Design and Performance Evaluation of a Photovoltaic Greenhouse as an Energy Hub with Battery Storage and an Electric Vehicle Charger

Abstract

This work presents a photovoltaic greenhouse’s design and performance evaluation as an energy hub in modern agriculture that integrates battery energy storage, an electric vehicle charging station, and non-controlled loads. The greenhouse roof comprises 48 semi-transparent photovoltaic panels with nominal transparency of 20% and 110 W capacity. The control of the photovoltaic greenhouse as an energy hub was approached as an optimization problem with the aim of minimizing the energy purchased from the grid. The simulation results indicate that the system is capable of balancing power transactions within the microgrid, thus enabling electromobility and, at the same time, achieving an average energy saving of up to 41%. Furthermore, it was found that the case of slow charging of the electric vehicle at night was less demanding on the battery system than fast charging during the day in terms of abrupt power transitions and average state of charge of the battery system, 61% vs. 53%, respectively. Empirical results also demonstrated the negative impact of soiling generated by agricultural activity on the performance of solar panels. For a period analyzed of three years, an average annual production loss of 6.8% was calculated.

1. Introduction

The increase in global population and food demand will tighten the competition between different sectors for energy, land, and water use. According to the Food and Agriculture Organization of the United Nations [1], feeding a global population that is expected to reach 9 billion people by 2050 will require a 60% increase in food production, where agriculture plays a key role. Therefore, sustainable agriculture practices are needed to promote water savings, soil conservation, sustainable land management, and conservation of natural resources to alleviate the pressure on the food–energy system. With a focus on the energy problem, the challenges of the agricultural sector are mainly in implementing energy efficiency measures and incorporating non-conventional renewable energy (NCRE) into production processes [2].

Agriculture is highly dependent on electricity and fossil fuels to produce primary foods such as fruits, vegetables, and cereals [3]. Significant electrical energy expenditure occurs in the early stages of the production chain, including pumping, filtering, and injecting water into modernized irrigation systems, purifying and disinfecting water for safe vegetables, and operating greenhouses [4]. In addition, climate change has led to an increase in the cultivation of crops in greenhouses, where the control of ambient conditions, irrigation, and the use of pesticides can be carried out more effectively. This change in the way agriculture is conducted implies an increased use of electricity. This is a major issue, as many agricultural operations take place in rural areas that lack an electricity distribution grid or have one that is unreliable and poor quality [5]. Therefore, diesel generators are often used to meet the demand for electricity, which is not ideal from a sustainability point of view, as fossil fuels should be partially replaced with renewable sources. Here, photovoltaic (PV) technology plays an essential role in creating a more sustainable agricultural system.

The rest of the paper is organized as follows: Section 2 presents and discusses relevant literature related to PV applied to agriculture. From this review, gaps in the knowledge are identified and, therefore, the contribution of our research is introduced. Section 3 presents the methodology used to evaluate the energy performance of the proposed PV greenhouse and the proposed control strategy based on model predictive control. Section 4 presents the empirical results of the energy performance study along with simulation results to analyze the performance of the proposed control strategy. Finally, in Section 5 and Section 6, a discussion of the research findings, conclusions, and future research work is presented.

2. Literature Review

The use of PV technology as an alternative to generate clean and renewable electricity has increased in recent years, particularly in the agricultural sector [6]. This has led to the emergence of agrivoltaics, a new field of research and technology development. Agrivoltaics is focused on the development of integrated systems that allow the simultaneous cultivation of crops and production of PV energy on the same piece of land [7,8,9]. Among the main advantages of agrivoltaics are the optimization of land and water use [10,11], crop protection against rare weather conditions such as severe solar radiation and hail storms, rural electrification, and stimulation of economic growth in the local community. In terms of productivity, research reported in [12] shows that partially covered crops with photovoltaic panels—with a reduction 57% in the received light—reduced their productivity only by 19% and could increase the overall productivity of the soil by 35–73%. A similar result of an increase in global productivity of up to 70% is reported in [13].

One of the most common uses of agrivoltaics is the installation of conventional PV panels in greenhouses [14], known as the photovoltaic greenhouse (PVG). As an example, we can look at China, a pioneering country in the research, innovation, and development of agrivoltaics. At the end of 2020, the installed capacity of photovoltaic agriculture projects connected to the grid in China was approximately 7% of the total installed photovoltaic power capacity (250 GW) [15]. It is clear that PV is becoming an increasingly popular renewable energy source in China and many other countries. This trend is likely to continue, leading to the construction of more PV plants on agricultural land. As a result, agrivoltaics is likely to become a common practice in the future. Hence, to ensure a successful, clean, and sustainable energy–food system, it is essential to understand how to combine PV and agriculture effectively.

Agrivoltaics commonly uses conventional PV panels, which are monofacial, meaning that they only capture photons on the front side where the photovoltaic cells are located [16]. The back of the panel is covered with an opaque protective sheet, which limits the amount of light that can pass through the greenhouse roof. This prevents the entire roof surface from being covered with conventional photovoltaic panels, thus reducing the amount of electricity that can be produced and potentially not meeting the energy demand of the energy-harvesting system. This presents an opportunity for other photovoltaic technologies to be used in greenhouses, either as a replacement or as a supplement to conventional PV systems, to maximize electricity production. The latest advances in solar cell technology make it possible today to reach power conversion efficiencies of about 20% with flexible and lightweight PV modules based on Perovskite [17,18]. These new solar cells are considered the future of agrivoltaics since their lower weight (around 1/10 of the weight of a conventional PV panel) and flexibility allow for the use of lighter structures for mounting the PV system. In addition, recent research has found that Perovskite cells exhibit about 70% transparency in the wavelength range 540–760 nm, which is more suitable for photosynthesis [19]. These results make Perovskite flexible solar modules the perfect candidate to provide protective coverage and electric power generation for greenhouses. However, they are relatively new technology compared to the mature and commercially established silicon-based cells.

Another relevant aspect of PV systems, in general, is quantifying the loss of energy production over time. Although manufacturers provide the nominal efficiency of their equipment and its nominal degradation rate, environmental factors, such as soiling, can cause a greater decrease in production [20]. In the case of a PVG, the solar panels are installed in agricultural settings where machinery regularly move the soil and chemical products are dispersed through the air, increasing the soiling of PV panels. Furthermore, it is assumed that the PV panels are not subjected to scheduled cleaning, as it is not a typical agricultural task and requires water, which is usually scarce.

After reviewing the specialized literature, we have identified three aspects that have received little attention in the research on agrivoltaics. These aspects will be addressed in the present work and are highlighted as the following research contributions:

• Most of the reported research on PV greenhouses entails the usage of roof-mounted conventional solar panels, dealing with optimal arrangement of modules and its effect on the shadow patterns at the inside of the greenhouse [14]. In the present study, we propose to evaluate the use of semi-transparent PV panels [16] as an alternative technology that can be designed to completely cover the roof (even the entire structure) of the greenhouse without producing shadow patterns and then facilitate the design.
• There is a vast amount of research on soiling and its effects on the energy production of conventional photovoltaic systems. However, the semi-transparent panels we propose to evaluate can be affected by soiling on both sides of the solar panel because the photovoltaic material is inserted in a glass–glass sandwich. Furthermore, to the best of our knowledge, no research has been reported on the soiling of semi-transparent photovoltaics applied to a greenhouse in an agricultural environment. This is one of the aspects that we are looking to evaluate in this work.
• The future of a more sustainable agriculture depends on the use of clean energy to increase the level of technology of different rudimentary processes involved in the chain of production. In this context, the PV greenhouse emerges as a potential center of energy transactions. Thus, this research introduces the concept of the PVG as an energy hub (EH) for modern agriculture, where the greenhouse plays a vital role [21,22,23]. Figure 1 illustrates the concept with the PVG at its core and the various tasks, loads, and resources connected to it. The arrows in the diagram show the direction of the electrical power flow between the components.

3. Methodology and Analysis

The greenhouse is located at the Colchagua Campus of the Universidad de O’Higgins (UOH) with the coordinates shown in

Table 1. Gepgraphical coordinates of UOH Colchagua Campus.

Parameter	Value
Latitude	34.6118° S
Longitude	70.9901° W
Elevation	351 m

. It has 48 glass–glass thin film 20% transparency 110 Wp panels that cover its entire roof. The PV panels are the model Lucid+ NB-110AS manufactured by NexPower, Hsinchu, Taiwan, and are arranged in two arrays of twenty-four panels connected to two inverters. The inverters used in this configuration are the model Galvo 2.0-1 manufactured by Fronius, Pettenbach, Austria. The greenhouse is referred to as the semi-transparent photovoltaic greenhouse (ST-PVG). Figure 2 shows an overview of the facilities: in the bottom right, the ST-PVG that is part of the present study; in the center, the PVG with conventional PV panels; and in the top left, a typical tunnel greenhouse. Note that the last two greenhouses are not part of the present study.

Using the online platform Solar Explorer, a simulation was carried out for ST-PVG, considering the inclination of the roof of 23°, an azimuth angle of −8° representing the orientation of ST-PVG, a temperature coefficient for PV panels of −0.45%/°C, an inverter efficiency of 96%, and total losses of 14%. The results are summarized in

Table 2. Estimated electricity generation of the ST-PVG.

Parameter	Value
Installed capacity	5.3 kW
Daily energy	22.2 kWh
Annual energy	8.1 MWh
Plant factor	17%

.

3.1. ST-PVG Energy Performance

The methodology for the analysis of the energy performance includes the following steps:

• Read energy production historical data from inverter manufacturer web platform.
• Compare the actual annual production of the ST-PVG with the theoretical annual production.
• Calculate the difference between the actual data and the theoretical estimate.
• Calculate the loss of energy production according to (1).

Energy production data from the same months but different years were compared to find the year-to-year trend in the difference in energy production. The average difference was then calculated according to the following equation:

$$E_{\mathrm{diff}}^{\mathrm{month}}=(E_{\mathrm{year}+N}^{\mathrm{month}}-E_{\mathrm{year}}^{\mathrm{month}})/N,$$

(1)

where $E_{\mathrm{year}}^{\mathrm{month}}$ is the energy produced for a specific month of a specific year, $E_{\mathrm{year}+N}^{\mathrm{month}}$ is the energy produced for the same specific month, but N years after, then $E_{\mathrm{diff}}^{\mathrm{month}}$ is the difference of the energy produced in the months previously indicated.

3.2. ST-PVG as Energy Hub

The methodology for analyzing the performance of the ST-PVG as an energy hub entails applying Model-Based Predictive Control (MPC) [24] to the microgrid [25] conformed by the elements and configuration shown in Figure 3: lighting, irrigation, and air conditioning as non-controllable loads, a PV system as non-controllable generation, and an electric vehicle (EV) charging station and a battery energy storage system (BES) as controllable loads and generation. The power flows related to the controllable loads, namely, the battery system and EV charging station, are denoted by $P_{BES}$ and $P_{EV}$, respectively. Regarding the non-controllable loads, campus lighting and greenhouse loads, their power flows are denoted $P_{Light}$ and $P_{GH}$, respectively. Finally, there is PV generation and a connection to the main grid, where the powers that these deliver to the microgrid are $P_{PV}$ and $P_{Grid}$, respectively. All power quantities are expressed in kilo-Watts, [kW].

3.2.1. MPC Problem Formulation

The discrete-time expressions determining the dynamics of the EV charging station and BESS are described by

$$\begin{array}{ccc}\hfill SO{C}_{BES}(k+1)& =& SO{C}_{BES}\left(k\right)+P_{BES}\left(k\right)\frac{\Delta k}{C_{BES}}\hfill \end{array}$$

(2)

$$\begin{array}{ccc}\hfill SO{C}_{EV}(k+1)& =& SO{C}_{EV}\left(k\right)+P_{EV}\left(k\right)\frac{\Delta k}{C_{EV}}\hfill \end{array}$$

(3)

$$\begin{array}{ccc}\hfill b^{av}\left(k\right)& =& \left\{\begin{array}{cc}1,\hfill & \mathrm{if}{t}^{in}\le k\Delta k\le t^{out}\hfill \\ 0,\hfill & \mathrm{otherwise}\hfill \end{array}\right.\hfill \end{array}$$

(4)

Equations (2) and (3) are based on the principle of energy conservation, since, in both cases, the state of charge for the next instant is determined from the sum of the current state of charge and power. On the other hand, expression (4) models the availability of the EV through a binary variable, which is later used to define the operational limits of the EV.

With respect to (2), it models the evolution of the energy state of the BES subsystem. Therefore, in the said expression, $P_{BES}\left(k\right)$ represents the power that enters or leaves the subsystem at instant k, $SO{C}_{BES}\left(k\right)$ symbolizes the state of charge of the batteries, $C_{BES}$ is the nominal storage capacity of the batteries, and $\Delta k$ is the discretization time-step.

Regarding (3), it describes the dynamics of the energy state of the cars that use the charging station. For this reason, $SO{C}_{EV}\left(k\right)$ represents the state of charge of the EV at instant k, $P_{EV}\left(k\right)$ the power that enters or leaves the vehicle’s battery at instant k, and $C_{EV}$ the nominal storage capacity of the EV.

With respect to (4), this models the availability to load/unload that the EV has at a certain instant k. For this reason, $b^{av}\left(k\right)$ is denoted as the binary variable that represents said availability, which is equal to 1 if the instant k is between $t^{in}/\Delta k$ and $t^{out}/\Delta k$ or 0 otherwise. In other words, a value of 1 represents that the car is connected and, therefore, available.

The quantities involved have the following units of measurement: the states of charge $SO{C}_{BES/EV}$ of batteries are measured in per unit (p.u.), nominal storage capacity of batteries $C_{BES/EV}$ is measured in kilo-Watt-hour (kWh), and discretization time-step $\Delta k$ is in the order of minutes but is expressed in units of hours (h) to cancel out the hours in the quotients $\Delta k/C_{BES/EV}$.

BES and EV charging dynamics have physical and operational limitations, which are represented as follows:

$$\begin{array}{ccc}\hfill SO{C}_{BES}^{min}\le & \hspace{1em}SO{C}_{BES}\left(k\right)& \hspace{1em}\le SO{C}_{BES}^{max}\hfill \end{array}$$

(5)

$$\begin{array}{ccc}\hfill SO{C}_{EV}^{min}\le & \hspace{1em}SO{C}_{EV}\left(k\right)& \hspace{1em}\le SO{C}_{EV}^{max}\hfill \end{array}$$

(6)

$$\begin{array}{ccc}\hfill SO{C}_{EV}^{req}\le & SO{C}_{EV}(t^{out}/\Delta k)& \le SO{C}_{EV}^{max}\hfill \end{array}$$

(7)

where (5) and (6) model the maximum and minimum limits for the battery bank of the BES subsystem and electric cars, respectively. Meanwhile, (7) models the output condition that the EV must meet to ensure a desired energy state at the instant $t^{out}/\Delta k$ when it is disconnected. It can be seen that (5) and (6) limit the energy stored in the batteries of the BES and the EV respectively. On the other hand, (7) guarantees that when disconnecting the EV from the charging station, it has to reached a desired energy state $SO{C}_{EV}^{req}$.

On the other hand, the restrictions associated with the power that BES and EV can deliver are defined as follows:

$$\begin{array}{ccc}\hfill -P_{BES}^{max,dis}\le & P_{BES}\left(k\right)& \le P_{BES}^{max,char}\hfill \end{array}$$

(8)

$$\begin{array}{ccc}\hfill -P_{EV}^{max,dis}{b}^{av}\left(k\right)\le & P_{EV}\left(k\right)& \le P_{EV}^{max,char}{b}^{av}\left(k\right)\hfill \end{array}$$

(9)

where (8) and (9) represent the maximum power that can be delivered and consumed by the battery energy storage and the EV charging station, respectively.

Now, the power balance of the greenhouse microgrid is established. The sign of each power term in this balance is given by the directions of the energy arrows depicted in Figure 3, where those that consume power have a negative sign, while those that produce power have a positive sign. The power balance is written as follows:

$$P_{PV}\left(k\right)+P_{Grid}\left(k\right)-P_{BES}\left(k\right)-P_{EV}\left(k\right)-P_{Light}\left(k\right)-P_{GH}\left(k\right)=0$$

(10)

where $P_{PV}$ represents the power generated by the PV panels, $P_{Grid}$ the power provided by the electrical grid, $P_{BES}$ the power received or consumed by the BES subsystem, $P_{EV}$ the power received or consumed by the charging station, $P_{Light}$ the power consumed by the campus lighting system, and $P_{GH}$ the power consumed by the greenhouse loads (air conditioning and irrigation).

Based on the MPC methodology, the control of the ST-PVG as an energy hub was approached as an optimization problem with the aim of minimizing the cost of the energy used. The problem was solved at each sampling instant for a horizon K, where the first term of the control sequence is applied to the system. Subsequently, the process was repeated at the next sampling instant. Based on the above, the optimization problem was modeled as follows:

$$\begin{array}{cc}{min}_{k\in \{0,...,K-1\}}& J\left(k\right)\hfill \end{array}$$

(11)

Subject to:

$$\begin{array}{ccc}& \left(2\right),\left(3\right),\left(4\right)& \mathrm{Dynamics}\hfill \\ & \left(5\right),\left(6\right),\left(7\right)& \mathrm{Operational}\hfill \\ & \left(10\right)& \mathrm{Power}\mathrm{balance}\hfill \end{array}$$

where the decision variables are the power from the battery bank and the power from the EV charging station. On the other hand, the cost function (12) has three terms, with the first representing the cost of the energy obtained from the grid, while the other two quadratic terms penalize large excursions of $P_{Grid}$ and $P_{EV}$ smoothing out their profile over time.

$$\begin{array}{c}J\left(k\right)=\sum _{i=1}^K\{P_{Grid}(k+i)c_e(k+i)+{(P_{Grid}(k+i)-P_{Grid}(k+i-1))}^2+\hfill \\ {\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}(P_{EV}(k+i)-P_{EV}(k+i-1))}^2\}\hfill \end{array}$$

(12)

3.2.2. Configuration of Parameters and Inputs

This section describes the parameters and inputs of the MPC problem.

A plain tariff was considered to deduce the final price of the electricity consumed from the grid. The tariff was obtained from information published in February 2023 by the local distribution company. The information is summarized in

Table 3. Electricity tariff.

Charge	$ CLP/kWh
Energy	85.285
Public service	2.526
Transmission	26.865
Total	114.676

.

The greenhouse generates electricity through its rooftop photovoltaic system. The data on electricity generation were then obtained from the inverter provider monitoring platform. Figure 4 shows the curves corresponding to photovoltaic generation of each season of the year (southern hemisphere), where the data correspond to one typical day of January for summer (Figure 4a), April for fall (Figure 4b), July for winter (Figure 4c), and October for spring (Figure 4d).

The greenhouse uses electricity for air conditioning (temperature control), lighting, and irrigation. In the following sections, the electricity consumption of these systems is estimated. The greenhouse has four air conditioning units (ac) of 18,000 BTU thermal rating each to perform temperature control. Each unit has an electricity consumption of 1.696 kW for cooling mode and 1.464 kW for heating mode. To determine the electricity consumption profile of the ac units together, it is necessary first to determine whether the ac units operate in heating and cooling mode. To do that, we first defined the greenhouse comfort zone, where the temperature inside the greenhouse is allowed to vary in the range of 14°C to 28° C. Then we compared the comfort zone with historical data on the average ambient temperature at the greenhouse location, shown in

Table 4. Historical seasonal average ambient temperature (°C) per hour of the day at the greenhouse location.

	00	01	02	03	04	05	06	07	08	09	10	11	12	13	14	15	16	17	18	19	20	21	22	23
Summer	17.7	16.9	16.2	15.6	15.0	14.3	13.8	14.3	15.5	17.2	19.4	21.8	24.1	26.0	27.6	28.6	28.9	28.1	26.8	24.8	22.7	21.1	19.9	18.7
Fall	11.9	11.3	10.7	10.1	9.7	9.2	8.7	8.3	8.9	10.4	12.3	14.6	16.7	18.6	20.2	20.9	20.8	19.8	17.7	16.3	15.1	14.2	13.3	12.5
Winter	6.7	6.3	6.1	5.8	5.6	5.4	5.1	4.8	4.8	5.6	6.6	8.1	9.6	11.0	12.0	12.3	12.3	11.1	10.1	9.4	8.8	8.4	8.0	7.6
Spring	11.5	10.8	10.3	9.8	9.3	8.8	8.3	8.5	9.4	10.9	13.0	15.2	17.1	18.6	19.8	20.4	20.4	19.7	18.4	16.5	15.1	13.9	13.0	12.2

. Note that we have marked in blue the temperatures above the comfort zone and in red the temperatures below the comfort zone. When the average ambient temperature is above the upper limit, the ac units must operate in cooling mode, consuming 3.4 kW of electricity per hour. On the contrary, when the average ambient temperature is below the lower limit, the ac units must operate in heating mode, consuming 2.9 kW of electricity per hour.

An irrigation system consisting of two water pumps with a power of 0.65 kW was considered. The lighting system consists of eight 32 W LED lamps for interior lighting and four 50 W LED lamps for exterior lighting. Based on sunrise and sunset times and irrigation requirements, the following is the scheduled daily operation:

• Summer (Sunrise 06:35 and Sunset 20:55)

  –

   Irrigation: Switched on from 08:00 to 08:30 and from 20:00 to 20:30.

  –

   Interior lighting: Not used.

  –

   Exterior lighting: Switched on from 00:00 to 06:35 and from 20:55 to 23:55.

• Fall (Sunrise 07:55 and Sunset 19:40)

  –

   Irrigation: Switched on from 08:00 to 08:30 and from 20:00 to 20:30.

  –

   Interior lighting: Switched on from 19:40 to 20:30.

  –

   Exterior lighting: Switched on from 00:00 to 07:55 and from 19:40 to 23:55.

• Winter (Sunrise 07:50 and Sunset 17:55)

  –

   Irrigation: Switched on from 08:00 to 08:30.

  –

   Interior lighting: Not used.

  –

   Exterior lighting: Switched on from 00:00 to 07:50 and from 17:55 to 23:55.

• Spring (Sunrise 07:20 and Sunset 19:45)

  –

   Irrigation: Switched on from 08:00 to 08:30 and from 20:00 to 20:30.

  –

   Interior lighting: Switched on from 19:45 to 20:30.

  –

   Exterior lighting: Switched on from 00:00 to 07:20 and from 19:45 to 23:55.

The greenhouse total load profile due to temperature control, irrigation, and lighting is shown in Figure 5. Note that the plotted data represent the average hourly power consumption of the greenhouse for a typical day of each season of the year. Later, when the system is simulated, this daily behavior must be replicated for each day of the corresponding season.

Another consumption considered part of the microgrid is the lighting of the main campus building. The lighting equipment entails sixteen 34 W LED lamps for each classroom and forty-four 80 W LED lamps for public spaces. Based on sunrise and sunset times and educational building requirements, a scheduled daily operation was designed to achieve proper illumination. The resulting load profile of the main campus building lighting is summarized and shown in Figure 6.

Finally, the total consumption corresponding to the main campus building was added to the total greenhouse consumption to generate the total electricity demand of the microgrid. Figure 7 shows the resulting daily load profile.

The EV charger considered is the model Terra AC Wallbox from manufacturer ABB, Zurich, Switzerland; it allows for a maximum and minimum charging power of 11 kW and 3 kW, respectively. The battery system considered is the model AXE 20.0L from manufacturer Growatt, Shenzhen, China, which has a storage capacity of up to 20 kWh and maximum charging and discharge power of 8.64 kW. For both devices, minimum and maximum SOC of 0.2 and 0.8 were considered, respectively. Parameters of BES are summarized in

Table 5. BES charging profile parameters.

Parameter	Value
$C_{BES}$	20 kWh
$SO{C}_{BES}^{min}$	0.2
$SO{C}_{BES}^{max}$	0.8
$P_{BES}^{max,dis/char}$	8.64 kW

. In the following sections, two scenarios are described for the use of the EV charging station.

4. Results

4.1. ST-PVG Energy Performance

This section reports the empirical results of the energy performance of the ST-PVG under study. Figure 8 shows the overall results of the energy production analysis. Figure 8a shows a summary of the predicted monthly energy production using historical data from 2004 to 2016 (blue bars) and measured data from 2019 to 2021 (orange bars). Figure 8b shows the difference between the measured energy production and the predicted production, along with its trend. The average obtained from the differences was −57.4 kWh. This value indicates that, on average, the ST-PVG produces 57.4 kWh less energy than theoretically predicted for each month. Moreover, the trend curve indicates that the negative difference tends to increase over time, showing that the ST-PVG would tend to produce less energy each month than theoretically expected.

Figure 8c shows the data plot used to characterize the energy production loss. The green bars represent the difference calculated with Equation (1), with the average production loss being −43.8 kWh/month. This monthly average adds up to an estimated annual loss of −525.5 kWh, which represents 6.8% of the energy produced in 2019. A summary of the main results obtained in this section is presented in

Table 6. Summary of ST-PVG annual performance.

Parameter	Value
Installed power	5.3 kW
Theoretical production	8.1 MWh
Measured production	7.72 MWh
Energy loss	0.526 MWh
Energy loss	6.8%

.

4.2. ST-PVG as Energy Hub

This section reports the results obtained from simulations of the proposed MPC strategy for the microgrid of the energy hub, considering the consumption and generation curves during the summer. Simulations were carried out with a discretization time of 5 min, an optimization horizon of 1 day (24 h window), and a prediction horizon of 3 days (72 h window). The simulations were carried out in Matlab_® r2022b, which is a programming and numeric computing platform, and using Gurobi 10.0.1, a commercial optimization software, for solving the optimization problem for the control of the PV greenhouse.

Two cases were simulated. Case 1 considers that a personal vehicle will use the EV charging station during working hours. The vehicle has a storage capacity ($C_{EV}$) of 40 kWh. The car arrives at the station at 08:30 with a $SO{C}_{EV}^{in}$ close to 0.3 and leaves the station at 17:30 (9 h of availability). This usage profile is repeated for every day of the simulation. In Case 2, an institutional vehicle of 54.9 kWh capacity will be considered to use the charging station under two charging regimes: from 17:00 to 23:00, the car is allowed to charge and discharge freely, and from 23:00 to 08:00, the car is allowed to charge only at constant power. As in Case 1, this charging behavior is repeated for every day simulated.

Table 7. EV charging profile parameters. Case 1: Personal vehicle. Case 2: Institutional vehicle.

Parameter	Case 1	Case 2
$t_{EV}^{in}$	08:30 h	17:00 h
$t_{EV}^{out}$	17:30 h	08:00 h
$C_{EV}$	40 kWh	54.9 kWh
$SO{C}_{EV}^{in}$	∼0.3	∼0.35
$SO{C}_{EV}^{req}$	1.0	1.0
$P_{EV}^{max,dis/char}$	11 kW	11 kW
$P_{EV}^{min,dis/char}$	3 kW	3 kW

summarizes the parameters for both study cases.

Figure 9 shows the results of the first 24 h for Case 1. Figure 9a shows the power flows of the whole system. Here it can be seen that the EV is charged at maximum power as soon as it is plugged into the microgrid. With regard to power transactions with the grid, it is observed that the microgrid consumes power during the morning and at night. In addition, it can be observed that the BES absorbs all the extra power demands and surplus power.

Figure 9b shows the power demanded by the EV and the evolution of its SOC over time. As mentioned above, the EV is charged in around three hours at maximum power, as expected. It is important to mention that before and after EV charging, there is no knowledge of the evolution of the SOC. This is marked as a gray zone in Figure 9b, where the SOC is kept constant only for simulation purposes.

Finally, Figure 9c shows the power flow through the BES and the evolution of its SOC over time. It can be observed that the most demanding operations for the BES are the charging of the EV, which empties the BES in nearly 2 h, and the storage of PV surplus power that recharges the BES in approximately 4 h.

Figure 10 shows the results of the first 24 h for Case 2. The general behavior is very similar to Case 1, as seen in Figure 10a. The EV is also charged at the maximum power allowed when connected to the microgrid. However, in this case, the power limit is much lower and it charges at night. Regarding the power transactions with the grid, it is observed that the microgrid also consumes power during the morning and at night. However, in this case, the power consumed during the night is higher due to the continuous charging of the EV.

Figure 10b shows the power demanded by the EV and the evolution of its SOC over time. As in Case 1, it is important to mention that between 08:00 and 23:00 h, there is no knowledge of the evolution of the SOC. This is marked as a gray zone in Figure 10b, where the SOC is kept constant only for simulation purposes.

Finally, Figure 10c shows the flow of power through the BES and the evolution of its SOC over time. Unlike in Case 1, the usage of the BES is less harsh and its average SOC is higher (61% vs. 53%). It can be observed that the most demanding operation is the storage of surplus PV power that recharges the BES in approximately 4 h. In the last part of the EV charging (from 06:00 until 08:00), the BES is emptied in nearly 2 h, but this is because this portion of the charging period overlaps with the power demanded by the campus building.

4.3. Energy Savings

Based on the simulation results, it is possible to estimate and compare energy savings for both cases. We have chosen to calculate in terms of energy since we have considered a plain energy tariff as indicated in Table 3. First, the numerical integration of curves from Figure 9a and Figure 10a is performed to calculate the energy balance.

Table 8. Microgrid daily energy balance (kWh).

System	Case 1	Case 2
PV	34.37	34.37
Loads	−58.57	−58.57
BES	2.36	−6.14
EV	−27.2	−36.23
Grid	49.04	66.57

compares the result for both cases.

Now, for the calculation of monthly energy savings, we considered a typical month of 30 days, of which 22 are working days. In these working days, the EVs charge from the microgrid. During the remaining eight non-working days, the EVs are not present in the microgrid and the balance is due to only non-controllable loads and PV generation.

Table 9. Calculation of monthly energy savings with microgrid.

Parameter	Case 1	Case 2
Working day energy consumption (kWh)	85.77	94.79
Working day energy from grid (kWh)	49.04	66.57
Nr. working days	22	22
Non-working day energy consumption (kWh)	58.57	58.57
Non-working day energy from grid (kWh)	24.2	24.2
Nr. Non-working days	8	8
Monthly energy consumed (MWh)	2.36	2.55
Monthly energy purchased from grid (MWh)	1.27	1.66
Monthly energy savings (MWh)	1.09	0.89

summarizes the results.

5. Discussion

As mentioned in the literature review, most of the reported research deals with conventional PV panels integrated into greenhouses. However, there is solid evidence that new PV technology has great potential to complement and even replace conventional PV in greenhouses. On the other hand, the ambient conditions of an agricultural setting pose a big question mark for the actual performance of a PV system. The latter has been less studied for the specific case of greenhouses, where there is also an enclosed atmosphere that affects panels from the inside of the greenhouse. In this sense, one of the main results of our research is the empirical verification of the loss of production for the proposed ST-PVG. An average production loss from one period to another of 6.8% was calculated for the period analyzed. This tendency to lose production is mainly governed by the inevitable progressive loss of efficiency of the PV system and by the dirt accumulated on the panels. As highlighted earlier, this empirically determined production loss factor is essential because it represents a realistic figure to estimate the production of a PVG under actual operating conditions and without considering routinary panel cleaning. However, because of the scope of the experiment, the results can only be used to explain the past performance of the proposed system. Therefore, the results should be compared with future energy production measurements to validate their ability to predict future performance.

Another relevant aspect highlighted in the literature is that the future of a more sustainable agriculture depends on using clean energy to increase the level of technology throughout the production chain. Here the role of the PV greenhouse is crucial to achieve the sustainability goal. In this sense, our research contributes to better understanding and evaluating the potential of the PVG as an energy hub for modern agriculture, focusing on integrating the EV charging capacity and minimizing the energy purchased from the grid. Two cases of different EV charging profiles were considered: a private car and an institutional car. Simulation results indicate that in both scenarios, the BES is capable of balancing power transactions within the microgrid. The case of the institutional car is less demanding on the BES than the private car in terms of abrupt power transitions and the average SOC, which resulted to be 61% in the case of the institutional car versus 53% for the case of the private car. However, for the private EV case, energy savings of 46.2% were achieved, compared to the institutional EV case, where energy savings were 34.9%. Despite the small differences between cases, these results show that significant energy savings can be achieved at the same time EVs are incorporated into the microgrid, highlighting the potential of PVGs to enable electromobility in agricultural settings. Finally, it must be noted that different assumptions were made to lower the complexity of the MPC formulation, such as considering a plain energy tariff and ideal charging and discharging characteristics of the EV and BESS systems.

6. Conclusions and Future Research Work

This work reports the results of a project consisting of designing and evaluating a photovoltaic greenhouse as an energy hub in modern agriculture. The system also integrates battery energy storage and an electric vehicle charging station. Empirical results from three years of study indicate an average annual energy production loss of 6.8%. On the other hand, the simulation results of the entire system controlled with a model-based predictive control strategy demonstrate the capacity of the system to accommodate EV charging achieving at the same time an average energy saving of up to 41%.

Future research work should consider incorporating the following effects to improve the model’s quality and its predictions: non-linear charging and discharging of BES and EV, differential energy tariff, and simulations for longer periods of time and different seasons of the year.