InGenio Journal  
Revista de Ciencias de la Ingeniería de la Universidad Técnica Estatal de Quevedo  
https://revistas.uteq.edu.ec/index.php/ingenio  
e-ISSN: 2697-3642 - CC BY-NC-SA 4.0  
PV Integration and Emission Offsets in a Campus Data  
Center: Evidence from Seven Years of Operation  
(Integración fotovoltaica y compensación de emisiones en un centro de  
datos universitario: evidencia de siete años de operación)  
Danny Ochoa-Correa , Emilia Sempértegui-Moscoso  
Universidad de Cuenca, Ecuador.  
danny.ochoac@ucuenca.edu.ec, emilia.sempertegui96@ucuenca.edu.ec  
Abstract: Data centers raise energy and environmental challenges for public and  
educational institutions. This paper evaluates an institutional data center at the University  
of Cuenca (Ecuador) supplied in part by a 35 kWp rooftop photovoltaic (PV) system  
operating without electrical storage since 2018. Seven years of data (2018–2024) were  
analyzed, including monthly records of electricity consumption, PV generation, and  
meteorological variables. Renewable-integration indicators are computed (demand  
coverage and on-site use of PV energy), and avoided CO2 emissions are estimated using  
official annual grid factors for Ecuador’s national power system, including a sensitivity  
case for higher carbon intensity. Annual PV generation ranged from 24.3 to 47.3 MWh  
(34.4 MWh average), while annual data center consumption increased from 63.0 MWh  
(2018) to 149.3 MWh (2024), with a step change starting in 2020. PV coverage at the  
shared distribution panel (laboratory + data center) declined from 52.8% (2018) to 23.4%  
(2023) and 15.6% (2024), driven mainly by demand growth; in 2024, PV output was  
further reduced by grid outages because anti-islanding protection disconnects a grid-tied,  
battery-less system. Over 2018–2024, avoided emissions are estimated at 64.9–72.1 tCO2  
using national emission factors (0.27–0.30 kg CO2/kWh), and 96.2 tCO2 under a 0.40 kg  
CO2/kWh scenario. The results support the green data center agenda by informing  
planning decisions for integrating on-site renewable generation in data centers, with  
emphasis on finer submetering, energy storage, and PV capacity expansion as digital  
workloads grow.  
Keywords: green-data centers, photovoltaic energy, self-consumption, CO2 emissions,  
renewable integration.  
Resumen: Los centros de datos plantean retos energéticos y ambientales en instituciones  
públicas y educativas. Este trabajo evalúa un centro de datos institucional de la Universidad  
de Cuenca (Ecuador), asistido por un sistema fotovoltaico (FV) en azotea de 35 kWp, sin  
almacenamiento, en operación desde 2018. Se analizaron siete años de datos (2018– 2024),  
con registros mensuales de consumo eléctrico, generación FV y variables meteorológicas.  
Se estimaron indicadores de integración renovable y las emisiones de CO2 evitadas  
mediante factores oficiales del sistema eléctrico nacional, incluyendo un análisis de  
sensibilidad para un escenario de mayor intensidad de carbono. La generación FV anual  
varió entre 24,3 y 47,3 MWh (promedio 34,4 MWh), mientras que el consumo anual del  
centro de datos aumentó de 63,0 MWh (2018) a 149,3 MWh (2024), con un cambio de  
nivel a partir de 2020. La cobertura FV del consumo compartido (laboratorio + centro de  
datos) disminuyó de 52,8 % (2018) a 23,4 % (2023) y 15,6 % (2024), principalmente por el  
crecimiento de la demanda; en 2024, además, la generación FV se redujo por  
desconexiones durante cortes programados de red. En el periodo 2018–2024 se estimaron  
64.9–72.1 tCO2 evitadas con factores nacionales (0.27–0.30 kg CO2/kWh), y 96,2 tCO2  
bajo un escenario de 0,40 kg CO2/kWh. El estudio aporta insumos cuantitativos para la  
agenda de green data centers y respalda decisiones de planificación orientadas a incorporar  
Volumen 9 | Número 2 | Pp. 107–120 | Julio 2026  
DOI: https://doi.org/10.18779/ingenio.v9i2.1268  
Recibido (Received): 2026/03/05  
Aceptado (Accepted): 2026/06/22  
generación renovable en centros de datos, con énfasis en submedición, almacenamiento e  
incremento de capacidad FV conforme crecen las cargas digitales.  
Palabras clave: centros de datos sostenibles, energía fotovoltaica, autoconsumo, emisiones  
de CO2, integración renovable.  
1. INTRODUCTION  
Data centers account for a growing share of electricity demand as digital services, cloud  
computing, and data storage expand. Although server power management and facility thermal  
design have improved over time, total consumption may still increase when computational  
capacity and service availability grow faster than efficiency gains. Sustainability assessments  
extend beyond electricity use to include cooling-related water consumption and electronic waste  
streams [1], [2]. These issues are relevant for university data centers, where infrastructure  
modernization is often shaped by budgetary and operational constraints while service demand  
continues to grow [3], [4]. In this context, the notion of green data centers has shifted from a  
broad aspiration to a set of design and operational practices tied to measurable sustainability  
objectives. Recent frameworks emphasize energy efficiency together with coordinated  
management of supporting resources and operational decisions (e.g., virtualization and  
consolidation policies, cooling strategies, and continuous energy monitoring) to reduce overall  
environmental impacts [1]. For medium-scale facilities, a recurring theme is that renewable  
integration alone is insufficient unless it is paired with instrumentation and data services that  
make energy use, thermal conditions, and workload behavior jointly observable and usable for  
operational control [5], [6].  
Photovoltaic (PV)-assisted data centers are feasible in regions with adequate solar resources,  
given the modularity of PV deployment and the opportunity to increase on-site self-  
consumption. However, solar generation is variable, whereas computing services and cooling  
requirements are continuous and often time-sensitive. Prior research has therefore emphasized  
integrated monitoring and cyber-physical management architectures that combine PV  
production, weather conditions, subsystem power measurements (information technology (IT)  
equipment and cooling), and workload descriptors—including container-level visibility—to  
support operational decision-making [5], [6]. The literature also examines algorithmic  
approaches to reduce grid electricity use and carbon emissions through workload shifting and  
demand-response participation. Deep reinforcement learning has been proposed as a controller  
to coordinate workload scheduling and cooling under stochastic conditions and onsite  
renewables [7], while hierarchical load models that capture coupled electrical–thermal–  
performance behavior have been introduced to improve demand-response quality without  
driving inefficient operating points [8]. Despite this progress, deployment in production  
environments remains limited; recent reviews report that overly complex frameworks, broken  
abstraction layers, narrow reliance on batch workloads, and weak incentives continue to slow  
adoption in operational settings [9].  
This study provides empirical evidence in this area by analyzing an operational institutional  
data center at the University of Cuenca, Ecuador, partially supplied by a 35 kWp rooftop PV  
system directly connected to the internal distribution panel. The facility has operated  
continuously since 2018, enabling a longitudinal assessment over a seven-year observation  
window (2018–2024). We characterize the temporal evolution of electricity consumption and  
PV generation, quantify the renewable contribution to the facility energy balance, and estimate  
the associated reduction in carbon emissions due to displacement of grid electricity. By focusing  
on an educational, medium-scale installation in Latin America, the work complements prior  
conceptual and algorithmic studies with measured, long-term performance evidence to inform  
planning and replication in similar public sector settings [1], [3], [5].  
InGenio Journal, 9(2), 107–120  
108  
2. MATERIALS AND METHODS  
2.1. Infrastructure description  
The data center analyzed in this study is located at the University of Cuenca (Cuenca,  
Ecuador) and has operated continuously since 2018, supporting institutional computing  
services, virtualized workloads, storage, and internal academic platforms. The facility is  
partially supplied by a rooftop photovoltaic (PV) plant with an installed capacity of 35 kWp,  
installed on the main building [10]. The PV installation comprises both polycrystalline and  
monocrystalline silicon modules. Most of the PV arrays are mounted in fixed structures with a  
tilt angle of 5° and a north-facing orientation, which is a common configuration for maximizing  
solar energy capture under local conditions in the southern hemisphere. The PV system is  
connected directly to the internal low-voltage distribution panel and operates without electrical  
storage (battery-less configuration); therefore, its contribution depends on real-time irradiance  
conditions.  
The information technology (IT) room follows a rack-based architecture. According to the  
as-built layout, the installation comprises eight racks with a rated capacity of 5 kW per rack,  
arranged in four pairs to support operational organization and cable and airflow management.  
Thermal control is provided by four precision air-conditioning (A/C) units (Air Conditioner–  
B1), each rated at 21 kW of electrical power and equipped with electric heating and  
humidification. These units maintain the indoor environmental conditions required for reliable  
IT operation and provide redundancy at the cooling-system level. Table 1 summarizes the  
nameplate electrical capacities used in this work to characterize the infrastructure.  
Table 1. Installed electrical capacity of the main subsystems in the Microgrid Laboratory  
(LMR) data center (nameplate values).  
Subsytem  
Qty  
Unit (kW)  
Total (kW) Notes  
IT racks (servers, storage,  
networking)  
Racks arranged in four  
pairs  
8
5
40  
84  
Precision A/C (Air  
Conditioner– B1)  
Includes electric heating  
and humidifier  
4
21  
Excludes auxiliaries  
(uninterruptible power  
supply (UPS) losses,  
lighting, etc.)  
Total (IT + cooling)  
-
-
124  
The values reported in Table 1 represent nameplate ratings (design capacities), not  
continuous operating demand. In practice, the instantaneous IT power depends on server  
utilization and the equipment population within each rack, while cooling demand varies with  
indoor and outdoor conditions and the control strategy of the A/C units (e.g., compressor  
cycling or modulation). In addition, redundancy policies may keep one or more units in standby  
or operating below their maximum rating. As a result, the measured electrical demand of the  
facility can be lower than the sum of nameplate ratings during normal operation.  
To document the physical arrangement and support interpretation of the energy  
measurements, this section includes: (i) photographic evidence of the data center room and the  
rooftop PV installation (Figure 1 and Figure 2); (ii) a schematic representation of the internal  
composition of the data center (IT racks and cooling subsystems) used as a reference model for  
the energy balance discussion (Figure 3); and (iii) the single-line diagram of the electrical layout  
of the studied installation (Figure 4), which summarizes the net-balance operating principle.  
Under normal grid availability, the data center and the remaining laboratory loads draw  
InGenio Journal, 9(2), 107–120  
109  
electricity from the distribution network and are supplemented in real time by the on-site PV  
production at the shared low-voltage panel. When PV generation exceeds instantaneous  
demand, the surplus can be exported to the public grid, whereas when demand exceeds PV  
output, the deficit is supplied by the grid. During public-grid outages, the grid-tied PV system is  
disconnected by anti-islanding protection and stops generating; in that condition, continuity of  
data center operation is maintained by thermal generator sets.  
Figure 1. Photograph of the University of Cuenca data center.  
Figure 2. Rooftop PV installation (35 kWp) connected to the internal distribution panel.  
Figure 3. Internal composition of the data center used as a reference for the energy assessment  
(IT racks and cooling subsystems).  
InGenio Journal, 9(2), 107–120  
110  
Figure 4. Electrical layout of the Microgrid Laboratory (LMR).  
2.2. Data sources  
The analysis uses a seven-year dataset spanning January 2018 through December 2024. It  
integrates electricity demand, PV production, and environmental records to characterize long-  
term behavior of the LMR facility and its onsite renewable supply. Electricity consumption was  
obtained from the internal metering infrastructure, which provides monthly and annual energy  
totals at the facility level for longitudinal assessment and interannual comparison. In parallel,  
PV generation data were retrieved from the inverter monitoring platform and cross-checked  
against local dataloggers connected to the DC/AC conversion stage. This cross verification  
reduces the likelihood of missing values and improves traceability when performance changes  
are observed over time.  
To provide context for PV production variability and to support interpretation of seasonal  
and interannual trends, local meteorological measurements were also incorporated. Solar  
irradiance and ambient temperature records were collected from a nearby weather station  
operated by the Faculty of Engineering. These environmental series are not used to predict  
generation; instead, they provide a physical reference for expected fluctuations and help  
distinguish resource-driven changes (e.g., irradiance patterns) from system-driven changes (e.g.,  
operational events).  
A relevant methodological aspect is that the PV plant supplies a shared low-voltage panel  
feeding both the data center and auxiliary laboratory loads. Therefore, the “total demand” used  
in the PV coverage metrics corresponds to the combined energy consumption of the Microgrid  
Laboratory and the data center. To isolate the operational demand of the data center for  
subsequent analyses, a dedicated measurement campaign was conducted at the data center three-  
phase service entrance. The Results section reports both views—aggregate facility demand and  
data center demand—to avoid conflating infrastructure-level behavior with IT-room operation.  
2.3. Renewable energy share assessment  
On-site PV integration is quantified using complementary indicators that describe (i) the  
renewable fraction of the total demand supplied by the shared panel and (ii) the extent to which  
PV production is used locally. First, the Energy Share (ES) is computed as the ratio between PV  
generation and total electricity demand over a given accounting window (monthly, then  
aggregated annually). This indicator summarizes the renewable contribution relative to the  
energy needs of the supplied loads.  
InGenio Journal, 9(2), 107–120  
111  
Because behind-the-meter PV generation can be consumed locally or exported to the public  
grid, the analysis distinguishes between PV generation and PV utilization. Self-consumption is  
defined as the fraction of PV energy used within the supplied loads, excluding any potential  
feed-in. Following the framework in [1], two indicators are reported:  
•
•
Green Share (GS): the proportion of electricity demand met by on-site PV generation  
within the same accounting period.  
Renewable Energy Utilization Rate (REUR): the fraction of total PV generation that is  
consumed locally within the supplied loads.  
All indicators are computed on a monthly basis to preserve seasonal information and then  
aggregated annually to identify long-term trends in energy integration, operational stability, and  
PV-to-load matching. When interpreting the indicators, the manuscript distinguishes between  
metrics computed at the shared-panel level (laboratory + data center) and those derived from the  
dedicated data center measurements.  
2.4. Avoided emissions estimation  
Avoided carbon emissions are estimated using a displacement approach in which PV  
electricity consumed on-site offsets an equivalent amount of electricity that would otherwise be  
supplied by the public grid. For consistency with institutional greenhouse-gas (GHG)  
accounting, this work adopts the official annual grid-average emission factor for Ecuador  
Continental reported in the national “Factor de Emisión de CO2 del Sistema Nacional  
Interconectado (SNI)” series coordinated by the Operador Nacional de Electricidad (CENACE)  
[11], [12], [13]. This choice aligns with the location-based method described in the GHG  
Protocol Scope 2 Guidance, which recommends using grid-average emission factors to  
represent the average emissions intensity of the electricity system where consumption occurs  
[14]. The same national reports also include project-oriented factors based on the UNFCCC  
combined-margin methodology (Tool 07) [15]; however, because the objective here is to  
quantify avoided emissions under operational self-consumption (behind-the-meter PV  
displacing grid electricity at the point of supply), the grid-average factor is the appropriate  
reference.  
Avoided emissions are calculated as shown in Equation (1):  
(1)  
is  
퐸푎ꢀꢁꢂꢃꢄꢃ = 퐸푃ꢅ,푢ꢆꢄꢃ × 퐸ꢇ  
푔ꢈꢂꢃ  
where 퐸푃ꢅ,푢ꢆꢄꢃ is the PV electricity consumed locally by the supplied loads and 퐸ꢇ  
푔ꢈꢂꢃ  
the annual grid emission factor (kg CO2/kWh). This formulation follows common reporting  
practice for renewable displacement assessments and is consistent with the methodologies  
discussed in [7], [8]. When presenting results, avoided emissions are reported at monthly  
resolution and summarized annually to compare environmental outcomes across years.  
2.5. Additional indicators  
Beyond energy and emissions, the study considers complementary metrics when the  
required information is available in the monitoring records. In particular, an economic reference  
can be provided through the levelized cost of energy (LCOE) of the PV system, computed from  
capital expenditure, operation and maintenance costs, and cumulative electricity production over  
the assumed system lifetime. In addition, an operational environmental-intensity indicator can  
be discussed using Carbon Usage Effectiveness (CUE), relating carbon emissions to IT energy  
use. If these variables cannot be supported with sufficient data granularity or traceability for the  
study period, the manuscript limits its reporting to the energy-share and avoided-emissions  
indicators described above and states explicitly which inputs were not available.  
InGenio Journal, 9(2), 107–120  
112  
3. RESULTS  
3.1. Photovoltaic Generation Analysis  
The annual and seasonal performance of the 35 kWp rooftop PV system was evaluated for  
2018–2024 using inverter-reported production data cross-checked with local dataloggers. Table  
2 summarizes the annual energy yield and the corresponding capacity factor, which represents  
the effective utilization of the installed capacity in each year. Over the study period, the PV  
system produced an average of about 34.4 MWh per year. Year-to-year variation reflects  
changes in operating conditions and solar resource availability. The highest annual output  
occurred in 2022 (47.3 MWh; capacity factor 15.4%), whereas the lowest occurred in 2024  
(24.3 MWh; capacity factor 7.9%). The 2024 value must be interpreted in context because it  
coincided with an unusual operating period in Ecuador, with extended scheduled power outages  
associated with the national electricity crisis. Since the PV plant operates in a grid-tied, battery-  
less configuration, inverter anti-islanding protection disconnects generation during grid outages.  
As a result, the PV energy recorded by the monitoring system can drop even when solar  
resource is available. During those periods, the data center was supplied mainly by on-site  
generator sets, and the 2024 PV output does not reflect typical operation under continuous grid  
availability.  
Table 2. Annual PV generation and capacity factor of the 35 kWp rooftop system (2018–2024).  
Year  
2018  
2019  
2020  
2021  
2022  
2023  
2024  
PV generation (kWh)  
34 665.1  
Capacity factor (%)  
11.3  
10.8  
11.2  
9.6  
33 004.5  
34 309.6  
29 545.2  
47 289.0  
15.4  
12.2  
7.9  
37 391.7  
24 257.0  
At the seasonal scale, the monthly profiles presented later follow the local climatic cycle,  
with recurring periods of higher production during months with greater solar resource and  
reduced cloudiness. To further support this interpretation, Figure 5 compares the annual PV  
energy production with the annual solar irradiation measured by a meteorological station  
installed on the rooftop of the same building where the PV system is located. The results show a  
strong correspondence between both variables throughout most of the study period, indicating  
that interannual variations in PV generation are largely explained by the available solar  
resource. Although the irradiation data are presented at annual resolution, the Andean climate of  
Cuenca is characterized by alternating sunny and cloudy periods, influenced by seasonal rainfall  
patterns and regional meteorological anomalies such as El Niño-related variability, drought  
conditions, and prolonged dry spells. In this context, the higher irradiation observed in 2024 is  
consistent with reduced rainfall and clearer-sky conditions during the national drought period.  
However, a noticeable divergence appears in 2024, when PV production decreased  
disproportionately relative to the measured solar irradiation. This behavior is consistent with the  
extended scheduled power outages experienced in Ecuador during the national electricity crisis.  
Thus, the mismatch between irradiation and PV production in 2024 is mainly attributed to  
power outages associated with the drought-driven low-water period, rather than to a reduction in  
the available solar resource. Because the PV system operates in a grid-tied, battery-less  
configuration, anti-islanding protection disconnects the inverters whenever the public grid is  
InGenio Journal, 9(2), 107–120  
113  
unavailable, preventing energy production even under favorable solar conditions. Therefore, the  
2024 reduction in PV output should not be interpreted as a consequence of diminished solar  
resource availability but rather as the result of operational restrictions imposed by prolonged  
grid interruptions. This seasonal structure is used in the following subsections to interpret PV-  
to-load matching and the resulting renewable utilization indicators.  
50000  
45000  
40000  
35000  
30000  
25000  
20000  
15000  
10000  
5000  
150  
145  
140  
135  
130  
125  
120  
115  
0
2018  
2019  
2020  
2021  
Years  
PV production Sol. Rad.  
2022  
2023  
2024  
Figure 5. Annual solar irradiation measured at the rooftop meteorological station and annual  
PV energy production of the 35 kWp system (2018–2024).  
3.2. Data Center Energy Consumption  
Electricity demand at the data center increased over 2018– 2024, consistent with the  
expansion of institutional digital services and the growing deployment of virtualized workloads.  
Although the overall trajectory is upward, the series presents distinct phases rather than a  
uniform, linear increase. Consumption remained relatively stable during the first two years of  
operation (2018–2019), followed by a pronounced increase in 2020–2021, consistent with the  
commissioning of additional computing and storage resources and the consolidation of new  
services. From 2022 onward, the facility operated at a higher demand level, with year-to-year  
variation associated with operational scheduling, workload dynamics, and infrastructure  
management practices.  
Seasonal behavior is also evident in the monthly records. Higher energy use typically aligns  
with active academic periods (March–July and September–December), when user access, hosted  
platforms, and institutional services intensify. Lower consumption is observed during semester  
breaks, when user activity decreases and some computational tasks are reduced or rescheduled.  
In addition to workload-related drivers, the cooling subsystem contributes to seasonal variability  
because heating, ventilation, and air conditioning (HVAC) operation depends on indoor  
setpoints and ambient conditions and may respond differently during warmer or more humid  
weeks.  
The PV plant supplies a shared low-voltage panel feeding both the data center and auxiliary  
laboratory loads. Historical energy records indicate that the data center accounts for the  
dominant share of the combined Microgrid Laboratory + data center electricity demand;  
therefore, changes in aggregate facility energy largely follow data center operation. To support a  
clear separation between data center demand and other laboratory loads, a dedicated  
measurement campaign was conducted at the data center three-phase service entrance. This  
campaign provides an operational reference for typical daily energy use and weekday–weekend  
differences, which is later used to validate the allocation assumptions applied in the energy-  
share indicators.  
Table 3 summarizes the annual electricity consumption of the data center for the analyzed  
period. A joint visualization comparing PV generation and data center demand is presented in a  
subsequent subsection to support the statistical interpretation of renewable contribution under  
evolving load conditions.  
InGenio Journal, 9(2), 107–120  
114  
Table 3. Annual electricity consumption of the University of Cuenca data center (2018–2024).  
Year  
2018  
2019  
2020  
2021  
2022  
2023  
2024  
Data center consumption (kWh)  
63 019.62  
66 220.96  
112 013.29  
111 484.00  
145 045.34  
153 153.91  
149 261.40  
In addition to the historical monthly and annual records, the measurement campaign at the  
data center three-phase service entrance was used to characterize a representative short term  
operating profile and to support the separation between data center demand and auxiliary  
laboratory loads. The campaign covered seven consecutive days (including working days and  
the weekend) and recorded the daily energy exchanged at the data center point of common  
coupling. Figure 6 shows the resulting weekly pattern: daily consumption remains relatively  
stable during weekdays, with a moderate reduction on Saturday and Sunday, consistent with  
lower user activity and reduced workload intensity outside regular institutional schedules.  
Figure 6. Typical weekly data center energy consumption measured at the three-phase service  
entrance during a 7-day campaign (weekday vs weekend operation).  
3.3. Energy Balance and PV Contribution  
The energy balance compares the annual PV production of the 35 kWp rooftop plant against  
the electricity demand supplied at the same low-voltage panel. Figure 7 summarizes this balance  
by displaying PV generation and data center consumption together, which supports  
interpretation of how the renewable contribution changes as institutional workload evolves over  
time. While PV production remains within the same order of magnitude across the monitoring  
horizon, data center demand shows a step increase starting in 2020 and remains higher  
thereafter. Consequently, the relative contribution of PV declines in years with higher IT  
activity, even when PV yield is stable.  
InGenio Journal, 9(2), 107–120  
115  
Figure 7. Annual energy balance between PV generation and data center electricity  
consumption (2018–2024).  
Following the indicators defined in Section 2, PV coverage of the total supplied demand  
(Microgrid Laboratory + data center) ranges from 52.8% (2018) to 15.6% (2024). The decline  
after 2019 is explained mainly by growth in electricity demand rather than by a persistent  
reduction in PV output. Over 2018–2024, cumulative PV generation reached 240.5 MWh,  
representing on-site renewable electricity produced during institutional operation.  
Results for 2024 require careful interpretation. That year coincided with an unusual national  
operating context in Ecuador characterized by extended scheduled power outages. Because the  
PV system is grid-tied and battery-less, inverter anti-islanding protection disconnects PV  
generation during grid interruptions, which can reduce recorded PV energy even when solar  
resource is available. During these periods, data center continuity relied mainly on on-site  
thermal generator sets. For this reason, 2024 is treated as an outlier year in the discussion, and  
results are also interpreted using the pre-crisis years (2018–2023) as a reference baseline.  
The observed PV contribution reduces grid electricity purchases and associated operating  
costs and increases avoided emissions. The corresponding economic and environmental impacts  
are quantified in the subsequent subsections using the tariff and emission-factor assumptions  
described in the Methods section.  
3.4. Emission Reduction  
Avoided carbon emissions were estimated using a displacement approach in which PV  
electricity used on-site offsets an equivalent amount of electricity that would otherwise be  
supplied by the public grid. Consistent with Section 2, avoided emissions over a given  
accounting period are computed according to Equation 2:  
(2)  
퐸푎ꢀꢁꢂꢃꢄꢃ = 퐸푃ꢅ,푢ꢆꢄꢃ × 퐸ꢇ  
푔ꢈꢂꢃ  
where 퐸푃ꢅ,푢ꢆꢄꢃ is the PV energy consumed behind the meter and 퐸ꢇ푔ꢈꢂꢃis the grid-average  
emission factor (kg CO2/kWh). As the PV system operates in a battery-less, grid-tied  
configuration and the supplied loads are large relative to PV capacity, self-consumption is close  
to 100% during normal operation. Therefore, for annual accounting, 퐸푃ꢅ, used is approximated  
by the total measured PV generation when export is negligible. If export data are available, the  
same formulation applies by subtracting exported energy from total generation.  
Baseline avoided emissions using official Ecuadorian factors  
Grid emission factors were obtained from the annual reports published by Ecuador’s  
electricity operator (CENACE) and environmental authorities (MAATE), following the GHG  
InGenio Journal, 9(2), 107–120  
116  
Protocol guidance for institutional inventories. These factors reflect the average carbon intensity  
of the national electricity mix (Sistema Nacional Interconectado, SNI) and are suitable for  
location-based emission accounting in public-sector facilities. For the 2018–2024 period, this  
study adopts an estimated range of 0.27–0.30 kg CO2/kWh to reflect interannual variability and  
reporting conventions [11], [12], [13].  
Table 4 reports the resulting annual range, computed by applying the lower and upper  
emission-factor bounds to each year’s PV generation. Over 2018–2024, cumulative avoided  
emissions fall within the range given by Equation (3):  
[
]
(3)  
퐸푎ꢀꢁꢂꢃꢄꢃ,푡ꢁ푡 휖 64,9, 72,1 ꢅ퐶ꢆ2  
which represents the net reduction in emissions associated with PV-based self-consumption  
under national grid conditions.  
Table 4. Avoided emissions estimated from annual PV generation using the national emission-  
factor range (0.27–0.30 kg CO2/kWh).  
Avoided emissions (tCO2)  
Year  
PV used (kWh)  
0.27 kg/kWh  
9.36  
0.30 kg/kWh  
10.40  
9.90  
2018  
34 665.1  
33 004.5  
34 309.6  
29 545.2  
47 289.0  
37 391.7  
24 257.0  
240 462.1  
2019  
2020  
2021  
2022  
2023  
2024  
Total  
8.91  
9.26  
10.29  
8.86  
7.98  
12.77  
10.10  
6.55  
14.19  
11.22  
7.28  
64.92  
72.14  
Interannual variability and interpretation (including 2024)  
Year-to-year variation in avoided emissions is driven primarily by changes in PV  
generation. In this dataset, the highest annual reduction occurred in 2022 (14.2 tCO2), while the  
lowest corresponds to 2024, which reflects an atypical operating context. During that year,  
Ecuador experienced extended scheduled grid outages related to a national electricity supply  
crisis. Because the PV system is grid-tied, inverter anti-islanding protection prevented operation  
during interruptions, even under available solar resource. Consequently, 2024 results do not  
represent standard PV performance under continuous grid availability. During those periods,  
data center operation relied mainly on on-site thermal generators, which fall outside the  
emission displacement calculation. For consistency, the Discussion section treats 2024 as an  
outlier and uses 2018–2023 as a baseline for typical grid-tied operation.  
Sensitivity to higher carbon intensity scenarios  
To assess the influence of grid carbon intensity, the avoided-emissions estimate was  
recalculated using a more conservative scenario aligned with thermal-dominated generation  
contexts. Under 퐸ꢇ  
= 0.40 kg CO2/kWh, cumulative avoided emissions rise to the value  
푔ꢈꢂꢃ  
given by Equation 4:  
(
)
(4)  
퐸푎ꢀꢁꢂꢃꢄꢃ,푡ꢁ푡 0.40 = 96.2 ꢅ퐶ꢆ2,  
InGenio Journal, 9(2), 107–120  
117  
illustrating the increased environmental benefit of PV when the displaced grid electricity has  
higher emission content. This sensitivity analysis supports comparisons across regions and  
underscores the relevance of PV integration under evolving generation mixes.  
4. DISCUSSION  
This study quantified the long-term interaction between a rooftop PV plant and an  
institutional data center operating continuously since 2018. The results offer a practical  
perspective on sustainability outcomes in a university setting: a measurable share of electricity  
demand is supplied by onsite renewables, PV generation is largely self-consumed due to the  
persistent base-load behavior of the data center, and avoided emissions can be estimated under  
the national electricity mix using recognized accounting practices. The use of complementary  
indicators such as Green Share (GS) and Renewable Energy Utilization Rate (REUR) (Murino  
et al., 2023) distinguishes between the share of demand met by renewables and the fraction of  
PV generation consumed onsite. This distinction is especially relevant in the LMR  
configuration, where the PV system supplies a shared low-voltage panel and the data center  
dominates the total demand. Additionally, the dedicated three-phase measurement campaign at  
the data center entrance provides a reference for typical weekday–weekend operation and  
supports load disaggregation.  
The PV–demand balance shows that the decline in PV coverage after the initial years is  
primarily driven by increased consumption (not reduced output), particularly following a step  
change in 2020. This is consistent with institutional growth through virtualization, storage  
expansion, and service consolidation. The analysis also clarifies why real operating demand is  
often below nameplate ratings: IT power depends on utilization levels and rack population, and  
precision A/C units modulate input based on thermal loads and setpoints, with redundancy  
strategies keeping some units partially loaded or on standby.  
The operating conditions in 2024 must be interpreted separately. Extended scheduled  
outages during Ecuador’s electricity crisis led to a temporary loss of PV contribution due to  
anti-islanding protections in the battery-less, grid-tied configuration. During these periods, the  
facility relied on thermal backup generators, which are excluded from the emission reduction  
calculation. This contrast highlights the difference between sustainability under normal grid  
operation and resilience during supply disruptions.  
Regarding emissions, while the grid in Ecuador has traditionally been low in carbon  
intensity due to hydropower dominance, cumulative avoided emissions remain relevant over  
multi-year operation. The estimates in this study are consistent with institutional reporting  
frameworks such as the GHG Protocol, using official annual grid-average emission factors  
published by CENACE and MAATE [11], [12]. The sensitivity analysis shows that avoided  
emissions increase in proportion to the grid’s carbon intensity, reinforcing the environmental  
value of PV as generation mixes evolve.  
The findings support two upgrade pathways that could enhance both indicator performance  
and operational resilience: (i) incorporating energy storage to increase PV utilization through  
time-shifting and to enable autonomous operation during outages via islanding-capable control;  
and (ii) scaling PV capacity to match post-2020 demand growth and recover earlier coverage  
levels as digital workloads continue expanding.  
Limitations and opportunities for future work are tied to data granularity and operational  
traceability. Extending submetering to separate IT loads, cooling systems, and auxiliary circuits  
would allow more refined energy allocation and enable metrics such as Carbon Usage  
Effectiveness (CUE). Incorporating inverter status logs and maintenance records would improve  
interpretation during abnormal periods. Beyond hardware upgrades, future studies may explore  
workload-aware scheduling and cooling setpoint optimization, supported by data center  
InGenio Journal, 9(2), 107–120  
118  
infrastructure management (DCIM) tools aligned with institutional service delivery and energy  
availability.  
5. CONCLUSIONS  
This work evaluated seven years of operation (2018–2024) of a university data center  
partially supplied by a 35 kWp rooftop PV plant. Historical energy records, PV monitoring data,  
and local environmental measurements were combined to characterize long-term performance  
and renewable contribution. The results indicate that on-site PV generation supplies a  
measurable share of the facility’s energy needs and supports sustained avoided emissions. They  
also show that the evolution of demand, in addition to PV yield, determines the renewable  
fraction as institutional digital services expand. The study further clarifies the practical  
difference between nameplate ratings and operating demand in data center environments, where  
IT utilization, cooling control, and redundancy policies lead to power levels below installed  
capacity during normal operation. The operating conditions observed in 2024, marked by  
extended grid outages, illustrate the limitations of a battery-less, grid-tied PV configuration and  
motivate upgrades oriented to resilience.  
Finally, the analyzed facility provides a reference for campus digital infrastructure supplied  
by on-site renewables, and the reported indicators point to improvement pathways through  
enhanced submetering, integration of energy storage to increase PV utilization and support  
operation during outages, and potential scaling of PV capacity to align with growing workloads.  
ACKNOWLEDGMENTS: The authors acknowledge the support of the Microgrid Laboratory  
at the Faculty of Engineering of the University of Cuenca for providing access to the  
experimental infrastructure and for the technical assistance that enabled the measurement  
campaign, data collection, and energy assessment reported in this work. The authors also thank  
the University of Cuenca Information Technologies Directorate (DTIC), led by Pablo Pintado  
Zumba (Acting Director of DTIC until January 2026), for facilitating access to operational  
information and historical records of the data center analyzed in this article.  
REFERENCES  
[1] T. Murino, R. Monaco, P. S. Nielsen, X. Liu, G. Esposito, and C. Scognamiglio,  
“Sustainable Energy Data Centres: A Holistic Conceptual Framework for Design and  
Operations”, Energies, vol. 16, no. 15, pp. 5764, ago. 2023. [Online]. Available:  
https://doi.org/10.3390/en16155764  
[2] N. Hinov, “From Energy Efficiency to Energy Intelligence: Power Electronics as the  
Cognitive Layer of the Energy Transition”, Electronics, vol. 14, no 23, pp. 4673, nov.  
2025. [Online]. Available: https://doi.org/10.3390/electronics14234673  
[3] M. Sarac, S. Adamovic, and D. Stamenkovic, “Experimental Analysis of Energy  
Efficiency of Server Infrastructure in University Datacenters”, Teh. vjesn., vol. 27, no 5,  
oct. 2020. [Online]. Available: https://doi.org/10.17559/TV-20160517155453  
[4] K. L. Veigas, A. Chinnici, D. De Chiara, and M. Chinnici, “Towards Energy Efficiency of  
HPC Data Centers: A Data-Driven Analytical Visualization Dashboard Prototype  
Approach”, Electronics, vol. 14, no 16, pp. 3170, ago. 2025. [Online]. Available:  
https://doi.org/10.3390/electronics14163170  
[5] M. D. Dikaiakos et al., “A cyber‐physical management system for medium‐scale solar‐  
powered data centers”, Concurrency and Computation: Practice and Experienc, vol. 35,  
no 10, pp. e7658, may 2023. [Online]. Available: https://doi.org/10.1002/cpe.7658  
InGenio Journal, 9(2), 107–120  
119  
[6] A. Tryfonos et al., “ENEDI: Energy Saving in Datacenters”, in 2018 IEEE Global  
Conference on Internet of Things (GCIoT), Alexandria, Egypt: IEEE, dic. 2018, pp. 1-5.  
[Online]. Available: https://doi.org/10.1109/GCIoT.2018.8620159  
[7] K. Zhang, P. Wang, N. Gu, and T. D. Nguyen, “GreenDRL: managing green datacenters  
using deep reinforcement learning”, in Proceedings of the 13th Symposium on Cloud  
Computing, San Francisco California: ACM, nov. 2022, pp. 445-460. [Online]. Available:  
https://doi.org/10.1145/3542929.3563501  
[8] Y. Zhou, F. Wei, S. Li, Z. Wang, J. Liu, and D. Yu, “Data center load modeling through  
optimal energy consumption characteristics: A path to simultaneously enhance energy  
efficiency and demand response quality”, Applied Energy, vol. 393, p. 126095, sep. 2025.  
[Online]. Available: https://doi.org/10.1016/j.apenergy.2025.126095  
[9] J. Xing, and B. C. Lee, “Datacenter Demand Response for Carbon Mitigation: From  
Concept to Practicality: Invited Paper”, in 2024 IEEE 15th International Green and  
Sustainable Computing Conference (IGSC), Austin, TX, USA: IEEE, nov. 2024, pp. 142-  
144. [Online]. Available: https://doi.org/10.1109/IGSC64514.2024.00034  
[10] J. L. Espinoza, L. G. González, y R. Sempértegui, "Micro grid laboratory as a tool for  
research on non-conventional energy sources in Ecuador", in 2017 IEEE International  
Autumn Meeting on Power, Electronics and Computing (ROPEC), Ixtapa, Mexico: IEEE,  
nov. 2017, pp. 1-7. [Online]. Available: https://doi.org/10.1109/ROPEC.2017.8261615  
[11] Operador Nacional de Electricidad (CENACE), "Factor de Emisión del CO2 del Sistema  
Nacional  
Interconectado  
del  
Ecuador  
-
Año  
2024".  
[Online].  
Available:  
https://arconel.gob.ec/wp-content/uploads/downloads/2025/09/Factor-de-emision-de-CO2-  
del-Sistema-Nacional-Interconectado-de-Ecuador-Informe-2024_compressed.pdf  
[12] Operador Nacional de Electricidad (CENACE), “Factor de Emisión de CO2 del Sistema  
Nacional Interconectado del Ecuador  
-
Informe 2023”. [Online]. Available:  
https://www.ambienteyenergia.gob.ec/wp-content/uploads/2024/09/Factor-de-emision-de-  
CO2-del-Sistema-Nacional-Interconectado-de-Ecuador-Informe-2023-comprimido.pdf  
[13] Operador Nacional de Electricidad (CENACE), “Factor de Emisión del CO2 del Sistema  
Nacional Interconectado de Ecuador  
-
Informe 2021”. [Online]. Available:  
https://www.cenace.gob.ec/wp-content/uploads/downloads/2022/12/Informe-Factor-de-  
CO2-2021.pdf  
[14] Greenhouse Gas Protocol Initiative, “GHG Protocol Scope 2 Guidance: An Amendment to  
the  
GHG  
Protocol  
Corporate  
Standard”.  
[Online].  
Available:  
https://ghgprotocol.org/sites/default/files/2023-03/Scope%202%20Guidance.pdf  
[15] Operador Nacional de Electricidad (CENACE), “Factor de Emisión del CO2 del Sistema  
Nacional Interconectado de Ecuador Informe 2022”. [Online]. Available:  
-
https://www.ambienteyenergia.gob.ec/wp-content/uploads/2023/08/wp-  
1692720103183.pdf  
Copyright (2026) © Danny Ochoa-Correa y Emilia Sempértegui-Moscoso.  
Este texto está protegido bajo una licencia internacional Creative Commons 4.0. Usted es libre para compartir, copiar y redistribuir el material en  
cualquier medio o formato. También podrá adaptar: remezclar, transformar y construir sobre el material. Ver resumen de la licencia.  
InGenio Journal, 9(2), 107–120  
120