Abstract

Soiling of photovoltaics (PV) panels is affected by various factors such as relative humidity, dust concentration, and panel tilt angle. The soiling can lead to significant losses in electricity production, especially in a place like Dubai, UAE. Soiling can also lead to long-term damage of the PV panels such as degradation and delamination due to the hot spots caused by dirt deposition. It is important to choose the right cleaning strategy (method and frequency) to maximize the electricity production and economic performance of the PV facility. An optimization algorithm was developed and tested for multiple PV panel configurations based in Dubai Water and Electricity Authority’s (DEWA) outdoor test facility (OTF) solar lab. The algorithm’s input included electricity production, soiling rates (SRs), electricity price, and cleaning costs. The output included number of cleaning events and the extra revenue as compared with the current practice of periodic (5-day cycle) manual cleaning. Four different cleaning scenarios were tested and compared with the current scenario. Three scenarios resulted in improved net cost benefit (NCB), up to 34% for the case of performance-based manual cleaning. The fourth scenario resulted in diminished NCB, down by 245% for the case of daily automatic cleaning. Other findings of the study included higher tilt angles that resulted in lower cleaning requirements and thin-film PV panels that required less cleaning than first generation PV panels (mono/polycrystalline). The algorithm is an effective yet simple tool to help operators optimize the NCB of their PV facilities.

1 Introduction

A crucial maintenance aspect of photovoltaics (PV) panels is dust management, as PV panels are affected by it in the short term and long term. In the short term, performance output is reduced by dust as it acts as an obstruction, in the form of both hard and soft obstruction, to incoming solar irradiation [1]. In the Gulf Cooperation Council (GCC) region and specifically in the UAE, lower optimum tilt angles may actually promote more dust accumulation as dust will not slide off the panels as easily as when tilt angles are higher [2]. In the long term, PV panels can be degraded permanently by soiling through the formation of hot spots as well as through erosion of the panels both of which would reduce the panels: overall lifecycle and age expectancy. When combined with humidity, soiling can turn into cementation as well as discoloration and delamination may occur [3].

The cleaning of PV panels, especially when addressing largescale applications, can be a demanding and challenging task. Current cleaning practices favor periodic-based cleaning frequency, e.g., Jiang et al. [4] suggested a 20-day-cleaning cycle. Still, period cleaning does not account for daily or seasonal changes in climatic conditions which have an impact on the need for cleaning the PV panels. Thus, fixed cleaning cycles can result in cleaning a panel when this is not needed or losing electricity production potential by not cleaning dirty panels because it is not due to cleaning. Using a performance-based criterion, which is linked to deviation in the PVs from the expected, will ensure that the panels are only cleaned when needed. The trigger for such a performance-based cleaning practice needs to strike the proper economical balance between cleaning costs and production losses due to soiling. Hence, different cleaning approaches need to be evaluated as they all have different associated costs with them. This includes both the cleaning technology, manual or automated, as well as the percentage drop in panel performance that triggers the cleaning action.

Calculating the optimum frequency of different cleaning approaches based on both technical and economic bases will, in turn, maximize the return on investment. Furthermore, in PV plants and installations, it can enhance their cleaning schedules by possibly reducing the number of cleaning events, which can be demanding when taking into account the sheer scale. Reduced cleaning cycles can also extend the life of the panels by reducing the potential scratches that might arise from the cleaning process. Moreover, this can further aid in developing a database to ensure that PV operators hire an adequate staff number as well as procuring the most effective equipment for the job.

2 Literature Review

2.1 Cleaning Schedule.

Sayyah et al. [5] argued that schedules are based on the costs of labor and water, which would allow a calculation on the minimum cleaning frequency to maintain an acceptable level of performance. These schedules can be further enhanced through the use of predictive energy yield loss models. Fathi et al. [6] addressed the cleaning schedules in Algerian sites. Using the helioscope computer simulation program, they concluded that the performance-based drop that warranted cleaning to be 7%. This was done by calculating the costs as well as soiling losses in a plant to compare them and extract the breakeven point. It is important to note that the values used in this study were based on the total initial investment in the PV plant rather than the running costs.

Naeem and TamizhMani [7] addressed the optimization of cleaning frequencies for a rooftop PV panel in Arizona. They found a daily soiling rate (SR) of 0.061% for the panels installed at a 20-deg tilt and concluded it was not economically justifiable to clean the panels. They used two panels and measured the PV performance as well as weather data. These data were then used to simulate a year of soiling and simulated three different cleaning periods (panel being cleaned once annually, twice, and thrice). This was then converted into economic terms based on cleaning subcontracting prices in the region as well as the production losses due to soiling. Comparisons of the cleaning costs and soiling losses for residential, commercial, and utility-scale PV installations showed that only utility-scale PV installations had economical gain from cleaning PV panels. They further recommended the ideal days for cleaning over the course of the year with respect to the three cleaning periods.

Zaihidee et al. [2] recommended more frequent cleaning for PV panels in locations with droughts, urban areas, polluted areas, and immediately after soiling events such as dust storms. They also recommended more frequent cleaning for panels with plastic or epoxy covers compared with glass covers. Mani and Pillai [8] addressed multiple cleaning schedules that may be adopted on the basis of various factors. For example, they noted a study that experimented on dust deposition for multiple tilts on various glass samples between 0 and 90 deg tilt in Egypt. They noted that tilts under 30 deg required regular periodic cleaning and recommended weekly cleaning with an adaptive approach of cleaning right after dust storms in Egypt and the Arabian Gulf regions. They summarized cleaning schedules based on climatic zones and corresponding PV installations characteristics.

Kimber et al. [9] conducted experimental research on various arrays installed on rooftops in California. They hired a window washing cleaning crew to clean the panels during the drought season, where they cleaned one array twice during the drought season (A2), one array once (B1), and one array was never cleaned (B3). They measured the soiling losses by logging the performance metric of each system. They conducted a cost–benefit analysis and recommended that cleaning is done when electricity can be sold at 0.25 dollars per kWh and that schedules should be based on a monitoring system that can aid operators in knowing when cleaning is economically justifiable.

A study [10] conducted in Saudi Arabia addressed optimizing cleaning frequencies on the basis of a cost–benefit analysis. They calculated all relevant costs to the construction and operation of a 100 MW PV plants. This analysis showed an average of 19.9 days of cleaning frequency for manual washing and 8.31 days for washing tractor-assisted cleaning over multiple tested PV technologies. The schedules proposed by the researchers were based on a performance-based approach that calculated the next optimal cleaning time from a cleaning event.

Jiang et al.’s [4] paper aimed at developing a model that can estimate the required cleaning frequency on the basis of soiling mechanisms and environmental conditions. They identified a cleaning criterion of 5% in soiling losses based on the literature review conducted by the researchers and addressed how dust particle size, wind speed, and tilt of panel. They found that cleaning periods in desert regions can be estimated at around 20.7 days. However, the developed model can only be applied in dry regions as it does not take into consideration the cleaning effect of rain. Larger dust particles, higher dust densities, and lower tilt angles experienced higher cleaning requirements. The wind had no effect on the cleaning schedules.

2.2 Cleaning Costs.

Kaldellis and Kokala [11] studied five identical PV panels positioned in an urban area in Athens, Greece, with a slightly intensified contamination. They found that even the slightest amount of dust accumulation influenced the performance of the PV panels. They measured the PV panel production via their lab experiment comparing two cleaned panels with three naturally polluted panels and found that they experienced 6.5% soiling losses in 2 months. They measured these losses against the PV plant installation prices and found that annually they would lose 40 Euros per kW annually if left uncleaned, which represents 1% of the turnkey price of the PV plant.

Jones et al. [10] used a cost–benefit analysis model to optimize the cleaning frequencies of PV panels in Saudi Arabia. They measured the values of energy losses due to soiling as well as the costs of cleaning via two different methods, manually and tractor assisted. They conducted the same analysis over multiple different types of technologies to further compare their performance to one another and how it translates into cleaning regimes. They first prepared their algorithm to calculate the optimum cleaning interval mathematically. They then prepared a cost estimate for the entire plant accounting for both manual and tractor assisted cleaning. Finally, they performed the calculations for both cleaning methods and compared these values. The same calculation was done for obtaining the optimal cleaning days on various PV technologies along with the costs associated with cleaning them with either method.

Fathi et al. [6] estimated the soiling rates that would economically justify cleaning of PV panels in Algeria. They collected data based on literature review studying two plant sizes with the production capacities of 1 MW and 5 MW, respectively, and estimated that cleaning of 1 MW of PV panels every week would cost 22,500 Euros while cleaning the plant twice fully in 1 year would cost 14,300 Euros per MW. Based on the soiling losses, it was found that the first protocol would recover its costs if the plant was cleaned at a soiling level of 10%, while the second would be at 7%.

3 Methodology

3.1 Mohammed bin Rashid Solar Park; Phase 1 (13 MW).

Located in Seih Al Dahal, Dubai, Mohammed bin Rashid Solar Park (MBRSP) phase 1, comprises 13 MW of PV panels. It is part of the MBRSP solar park development that aims at producing clean and renewable electricity from both PV technology and concentrated solar power. The location of the plant is deep inland away from urban centers, whereas the environment is a sandy generally flat desert terrain with sparse vegetation and small dunes. However, it is important to note that construction for the future phases of the solar park is ongoing in the neighboring plots, which could affect dust deposition in the 13 MW plant due to construction activities or heavy vehicles movements.

The plant consists of 152,880 thin-film CdTe frameless PV fixed at a 20-deg tilt. The entire plant is operated and maintained by the Dubai Electricity and Water Authority (DEWA) but cleaning is done via a subcontractor. As of 2018, the yearly fee for subcontracting the cleaning is 266,000 Emirati Dirham (AED) where the subcontractor cleans the entire plant once every 5 days periodically. Cleaning is done by one crew consisting of four workers and a supervisor that use two brush trolleys, supplied by the Engineering, Procurement and Construction (EPC) contractor. Cleaning is done between 8 p.m. and 4 a.m. daily, whereas client inspectors visit the plant regularly to perform inspections and safety checks. Figure 1 shows the typical manual cleaning event as used by the plant as well as a close up of the brush trolley. Furthermore, an automated cleaning robot that is being tested on 336 panels is powered by an integrated PV panel (cleaned manually too) as also shown in Fig. 1. The robot passes over the panels twice daily at 7 p.m. and cleans them with brushes, which takes 3.5 min for the forward pass and 3.5 min for the backward pass. The robot was purchased for 3650 US dollars or 13,395.5 AED, with a yearly maintenance cost of 10% or 1339.55 AED. The device was custom made by the supplier company and comes with a 2-year warranty and an expected lifespan of 10 years.

3.2 Dubai Electricity and Water Authority Outdoor Test Facility Solar Lab.

The solar lab is a research and development asset that is used by the Dubai Electricity and Water Authority, or DEWA, to conduct research on solar technology. It is located adjacent to the MBR 13 MW solar park and experiences much of the same geographical, topographical, and weather conditions. Similar construction activities are going on around the lab and thus soiling rates in the lab can be considered similar to the solar park.

The lab houses multiple different setups that span different PV technology types: Mono-Si, Poly-Si, Thin-Film, Building Integrated PV (BIPV), Copper Indium Gallium Selenide (CIGS), and Cadmium Telluride (CdTe). The panels are all placed on different tilts, some even on motion trackers. Further to the free-standing panels, building integrated PV panels are also being tested at the site as well as rooftop installed panels, whereas all data are collected directly at the site by the on-site research team. The panels and setups selected for this study include six different technologies placed at two different tilts, 5 and 25 deg, for a total of 12 different panels.

The time frame selected to be studied spanned the duration of June 1, 2016, till May 31, 2017, to cover 365 days or in other words a full year. Data are collected at the site through the use of electronic loads, whereas multiple installed weather stations measure the weather conditions that include ambient temperature, solar irradiation, humidity, and wind speed/direction. The pyranometers are installed on the plane with the solar panels as well as temperature meters on the panels themselves provide more accurate data on the exact operating conditions of the panels.

3.3 Inputs.

Due to the strong correlation between PV panels and weather, many variables are difficult to control in real-life situations. Furthermore, cleaning methods can differ greatly due to the introduction of human factors. This is further aggravated by the size of PV installations as well as repetition of work. Data that will be collected will be real values measured at existing PV installations. Cleaning methodologies too will be based on actual procedures being implemented in the UAE to account for the unique geographical and economic characteristics of the region that better represent the theory in question.

3.3.1 PV Output.

The foremost input data will be the daily PV output for the whole installation measured in kWh. This will be based on real-life values as measured at the installations and will be the basis of identifying the gains of the project. This will be correlated with the day and date of the data being obtained. This will also quantify all other parameters that affect PV generation such as solar irradiation and temperature. The power of the panel will be measured via electronic load using the maximum power point tracker. This will provide the output of the panel for every 30 s in Watts, which is then multiplied by 30 s to get the Watt per second values. This is then summed up for the duration of the day and is converted into kWh. All these data will be obtained based on historically collected data in the Dubai Electricity and Water Authority OTF solar lab for the years of 2016 and 2017.

3.3.2 Electricity Price.

In addition to the PV output, the price at which electricity is sold at will be used to determine the actual financial income of the electrical generation as well as the financial value of the electrical generation lost due to soiling. This may be in some cases based on an independent power producer agreement and/or the utility company pricing. As of 2018, DEWA implemented a slap tariff system for the electricity sold to residential and commercial customers as per Table 1. In addition to these rates, there is a fuel surcharge rate, 0.065 AED/kWh as of July 2019, plus the value-added tax, 5% as of July 2019; the fuel surcharge do not apply to PV generated electricity. For the purpose of this paper, the lowest price of 0.23 AED/kWh will be considered.

3.3.3 Soiling Rate.

The second important input will be the soiling rate. This will be a daily value that represents the loss of electrical generation due to the various factors such as dust, animal droppings, and dirt. This soiling rate will be based on the Mohammed bin Rashid 13 MW Solar Park (MBR13SP) values as calculated over the course of 2016 and 2017 by a team of researchers at DEWA [13,14]. These values were obtained via two different processes, the first which was deemed more accurate due to incorporating field values while the second incorporated theoretical modeling.

The first method [13] used involved deriving the soiling rate from the slope of daily performance metrics for each month. These data were used to collect the 25-deg tilt data on all six types of panels selected for this study, for the duration of June 1, 2016, till Dec. 31, 2016.

The second method [14] used a stochastic rate and recovery methodology. This was used for all 5-deg tilt data collected in this paper as well as the 25-deg tilt data for the duration of Jan. 1, 2017, till May 31, 2017, for all six types of panels selected.

It is important to note that in both papers, the soiling rate was found to behave in a linear trend shown in Fig. 2(a). That is, over a certain period (in this case over the month), the soiling rate can be estimated linearly between two cleaning events using the Theil-Sen estimator as evident by Fig. 2(b).

3.3.4 Cleaning Cost.

Further to the soiling rate, the cost of cleaning will be obtained based on the method used. It will be the cost of cleaning the entire installation (as it will be compared with the PV production cost of the installation as a whole). This is a major variable when considering the number of approaches and methods that are used to clean the panels. In terms of practices in the UAE, the most common approach is manual based dry cleaning of panels. This is generally the case in places where low labour costs exist. Furthermore, manufactures recommendations generally recommend dry cleaning using brushes. In terms of the cost of the manual cleaning, the data were collected based on the MBR13SP where the plant was subcontracted to be cleaned every 5 days for an annual cost of 266,000 AED covering the 152,880 panels that make up the MBS13SP. In other words, the cost of manually cleaning the entire plant once can be estimated at 3644 AED, whereas cleaning one panel manually, taking into account economies of scale, can be estimated at 0.024 AED. Furthermore, the cost of using automatic cleaning tools was based on the price of purchasing, operating, and maintaining automatic cleaning mechanisms. The device being tested and used in DEWA was bought at for 4000 US$ or 14,692.4 AED. It is powered by a PV panel of its own. The expected lifespan of the device is set at 10 years with a warranty from the manufacturer for 2 years and an annual maintenance cost of 10% of the purchase price. One such device may be used to clean 336 panels, though it may be customized to clean even more. This puts the price of cleaning individual panels at 0.0197 AED per cleaning event that is done daily over the 10 years with operation and maintenance costs considered for 8 years (as 2 years remain under warranty).

3.4 Assumptions.

Various assumptions were made to streamline the calculations either because they were already intrinsically included in the input data or because comparing between different cleaning methodologies negates their impact and effects. Weather conditions were major factors that were important to include in the study; however, these were assumed to be factors already taken into account when obtaining PV power output and soiling rates. Various factors such as temperature, solar irradiance, and atmospheric pressure were also assumed to already have been included in the power measurements. Other factors such as wind, humidity, and dust storms were taken into account when collecting soiling data and accounted for in the PV panel soiling model

Furthermore, soiling data collected over the course of a year may vary over consecutive years due to changing weather and climatic conditions. This means that although the developed algorithm in this paper looks at cleaning procedures in retrospect, any developed algorithm to be used in operating plants requires further enhancement to include predictive capabilities to calculate approximate future soiling rates based on trend analysis and historical data both as well as weather forecasts.

Another assumption made pertains to the cleaning efficiency of PV panels by either manual or automatic cleaning. In terms of manual cleaning, training of personnel, as well as the use of effective tools, can greatly affect the efficiency. This is further aggravated by the fact that repeating the same job might create less efficiency due to boredom, exhaustion, or sloppiness. Whereas automatic cleaning devices may become less effective, the older they get or may be installed or operated incorrectly. However, since the comparison is being done between different frequencies, the efficiency can be assumed to be more or less the same in both cases, hence its impact on the conclusions will not be significant. One researcher noted that the cleaning efficiency of panels is 99% with brushes and compressed air and 100% if cleaned with water [15]. The 1% increase in efficiency does not warrant spending a valuable resource like water, hence for the purpose of this study 99% was taken as the cleaning efficiency irrespective of automatic or manual cleaning.

Some data points pertaining to the output and soiling rates of some of the PV panels were not measured due to various events and thus assumed values were used to ensure data was available for the course of an entire year for effective comparison. The missing data were filled by considering the behaviors of other panels in terms of average soiling rate and technology.

Finally, the capital cost in purchasing and setting up the PV plant is not to be considered as it does not contribute into the decision making the process for when a panel should be clean nor in the economic valuation of it. This is because these initial costs are considered a sunk cost, whereas cleaning protocols are governed by the relationship between the actual output of the PV panel in comparison with the theoretical cleaned output of the same day.

3.5 Output.

To generate an all-encompassing algorithm, quantitative and qualitative analysis of the results will be conducted in order to evaluate the best approaches to be followed. This will be done from both an economic point of view and a technical point of view based on real data input into the algorithm, the assumptions made, and the predicted generation output. The same methodology will be applied to data collected from multiple sites and setups in these sites addressing various cleaning cases that include automatic and manual cleaning.

In terms of quantitative analysis, the number of cleaning events in each approach will be quantified directly from the algorithm’s output. A cleaning event within this study is defined as cleaning the panel once on that particular day. These cleaning events can then be correlated to the cost of cleaning to obtain the total cleaning cost over the year. In addition, the cumulative predicted production loss over the year will be calculated as kWh and translated into an economical value. The production loss and cleaning cost can then be added to obtain the total costs for each approach to determine the effectiveness of each approach. As the cleaning protocol is the focus of this study, only the incremental cost of cleaning will be compared with the incremental revenue from cleaning the PV panels

Additionally, a qualitative analysis will be done to further address all issues pertaining to each scenario and case. The theoretical periodic and performance-based cleaning frequencies as calculated via the algorithm will be compared with the actual frequencies currently being used at the MBR13SP. In some cases, it might be found that even when approaching cleaning on a performance basis, we can identify certain recurring cleaning periods/frequencies. This can be used to enhance cleaning carried out on periodic bases by estimating and recommending a better cleaning schedule should performance-based cleaning is not implemented. Furthermore, the scheduling difficulties that may arise due to irregular cleaning frequencies will be addressed based on the algorithm to determine how it will affect staffing and equipment requirements, as well as how feasible would the approach be. The discussion will also cover different PV technologies, cleaning technologies, and panel tilt angles all of which can impact the cleaning schedule.

3.6 Simulation Setup.

To perform the numerical calculations, Microsoft Excel was used to develop an algorithm that will perform multiple iterations of equations for multiple cases and scenarios. This algorithm requires, as input data, the PV output, daily soiling rates, price of electricity, and cost of cleaning, whereas the outputs generated will be evaluated in terms of total costs (cleaning costs and production loss costs), as described in Sec. 3.5.

The decision of when a cleaning event should occur was based on four different approaches:

  1. The first approach is that of a performance-based approach, in which panels are cleaned only when the economic value of the electricity production lost due to soiling exceeds the cost of cleaning the panels.

  2. The second approach is that of a periodically based approach to cleaning in which panels were routinely cleaned after a specific number of days regardless of any significant weather conditions or the state of the PV panel; this is based on the actual practices used in the MBR13SP.

  3. The third approach is that of an optimized periodic-based approach, which uses the same periodic calculation as the second approach; however, the period selected is based on the overall average cleaning interval as calculated from the performance-based approach.

  4. The fourth approach is that of an adaptive periodic-based approach, which uses periodic calculations based on the seasonal average cleaning interval of the performance-based approach. To do this, the standard deviation of the total cleaning intervals was calculated for the performance-based approach. This was then used to modify the cleaning interval whenever two consecutive cleaning intervals from the performance-based approach surpass 1 standard deviation of the average of that season. This new period is calculated for those events as an average of those events. This may happen more than once and ensures that any significant changes in the cleaning trends are reflected in the final cleaning plan.

Conducting such economic analysis requires a comparison between the “dirty” or “actual” output from the PV panels, measured real-time at the site, against the “clean” output, calculated using the soiling rate. Calculating the clean output can be done by multiplying the dirty output with the effect of soiling as developed by the author of this paper as such 
PVclean=PVdirty*(1+SRcum/100)
(1)
where PVclean is the total potential output in kWh of the PV panel assuming 0% soiling, PVdirty is the actual data as collected from the installation in kWh under the effect of soiling, and SRcum is the cumulative effect of soiling for that particular day. It is important to remember that the calculated soiling rate from the site is a daily value, and thus the soiling rate used in the equations should be the cumulative soiling rate since the last cleaning event. The difference between the outputs of the clean and dirty PV panels will give us the lost output due to soiling, which is then multiplied by the price of electricity to obtain its financial value.

In the case of performance-based cleaning, this will be done over multiple iterations where it will accumulate the cost of production loss until it is greater than the cost of cleaning. It is at this point that a cleaning event will be triggered. When this happens, the next day’s cumulative soiling rate will reset back to the daily value (in addition to the remaining cumulative soiling rate due to imperfect cleaning which is the cumulative soiling rate multiplied by the cleaning efficiency). The cumulative production loss cost will also reset to include only the next day’s production loss. This approach ensures that economics dictate the action to be taken (clean or not clean) with the aim of generating the maximum rate of return on the investment.

On the other hand, periodic-based cleaning will use the same equations to calculate production losses as well as cleaning costs. However, instead of triggering a cleaning event when the production loss is higher than cleaning cost, a cleaning event is executed once over a predefined fixed cycle. For example, if the period identified for the study was 1 week then regardless of the results of Eq. (1), the panels will be cleaned every 7 days. This periodic-based period will use both the existing current practices of both manual cleaning every 5 days and automatic cleaning every day. The optimized periodic-based cleaning approach will also use this equation, but instead of using a predefined fixed period it will use the average cleaning interval (rounded to the nearest day) obtained from the performance-based method as calculated for the full year.

The adaptive periodic-based cleaning will also rely on the average cleaning interval as obtained by the performance-based method. The standard deviation of all cleaning intervals (from the performance-based method) will be calculated, whereas 1 standard deviation will dictate the cleaning interval change all values rounded to the nearest day. Whenever two or more consecutive cleaning events fall between, beyond, or below the 1 standard deviation bounds, this will signify a season with its own unique cleaning period. Each season’s period will be determined by the average of all cleaning intervals (from the performance-based method) within that season and will continue to recur until a season change is initiated. That is until two or more cleaning events lie in a different set of bounds. For example, if five cleaning events fall within 1 standard deviation of the period and then the sixth and seventh cleaning event interval is greater than 1 standard deviation, a separate period will be calculated for the first five and a second period will be calculated for the second two.

4 Simulation Results

To run the simulation, the data for each of the 12 setups as elaborated in Sec. 3.3.1 were collected and imported into excel. These data included the daily PV panel electrical energy output in Wh which was obtained from historical data in the DEWA OTF solar lab along with the corresponding daily soiling rate obtained from previous work done by DEWA [13,14].

As an input, the existing practices were considered as the baseline case from the MBR13SP as shown in Table 2:

  1. The price of energy is 0.23 AED/kWh [12].

  2. The manual cleaning cost for a single cleaning event is 0.024 AED per panel (as calculated in Sec. 3.3.4).

  3. The automatic cleaning cost for a single cleaning event is 0.0197 AED per panel (as calculated in Sec. 3.3.4).

  4. The manual cleaning period was set as 5 days (as described in Sec. 3.1).

  5. The automatic cleaning period as 1 day (as described in Sec. 3.1.

  6. The efficiency of both cleaning methods is 99% [15].

Due to the large number of simulation runs, a sample of 14 days from June 1, 2016, to June 10, 2016, will be shown for the polycrystalline panel at a 5-deg tilt for each of the different approaches.

The input data are then automatically imported into the simulation run, in this case, the PV daily output and its corresponding date and day number are shown along with the corresponding daily soiling rate in Table 3. For day 1, the output is 1.7647 kWh with a daily soiling rate of 0.37. The cumulative soiling rate is simply the sum of all daily soiling rates since the last cleaning event. For day 1, the daily soiling rate and cumulative soiling rate is equal as there are no previous soiling rates 
SRcum=SRdailySRcum=0.37SRcum=0.37
Calculating the theoretical cleaned output of the PV panel (if the panel where to be cleaned that day) uses the cumulative soiling date of that particular day. In the case of day 1, the PV cleaned output is calculated as 
Pclean=Pdirty*(1+SRcum/100)Pclean=1.7647kWh*(1+(0.37/100))Pclean=1.7712kWh
The difference between the PV clean output and the PV actual output gives the production loss of that day due to noncleaning. By multiplying this difference by the selling price of power, the economic value of power not generated (and thus not sold) can be calculated for each day. For day 1, the production loss in kWh and its economic value in AED is calculated as 
ΔP=PcleanPdirtyΔP=1.7712kWh1.7647kWhΔP=0.0065kWh
 
ΔP=0.0065kWh*0.23AED/kWhΔP=0.0015AED
where P is the clean panel output. The cumulative production loss is calculated for each day by adding the sum of all production losses since the last cleaning event. For day 1, the cumulative production loss equals the daily production loss as it is the first day 
ΔPcum=ΔPdailyΔPcum=0.0015AEDΔPcum=0.0015AED
For day 2, the same calculations would be done as follows: 
SRcum=SRdailySRcum=SRday1+SRday2SRcum=0.37+0.37SRcum=0.74
 
Pclean=Pdirty*(1+SRcum/100)Pclean=1.7764kWh*(1+(0.74/100))Pclean=1.7895kWh
 
ΔP=PcleanPdirtyΔP=1.7895kWh1.7764kWhΔP=0.0131kWh
 
ΔP=0.0131kWh*0.23AED/kWhΔP=0.003AED
 
ΔPcum=ΔPdailyΔPcum=0.0015AED+0.003AEDΔPcum=0.0045AED

The above calculations will be done for each day until a cleaning event is triggered and is the same process done irrespective of the cleaning approach being studied.

4.1 Periodic-based Manual Cleaning Approach.

As explained in Sec. 3.6, the periodic-based manual cleaning approach will wait for a predetermined period before initiating the next cleaning event. The cleaning interval was taken as 5 days to reflect the actual current practices within the MBR 13 MW solar park and will be considered as the baseline case. On day 5, a cleaning event is initiated where the same equations used in Sec. 4 are used here for day 6: 
SRcum=SRdailySRcum=SRcumday15*(1manualcleaningefficiency/100)+SRday6SRcum=1.85*(199/100)+0.37SRcum=0.3922
 
Pclean=Pdirty*(1+SRcum/100)Pclean=1.6839kWh*(1+(0.3922/100))Pclean=1.6906kWh
 
ΔP=PcleanPdirtyΔP=1.6906kWh1.6839kWhΔP=0.0067kWh
 
ΔP=0.0067kWh*0.23AED/kWhΔP=0.0015AED
 
ΔPcum=ΔPdailyΔPcum=0.0015AEDΔPcum=0.0015AED

The same calculations will be repeated every 5 days, irrespective of any panel performance or cleaning cost values.

4.2 Daily Automatic Cleaning Approach.

Just like Sec. 4.1, the same approach will be taken for the periodic-based automatic cleaning approach. However, the values used would be different as the cleaning interval will be done daily and the cost of one cleaning event is taken as 0.0197 AED as per the current device being used in the MBR 13 MW solar park. The efficiency of the automatic cleaning method was taken as 99% as well. On day 1 (and every subsequent day), a cleaning event is triggered where the same equations as in Sec. 4.1 are applied for day 2 
SRcum=SRdailySRcum=SRcumday1*(1automaticcleaningefficiency/100)+SRday2SRcum=0.37*(199/100)+0.37SRcum=0.3774
 
Pclean=Pdirty*(1+SRcum/100)Pclean=1.7764kWh*(1+(0.3774/100))Pclean=1.7832kWh
 
ΔP=PcleanPdirtyΔP=1.7832kWh1.7764kWhΔP=0.0067kWh
 
ΔP=0.0067kWh*0.23AED/kWhΔP=0.0015AED
 
ΔPcum=ΔPdailyΔPcum=0.0015AEDΔPcum=0.0015AED

The same calculations will be repeated every day, irrespective of any panel performance or cleaning cost values.

4.3 Performance-Based Manual Cleaning Approach.

As described in Sec. 3.7, the performance-based manual cleaning approach will compare the ΔPcum at each day against the cost of cleaning of 0.024 AED per panel. In the case of the polycrystalline panel at 5 deg tilt, at day 6 the ΔPcum is 0.0307 AED which is greater than 0.024 AED. This signifies a cleaning event at day 6 where 0.024 AED will be spent to clean the panel. Due to this cleaning event, day 7’s SR cum is calculated as follows: 
SRcum=SRdailySRcum=SRcumday16*(1manualcleaningefficiency/100)+SRday7SRcum=2.22*(199/100)+0.37SRcum=0.396
It is important to note that since the manual cleaning efficiency is 99%, the SRcum will not reset to zero but rather to 1% of the previous SRcum in addition to the day’s daily soiling rate. The ΔPcum will also reset to zero and will include only the effect of the soiling rate on that day as follows: 
Pclean=Pdirty*(1+SRcum/100)Pclean=1.7479kWh*(1+(0.396/100))Pclean=1.7549kWh
 
ΔP=PcleanPdirtyΔP=1.7549kWh1.7479kWhΔP=0.007kWh
 
ΔP=0.007kWh*0.23AED/kWhΔP=0.0016AED
 
ΔPcum=ΔPdailyΔPcum=0.0016AEDΔPcum=0.0016AED

The same calculations will continue for the duration of the year, each day’s ΔPcum being compared with the cleaning cost.

4.4 Optimized Periodic-based Manual Cleaning Approach.

The same procedure from Sec. 4.1 will be followed; however, the cleaning interval will be taken as the average cleaning interval from the performance-based manual cleaning method. In the case of the polycrystalline panel at 5 deg tilt, the average was found to be 7 days and thus the cleaning interval was taken to be every 7 days. On day 7, a cleaning event is triggered, whereas the calculations for day 8 are as follows: 
SRcum=SRdailySRcum=SRcum day17*(1manualcleaningefficiency/100)+SRday8SRcum=2.59*(199/100)+0.37SRcum=0.3996
 
Pclean=Pdirty*(1+SRcum/100)Pclean=1.7344kWh*(1+(0.3996/100))Pclean=1.7413kWh
 
ΔP=PcleanPdirtyΔP=1.7413kWh1.7344kWhΔP=0.0069kWh
 
ΔP=0.0069kWh*0.23AED/kWhΔP=0.0016AED
 
ΔPcum=ΔPdailyΔPcum=0.0016AEDΔPcum=0.0016AED

The same calculations will be repeated every 7 days, irrespective of any panel performance or cleaning cost values. Figure 3 shows the sample calculations for the first 10 days.

4.5 Optimized Periodic-based Manual Cleaning Approach.

The same procedure from Sec. 4.1 will be followed; however, the cleaning interval will be taken as the average cleaning interval from the performance-based manual cleaning method. In the case of the polycrystalline panel at 5 deg tilt, the average was found to be 7 days and thus the cleaning interval was taken to be every 7 days. On day 7, a cleaning event is triggered whereas the calculations for day 8 are as follows: 
SRcum=SRdailySRcum=SRcumday17*(1manualcleaningefficiency/100)+SRday8SRcum=2.59*(199/100)+0.37SRcum=0.3996
 
Pclean=Pdirty*(1+SRcum/100)Pclean=1.7344kWh*(1+(0.3996/100))Pclean=1.7413kWh
 
ΔP=PcleanPdirtyΔP=1.7413kWh1.7344kWhΔP=0.0069kWh
 
ΔP=0.0069kWh*0.23AED/kWhΔP=0.0016AED
 
ΔPcum=ΔPdailyΔPcum=0.0016AEDΔPcum=0.0016AED

The same calculations will be repeated every 7 days, irrespective of any panel performance or cleaning cost values.

4.6 Adaptive Periodic-based Manual Cleaning Approach.

To perform an adaptive periodic-based cleaning approach, an analysis is to be performed on the performance-based manual cleaning approach to obtain bounds that will determine when the period will change. To do this, the standard deviation of all cleaning intervals from the performance-based approach is taken, which in the case of the polycrystalline panel at 5 deg tilt is calculated as 3 days (rounded to the nearest day). The standard deviation of 3 days is then added and subtracted from the performance-based average of 7 days (rounded to the nearest day) to get the upper and lower bounds, which comes out to be 10 days and 4 days, respectively.

After applying these bounds to the performance-based manual cleaning approach cleaning intervals, the trend can be studied. As evident in Figs. 4 and 5, starting from June 1, 2016, up until Feb. 1, 2017, all cleaning intervals fall between the bounds of 4 days and 10 days hence this will be labeled as a season. It is important to note that on Oct. 15, 2016, the cleaning interval becomes 14 days which is higher than 10 days; however, this value is considered an outlier as the consecutive interval of 6 days fell back within the bounds of 10 days and 4 days, and not another interval higher than 10 days. After Feb, 1, 2017, until the Apr. 6, 2017, the intervals are above 10 days (as they are 19, 16, 17, and 12 days) hence everything between these dates will have a different period and hence a second season. After Apr. 6, 2017, the interval again falls within the bounds of 10 and 4 days with intervals of 6s and 5s which is the third and final season.

To obtain the period of each of these three seasons, the average of all the cleaning intervals will be calculated. For season 1, the average is calculated as 
Cleaningperiodaverage=Cleaningintervaldays/cleaningeventsCleaningperiodaverage=(2*5+16*6+7*7+6*8+2*9+1*10+1*14)/35Cleaningperiodaverage=7
Hence, the cleaning period for the season from June 1, 2016, to Feb. 1, 2017, is 7 days, where a cleaning event is triggered on day 7 and the calculations for day 8 are as follows: 
SRcum=SRdailySRcum=SRcumday17*(1manualcleaningefficiency/100)+SRday8SRcum=2.59*(199/100)+0.37SRcum=0.3996
 
Pclean=Pdirty*(1+SRcum/100)Pclean=1.7344kWh*(1+(0.3996/100))Pclean=1.7413kWh
 
ΔP=PcleanPdirtyΔP=1.7413kWh1.7344kWhΔP=0.0069kWh
 
ΔP=0.0069kWh*0.23AED/kWhΔP=0.0016AED
 
ΔPcum=ΔPdailyΔPcum=0.0016AEDΔPcum=0.0016AED

The same calculations will be repeated every 7 days, irrespective of any panel performance or cleaning cost values, until Feb. 1, 2017, where a new period will be calculated and followed. Figure 4 shows the individual cleaning events from the performance-based method, while Figure 5 shows the calculated seasonal average cleaning events.

5 Results and Discussion

5.1 Comparison of Approaches.

This section compares the different approaches on the basis of the results in Sec. 4. This includes a quantitative comparison based on the percentage improvement from the base case of periodic base manual cleaning approach as well as a qualitative comparison based on the pros and cons of each approach, specifically in terms of operation and maintenance requirements. A discussion as well on how these approaches would apply in comparison with the case study of the MBR 13 MW solar park will be discussed to apply these recommendations in a real-life case.

The summary of increase (or decrease) in improvements in total costs, that is the sum of cleaning costs and the production losses cost due to soiling, are summarized in Table 4 for each approach when compared with the baseline case of the periodic-based manual cleaning approach:

5.1.1 Performance-Based Manual Cleaning Approach.

In all cases, which include the six different technologies at two different tilts for a total of 12 cases, the performance-based manual cleaning approach exhibited the best improvement over existing periodic manual cleaning practices as expected. These improvements ranged from 9.91% up to 33.87% in total costs depending on the tilt and technology in question.

The only exception is with the CIGS thin-film panel at 5 deg tilt, in which the adaptive periodic-based manual cleaning approach showed the best improvement over the rest with 20.46%. Although only marginally by 0.04%, the performance-based manual cleaning approach performed as a close second with 20.42%, where the difference can be attributed to rounding during calculations.

Despite the improvement, it is important to take into account the cons when using a performance-based approach. Most importantly, scheduling and resourcing effectively for such an approach would be very difficult as the time for cleaning would be unknown. The adaptive cleaning approach, for example, would not face the same issues, which is important to consider especially when the percentage improvement between performance-based against adaptive based manual cleaning is at a maximum of 3.25%, with most values ranging at 1–2.5%. However, this can be mitigated by relying on historical data and/or predicting cleaning events, which would give a better indication of when to expect a cleaning event to be required. To better manage manpower requirements in larger plants, sectioning parts of the plant with which a crew may clean in one night and then staggering the cleaning schedules of each of these sections. This can help maintain a reasonably sized cleaning crew while ensuring they remain productive over time as they will be deployed at different sections at different times.

5.1.2 Adaptive Periodic-based Manual Cleaning Approach.

The adaptive periodic-based manual cleaning approach showed the next best improvement percentage over the baseline case of periodic-based manual cleaning. The only exception to the above is CIGS thin-film panel at 5 deg tilt, which showed a very slight difference in improvement that amounted to 20.46% in total costs against the performance-based 20.42%. The improvements in the adaptive periodic-based manual cleaning approach ranged from 8.75% to 31.75% in total costs depending on the tilt and technology in question.

It is important to note that even though it showed the second highest improvement against the baseline case, the difference in improvement from the adaptive periodic and performance-based manual cleaning approaches against the baseline was only by 0.04% to a maximum of 3.25%. In case an even more efficiency adaptive periodic is required, the standard deviation that defines the bounds of the season can be taken as lesser than 1.

This method allows a more organized resource and manpower approach, as the cleaning requirements during each season can be identified beforehand and thus planned ahead. However, historical data would be required to perform such an analysis, where such an approach would be better suited for older more established larger PV installations (such as PV solar parks). Another approach could take this same analysis and correlate it with weather data that can provide cleaning frequencies based on that weather data. Dry seasons with dusty hazes can expect more cleaning frequencies, whereas wet seasons can expect lesser cleaning frequencies. Such recommendations can be shared with smaller privately owned PV installations (such as commercial or residential rooftop installed panels).

5.1.3 Optimized Periodic-based Manual Cleaning Approach.

The optimized periodic-based manual cleaning approach showed the third best improvement percentage over the baseline case of periodic-based manual cleaning in all cases. The improvements in the optimized periodic-based manual cleaning approach ranged from 5.78% to 28.39% in total costs depending on the tilt and technology in question.

This method was the simplest improvement against the baseline periodic-based manual cleaning, in which it aimed to provide a different period that is more optimized based on the performance-based manual cleaning approach. In terms of cleaning protocols, this is the easiest approach as throughout the year the cleaning times are defined well ahead of time and the crew size will remain the same irrespective of performance or weather conditions. Such an approach can best be used in privately owned small PV installations such as those found in residential units or on commercial buildings.

It is important to note however that the optimum cleaning period is highly dependent on the technology and tilt angle installed. For the technologies and tilts in this study, the optimum frequency ranged between 7 days and 13 days. All these frequencies were higher than the current existing practices used in the MBR 13 MW solar park.

5.1.4 Periodic-based Manual Cleaning Approach.

The periodic-based manual cleaning approach was taken as the baseline case as it was the currently used practices in the MBR 13 MW solar park. Based on the analysis done, the current practices are inadequate especially when taking into account the specifics of the solar park. The 13 MW solar park currently has installed 152,880 thin-film CdTe frameless PV fixed at a 20-deg tilt. When compared with the results from the CdTe thin-film at 25-deg tilt, we can see that the optimized cleaning period would be 13 days which is a full 8 days difference from the existing practices, more than twice the duration.

Over the course of the year, this means that current practices include cleaning the entire PV plant 73 times per year at a total cost of 324,105.6 AED (2.12 AED per panel) while if an optimized period is maintained only 28 cleaning events per year is required at a total cost of 232,377.6 AED (1.52 AED per panel). If a performance-based approach is instead opted for, we would still require 28 days but at a total cost of 214,032 AED (1.4 AED per panel), and if an adaptive periodic approach is taken then it would require 28 days at a total cost of 221,676 AED (1.45 AED per panel). This means that irrespective of what approach is used, it would still generate improvement and cost saving for the entire plant when compared with the periodic-based manual cleaning approach.

5.1.5 Periodic-based Automatic Cleaning Approach.

In all cases, the periodic-based automatic cleaning approach underperformed greatly with decreases in improvement between 164.64% and 245.34% in total costs when compared with the existing periodic manual cleaning practices, depending on the technology and tilt. This is mainly due to the daily cleaning done by the robot which results in only minor performance gains.

It is also important to note that these devices differ greatly depending on the mechanism used and the environment they are being used in. More resource consuming cleaning mechanisms such as detergent or water-based cleaning would require more costs to operate and maintain. On the other hand, exposure to the elements and direct sunlight may lead to a reduced lifespan of the device and more repairs. This would ultimately lead to an even less feasible implementation of the device.

Based on the above algorithm, robotic cleaning is not an economically feasible approach to cleaning PV panels. Furthermore, it does not remove the human element to operation and maintenance needs as staff is still required to be available at the site to address any breakdowns or perform preventive maintenance. In the case of the robot used in the MBR 13 MW solar park, the integrated PV panel used to charge the robot is required to also be cleaned manually to ensure robot has enough charge to perform its cleaning route. Installing a robot that can clean a larger number of PV panels may enhance this method’s economic feasibility; however, lifespans and reliability must increase while capital costs of purchasing them must decrease.

5.2 Comparison of PV Technologies.

This section builds on the previous section to further compare in detail the different technologies used on the basis of their cleaning requirements. This comparison will be done at each different cleaning approach at each tilt to discuss how different technologies may require different cleaning frequencies despite being exposed to similar environmental conditions, cleaning approaches, cleaning methods, and tilt.

The CIGS and CdTe thin-film panels showed the highest soiling rates and the lowest energy output, where the soiling rates increase can be attributed to the textured polymeric film front cover where the textured gaps may allow dust to get trapped thus more dust will adhere to the surface in this case [13]. The other four first generation panels (Poly, Mono, BIPV, and Bifacial) behaved similarly at either tilt with the summary of soiling rates and PV output are shown in Table 5 and Figures 6 and 7:

In all cleaning scenarios, CdTe thin-film panels showed the least cleaning frequencies while the polycrystalline panel required the highest cleaning frequencies in both the 5-deg tilt and the 25-deg tilt. Bifacial monocrystalline panels also showed consistently the second highest cleaning frequencies at both 5 and 25-deg tilt, in all different cleaning approaches. On the other hand, CIGS thin-film, monocrystalline, and BIPV double glass polycrystalline panels cleaning frequencies differed in comparison with the other technologies depending on the tilt and approach used to clean them. CIGS thin-film panels showed the second least cleaning frequencies at 25-deg tilt while monocrystalline panels showed the second least cleaning frequencies at 5 deg, pushing CIGS to the third least.

Despite having the highest soiling rates and lowest output, the thin-film panels showed low cleaning requirements especially at 25 deg tilt where they held the lowest cleaning frequencies. This leads to the conclusion that cleaning frequencies do not necessarily increase solely based on the soiling rate, but rather they increase based on the relationship between the PV output, the soiling rate, and the cleaning cost.

5.3 Comparison of Tilt.

This section compares the two different tilts for each technology on the basis of how cleaning frequencies differ for each. This comparison will be done for different technologies and addresses the cleaning requirements at the different cleaning approaches. Ultimately this comparison will determine whether tilt plays a role in how much soiling accumulates on the panels and thus how it translates into how many cleaning events would be required under the same environmental conditions.

In almost all cases, especially when considering the same technology and comparing across the same cleaning approach, higher tilts translated into less cleaning frequencies. In many of these cases, all three cleaning approaches for the 25-deg tilt (performance, adaptive periodic, and optimized periodic-based manual cleaning) would require less cleaning than all three cleaning approaches for the 5-deg tilt. That is even when using an optimized periodic-based manual cleaning approach for the 25 deg, this would require less cleaning than using the performance-based manual cleaning approach for the 5 deg. The only exception to the above was the monocrystalline panel that showed the 5-deg tilt requiring less cleaning frequencies than the 25 deg when considering each different approach.

Research done indicates that soiling is proportional to the cosine of the panel’s tilt angle. It was further found that at lower wind speeds, the panel’s tilt had less effect on the deposition of dust as opposed to higher wind speeds [16]. It is also important to note that despite the higher soiling rates of lower tilts, cleaning behavior was governed by the relationship between the soiling rate and the PV power output as well as the cleaning cost. Furthermore, although on average soiling rates are higher on average with lower tilts, some months might experience slightly more soiling rates on higher tilts.

6 Conclusions

In conclusion, multiple outcomes may be drawn out of the 12 simulations run that. In terms of the ideal approach to adopt, the following was concluded when compared with the baseline case of the existing 5 days periodic-based manual cleaning approach:

  1. Performance-based manual cleaning approach is the ideal approach to adopt with improvements of 9.91–33.87% in total costs against the baseline case; however, procedures should be in place to ease the cleaning schedule such as staggering the cleaning of different sections at different times or predicting cleaning needs ahead of time.

  2. Adaptive periodic-based manual cleaning approach is the next best approach to implement with improvements of 8.75–31.75% in total costs against the baseline case, as it combines the cleaning efficiency of the performance-based method while mitigating the disadvantages that come with the cleaning schedule.

  3. Optimized periodic-based manual cleaning approach based on the average cleaning interval found in the performance-based method can still provide a more efficient cleaning schedule compared with other more generic cleaning periods with improvements of 5.78–28.39% in total costs against the baseline case.

  4. The additional costs associated with daily periodic automatic cleaning were more than the additional revenue from the extra electricity generated from the clean PV panels.

In terms of the effect of cleaning procedures on different technologies and tilts used, the following conclusions were drawn:

  1. Higher soiling rates do not correlate to higher cleaning frequencies, as cleaning frequencies are instead dependent on the relation between PV output, soiling rates, and cleaning costs.

  2. Thin-film panels generally have lower cleaning requirements when compared with first-generation PV panels.

  3. Higher tilt angles generally mean lower cleaning frequencies.

Finally, the above analysis done on the CdTe thin-film panel at 25 degr was compared with the existing practices in the MBR 13 MW solar park and the below recommendations were established when compared with the baseline of periodic-based manual cleaning approach:

  1. 13 days was found to be the ideal period, which is more than twice the current practice of 5 ays, where this optimized approach showed a total cost of 1.52 AED per panel compared with the baseline case of 2.12 AED per panel.

  2. The current robot being used in the plant for periodic-based automatic cleaning approach is greatly inefficient when compared with the baseline, with a total cost of 7.31 AED per panel; hence, the current model is not a viable option at its current operating conditions.

  3. The ideal economic approach would be a performance-based manual cleaning approach with a total cost of 1.4 AED per panel, followed by the adaptive periodic-based manual cleaning with a total cost of 1.45 AED per panel.

Following the above approaches of performance, adaptive periodic, and optimized periodic-based manual cleaning, we can realize yearly savings at the MBR 13 MW solar park of 110,073.6 AED, 102,429.6 AED, and 91,728 AED, respectively, in the operation and maintenance procedures.

At first glance, it would be evident that performance-based manual cleaning approach would be the ideal approach to take due to economics where it results in the least total cost out of all the methods. However, the difference in percentage improvement over the adaptive periodic manual cleaning approach is less than 3.25%, usually ranging between 1% and 2.5%, does not significantly impact economics (in the case of the CIGS panel at 5-deg tilt, it showed an improvement of 0.04%). In the case study of the MBR 13 MW solar park, the difference between the two approaches only generated a yearly saving of 7644 AED for the performance-based approach.

When comparing these savings with the qualitative advantages of an adaptive periodic manual cleaning approach, the adaptive periodic manual cleaning approach has more advantages. As discussed in Sec. 5.1, the adaptive periodic manual cleaning approach allows a set period to be placed for each season that can be further adapted to account for irregular events such as raining periods, nearby construction activity, or dust storms. This allows maintenance teams and operators to have fixed cleaning schedules as opposed to the performance-based irregular schedules. It can allow for more optimized resource leveling as well as easier labour requirements. Furthermore, such adaptive approaches can be more flexible to incorporate PV panel manufacturer recommendations and requirements.

In conclusion, the adaptive periodic-based manual cleaning is the recommended approach as it results in only a minor reduction in economic savings while providing added value in the form of qualitative advantages. On the other hand, despite the advancements in automatic cleaning methods manual cleaning is still recommended in the country of UAE due to the low labor costs and harsh weather that may damage and reduce the lifespan of the robotic devices. Higher tilts reduced cleaning as well as thin-film panels, however, other factors come into play in these decisions.

References

References
1.
Maghami
,
M.
,
Hizam
,
H.
,
Gomes
,
C.
,
Radzi
,
M.
,
Rezadad
,
M.
, and
Hajighorbani
,
S.
,
2016
, “
Power Loss Due to Soiling on Solar Panel: A Review
,”
Renewable Sustainable Energy Rev.
,
59
, pp.
1307
1316
. 10.1016/j.rser.2016.01.044
2.
Zaihidee
,
F.
,
Mekhilef
,
S.
,
Seyedmahmoudian
,
M.
, and
Horan
,
B.
,
2016
, “
Dust as an Unalterable Deteriorative Factor Affecting PV Panel’s Efficiency: Why and How
,”
Renewable Sustainable Energy Rev.
,
65
, pp.
1267
1278
. 10.1016/j.rser.2016.06.068
3.
Lopez-Garcia
,
J.
,
Pozza
,
A.
, and
Sample
,
T.
,
2016
, “
Long-Term Soiling of Silicon PV Modules in a Moderate Subtropical Climate
,”
Sol. Energy
,
130
, pp.
174
183
. 10.1016/j.solener.2016.02.025
4.
Jiang
,
Y.
,
Lu
,
L.
, and
Lu
,
H.
,
2016
, “
A Novel Model to Estimate the Cleaning Frequency for Dirty Solar Photovoltaic (PV) Modules in Desert Environment
,”
Sol. Energy
,
140
, pp.
236
240
. 10.1016/j.solener.2016.11.016
5.
Sayyah
,
A.
,
Horenstein
,
M. N.
, and
Mazumder
,
M. K.
,
2014
, “
Energy Yield Loss Caused by Dust Deposition on Photovoltaic Panels
,”
Sol. Energy.
,
107
, pp.
576
604
. 10.1016/j.solener.2014.05.030
6.
Fathi
,
M.
,
Abderrezek
,
M.
, and
Grana
,
P.
,
2017
, “
Technical and Economic Assessment of Cleaning Protocol for Photovoltaic Power Plants: Case of Algerian Sahara Sites
,”
Sol. Energy
,
147
, pp.
358
367
. 10.1016/j.solener.2017.03.053
7.
Naeem
,
M.
, and
TamizhMani
,
G.
,
2015
, “
Cleaning Frequency Optimization for Soiled Photovoltaic Modules
,”
2015 IEEE 42nd Photovoltaic Specialist Conference (PVSC)
,
New Orleans, LA
,
June 14–19
.
8.
Mani
,
M.
, and
Pillai
,
R.
,
2010
, “
Impact of Dust on Solar Photovoltaic (PV) Performance: Research Status, Challenges and Recommendations
,”
Renewable Sustainable Energy Rev.
,
14
, pp.
3124
3131
. 10.1016/j.rser.2010.07.065
9.
Kimber
,
A.
,
Mitchell
,
L.
,
Nogradi
,
S.
, and
Wenger
,
H.
,
2006
, “
The Effect of Soiling on Large Grid-Connected Photovoltaic Systems in California and the Southwest Region of the United States
,”
Photovoltaic Energy Conversion, Conference Record of the 2006 IEEE 4th World Conference
,
Waikoloa, HI
,
May 7–12
.
10.
Jones
,
R.
,
Baras
,
A.
,
Saeeri
,
A.
,
Al Qahtani
,
A.
,
Al Amoudi
,
A.
,
Al Shaya
,
Y.
,
Alodan
,
M.
, and
Al-Hsaien
,
S.
,
2016
, “
Optimized Cleaning Cost and Schedule Based on Observed Soiling Conditions for Photovoltaic Plants in Central Saudi Arabia
,”
IEEE J. Photovoltaics
,
6
(
3
), pp.
730
738
. 10.1109/JPHOTOV.2016.2535308
11.
Kaldellis
,
J. K.
, and
Kokala
,
A.
,
2010
, “
Quantifying the Decrease of the Photovoltaic Panels’ Energy Yield Due to Phenomena of Natural Air Pollution Disposal
,”
Energy
,
35
, pp.
4862
4869
. 10.1016/j.energy.2010.09.002
12.
Dubai Electricity & Water Authority | Tariff
,
2018
, https://www.dewa.gov.ae/en/customer/services/consumption-services/tariff, Accessed May 5, 2018.
13.
John
,
J.
,
Elnosh
,
A.
,
Safieh
,
A.
,
Almheiri
,
A.
,
Stefancich
,
M.
, and
Banda
,
P.
,
2016
, “
Estimation of Soiling Rates From PV Modules in the Desert Climate of Dubai
,”
EUPVSEC
.
14.
Elnosh
,
A.
,
John
,
J.
,
Alnuaimi
,
A.
,
Quadir
,
J.
,
Stefancich
,
M.
, and
Banda
,
P.
,
2018
, “
Evaluation of Soiling Rates for PV Modules Installed at Different Tilt Angles in Dubai, UAE
,”
Dubai Electricity and Water Authority
.
15.
John
,
J.
,
Tatapudi
,
S.
, and
Tamizhmani
,
G.
,
2014
, “
Influence of Soiling Layer on Quantum Efficiency and Spectral Reflectance on Crystalline Silicon PV Modules
,”
2014 IEEE 40th Photovoltaic Specialist Conference (PVSC)
,
Denver, CO
,
June 8–13
.
16.
Figgis
,
B.
,
Ennaoui
,
A.
,
Ahzi
,
S.
, and
Rémond
,
Y.
,
2017
, “
Review of PV Soiling Particle Mechanics in Desert Environments
,”
Renewable Sustainable Energy Rev.
,
76
, pp.
872
881
. 10.1016/j.rser.2017.03.100