Assessing the nearly zero-energy building gap in university campuses with a feature extraction methodology applied to a case study in Spain

Selection of a case study

For a thorough description of the methodology, the quality and amount of energy data available from university buildings should be known. In this context, the University of Lleida (UdL) has been selected in this work for several reasons. First, the access to a recent energy building data is facilitated as most of the authors are professors and/or researchers of the same university making communication with the responsible university energy managers easier. Second, the climate of Lleida (BSk for Köppen-Geiger classification) is a dry semiarid climate and has more extreme winters and summers requiring more cooling and heating compared to other Mediterranean cities. Third, the UdL represents an average-size university in the Spanish system, and fourh it has been pioneer in Spain and Catalonia in installing PV arrays in several buildings, so the required presence of renewable systems for NZEB is accomplished.

Compilation of building data

Collection of energy use and self-generation data

Gas and electricity consumption have been compiled with the maximum time resolution available. In the case of gas, monthly bills for the last 7 years (from 2010 to 2016) are available. For electricity, a new monitoring system enables us to get energy readings every 15 min for the last 2 years (2015 and 2016). Power meters are installed in each building and the data loggers of each building send the information via RS-485 to campus concentrators, which in turn transmit all data to a general server in the cloud. This server can be accessed through an online platform. This software is called DEXCell Energy Manager, of DEXMA Company. The same system is planned to be installed for the gas metres by 2019. The renewable energy production is also monitored every hour since the year 2010 and has been collected for analysis. All the existing PV installations consists of a total of five polycrystalline PV systems located in two different campuses, 2 in Campus 1, with a peak nominal power of 96.6 kW (on E1 roof) and 95.9 kW (on E4 roof); and the other 3 in Campus 3, with peak nominal powers of 79.2 kW (on roofs of E9, E10 and E11), 47.95 kW (on E12 roof) and 95.9 kW (on roofs of E13 and E14). All the PV modules are mounted on flat roofs, with inclined supports. They are non-tracking modules, oriented to the South, in the range 135°–225°, and with tilt angles between 20 and 25 °C, depending on the building. The total area of PV modules installed is of 3029.1 m² with a total installed power of 416.3 kWp. To be able to assign a particular annual PV energy generation to the buildings that share the same installation, weighting factors based on installed peak power on each roof are applied. These factors are 20, 40 and 40% for roofs E9, E10 and E11, and 83 and 17% for E13 and E14, respectively.

Overall and detailed analysis of energy data

Monthly analysis of gas data

To achieve a better understanding of the use of gas for heating and other possible uses such as domestic hot water production, performance lines for every building are generated using the degree-days theory [23]. These performance lines are useful for measuring savings derived from energy conservation retrofits and to identify and correct operational and maintenance problems [24]. The estimated gas consumption for a heating period can be calculated as (Eq. 1):

$$F = \frac{{24 \cdot {\text{HLC}} \cdot \left( {\text{HDD}} \right)}}{\eta } + B,$$

(1)

where F is the fuel consumption for the period (kWh), η is the overall heating system efficiency for that period, HLC is the overall building heat loss coefficient (kW/K), which includes the heat conduction losses, the air infiltration and the ventilation losses, and B is a constant value that corresponds to other possible non-weather dependant gas consumptions. The degree-days for the period, HDD, are found as the sum of the daily degree days, D_d, over the considered billing period (about a month) using the most rigorous method, with exterior hourly temperature data. The daily degree days can be expressed according to (Eq. 2):

$$D_{\text{d}} = \frac{{\sum\nolimits_{1}^{24} {\left( {T_{\text{hb}} - T_{\text{ext,i}} } \right)_{{\left( {\left( {T_{\text{hb}} - T_{\text{ext,i}} } \right)\; > \;0} \right)}} } }}{24},$$

(2)

where T_hb is the heating base temperature and T_ext is the outdoor air dry-bulb temperature. For determining the value of T_hb in each building, several guess values for every 0.5 °C are used, in the range from 12 to 19 °C, and the base temperature yielding to the best fitting (best R²) is selected.

As the billing periods for gas are not always regular in the number of days between gas readings, for each period, the gas consumption per day has been correlated with the HDD per day, by dividing both the gas readings and the HDD by the exact number of days of that particular billing period [25]. Thus, for each building, the slope (m) and the intercept (b) of a regression line (Eq. 3) is determined:

$$\left( {\frac{F}{n \cdot A}} \right) = m \cdot \left( {\frac{\text{HDD}}{n}} \right) + b,$$

(3)

where n is the number of days for each billing period and A is the gross floor area (m²).

Monthly gas consumptions are also divided by the gross floor area of each building for the purpose of comparison. So, the values of m and b correspond to the following physical interpretation (Eq. 4):

$$m = \frac{{24 \cdot {\text{HLC}}}}{\eta \cdot A}; \cdot b = \frac{B}{n \cdot A}.$$

(4)

A summary of the procedure followed to determine the heat loss coefficient, HLC, the baseload gas consumption, B, and the heating base temperature, T_hb is shown in the flowchart of Fig. 2:

Open image in new window

Most of the readings for gas consumption for the 7 years of data are actual values, according to the statement of the gas distributing company in the bills. The ones which are estimated have been removed to avoid possible outliers with no physical meaning.

The comparison of the values of the slopes, m, of the different buildings will be used to rank them with respect to the level of building heating efficiency. The larger is the slope, the higher the dependence (less efficient building) on the external temperature.

NZEB gap and proposals for energy improvements

Annual energy overview

Absolute annual electricity use values for the twenty university buildings studied in 2015 and 2016 are shown in Fig. 3a. They are ordered from largest to smallest figures starting from the left, for each campus. The three main consumers are clearly highlighted, with electricity uses above the 1000 MWh/year. The rectorate building, E8, is comparatively larger than the average size of the buildings (21,000 vs 5250 m²), so this may explain this extra energy. On the other hand, E19 and E20 buildings have similar or even smaller gross floor areas than the average. In this case, other reasons, such as the extra power required for animal labs and other bio-research equipment, may explain these large consumptions results. The rest of the buildings are in the range 50–600 MWh/year, with very small differences between the 2 years studied. The exception is building 19, which has doubled its consumption, probably due to the opening of new, high energy intense research labs. The historical evolution of gas use at the UdL is shown in Fig. 3b. Note that two buildings, E2 and E15, do not have gas use, as their heating needs are covered by an electrically driven heat pump (see Table 1). Again, E8 building is the highest consumer, probably due to its size, followed by the same Health Science campus buildings found in the electricity analysis, E20 and E19. Average gas use in 2016 is around 320 MWh/year, being all the buildings in the range 150–1000 MWh/year. In general, a decreasing gas consumption trend is observed until year 2014, whereas an increasing trend is detected in the last 2 years in many buildings. Whether or not these ups and downs in gas annual use are due to different climatic severities along these years is illustrated in Fig. 4. The historical evolution of gas consumption for each campus is presented in a weather normalised way, dividing the annual MWh per the calculated heating degree days of Lleida each year. Campus 1 and Campus 4 have a slight increase of gas use along the years, with a pronounced increase in years 2015 and 2016. On the other hand, campuses 2 and 3 show a decreasing trend in the first 4 years, followed by an increasing trend in the last 3 years, again with a peak in year 2016. Unfortunately, this weather normalised behaviour indicates a deterioration in the envelope transmission and infiltration heat losses and/or a decrease in the campuses gas heating efficiency in the last years, which should be further analysed and reversed, if the path for sustainability and NZEB targets is to be followed.

Open image in new window

Open image in new window

Figure 5 shows the historical evolution of the annual renewable energy production in the two campuses with installed PV systems. Most of the PV systems started operation in late 2010 and some in January 2011. For that reason, the energy generation values of 2010 are very low and have not been included in the linear regression. Note that the two campuses have a decreasing trend in PV output production, with about 3% reduction per year. This observed efficiency drop per year is higher than the typical 1% drop given by many PV manufacturers and should be studied in more detail. The oscillating, rather declining, annual solar radiation values during this 6-year period (Fig. 5) cannot explain this behaviour. Thus, the deterioration of the PV cells is the most likely reason.

Open image in new window

Figure 6 shows annual aggregated gas and electricity use per gross floor area for all the buildings. With this normalisation per floor area, the two Health Sciences buildings are highlighted as the major consumers even more than in the absolute comparison shown before. On the other hand, the other big consumer, the E8 building, moves down several positions, with values 20% smaller than the average, which is 140 kWh/m² year. Excluding the two high energy consuming buildings E20 and E19, with values above the 300 kWh/m² year, the rest of the UdL buildings are in the range 50–175 kWh/m² year, comparable to the values reported by the Polytechnic University of Barcelona (UPC) for up to sixty buildings in different campuses, which are between 40 and 200 kWh/m² year [19]. Another study for eleven buildings in a university campus in Korea shows energy use intensities for gas and electricity in the range 106–399 kWh/m² year [17]. Important differences are observed between the larger and smaller consumers in each campus. These are 60, 67 and 80% for campuses 1, 3 and 4, respectively.

Open image in new window

If the same comparison is made by the number of users per campus (Table 1) as the normalisation parameter, results are different (Fig. 7). The energy use gap between Campus 4 and the rest of the campuses is emphasised, as the number of estimated users is relatively small, compared to Campus 1 and 3, and absolute energy consumption is higher. Based on the number of users, which is only an estimate of maximum occupation given by the UdL self-protection and evacuation plan, the annual energy use per user of campus 1, 2, and 3 is between 500 and 1100 kWh/user/year, whereas Campus 4 achieves almost 1600 kWh/user/year. Campus 1 is the second energy consumer in absolute terms, but the high estimates of users, above 7000 people, moves it to the last position in energy per user, with a 67% reduction compared to Campus 4.

Open image in new window

Monthly analysis of gas data

Figure 8 shows the heating performance lines for two buildings with extreme slope values in the Cappont campus (Campus 1). Similar curves are obtained for the rest of the buildings (not shown). The results of these linear regression models are summarised in Table 3. Buildings E2 and E15 are not included in the analysis because they have electricity driven systems for both space heating and cooling. The availability of 7 years of monthly bills has increased significantly the number of data points with respect to a typical performance line with gas readings for only 1 year. For this reason the coefficient of determination, R², for the regression lines is in general quite high, above 0.8 in most of the buildings. The only exception is E13 building, where the goodness of fit is rather low (R² = 0.50%), suggesting that many gas monthly bills are not actual readings, but company estimates. Standard deviations of the slopes, m, for the linear regression lines obtained are in the range 3–6%. Thus, the observed almost threefold differences among extreme building performance lines cases are statistically significant.

Open image in new window

Table 3

Parameters for heating performance lines from gas monthly data

Campus	Building ID	Base temp., Thb (°C)	Slope, m (kWh/m2 K day)	UH = HLC/A (W/m2 K)	Intercept, b (kWh/day m2)	R 2
Campus 1	E1	16.5	0.021 ± 0.002	0.613 ± 0.06	− 0.011	0.877
	E3	18.0	0.051 ± 0.005	1.750 ± 0.15	− 0.045	0.890
	E4	14.5	0.037 ± 0.003	1.079 ± 0.09	− 0.004	0.879
	E5	15.5	0.052 ± 0.004	1.517 ± 0.12	− 0.016	0.900
	E6	17.5	0.032 ± 0.004	0.933 ± 0.12	− 0.011	0.797
	E7	14.5	0.029 ± 0.002	0.788 ± 0.06	− 0.006	0.879
Campus 2	E8	17.0	0.029 ± 0.002	0.846 ± 0.06	− 0.016	0.887
Campus 3	E9	16.0	0.036 ± 0.003	1.050 ± 0.09	0.022	0.854
	E10	18.0	0.033 ± 0.003	0.963 ± 0.09	− 0.004	0.842
	E11	17.5	0.033 ± 0.004	0.963 ± 0.12	− 0.003	0.770
	E12	18.5	0.028 ± 0.003	0.817 ± 0.09	− 0.022	0.835
	E13	18.0	0.014 ± 0.003	0.408 ± 0.09	0.019	0.499
	E14	19.0	0.032 ± 0.005	0.933 ± 0.15	− 0.003	0.824
	E16	17.0	0.036 ± 0.004	1.050 ± 0.12	− 0.012	0.821
Campus 4	E17	16.0	0.045 ± 0.004	1.313 ± 0.12	0.022	0.883
	E18	17.0	0.048 ± 0.003	1.400 ± 0.09	− 0.022	0.912
	E19	18.0	0.053 ± 0.005	1.546 ± 0.15	0.049	0.866
	E20	19.0	0.054 ± 0.009	1.575 ± 0.26	0.005	0.735

The overall heat loss coefficient per gross floor area, UH = HLC/A, W/m² K) can be isolated from the transformation of Eq. 4. To determine its value, an approximate estimation of the seasonal efficiency of the gas boilers of the buildings should be obtained. Although each building has a different boiler and space conditioning system, a simplified homogenous seasonal gas heating boiler efficiency of 70% is used. This energy efficiency is determined according to the official annex document of the Spanish energy certification scheme [32]. This is a first estimation of the efficiency of the boilers in real use; however, since no more in situ measurements are available, it will be used to facilitate a practical example on how to obtain the overall heat loss coefficient of buildings based on data obtained from smart meters. As it can be seen in Table 3, that building E13 has the lowest UH value, this is not representative since its R² is too small. The next building with the lowest value of UH is building E1, which corresponds to the library of Campus 1. The E20 has the highest UH value, which is in contradiction with its recent date of construction. This building hosts many different medicine labs and further detailed analyses are needed to understand if thesee high values are due to the envelope heat losses, poor control of HVAC set-points, or ventilation loads. The E3 building in Campus 1 exhibits a higher overall heat loss coefficient than E1 building and/or the efficiency of E1 heating system is higher. As the boiler and the heating distribution systems in both buildings are very similar, the hypothesis of higher heat losses in E3 is more likely to be the reason. In any case, more energy efficiency insights should be gained to better understand the differences in these two buildings.

In Fig. 9, the correlation matrix among the estimated UH, the year of construction and the number of users is shown. This graphical visualisation shows the Pearson’s correlation coefficients among each pair of variables (lower section), the density functions of each variable (diagonal) and the regression lines with their corresponding confidence intervals (upper section). As it can be seen, there is no correlation between the UH and the year of construction, which means the overall heat loss coefficient, including the ventilation losses, is not lower for newly constructed buildings. It can also be observed that there is a relatively small negative correlation between the UH and the maximum number of users. This low correlation could be due to the increase of the gross floor area with the number of occupants, which is the denominator of the UH. Indeed, given the same U values and air changes per hour, the theoretically calculated UH value gets smaller in a logarithmic trend with the size of the building (and so with the gross flow area and the number of users).

Open image in new window

In Table 3, it can be seen that all the values for the intercept point, b, are around zero, meaning that there are no weather independent consumptions (gas base loads), such as domestic hot water.

Besides the HLC, intercept and goodness of fit (R²) values, this degree-days methodology has enabled the determination of the heating base temperatures in each university building, showing values in the range 14.5–19 °C. This base temperature or balance-point is defined as the outside temperature above which the building does not require heating. The slightly higher base temperatures found in Campus 3 and 4 suggest that these buildings have smaller average internal gains than the ones in campus 1 and 2.

Figure 10 illustrates the variation of the slopes for the building performance lines in the different campuses, as well as the base temperature found for each building. Note how Campus 1 buildings have a larger variation in extreme slope (53%), whereas the other campuses with more than one building (3 and 4) show more similar normalised weather dependence (excluding E13), with slope differences of 17% (Campus 3) and 13% for (Campus 4).

Open image in new window

Hourly analysis of electricity data

Figure 11 presents a compilation of weekly plots of hourly electric consumption for years 2015 and 2016 (104 weeks) in the Rectorate building (E8), which is the largest electricity consumer. The vacation periods are easily identified, with a baseload profile of about 100 kWh. For year 2015, these include weeks 0 and 1 (Christmas Holidays), week 13 (Easter), and weeks 31, 32 and 33 (summer vacation). Similar trends are seen in 2016. The other local or national holidays are clearly identified as well, for instance, the two-day period on the 28th and 29th of September 2015 (two first days in week 39). Electric profiles along the weekdays with no cooling are very similar, with peaks in midday between 200 and 300 kW and valleys during night, reaching the baseload of about 100 kW. Weekends’ behaviour indicates very low occupancy, with some slight increase from base load during the day. The activation of the refrigeration chillers for the cooling season is clearly observed during the second or third week of May for both years, with a significant increase in the peak loads, reaching values above 400 kW for the weeks of July (weeks 27–30). Electricity use in Saturdays represents almost 11% of the total electricity annual consumption. This is a rather high contribution for a day without activity and suggests that substantial energy savings could be achieved with an energy audit that points out unnecessary consumptions during nights, weekends and holiday periods and achieves a reduction of the baseload. Although not shown, similar trends are observed in the rest of the buildings, but with peak and baseloads proportional to the level of annual electricity consumption of each building. These observed similarities in the daily load profiles for weekdays, weekends and holidays in all the buildings studied may be explained because all of them are managed by the same institution and are of the same building type, educational buildings. Saturdays’ share of electricity use is in the range 4 to 13% of the total annual use.

Open image in new window

The EER can be calculated from the Eq. 7. As previously mentioned, this is a very approximate way to obtain an estimation of the EER for each building. This is shown here to illustrate a practical example on how to determine HVAC systems efficiency in function of smart metering data. If monitored data of the delivered energy for space cooling were available, the reliability of the estimated results would be improved. In Fig. 12 a bar plot of the estimated values of the EER for each building, as well as the histogram of frequencies are shown. The building E6 has not been included in the analysis since the R² of the segmented model is too small. The buildings E2 and E15 are neither included in the analysis since their UH could not be determined due to the lack of gas data. As can be seen, the mean value is EER = 2 and the median is EER = 1.85, meaning 11 of the 17 analysed buildings are below this level. It can also be seen than only 4 buildings have EER over 3. Considering a high efficient air conditioning unit should have an EER around 2.5-3, it is clear that at least 14 buildings need some improvements in their energy systems efficiency, including better control and regulation systems. Nevertheless, these simplified models for energy consumption cannot capture all the complexities of the real building physics and more insights should be gained to confirm these cooling efficiency estimates.

Open image in new window

Figure 13 illustrates the resulting 4P linear model for the daily values of electricity consumption over average daily external temperature, for the E5 building, in Campus 1. Weekends and holiday days have been discarded, as occupancy is very low. The adjusted R² for the 4P regression model is 0.77. The resulting change point temperature obtained is 17.4 °C. For temperatures above this cooling base temperature, there is an important dependence of daily electricity consumption with average temperature, mainly associated to cooling loads required in summer period. The slope is also showing an increasing trend with decreasing temperatures below the change point, but the dependence is threefold smaller. This may indicate an increase in consumption for lighting in winter days and/or the use of electric heaters in offices to complement the central heating with gas boilers.

Open image in new window

Figure 14 shows an example of 5P regression model adjusting the daily electricity consumption as a function of daily average external temperature. This corresponds to E2 building, in Campus 1, where electricity is used both for cooling and heating purposes. In this case, the better fit corresponds to a 3-region model, and two change points, at 16.8 and 21.4 °C, being the former the heating base temperature, and the latter the cooling base temperature. The adjusted R² for the model is 0.76. Strong dependence of consumption with temperature is found both in the cooling and heating regions, having the heating region at the left a slightly higher slope. This may point out a lower average COP of the heat pump during the winter season. The inter-seasonal region daily electricity consumption shows a weak correlation with temperature.

Open image in new window

NZEB gap and proposals for energy improvements

An analysis of the fulfilment or failure of the four requirements proposed in the draft prEN ISO/DIS 52001-1 and explained in the Methodology section is performed below:

First requirement: building fabric

The existing Spanish building regulation does not incorporate a maximum threshold in relation to the overall heat transfer coefficient. Besides, the estimated heat loss coefficient by gross floor (UH), cannot be assimilated to an overall heat transfer coefficient because the ventilation and air infiltration losses are included and the gross floor area is related but not equal to the overall envelope heat transfer surface. Because of these limitations, this first step is not checked in this research. In any case, most of the buildings were constructed under less stringent regulations in terms of U values than the current Spanish ones. For instance, for the climatic zone of Lleida (D3), the U value of walls should be lower than 0.66 W/m² K and the U value of roofs lower than 0.38 W/m² K. It is likely then that the actual U values for the UdL buildings are higher than current Spanish thresholds, and these thresholds will be probably higher than the future ones established for Spanish NZEBs. Thus, the probability that all the UdL buildings studied do not meet this first requirement of minimum building fabric quality is very high.

Second requirement: total primary energy use

In Fig. 15 the calculated total primary energy consumption of each building is shown. The horizontal red lines correspond to the maximum threshold range before estimated for the Spanish building regulation. As can be seen in Fig. 15, the buildings E2, E6, E12, E13, E15 and E18 have primary energy consumption values close to the maximum thresholds (135.68, 119.0 3, 125.49,157.45, 136.14 and 133.38 kWh/m² year, respectively). It can also be seen that buildings E19 and E20 are very far from the NZEB concept. The high values of primary energy consumption of these buildings indicate that the non-EPB services, such as specific labs, have a critical impact over the energy performance of these buildings. A more detailed analysis of these other electrical services should be performed. The other 12 analysed buildings have total primary energy consumption values around 2.5–3 times higher than the maximum thresholds, indicating that many improvements should be implemented to reduce this gap.

Open image in new window

Third requirement: non-renewable primary energy

As previously mentioned, the maximum threshold for the non-renewable primary energy portion should be in the range of 45–55 kWh/m² year. In Fig. 16 the two components of the primary energy, the non-renewable (nren_PE) and the renewable (ren_PE) portion, is shown. The red line corresponds to the maximum threshold of non-renewable primary energy while the green line corresponds to the minimum renewable primary energy contribution. As can be seen in Fig. 16, the buildings E12 and E13 are the ones with lowest non-renewable primary energy consumption; however, their values (97.97 and 97.08 kWh/m² year, respectively) are around 2 times bigger than the maximum threshold. Again, the buildings E19 and E20 have the worse building energy performance. The other buildings oscillate between 106 and 284 kWh/m² year which is around 3–5 times bigger than the maximum thresholds for non-renewable primary energy.

Open image in new window

Fourth requirement: renewable primary energy and overall NZEB balance

The renewable primary energy contribution is shown in Fig. 16. It should be highlighted that the PV systems are installed in the buildings E1, E4, E9, E10, E11, E12, E13 and E14. However, only the building E13 has a RES contribution higher than the thresholds (60 kWh/m² year). The buildings without PV systems and with highest non-renewable primary energy also have the highest contribution of renewable energy. This is due to the primary factors defined for Spain which include a relatively high percentage (17%) of renewables coming from main electricity grid. Because of these primary factors, some buildings have values of renewable primary energy contribution over the maximum threshold. However if we calculate the percentage of RES generated on-site, by removing the renewable energy coming from the grid (see Fig. 17, left graph), we obtain a maximum value of 25% of on-site-renewable primary energy contribution for the building E13, which is almost half of the range of 45–52% established before for the new Spanish building regulation. The other buildings with PV systems have contributions ranging from 3 to 14%. The right plot of Fig. 17 shows the overall primary energy balance of all the analysed buildings. As it can be seen, only the building E13, with a primary energy balance of 36 kWh/m² year, is close to the NZEB balance (threshold of 10 kWh/m² year). The buildings E1, E2, E6, E12 and E15 are the next buildings with lower primary energy balances, with 68, 88, 93, 70 and 88 kWh/m² year, respectively. The other buildings have values in the range of 116–242 kWh/m² year which is very far from a nearly zero balance. Again, the buildings E19 and E20 have the worse energy performance with very high values of their primary energy balance.

Open image in new window

Although not included in the proposed requirements for NZEB, net CO₂ emissions indicators are worth to be analysed. In this sense, using the Spanish conversion factors from natural gas to CO₂ emissions and from electricity to CO₂ emissions [34], the UdL annual CO₂ emissions associated to energy use are determined. They are above 4700 tonnes of CO₂ in year 2016. This figure takes into account the saved electricity coming from PV production, around 6% of the total electric energy consumption. Electricity use contributes with a 66% and the remaining 34% is associated to gas use. Achieving or at least getting closer to the NZEB limits will of course reduce significantly the current emissions.

Recommendation