Four significative days of the year corresponding to the two equinoxes, the 21st of December and the 21st of June, and the two solstices, the 21st of March and the 21st of September, have been taken into account. For each day simulations have been performed every two hours, beginning from 8:00 a. m. (library opening time) till 18:00 p. m. (closing time). Each simulation has been carried out both under uniformly overcast sky conditions and under clear sky ones with sun.

The preliminary analysis has concerned the current state. In this case the study module corresponds to a lecture hall, the light sources are essentially represented by the lantern and the portals on the west facade.

The lighting analysis of the design proposal has concerned the modified state. In this case the module corresponds to one of the two side bays of the lecture hall, the large glass facade towards the medieval walls is added to the light sources of the current state. In particular, the lighting analysis has developed in three consecutive stages in order to understand the variations induced on the illuminance distribution by the most significative recovery interventions on the existing pavilion. The first stage corresponds to the introduction of the glass facade towards the medieval walls; the second corresponds to the introduction of the mezzanine floor, at a height of +3,40 m, and the third to the introduction of a white screen placed below the lantern base at a height of +7,50 m. In this paper the third stage, representative of the definitive design solution, will be analyzed in detail.

The screen has a double function: shielding the sun rays entering into the room through the lantern southern glass window during the central hours of the day, that is, when the sun is high above the horizon, and, on the other hand, making them filter when the sun is low (first hours of the morning and late afternoon); avoiding the dazzlement phenomena which, in absence of screen, could occur for the direct sight of a sky portion from the mezzanine floor. The screen can be realised using a white cloth, can slide, twisting around itself, by runners; according to the requirements, it can be stretched out from a room side to another by means of a remote-controlled system or twisted around itself so as to annul its function. The screen has been schematized with a material having a reflection coefficients of 0.75.

On the work plane placed at a height of +4,20 m, corresponding to the reading desks on the mezzanine floor, the presence of the white screen below the lantern is perceivable owing to the optimal illuminance uniformity both under overcast sky conditions and under clear sky ones; the values of the uniformity ratio approach the optimal value 0.80, advised by the rules [8].

Under overcast sky conditions (see Figs. 5a-d) the illuminance distribution shows a locus of maximum values coinciding with two straight lines parallel to the axis of the lantern (east — west axis) and symmetrical to the same axis, on which the minimum values are noticed, owing to the shaded area created by the screen below itself. The tridimensional graph points out a slight "zoom”, proceeding from west to east, evidencing the influence of the glass facade towards the medieval walls on the mezzanine floor lighting. The presence of the screen remarkably reduces the illuminance values. In this case it seems to be to shut off the screen in order to allow the upper lecture hall to be more lighted.

Under clear sky conditions with sun (see Figs. 6a-e), the screen fully performs its function: during the first hours of the morning and the late afternoon the screen makes the sun rays
filter. On the contrary, during the midday hours, the sun rays are intercepted by the screen, reflected by it on the vault intrados and by the vault itself on the mezzanine floor, so as to generate an illuminance distribution analogous to the overcast sky case. In that way an increase, even slight, in the uniformity ratio U can be observed.

The Figs. 5 and 6 show the trend of the mean illuminance values all the year round under overcast and clear sky conditions. Under overcast sky, in each of the four simulated days, the trend turns out to be parabolic with the peak connected with 12:00 a. m. (values variable within a range from 500 to 1200 lux). It is interesting to observe that, inside the room, the mean illuminance falls below the minimum value recommended by the rules [8] only before 9:00 a. m. and after 15:00 p. m. of the 21st of December and after 17:00 p. m. on the 21st of March and September. Under clear sky, it is not possible to recognize an univocal trend in the four examined cases, but the luminance maximum value can be observed, except for the 21st of December, still compared to 8:00 a. m. Moreover, the recorded values exceed the 500 lux provided for by the rules [8].

On the reference plane placed at a height of + 0,80 m, under overcast sky conditions, the luminance trend does not vary qualitatively either during the year or during the day (see Fig. 7); the distribution of the isolux curves is characterised by two illuminance peaks compared to the portals on the west facade, by maximum values on the east side compared to the glass window, and by a low illuminance value site in the central zone below the mezzanine floor. The tridimensional graph points out the graph "zoom” going from west to east and its deflection in the central section, due to the mezzanine floor section. Since this standard distribution can be find out all the year round, a generalisation of such a trend can be made individualizing three areas with different lighting characteristics: an "east area” behind the large glass facade, characterized by high illuminance values, a "central area” below the mezzanine floor, where the minimum values are recorded and, finally, a "west area” along the main fagade, characterized by the illuminance peaks of the portals.

Under clear sky conditions with sun (see Figs. 8a-e) the illuminance trend on the same horizontal plane is qualitatively analogous to that one found out under overcast sky conditions. It is, therefore, possible even in this case, to recognize the three above-identified areas. However, it is possibile to see, by the slight changes of the illuminance distribution, the variation of the sun position during the day. Following the sequence of simulations relating to a reference day (21st March), carried out every two hours, it is interesting to observe how the east area (towards the medieval walls), is characterized by good illuminance values the whole day, since it receives light both from the glass facade and the roofing lantern, while the west area shows good illuminance values only during the afternoon, since it receives light almost exclusively from the portals; during the morning, in fact, the maximum values on the west side are only just traced, since the sun is positioned on the south-east sun-dial.

The analysis of the graphic and numerical results of the simulations have led to revise the initial assumption on the disposition of the furnishings proposed during the preliminary design stage. The new solution provides for the following: at the ground floor, in accordance with the previously mentioned subdivision into the three areas, a central zone below the mezzanine floor designed for the shelves, characterized by quite low illuminance values, but anyway, higher than the minimum value provided for by the rules [8]; two side zones, one protruding from the walls and the other on the main fagade, designed for study, reading and consultation, characterized by higher illuminance values. The upper gallery (mezzanine), characterized by an optimal illuminance uniformity is expected to be exclusively designed for reading and consultation, limiting the shelves down along the perimeter.

SHAPE * MERGEFORMAT

Fig. 7 — Overcast sky conditions. Illuminance levels on the work plane at +0.80m (ground level), 3D isolux curves on the 21st of March at 12:00a. m.

Conclusions

Daylight requires, owing to its extreme variability during the day as well as in the course of the year, a more accurate lighting analysis than the one necessary in the case of artificial lighting, generally considered "static”. The use of software tools allows overcoming the modelling difficulties and simulating the contributions due to daylighting in order to exploit it in the best way, reducing the recourse to the artificial one.

By means of the software ADELINE a spatially and functionally significative module has been represented, being part of the reading room of the new university library designed inside the Marzotto Scientific Pole. The lighting design has been composed of a preliminary analysis of the current state and of an analysis of the changes introduced by various recovery interventions on the existing building: the replacement of the opaque building envelope towards the medieval walls with an entirely glazed facade; the realization of a mezzanine floor; the introduction of a white screen below the lantern to assure a higher illuminance uniformity on the mezzanine floor and to eliminate the dazzlement phenomena.

The results relating to the definitive design solution, showed in this paper, have allowed a thorough knowledge of the lighting conditions due to the daylight inside the studied room; this has allowed to evaluate and to revise the architectural choices made in the preliminary design stage in order to realise the best, or one of the best, disposition of the reading desks and of the shelves. In this sense the "lighting design” has showed a high feed-back level inside the design pathway.

For the lighting of the reading room, in addition to a general lighting system to which it will be necessary to resort in winter and in the late afternoon, a "punctual” lighting system has been provided for on the reading desks, to which each user can resort whenever he considers the quality of the lighted room insufficient. For the lighting of the shelves a system of light fittings, placed in the mezzanine floor intrados and able to generate a "light cascade” on the shelf allowing the book back to be read, is expected to be used.

Taking into account that the daylight contribution would be sufficient (with respect to the prescriptions of the rules), inside the reading room, for a long time interval of the day (on an average, from 8:00 a. m. to 17:00 p. m.), the adoption of an opening time of the structure being flexible depending on the daylight availability (variable during the year) could be a valuable proposal as for the reducing of energy consumption.

Acknoledgements

This research was supported by Italian Ministry of Education, University and Scientific Research (MIUR) and by University of Pisa within the National Relevant Interest Project (PRIN 2003-2005): "Energy and environmental diagnosis on existing buildings: research methodologies, determination of qualification parameters and technico-economic assessments”.

The effect on artificial lighting demand of light shelves, fixed and movable reflecting lamellas, prismatic lovers and laser cut transparent devices was simulated, taking into account the actual number of hours of clear and overcast sky in Beijing. The final solution is the result of an iterative process involving energy and architectural issues.

Artificial lighting will be based on high efficiency lamps and fittings, controlled by a dimming system capable to adjust the lamps power to the actual local lighting needs, in combination with the natural light contribution. The geometrical positions of the lamps are optimised too. A presence control system will switch off lights in empty rooms. The integration of the envelope components chosen and the controls systems will reduce by several times the electric energy consumption for lighting.

Reflecting and semi-reflecting lamellas and louvers will also allow for sunshine to penetrate in the rooms in winter and to be rejected in summer, reducing the energy consumption of the building.

The wings of the building are covered by a light canopy made of louvers protecting the terraces from the direct sun in summer and allowing the penetration of natural light in winter.

Future work will put emphasis on the following fields:

• Continuation of the system monitoring (first and second generation prototype) in order to implement and test the developed control algorithms.

• Technical redesign of the storage modules.

• Production of a second generation prototype module.

• Development of stable modified sorption materials.

The newly developed second generation prototype needs to be extensively tested, analysed and evaluated under laboratory and practical conditions as input for the finalisation of the product development and identification of the most attractive market.

Acknowledgements

We are grateful to our project partners from Fraunhofer Institute for Solar Energy Systems (Freiburg/Breisgau, Germany) and Sortech AG (Freiburg/Breisgau, Germany) for their valuable work which contributed to the success of the project. We thank the European Commission and the Austrian Federal Ministry of Transport, Innovation and Technology (BMVIT) for financial support of the projects HYDES and MODESTORE.

References

• W. Mittelbach: Final Report HYDES Project. Final Report of a Joule III Project, European Comission 2001

• Tomas Nunez, Hans-Martin Henning, Walter Mittelbach: High Energy Density Heat Storage System — Achievements and Future Work, ISES Solar World Congress 2003, Goteborg, Sweden

measurements

M. J. Jimenez; B. Porcar; M. R. Heras

Department of Renewable Energies, "Bioclimatic Architecture Research Programme”, CIEMAT, Madrid, E-28040, Spain; Tel:+34 91 3466305,fax: +34 91 3466037, e-mail:

miose. iimenez@psa. es

Introduction

The use of passive design strategies to benefit from the solar energy in order to reduce heating and cooling loads in buildings but maintaining indoors comfort is a very important concern regarding environmental aspects.

The application of these design strategies requires an accurate knowledge on the parameters that characterise the thermal behaviour of each building component used, such as the heat transmission coefficient, U, and the total solar energy transmittance, g, values and time constants. These parameters are required to apply the mandatory national standards and regulations that impose the minimum requirements regarding energy consumption in buildings (eg. The Spanish NBE-CT- 79). The availability of these parameters for each component included in the building envelop is also important because they are required as input for most of the simulation programs which estimate heating and cooling loads and evaluate the thermal behaviour of buildings that include these components in its envelop.

In both cases, the accuracy in the final results depends on the accuracy on the estimation of the parameters of each individual component. It is possible to estimate these parameters, according to national standards, from tabulated values of the integrating parts of the component, but in this case their accuracy depends on the degree of knowledge of the composition of the enclosure. Outdoors testing (Vandaele, 1994) and in-situ measurements (ISO 9869:1994) are very useful and reliable alternative to provide more accurate estimations of these parameters.

This paper presents the experimental determination of the U value of each wall of a quite simple mono-zone building, located at the Plataforma Solar de Almerfa (Tabernas, Almerfa, Spain), by in-situ measurements and dynamical analysis. These values have been compared with those obtained from tabulated values according to national standards. Comparisons have been considered in order to obtain information on the predictable degree of agreement between both approaches.

The studied building is considered especially suitable for analysis as far as its simplicity and the high degree of knowledge about its construction and components allow undertaking relatively good estimations from tabulated values according to national standards in comparison to the usual quality of these calculations from tabulated values. Its performance is also interesting because it includes some passive strategies to reduce heating and cooling loads and to improve thermal comfort.

For the purposes of this study two different training algorithms were used. The first is the standard back-propagation (BP) algorithm. The second is the Lavenberg Marquardt (LM) algorithm. Mathematically, back-propagation is gradient descent of the mean-square error as a function of the weights. The mean square error of the network training process can be calculated as follows:

MSE = — T)2/L

LM optimization technique is a more sophisticated method than gradient descent. It is based on Gauss-Newton method, and it is very powerful and fast. The difference between these two algorithms is that for the back-propagation the weights are updated for each input, while for the LM algorithm, all inputs are presented to the network at the same time on each epoch. Many tests have been also conducted in order to identify the training patterns that yield to the best results. In general, too large training data sets contain obsolete data and require high

training times. Furthermore, training of the ANN to a very small error may result in data overfitting, i. e. the neural network loses its ability to generalize.

The training data set that was chosen consists of two months, that is, if the forecast day is 1 August, the training data set consists of input-output pair created from 1 June to 31 July.

1.2 Results

The main objective of this study is to investigate the capability of using neural networks to estimate the next day hourly cooling load profile, without knowing in advance the weather conditions. In order to use neural networks, the simulation algorithm TRNSYS was initially used to generate a database for three years. This database where divided into two sets. The first set consisted of the data from the two first years and was used for training and testing the ANN. The second set consisted of the third year data and was used for testing the trained networks, in order to evaluate ANN performance with a data set that the networks have never seen before.

Figures 4 and 5 gives the comparison for five representative days and for all considered buildings of the ANN predictions and the actual values of the hourly cooling load (calculated by using TRNSYS). It is clear that there is a good agreement between prediction and actual values. Also, concerning the whole cooling period, Figure 6 gives the frequency distribution of the hourly absolute percentage forecast error. According to this figure the prediction of the cooling load is enough accurate giving an error of 1% in 60-80% of the compared values, while 6-7% error is presented in less than 1% of the values.

Fig 6. Absolute Percentage forecast error frequency distribution

Further analysis of the correlation between actual values and ANN predictions were performed. Figure 7 shows another plot of actual and predicted values for all houses. Ideally the data should fall on a line of slope 1.0. Linear least square fit through the data show the slope of the best-fit line is 1.0009 while the R-square value (R2) is 0.9958 (building: House1). The same analysis was performed for each hour and the results are listed in Tables 3 and 4.

hour 12:00 13:00 14:00 15:00 16:00 17:00 18:00 18:00 19:00 20:00

R2 0.958 0.958 0.876 0.964 0.969 0.941 0.937 0.902 0.905 0.858

Table 4: Average R2 for hourly cooling load predictions

2. Conclusions

Feed Forward ANN were designed and trained to investigate the application of this technique to forecast cooling load for residential buildings within a prediction horizon of 24 hours. To evaluate ‘real’ performance of the neural networks all the tests were applied to a data set that was not used during the training phase and the networks were found to perform with accuracy of the order of 0.98 in predicting the inputs samples correctly. The small differences between the reported error in this study are due to the use of the same ANN architecture for all buildings, which is important when developing a general methodology. The optimum Feed Forward ANN, that gives the minimum error for each case, differs slightly between different buildings. In any case this is not a restriction as the accuracy of the presented methodology is satisfactory.

The results clearly show that ANNs can be used to predict cooling loads in residential buildings with reasonable accuracy. Due to the lack of experimental data, this study was mainly focused on investigating the suitability of ANN as a prediction tool in residential

buildings. In future, we intend to train ANN on experimental data for better and realistic performance of the network.

In this article a new building envelope component was presented. It is a very efficient insulated glass associated with a solar collector.

The glass collector was described. Its main characteristics were quantified. Its benefits have been evaluated by simulation. The performances quantified in this article showed the interest of the glass collector as thermal insulation, variable solar control, benefits on sun and daylight distribution and as a solar collector.

Over the present validation tests, we plan to go further with reference buildings integration. Acknowledgment

The work presented here was supported by the Region Alsace, France.

Nomenclature

a0 Intercept efficiency, [-]

a1.ext First-order coefficient of the efficiency curve for outside, [W/m2°C]

a1.int First-order coefficient of the efficiency curve for inside [W/m2°C]

a2 Second-order coefficient of the efficiency curve [W/m2°C2]

e Thickness, [m]

g Solar factor [%]

hiie Global heat exchange coefficient, [W/m2K]

Nu Nusselt number

It Incident radiation, [W/m2]

Ті; Tj Temperature of the two surfaces (i, j respectively), [K]

Q Solar flux, [W]

S Glass collector area, [m2]

Greek Symbols

ai Absorber area vs. Glass Collector area, [-]

a2 Transparent Glass Collector area vs. total Glass Collector area, [-]

eij Emissivity of i an j surface respectively [-]

П Overall collector efficiency, [-]

a Stefan Boltzman Constant = 5.67 10-8 [W/rnFK4]

t Transmittance, [-]

9a Outdoor ambient temperature, [°C]

9c Solar collector temperature, [°C]

9i Indoor ambient temperature, [°C]

9m Fluid mean temperature, [°C]

9v Inside glass temperature, [°C]

The selection of the most appropiated plants is an important question. The aspects that are being considered in the project are : preferent growing directions, speed of growing, maximum size and structural loads, selection of suitable decidous/evergreen species (the periods without leaves must be selected according to the heating/cooling demands of the
buildings), volume of earth substrate, attraction of animals (birds, insects), low toxicity, main­tenance works needed and their frequency and commercial availability.

It has been shown before that microstructures are suitable for the use as light-shading devices. Also the technical feasibility was proved and prototypes with structured surfaces with periods of approx. 20 pm were generated and replicated in transparent plastic up to a size of 375 x 375 mm[28] [29] [30] [31] [32]. The structures have been coated successfully, both all over and face-selectively. It has been shown that they add a new free design parameter to the conception of microstructured sun shading devices.

Switchable coatings have been attached successfully, too. This offers a wide range of concepts for new controllable light guiding devices like the example shown above. In further research, several of these concepts will be investigated and optimised. Also other ways to combine microstructured surfaces with an optical switching e. g. the combination with thermotropic layers will be analysed

Acknowledgements

The authors wish to acknowledge the contributions of C. Buhler, B. Blasi (Fraunhofer ISE) and of J. Mick (of IMTEK, Univ. of Freiburg, Germany) to the work presented here.

The samples build with a ultra precision milling cutter (see figure 8) where manufactured at the Institut for Microsystem Technology (IMTEK), A.-L.-University Freiburg by J. Mick and C. Muller

The work presented here was funded by the German "Bundesministerium fur Wirtschaft und Arbeit BMWA" under reference number 0327312 A and B. The authors assume the responsibility for the contents of this paper.

The design of the two double fagade sections is based on an earlier simulation study by Charron and Athienitis [4]. The double fagades in the outdoor test-room were fully instrumented with thermocouples attached at several points on the cavity surfaces, hot wire anemometers for velocity measurement, pressure sensors and a weather station. Note that the back panel in each fagade consists of polystyrene enclosed on both sides with 1 cm thick plywood panels; the total RSI value is equal to 1 and it has negligible heat storage. The heat transfer through the panel to the room was negligible compared to other energy flows.

Figures 2 and 3 below compare results from the two fagades for January 26, 2004. This day was cold and clear with negligible wind. The data were collected every minute and averaged for about half hour from 11:20 am to 11:50 am. Quasi-steady state conditions existed with no major (not more than 5%) change of any of the parameters measured.

As can be seen from Figure 3, much higher thermal efficiencies are obtained when we have air flow on both sides of the PV than when the PV panel is exposed. For the case of Figure 2, the electricity generated was 87 W and the thermal energy (heating of air) 337 W without taking into account the significant heating at the inlet; as can be seen from Fig. 2 the air at the inlet is heated about 3 °C and this effect may also be due to some air leakage from the room into the cavity. The resulting electrical efficiency was about 10% and the thermal efficiency 37% for a total efficiency of 47%.

By comparison, in the configuration of Figure 3, the electrical efficiency of the PV was only 6% (but it was not at its maximum power point) and its thermal efficiency was 65% for a total of 71%. One disadvantage of the double cavity configuration of Fig.3 is that the PV operates at a higher temperature — a maximum temperature of 40.7°C when the outside temperature is -17 °C. The temperature of the air exiting the cavities may be controlled by varying the flow rate and mixing with indoor air.

Studies of the temperature and velocity profiles across the horizontal (air gap) were also performed. Figure 4 shows a typical inlet and outlet temperature profile corresponding to the measurements in Fig. 2. The velocity profiles that were used to compute the average velocity (equal to 0.6 m/s) showed a buoyancy-induced peak near the PV followed by flat region in the middle. The flow is complex, definitely a mix of natural and forced (fan — induced) convection, laminar at the inlet and turbulent near the top of the cavity.

In the course of experiments, it has been obtained that the pressure of desorbed hydrogen molecules is approximately by 21 times as much as the calculated pressure of moist air being supplied through the microleak.

So a multiplicative effect [6,7] is realised on the semiconductor surface of the superinsulation. It should be noted that similar multiplicative phenomena have been observed
in the work [41] where the argon desorption stimulation has been investigated by means of the addition of oxygen. The relative enhancement of the desorption in the presence of oxygen has been increased approximately by 20 times as compared with pure argon crystals.