Data-Driven Energy Analysis for a Set of Rooftop Photovoltaic Systems Featuring Different Installation Characteristics: A Pairwise, Early-Life Performance Comparison

Abstract

The specific research presents a detailed, data-driven energy analysis on the early-life operation of two rooftop photovoltaic (PV) systems. The two PV systems examined are operated under the Net Metering scheme and are found in close proximity, within the geographical boundaries of a small-scale remote island on the Southeastern part of the Aegean Sea, Greece. The systems feature similar PV technology and capacity, i.e., 3.5 kW_p vs. 5 kW_p, but differ in terms of installation characteristics; thus, they offer an interesting case for pairwise comparison. Supported by the exploitation of a wealthy set of data, a detailed energy analysis is conducted, with our results providing useful insights on the seasonal performance variation of the two systems and its determinants, reflecting on the different siting and installation characteristics of the former.

1. Introduction

The Aegean Sea, located in the SE Mediterranean, includes numerous remote islands that are not interconnected to the mainland electricity system of Greece. These non-interconnected islands (NIIs) rely strongly on oil-based power generation [1], with local thermal power stations ranging from 150 kW to several MW, depending on the size of the NII [2]. To that end, according to recent official data of the Hellenic Electricity Distribution Network Operator (HEDNO) for the year 2023 [3], oil-based power generation represents 87% of the local electricity production in NIIs (i.e., 173 GWh_e out of 200 GWh_e), with the remaining 13% covered by existing RES installations (wind and PV). The given RES shares remain persistent over the last decade, illustrating the lack of flexibility means that could enable higher RES participation, especially for the smaller scale islands of the region [4]. This also reflects the extremely high electricity production cost for islands of this caliber [5], ranging, in the case of the Aegean Sea, between ~800€/MWh_e and ~1400€/MWh_e (Figure 1). The situation is similar to that encountered in other small European islands and island-developing states [6] featuring structural barriers such as limited land availability, insufficient infrastructure, etc., which, combined with intense energy demand variations across seasons, deem them more prone to energy vulnerabilities [5,6,7].

Meanwhile, and despite the benefits of RES solutions [8,9] and the high-quality solar potential appreciated in the Aegean Sea area (1500 to 1850 kWh/m² at the horizontal plane (from north to south)) [10,11], PV contribution in the local RES mix is relatively limited; residential-scale PVs, on the other hand, hold almost 5% of the overall RES capacity (Figure 2). Following the issuance of Law 4203/2013 and the application, until recently, of the Net Metering scheme, the development of small-scale, behind-the-meter PV installations at the building level (normally on building rooftops) saw an uprise in Greece, extending also to the region of the NIIs. By adapting to that scheme, PV prosumers may offset their electricity consumption, making use of PV installations with a capacity that is lower or equal to 10 kW_p, while treating the local grid as storage [12,13].

To that end, although land availability suggests an important limitation for the development of centralized PV installations in small-scale islands, challenges may also be noted in the integration of small-scale PV systems in the local built environment. Cultural heritage aspects, aesthetics, building density, and building topologies in island settlements entail limited building area availability, and often times lead to sub-optimal system design. Relevant to this and worth pointing out is that, to a certain extent, such limitations also apply in the case of urban settings in general, where distributed, behind-the-meter PVs, and battery storage are expected to grow rapidly in the coming years, supporting the emergence of self-consumption schemes [14] and the need to cope with grid congestion management issues.

Acknowledging the above, the current research aims to provide insights on the early-life operation of two small-scale rooftop and island-environment PVs. In doing so, we put emphasis on the performance impact of key installation parameters, arguing that a deeper understanding of this aspect is critical in order to enable a massive rollout in the sector. The two systems examined are installed on the small and remote island of Tilos, at the SE part of the Aegean Sea, and are found in close proximity with one another. Moreover, they feature similar PV technology and capacity but differ in terms of installation characteristics; thus, they offer an interesting case for pairwise comparison and energy performance analysis. The remainder of the paper is structured as follows: In Section 2, the case study under examination is described, and in Section 3 the methodological framework of the research is detailed. Accordingly, application results are presented and analyzed in Section 4, while in Section 5, the main research findings and conclusions are given.

2. Area and Scope of Study and Systems’ Characteristics

2.1. Area and Scope of Study

The under-study PV systems were installed in 2019 on the island of Tilos, i.e., one of the most remote islands of the Dodecanese complex, located NW of Rhodes (Figure 3a), and are both operated under the Net Metering scheme. In this context, the first system is hosted on the roof of a municipal building in the settlement of Livadia, i.e., one of the two main settlements of the island. The second PV system is found in close proximity (about 5 km NW of the first one), in the settlement of Megalo Chorio, i.e., an area with approximately the same elevation with Livadia (Figure 3b), installed on the roof of a local residence. It is worth mentioning that the first PV system is part of an integrated solar-EV charging station aiming to provide clean and cost-free energy to the local community, while both installations also serve the purpose of scientific research regarding the investigation of emerging, prosumer-level PV schemes.

As further analyzed in Section 2.2, the two systems are of similar capacity; nevertheless, they differ in terms of installation characteristics; thus, they offer an interesting case for pairwise comparison. Also considering that the two systems are found in close proximity and share similar elevations, they also appreciate a common solar potential profile, corresponding to 1750 kWh/m² per annum at the horizontal plane, as also presented in Figure 4.

Supported by the exploitation of a wealthy set of data over the early life operation span of the installations (first three years of operation), a detailed energy analysis is conducted next, with our results providing useful insights on the seasonal performance of the two systems and its determinants, reflecting the different siting and installation characteristics of the former. In more detail, the first of the systems is treated as a configuration limited by shading and sub-optimal installation aspects, while the second one is considered as a reference, shading-free system, featuring optimal tilt and azimuth angles.

2.2. Systems’ Characteristics

In the following, we present the detailed characteristics of the two systems examined. Concerning the Livadia PV system, the installation is hosted on the roof of a municipal building, near the coast and next to the pedestrian street of the island’s port (Figure 5a). The PV system consists of 17 PV panels; each witha nominal power of 290 W_p, having a rated efficiency of 17.72% (STC) and a total rated power output of 4.93 kW_p. The installation’s tilt and azimuth angles were selected at 15° and −70° (ESE), respectively, as a result of a design simulation exercise [16,17,18], seeking to balance visual impact minimization and annual energy yield maximization and also considering shading limitations and roof space availability (Figure 6).

On the other hand, the second PV system, located in Megalo Chorio, is installed on a residential rooftop and is not affected by shading (Figure 5b). The given system consists of 12 PV panels of 280 W_p each, determined by similar technology to that of the first system, with a rated efficiency of 17.11% and a total rated power output of 3.36 kW_p. Contrary to the Livadia system, the system in Megalo Chorio features optimal inclination (30°) and azimuth (0°) angles. As such, it can be determined as a reference system, suitable for pairwise comparison with other resembling installations.

Both PV systems consist of a single string comprising 17 and 12 PV panels, respectively, with the energy produced being either self-consumed or injected to the local electricity grid through commercial, single-phase voltage inverters. The inverters are of similar technology and of the same trademark, with a rated power of 5.0 kW and 3.6 kW, respectively. At this point, it is important to mention that the efficiency of the two inverters is comparable (97.0% and 98.4%, respectively), which, in combination with the corresponding PV panels’ efficiency, deems the overall efficiency of the two systems to be equal. Concerning the operational data of the two systems, it is retrieved through a dedicated monitoring software of the inverters’ manufacturer, providing data of different granularity, e.g., 15-min, hourly, daily, and annual power data.

3. Methodology

Under the current section, we define the main evaluation indices used in order to perform our analysis and elaborate on shading implications regarding the first system, with our research relying on the exploitation of data retrieved from the two systems over their early-life stage operation, marking a three-year period from 2019 to 2022.

To that end, in order to compare the energy performance of the two PV systems for the entire three-year period, one may divide the available measurements of electricity production (kWh_e) (expressed in 15 min time steps) with the corresponding nominal (peak) power of each system (kW_p). In this context, the Specific Energy Production (E_specif.,i, kWh_e/kW_p) and the Capacity Factor (CF), regarding the two PV systems’ operation, can be obtained. Comparing the available results on a seasonal basis for the entire three-year period, the impact of different inclination and orientation angles on the energy yield of the PV installations for each period of the year can be assessed.

As already mentioned, the two systems can be compared on the basis of their Specific Energy Production “E_specif.,i.”, expressing the energy yield per unit of installed power (kWh_e/kW_p), which results by using the actual output “E_i” (kWh_e) over a specific time period and the rated (or peak) power (kW_p) of the installation “P_nominal”, i.e.,

$${{\rm E}}_{specif.,i}={\displaystyle \frac{E_i}{P_{nominal}}}$$

(1)

Subsequently, the PV plants’ Capacity Factor “CF” is defined as the ratio of the actual energy production “E_i” (kWh_e) during a certain time period “Δt” (hours), to the corresponding maximum theoretical energy production over that period, which is equal to the product of the installation rated power (kW_p) and the time period examined. To that end, “CF” is a representative utilization factor of the installation over a given time period, i.e.,

$$CF={\displaystyle \frac{E_i}{P_{nominal}·\Delta t}}$$

(2)

Moreover, and according to the shading diagram presented in Figure 7, one may obtain the seasonal variation of shading concerning the system of Livadia (Figure 6). As it can be noted, except for the losses deriving from the panels’ sub-optimal angles, shading losses should also be taken into account, especially during the autumn and winter periods. It is important to mention at this point that the shading of the installation is purely exogenous, i.e., there is no shading between panels. Furthermore, as it can be obtained from the relevant diagram, shading is predominantly caused by a mountain range on the WNW of the installation. This becomes more intense during the winter months of the year, i.e., when the maximum solar altitude is low, and when surrounding, nearby obstacles also impose shading of the system.

In the following section, the energy performance of the two PV installations under investigation is examined first. For this purpose, we use the corresponding energy production from March 2019 to February 2022. Subsequently, using Equations (1) and (2), we assess the negative effect of sub-optimal inclination and orientation angles for the Livadia system, relevant to the performance of the installation in Megalo Chorio. With this in mind, monthly, daily, and intraday production profiles are studied, while special attention is given to assessing the impact of the different siting (shading) and installation characteristics on the energy performance of the two systems.

4. Results

4.1. Monthly Basis Comparison of Energy Production and Specific Energy Production

In the present section, the results of our analysis are presented, assuming an average year of operation for the two systems.

Table 1. Monthly operational comparison of the two PV Systems in terms of energy production, specific energy production, and capacity factor.

Month	Energy Production (Ei) (kWhe)		Specific Energy Production (Especific. i) (kWhe/kWp)		Capacity Factor (CF)
PV System	Livadia	M. Chorio	Livadia	M. Chorio	Livadia	M. Chorio
March	637	516	129	154	17.4%	20.7%
April	705	522	143	155	19.9%	21.6%
May	850	581	172	173	23.2%	23.2%
June	876	586	178	174	24.7%	24.1%
July	860	602	174	179	23.4%	24.1%
August	765	592	155	176	20.8%	23.7%
September	636	550	129	164	17.9%	22.7%
October	517	476	105	142	14.1%	19.0%
November	375	368	76	110	10.6%	15.2%
December	275	270	56	80	7.5%	10.8%
January	335	325	68	97	9.1%	13.0%
February	421	365	85	109	12.7%	16.2%
Total (annual):	7251	5751	1471	1712	-	-
Average (annual):	-	-	-	-	16.8%	19.5%

provides the monthly energy performance of the two systems in terms of actual production “E_i” (in kWh_e), specific production “E_specif.,i” (in kWh_e/kW_p) and Capacity Factor “CF”. As expected, the bigger installation presents higher values of absolute energy yield. In more detail, the total annual energy production of the Livadia installation exceeds 7.2 MWh_e, while, for the smaller installation in Megalo Chorio, it approaches 5.8 MWh_e. On the other hand, if we examine “E_specif.,i” (total annual) and then the annual average “CF”, it becomes clear that the system in Megalo Chorio (E_{specif., annual_M.Ch.} = 1712 kWh_e/kW_p, CF = 19.5%) operates more efficiently (E_{specif., annual}__Liv. = 1471 kWh_e/kW_p, CF = 16.8%). Moreover, it can be noted that June and July stand out as the most efficient months for the system in Livadia and Megalo Chorio, respectively. More specifically, during the first three years of operation, the monthly-based specific production maximizes at 178 kWh_e/kW_p for Livadia and at 179 kWh_e/kW_p for Megalo Chorio, while at the same time, the monthly CF reaches 23.5% and 24.1%, respectively.

4.2. Daily Basis Comparison of Energy Production and Specific Energy Production

Under the specific sub-section, we provide a graphical comparison concerning the daily-basis energy production and specific energy production of the two systems. In this context, Figure 8 compares the daily energy production of the two systems over the three years examined, designating the system of Livadia (4.93 kW_p) as the most productive. The latter presents a pronounced output during the spring and summer months, followed by an analogous decrease during autumn and winter. At the same time, the second system follows a similar pattern, determined, however, by a smoother transition between seasons. This is also consistent with the results of Table 1, where, as one may note, November, December, and January carry a similar production for the two systems, despite the 1.5 kW_p difference in installed capacity.

To better understand this, one may also compare the average daily energy yield of the two PV installations for representative months of the year (Figure 9). More precisely, Figure 9 allows us to capture the seasonal impact on energy production, caused by the sub-optimal inclination of the bigger PV installation. As expected, in April and July, when the solar altitude is still high, i.e., between 54° and 78° for the latitude of Tilos island [19], the energy production of the Livadia system outweighs the production of the second system, owing to the difference in capacity. On the contrary, during autumn and winter, when the solar altitude decreases between 30° and 54°, the energy produced by the two systems is comparable, despite their size difference, as also seen in the results of Table 1.

Using the index of specific energy production accordingly, the energy performance of the two systems is better articulated. Analyzing the respective daily values for the entire three-year period in Figure 10, what can be noted at first is that the installation of Megalo Chorio operates more efficiently for the biggest span of the examined period.

This becomes more pronounced during the autumn and winter months, i.e., from September to March, with the Megalo Chorio system outperforming the system of Livadia, demonstrating values of specific energy production that may, for certain days, even turn out to be double the ones of the system in Livadia. This reflects the analysis of energy production seen earlier (Figure 8) and justifies the bridging of the capacity gap between the two systems in terms of anticipated energy production for a great part of the year.

4.3. Pairwise Comparison for the Determination of Energy Losses

As mentioned earlier, by considering the relevant specific energy production data of Table 1, the average annual energy output per installed kW_p for the PV installation of 3.36 kW_p equals to E_{specif., annual_M.Ch.} = 1712 kWh_e/kW_p, while, for the same time period, the installation of 4.93 kW_p produces E_{specif., annual_Liv.} = 1471 kWh_e/kW_p. This difference of almost 240 kWh_e/kW_p between the two PV systems (ΔΕ_{specif., annual}) is attributed partially to shading losses (ΔΕ_{specif., shading}) and partially to losses due to the sub-optimal tilt and azimuth angles (ΔΕ_{specif., tilt&azimuth}) adopted in Livadia. Respectively, the percentage difference that results by comparing the Livadia installation with the reference system of Megalo Chorio (E_{specif., annual_M.Ch.}) can be defined as follows:

$$\Delta {{\rm E}}_{specif.,annual}={\Delta {\rm E}}_{specif.shading}+{\Delta {\rm E}}_{specif.tilt\&azimuth}={\displaystyle \frac{{{\rm E}}_{specif.,annual\_M.Ch.}-{{\rm E}}_{specif.,annual\_Liv.}}{{{\rm E}}_{specif.,annual\_M.Ch.}}}=14.1\%$$

(3)

Acknowledging the above, it is important to mention that according to the shading diagram of Figure 7, shading emerges in Livadia after 12:00 pm, regardless of the season, and it persists until sunset, especially during winter, when the energy production of the PV panels decreases due to shading caused by surrounding obstacles (Figure 6) and the mountain range at the WNW side of the Livadia installation.

Regarding the determination of shading losses, these are currently estimated theoretically with the use of commercial software [18] and expressed as a fragment of non-produced energy. According to the results obtained, the total loss is approximately equal to 5.7%.

Given that and according to Equation (3), there is an additional energy deficit of 8.4%, which describes the total effect of the sub-optimal tilt and azimuth angles (ΔΕ_{specif., tilt&azimuth}). Using the theoretical analysis results, the energy gains of the 15° tilt in Livadia, compared to a horizontal plane installation, reach 7.8%, while the corresponding percentage in Megalo Chorio, for the tilt of 30°, is 9.9%. This difference leads to tilt-driven energy production losses (ΔΕ_{specif., tilt}) of 2.1% for the Livadia system, which in turn, suggests azimuth losses of 6.3% (see also Equation (4)):

$${\Delta {\rm E}}_{specif.azimuth}={\Delta {\rm E}}_{specif.tilt\&azimuth}-{\Delta {\rm E}}_{specif.tilt}=8.4\%-2.1\%=6.3\%$$

(4)

4.4. Intraday Basis Comparison of Specific Energy Production

Subsequently, we further analyze production differences between the two systems, using the intraday specific energy output and representative days of the year, applying power data with a granularity of 15 min. In this direction, a first overview is given in Figure 11, where four, clear-sky days have been selected as representative days for the four seasons of the year.

For the summer period, the two installations demonstrate the same daily specific production, since neither shading nor tilt-driven deviations are recorded. Therefore, the only visible difference is orientation driven. In more detail, the system in Livadia, facing 70° east of south, operates more efficiently during morning hours and less efficiently during afternoon hours, with the daily balance between morning gains and afternoon losses being practically zero.

A similar situation is encountered for spring. As it can be seen, there is a small increase in losses for the system of Livadia during the afternoon, caused by shading and by the difference in orientation, while for the morning period, the variation is similar to that of the summer day. These differences in the energy performance of the two PV systems are better captured in Figure 12, where, for comparison purposes, we include the autumn and spring profile and mark the areas generating the gains and different types of losses, in comparative terms between Livadia and Megalo Chorio. In conclusion, the energy behavior of the two systems during the spring and summer periods proves to be similar, which is consistent with the result of previous research [20].

Contrariwise, the situation is notably different during winter and autumn (Figure 11 and Figure 12). Over that period, losses for the system in Livadia increase considerably. Primarily owing to the different tilt angle, a notable share of losses is evenly distributed in a broad time span before and after noon-time, while in the afternoon, shading and orientation losses stand as the main driver for the sharp reduction in production curves concerning the system in Livadia.

A more illustrative view of the trends identified above is given in Figure 13, which provides representative intraday profiles of specific energy consumption for each month of the year, organized in quarters.

The sequence of profiles to that end allows for a better understanding of the step-wise variation in the balance between gains and losses for the system of Livadia across all months of the year. Starting with summer, a diminishing trend is noted concerning orientation-driven gains until the beginning of autumn. Therefrom, both tilt-driven and shading losses emerge and gradually amplify, with the difference in orientation between the two systems further aggravating the performance of the Livadia system. The given trend is maintained until the beginning of springtime, with a clear recovery in trends appearing from April onwards.

Finally, we conclude the presentation of our results with the graphs in Figure 14, which provide the average (for all months of the three-year period) intraday profiles of the specific energy yield for the two PV systems, this time, applying an hourly time step of analysis. In accordance with previous results, the overall effect of sub-optimal angles and shading is currently revealed. Given that, it is worth emphasizing that the higher specific energy production noted in Megalo Chorio mainly results from the efficient operation of the system during winter and autumn. Moreover, the difference in specific energy production between monthly profile curves is higher in Livadia, determined by a noontime span that ranges between ~450 Wh_e/kW_p and ~720 Wh_e/kW_p; for the case of Megalo Chorio, this varies between ~550 Wh_e/kW_p and ~790 Wh_e/kW_p.

5. Discussion and Conclusions

The present study provides a data-driven energy analysis on the early-life performance of two small-scale rooftop PV systems. The analysis carried out puts forward a pairwise comparison, exploiting the fact that both systems operate in the same environment and that the second of the two carries optimal installation characteristics; thus, it can be treated as a reference installation. The first system on the other hand, although of relatively higher capacity, is determined by sub-optimal installation features that limit its performance for a considerable share of the year.

Conduction of a pairwise comparative analysis to that end is value-adding, in the sense that we assess actual system performance in a relative fashion, opposite to the more common approach reflecting simulation-based, theoretically anticipated results. In a broader and more abstract context, the given approach contributes also to the development of holistic evaluation frameworks for the promotion of rooftop PVs [21], especially in community settings. It does so by allowing for a better, quantified understanding of performance variation and its dynamics within the built environment, exploiting in-situ data, identifying the main drivers of performance variation across seasons and emphasizing the balance of relative energy gains and losses between resembling PV systems.

Acknowledging the above, and for the given pair of systems examined, three-years of detailed energy data were used in the current research, corresponding to the early-life operation of the two PV installations. Main findings of the research reach to the conclusion that sub-optimal installation characteristics and shading-driven limitations have an immediate impact on the system of Livadia in actual terms. This manifests mainly over the autumn and winter period of the year, where all three loss mechanisms, i.e., shading, orientation, and inclination, lead to a decrease in the energy performance of the system.

In more detail, and according to the intraday analysis carried out, sub-optimal inclination affects the energy output, following an even loss distribution before and after noon time, while orientation and shading emerge from early to late afternoon. During the spring and summer periods of the year, inclination losses diminish, with the difference in orientation generating not only losses, but also gains for the system of Livadia during the morning hours. The latter are balanced by both shading and orientation-driven losses during the afternoon hours of the day.

Acknowledging the above, it is worth mentioning that seasonal and intraday generation profile variations come with implications concerning the ability of rooftop PVs to meet a given building’s electricity needs, especially if considering that emerging, beyond net metering prosumer schemes, calls for higher levels of self-consumption. This is also relevant for small island regions, where load demand varies considerably across seasons, and where building occupancy could be limited to a fraction of time over the year, e.g., during the summer season.

With this in mind, insights provided by the current research contribute to a better understanding of similar implications, with conclusions drawn supported by actual operational data, thus gaining reliability. At the same time, and in the context of future research, findings of our research could apply in the design of new systems, either at the end-user level, or in relation to the development of aggregate PV schemes, as well as in the evaluation of additional energy performance factors, such as PV panels’ ageing mechanisms through, e.g., the comparison of resembling PV systems found in later stages of their operational lifetime.