of two 5 mm thick pre-bend glass plates laminated with resin. The back panel is colored blue. This prototype panel has been made with aid of ‘Tetterode Curved Glass BV, in Voorthuizen and Energy research Centre of the Netherlands (ECN) in Petten.

Ad.2.

Thin amorphous Silicon PV cells can be manufactured for instance with Chemical Vapour

Deposition (CVD) techniques. Curved PV surfaces can be obtained by using for example ridged curved glass plates as substrate for these CVD techniques [Curvet, 2004]. These modules are mmanufactured in co-operation between Curvet Spa from Pesaro and Bekaert ECD Solar Systems LLC ( modules with brand name Uni-Solar®) An example of such a multi-junction amorphous Silicon PV can be seen in figure 12.

Figure 12: Amorphous Silicon PV module on curved glass substrate

Ad. 3

Thin film PV cells deposited on flexible substrates such as metal foils. For example Flexible CIS (copper, indium and selenium) PV cells [Solarion, 2004]

Figure 13: Example of a bendable thin CIS PV module

Ad.4

Dye Sensitised PV cells can be manufactured as laminates between plastic sheets [Boschloo and Hagfeldt, 2003]. These PV devices can even be made as flexible ribbons. Still performing in indoor conditions such as for example: at 250 lux yielding 3 V and 8 uA with an active area of 10,8 cm2 The core of the dye-sensitized technology consists of nanometer-scale crystals of TiO2 semiconductor coated with a monolayer of light-absorbing dye and embedded in an electrolyte between the front and back electrical contacts.

Ad 5

Other possible solutions for curved PV cells are provided by the so-called ‘Semiconductor on Insulator by Enamel Techniques’ [Poelman and Kan, 2000] and the ‘Concrete Solar Cell’ [Arthur et. al., 1997]. In the first one, doped Silicon granules are imbedded in Indium doped Emails. The Indium has two functions as doping reservoir and as conducting interconnection, resulting on large surface PV cells. In the second solution doped Silicon granules was embedded in concrete resulting in Building Integrated PV walls.

Thin films of ZnS and ZnO/ZnS are obtained by chemical spray pyrolysis in air atmosphere. The experimental set-up used for the preparation of pyrolytically spray deposited films is described elsewhere in Ref [13,14]. Two different set of solutions have been used as precursors for the ZnS film: a) zinc acetate dehydrated (10-3 M) and thiourea (4-10"2M) in bidestillated water; b) zinc chloride (10"3M) and thiourea (4-10"2M) in bidestillated water. The set of solutions are sprayed at 30 ml/h onto the clean silicon substrates heated at 500°C for the first solution and 450°C for the second one. Compressed air is used as carrier gas. Some of the ZnS thin films were coated with a ZnO film by spray pyrolysis also using aqueous solutions of zinc acetate dehydrated as precursor. In that case, during spray deposition the substrate was held at 400°C.

Observation of surface morphology was performed using a scanning electron microscope (SEM JEOL model JSM5300). Optical measurement data were obtained with an UV-Visible Shimadzu 3101 double beam spectrometer, thickness calculations have been carried out using the method described in bibliography [15]. The chemical composition of the films and sublayers was studied by XPS combined with 4KeV Ar+ and 1.5KeV Ar+ sputter depth profiling using PHI 5700 equipment. The sputter rate at 4KeV Ar+ and 3nm/min (the rate is assumed to be 3nm/min) has been used in the monolayers, during the analysis of the outer ZnO film of the bilayer and when the sulphur content is
higher than the oxygen in it. The sputter rate an 1.5 KeV Ar+ and 1nm/min (the rate is assumed to be 0.5nm/min) has been used in the bilayer during the analysis of the interface to reduce charge effects. A standard X-ray source of 15KV, 300W and Mg Ka (1253.6eV) was used. Binding energies (BEs) were referenced to the Zn2p3/2 peak at 1022eV due to ZnS and ZnO in unsputtered surfaces. Spectra were handled with PHI — Access V.8 and Multipak software, both from Physical Electronics.

The expected yearly energy yields have been calculated using a specially developed software (Real Reporting Conditions Rating) [10]. In contrast to the power evaluation in standard test conditions (STC), it provides a rating of module performances under real conditions over an extended period of time. In addition to module parameters (see below), the software uses climate data bases to account for temperature and illumination conditions at the different geographical locations. The simulations have been realised for four different places (see table 2).

The input data for the modules were:

1) The efficiency in standard test conditions (STC)

2) The normalised spectral response

3) The variation of the short-circuit current as a function of the angle

4) The variation of the efficiency as a function the temperature

5) The variation of the efficiency as a function of the irradiance

6) The NOCT of the modules

For both the module with AR glass (Module AR) and for the module without AR (Module NG), the orientation was the same, chosen at values yielding the optimum energy output at each location. The data for the points 1)-3) have been directly taken from the results described in this paper. For point 4), a temperature dependence of the efficiency of —

0. 5%/°C has been assumed. For the point 5) we have first assumed that the short-circuit current Isc was proportional to the irradiance. For the module AR, Isc was set 2.65% higher than for the module Normal for perpendicular illumination. The efficiency as a function of the short-circuit current has then been calculated using the I-V characteristics obtained

from a two-diode model (see ref. [11], or [14] for a detailed application of the model), taking into account typical values for the series and parallel resistances of the solar cells. The efficiency as a function of Isc is then given by maximising the output power density VI. This procedure permits to reproduce accurately the fact that efficiencies at low irradiance in modules with crystalline Si solar cells are lower than at high irradiance for a given temperature. In particular, at low light level, this procedure leads to an additional efficiency gain for the module AR, because Voc and the fill factor FF are also improved when the current is higher. For the simulations, 3°C was added to the outdoor measured backskin temperature to account for temperature gradients through the module [12]. An NOCT of 45.2°C was hence taken for the modules Normal and of 46.4°C for the module AR. In summary, besides the higher short-circuit current and the better angular dependence, the module AR describes a realistic situation with one advantageous factor (increased efficiency because of the current increase) and a detrimental one (higher NOCT).

Table 2 gives the estimated module yearly energy outputs at 4 different locations: Wuerzburg, Freiburg and Essen in Germany and Miami in USA. Essen was taken as a prototype place with an important part in diffuse light and moderate temperature, whereas Freiburg is a slightly warmer and sunnier location. Miami represents a more equatorial position with a higher irradiation level. We have indicated for the module Normal the yearly efficiency n and the energy

yield per installed nominal Wp of power. The yearly efficiency is lower in Miami because of the higher operating temperature of the module. The most important results are given in the last two columns (energy output for the modules Normal and AR), and in Fig.4, where the relative improvement of the yearly energy yield in % is displayed for the module with the AR layers when compared to the performance of the module Normal. For the module AR, the gain in

Miami is 3.4% which is around 0.2% lower than the sum of the gain in perpendicular illumination (2.65%) plus the integral of the extra-gain in current integrated between -90° and +90°. The higher component of diffuse light and the lower temperature in Essen are favourable for an extra energy yield gain for the AR module. An energy yield gain give a gain of 3.54 to 3.67% for the three German locations. Although the result is slightly lower in Miami with 3.4%, the absolute gain in produced kWh remains however the highest.

1.1

PV module

The PV module used in the tests is the Solon P160/6, which is produced by the company SOLON Photovoltaik GmbH. The module consists of 50 (5×10) stk. polycrystalline 6” (150×150 mm) cells, [6]. The total solar cell area is 1.125 m2. Figure 1 shows a picture of the PV module used for the experiments.

The total measures of the PV module are 793x1593x5 mm (width x height x thickness) without frame. According to the product information from the manufacturer the rated maximal power of the module, Pmax, moduie, is 160 Wp (±3%), normalized 142 W/m2 under standard test conditions. The efficiency of the cells is 14.2 %, whereas the module efficiency is 12.5%. The front glass is a 4 mm ESG white glass.

First a number of efficiency measurements were carried out with the PV module. Then the outer surface of the glass cover of the PV module was equipped with an antireflection treatment by SunArc A/S. Finally; a number of efficiency measurements were carried out with the PV module with the new glass surface.

Modelling of the QWSC is performed using the device simulator, Simwindows [8-9]. Simwindows incorporates interactive solvers for electrical, optical and thermal equation sets. The electrical model uses the drift diffusion equations, the two dimensional density of states in the quantum wells, thermoionic emission at the quantum well interfaces and thermoionic emission and tunnelling across abrubt interfaces between wells and bulk materials. This quantum well model is presently limited to the inclusion of only the lowest of each electron and hole energy levels, which are estimated from analytical expressions for a finite height well rather than a solution of Schrodinger’s equation [10].

The simulated control cell is AlGaAs p-i-n solar cell with material and device parameters listed in table(1).

Table (1) Material and device parameters of AlGaAs p-i-n control cell.

are increased in the active region within the intrinsic region with different barrier to well width ratio between 1 to 0.25 and spacer thickness between 100 and 300 A. In both cases the target is to optimize all parameters in order to maximize the cell efficiency.

Radial basis function (RBF) networks become very popular due to several important advantages over traditional Multi-Layer Perceptrons (MLP)[9]:

1. Locality of radial basis function feature extraction in hidden neurons, that allows usage of clustering algorithms and independent tuning of RBF network parameters.

2. Sufficiency of one layer of non-linear elements for establishing arbitrary input-output mapping

3. Solution of clustering problem can be performed independently from the weight in output layers

4. RBF network output in scarcely trained areas of input space is not random, but depends on the density of the pairs in training data set.

These proprieties lead to potentially quicker learning in comparison to multi-layer perceptrons trained by Back-Propagation (BP)[8]. In some extent, RBF networks allow us to actualize a classical idea about training layer by layer. Before beginning tracking operation using an adaptive neural network model (RBF-IIR), the unknown non-linear plan must be estimated according to the certain model. In this particular estimation process, the model consists of an adaptive neural network topology the Radial Basis function embedded in the hidden unites. In cascade with the network is a local IIR block structure as shown in figure 2. The IIR (Infinite Impulse Response Filter) synopsis network is used
to create double local network architecture that provides a computationally efficient method of training the system, and accordingly resulting in quick learning, and fast convergence. The approximated signal of the network v(«) can be modeled by:

yiky^a, 2{к-і)и(к)+^р, y(k-j)v(k) )

(=0 j= i

Where M оми a ше ню iiuiiiuei ui itstsu-iuiward delays and coefficient of the HR filter, respectively, N and bj are the number of feedback and delays and recursive filter coefficients, respectively. The u, v are the input and the co-input to the model at example k, respectively. Input v(k) is usually kept small for the feed back stability purpose, b is the bias value. The RBF network (Fig. 2) have a same structure as the MLP having only one hidden layer, the RBF is applied to the hidden layer [8] it is chosen as being Gaussian defined by its average ‘m’ and its a2 variance, the output layer can be linear or non-linear function. The determination of the network parameters has the same procedure as the MLP, it is also a universal approximator. That is to say a vector u having ‘i’ components Xj formed the input layer of the RBF and that is to say a hidden layer contained ‘h’ neurons and output layer, the output layer is given by:

Where my is the vector average of the hidden neurone ‘i’, and which the element mij is the weight between input ‘j’ and the hidden neurone ‘i’ . Ы2 is the variance from the hidden neurone ‘i’ and wy is the synaptic weights. Determination of the parameters my, ai, and wi is done by using the BP algorithm. The neural network parameters wi, mi, y, ai, ai and bi can be optimized in the LMS sense by minimizing the energy function, E over all example. Thus e(k)=y(k)-y(k). The energy function is defined by:

£.І2><Ц’ (3)

To minimize E we may use tne method of steepest descent which requires the gradients

,-£E-, 4е, 4е and 4е for updating the incremental changes to each particular

uwt omt, j uu i cat ub

parameter wi, y, a and b respectively. Gradient of E are:

UWi M

-@£—=—2и*Х. ejz’Qptj—/№,/11 )ai(uj—mj)

от,,] m

■4Е=-шУ’е72,|гу, — m,,j If )(uj — m,, )(uj — m,, )T

от, ti 11 "

ІЕ-=-%(к)еШк-]) %—’ZgikMklHk-j)

The incremental changes of each parameter are simply the negative of their gradients,

Aw^F, Am=-$E, Дст =_Ж

ow cm ca

Аа=-Щ and АЬ=-Щ-Thus each oa ob

coefficient vector w, m, a, a and b of the network is update in accordance with the rule:

wij(k+1)=wt, j(k)+^.wAwt, j

m, j(k+1 )=mt, j(k)+2p. mAmt, j a t(k+1 )=at(k)+poAat, j at(k+1)=at(k)+paAat bt(k+1)=fo(k)+pbAfo

Where the subscripted p-values are fixed learning rate parameters.

■Kn)

v

Let y(n) and y(n) represent the current and predicted samples, respectively. We evaluate to estimate y(n) by a function g(y(n)

y(n-1), y(n-2),…………………. y(n-k)) so as to minimize the mean square error (MSE).

E{[y(n)-g(y(n-1), y(n-2), y(n-3,……… y(n-k))]2}, where E is the expectation operator and k is

the order of the predictor the optimal non-linear prediction of the sample y(n) given the k previous sample y(n-1), y(n-2),….,y(n-k) is given by the conditional expectation [2].

y(n) = [y(n) 1 y(n -1). y(n — 2),……………. , y(n — k)] (9)

The conditional probabilities needed in Eq. (9) are not a priori. Sub-optimal solutions could be found using a suitable model for the source signal (data), and then computing the joint conditional probabilities. However, the lack of precise modeling precludes any useful solution by the direct use of Eq. (9). A neural network predictor does not minimize the conditional expectation directly, however, it does approximate this minimization to any desired precision [2]. In modeling the number of input and output neurons is determined by each application. However, the number of hidden layers and the number neurons within each layer must be adjusted during the learning phase so that the network can be trained efficiently. While, in theory, one hidden layer suffices for approximating a general input — output mapping, in practice however, the number of hidden layers and neurons is specific to the problem at hand. A constructed neural network must be trained to learn or model relationship between given input-out put data or measurement. Usually, a portion of the available data is used for training set. To be effective, the resulting model must be able to generalize, i. e must produce accurate results a new samples outside of the available training set. The performance of the network in processing the test set is used a measure of the network’s generalization capability and can thus be used on line to stop the training process. Figure 3 shows the diagram block of developed model. In this study from 365 patterns, set of 200 patterns was used for training of the network and 165 were used as for testing and validation for our model for each signal.

Results

Once a satisfactory degree of input-output mapping has been reached, the RBF-IIR network training is frozen and the set of completely is an unknown test data that was applied for validation. After simulation of many different structures, we found that the best performance is obtained with a one hidden layer with 4 neurons. The best model is validated by comparison between predicted results and actual values for different signals that are shown in figure 4. We observe that there is almost a complete agreement between the two series for different signals.

Tablel, displays the statistical features (mean, variance, correlation coefficient and RE) between the calculated data and those predicted by our model, it is found that there is no significant difference between the predicted and the measured parameter from the statistical features point of view. The correlation coefficient obtained for the validation data set is varied between 82 to 99%. In this respect, the closer to unity these values are the better the prediction accuracy. In order to illustrate the advantage of this model we made a comparison between different neural network structures like Multi-layer perceptron (MLP), Radial Basis Function (RBF), Recurrent Neural Network (RNN), Modular Network (MN). Therefore we have trying to finding the temperature model of the battery the results obtained is listed in table 2. According to this table one notices that this model have less low time calculation compared to the other neural network structures and also gives good results, which are approaches with the real results.

Conclusion

The main of this work is to train the RBF-IIR model to learn the prediction and modeling of the signals from stand-alone PV system. Once trained, the RBF-IIR estimates these signals faster. The validation of the model was performed with unknown signals data, which the network has not seen before. The ability of the network to make acceptable predictions even in an unusual day is an advantage of the present method. The estimation with correlation coefficient varied between 82 to 99 % was obtained. This accuracy is well within the acceptable level used by design engineers. The advantage of this model is to predict of different signal coming from the stand-alone prediction signals allow to analyzing and studying the performance of the PV systems and the sizing of PV system. Also this model have been compared between different neural networks structures, and given good results.

References

[1] S. A.Rizvi, L. Wang and N. M Nasrabad, "Neural network architectures for vector prediction”, Proc. of the IEEE, 84, pp 1513-1528, 1996

[2] A. Papouli, Prpbability random variables and stochastic processes. McGraw-Hill New York, 1983.

[3] U. Amato, A. Andretta, B. Bartolli, B. coluzzi, V. Cuomo, F. Fontana, and C. Serio, "Markov process and Fourier analysis as a tool to describe and simulate solar irradiation”, Solar Energy, 37, pp197-2011986.

[4] M. Benghanem et A. Maafi, "Ajustement des donnees d’insolation d’Alger par un

modele autoregressive", ueme journees tunisiennes d’electrotechnique et

d’automatique, pp 53-56 , 1997.

[5] S. Liu and S. Chang. "Dimension estimation of discrete-time fractional Brownian motion with applications to image texture classification", IEEE trans. On Image Processing, Vol 6 ,N 8, pp 1176-1184, 1997.

[6] A. Guessoum, S. Boubkeur and A. Maafi, A global irradiation model using radial basis function neural network. WREC, UK, pp 332-336. 1998

[7] Mellit, A., Benghanem, M., Hadj Arab, A., and Guessoum, A. Modeling of sizing the photovoltaic system parameters using artificial neural network. Proc. of IEEE, Conference on Control Application, Vol 1, pp 353-357. 2003.

[8] S. Haykin, "Neural Networks, A comprehensive foundation". MacMillan, New York 1994.

[9] Leonard, J. A., Kramer, M. A., and Ungar, L. H, . "Using radial basis function to approximate the function and its error bounds". IEEE Trans. In Neural Networks. 4, pp 624-627.1992.