As the global demand for clean energy continues to rise, photovoltaic systems have become one of the most promising renewable energy technologies. Solar panels are widely deployed in various climates, yet their efficiency is significantly affected by environmental factors, particularly the accumulation of dust on the surface. In this study, I aimed to investigate the impact of dust deposition on the performance of solar panels and to develop a practical monitoring system based on visual analysis. The research was motivated by the need for a reliable, non-contact method to evaluate the degree of soiling and to optimize cleaning schedules for large-scale photovoltaic plants. In this paper, I present an experimental setup, an image-processing algorithm, a predictive model linking image gray values to power loss, and an economic optimization framework for determining the optimal cleaning cycle. Throughout this article, I repeatedly refer to the performance of solar panels as the central subject, because the degradation caused by dust is one of the most critical operational challenges in photovoltaic installations.
1. Introduction
The rapid expansion of photovoltaic capacity worldwide has brought attention to the operational reliability and maintenance of solar panels. Among the many factors that reduce the output of solar panels, dust deposition is perhaps the most pervasive and site-dependent one. Dust particles accumulated on the transparent cover of solar panels scatter and absorb incoming sunlight, thereby reducing the amount of solar radiation reaching the photovoltaic cells. This leads to a decrease in the generated power and a corresponding economic loss for the plant operator. Statistics indicate that the power output of solar panels can drop by 2% to 25% even with moderate soiling, and in severe cases, the loss may exceed 50% within a few weeks without cleaning. Therefore, effective monitoring of the soiling status of solar panels is essential for timely maintenance and cost minimization.
Traditional monitoring methods often rely on measuring the electrical output of solar panels and comparing it with a reference model that accounts for irradiance and temperature. However, such approaches cannot directly indicate the spatial distribution or the severity of dust on the surface. With the advancement of digital imaging technology, visual methods have emerged as a promising alternative. By analyzing images of solar panels, one can quantify the degree of surface soiling based on color and texture changes. In this work, I developed a vision-based system that captures images of solar panels under controlled conditions, processes them to remove the metallic grid lines (busbars and fingers), and computes an average gray value that correlates with the dust density. This value is then used to predict the power loss rate of the solar panels, enabling a cost-effective cleaning strategy.
In the following sections, I first review the fundamental theory of photovoltaic systems and the influence of environmental factors on the performance of solar panels. Then, I describe the hardware and software design of the experimental monitoring system. Next, I present the simulated dust deposition experiments and the improved interpolation algorithm used for image denoising and gray-value extraction. Subsequently, I build a prediction model that relates the image gray value to the power loss rate, and I develop a power output degradation model based on the exposure time. Finally, I propose a cleaning-cycle optimization framework that minimizes the total annual cost, which includes the cost of electricity loss and the cost of cleaning maintenance. The results are demonstrated using a 50 MW photovoltaic power plant as a case study, and the optimal cleaning interval is determined to be approximately 10.14 days.
2. Theoretical Background
2.1 Photovoltaic System Fundamentals
A photovoltaic system converts sunlight directly into electricity via the photovoltaic effect. The basic units are solar cells, typically made of silicon, which are connected in series and parallel to form modules. The output power of solar panels depends on several parameters, such as the irradiance, the cell temperature, and the spectral distribution of the incident light. In practice, solar panels are exposed to varying weather conditions, and their performance deviates from the nominal values measured under standard test conditions (irradiance of 1000 W/m², cell temperature of 25°C, and air mass 1.5). The power output of a photovoltaic module can be expressed by
$$P_{\text{max}} = V_{\text{oc}} \cdot I_{\text{sc}} \cdot FF$$
where \(V_{\text{oc}}\) is the open-circuit voltage, \(I_{\text{sc}}\) is the short-circuit current, and \(FF\) is the fill factor. The fill factor depends on the shape of the current-voltage curve and is sensitive to changes in series and shunt resistances. When dust accumulates on the surface of solar panels, the short-circuit current is reduced because less light reaches the cells, while the open-circuit voltage remains almost unchanged. Consequently, the overall maximum power output of solar panels declines proportionally to the loss in \(I_{\text{sc}}\).
2.2 Environmental Factors Affecting Solar Panels
The performance of solar panels is influenced by a variety of climatic and environmental factors. Wind can have a dual effect: it may remove dust particles from the surface, but it can also transport and deposit additional particles. The wind speed affects the rate of dust accumulation, with high wind speeds generally increasing the density of particles on the surface until a certain threshold is reached. The temperature of solar panels also plays a crucial role. As the temperature increases, the bandgap of silicon narrows, which slightly increases the short-circuit current but significantly reduces the open-circuit voltage. The net effect is a decrease in the maximum power output. The temperature coefficient of power for crystalline silicon modules is typically around \(-0.4\%/\text{K}\), meaning that each degree Celsius rise in temperature causes a power drop of approximately 0.4%. This is why cooling effects from wind can actually improve the efficiency of solar panels to some extent.
Dust is arguably the most site-specific factor. The impact of dust on solar panels depends on the size, composition, and density of the particles. Fine particles are more detrimental than coarse particles because they cover the surface more effectively. The chemical composition of dust may also contribute to permanent etching or corrosion of the glass cover, especially when moisture is present. For example, the dissolution of calcite and salt particles in humid air can form a chemically active mud that adheres strongly to the surface. Therefore, cleaning strategies must consider not only the immediate power loss but also the long-term degradation of solar panels.
3. Experimental Setup for Monitoring Dust on Solar Panels
To study the effect of dust accumulation on the performance of solar panels, I designed and built an online monitoring experimental system. The system was installed at a photovoltaic power plant located in a region with frequent dust events in northern China. The setup consisted of seven identical polycrystalline silicon photovoltaic modules from the same production batch, each with 60 cells arranged in series. All modules were mounted at the same tilt angle of 43° and oriented due south, ensuring that they received similar irradiance. One module was kept clean as a reference, while the other six were exposed to different controlled dust densities.
3.1 Hardware Components
The hardware architecture included a data acquisition unit, power monitoring sensors, temperature sensors, a high-resolution camera, and a weather station. The electrical parameters (voltage and current) of each solar panel were measured using a smart DC power meter PZ96-DE. The meter communicated with the host computer via RS485 using the Modbus-RTU protocol. To measure the backsheet temperature of solar panels, I used DS18B20 digital temperature sensors with an accuracy of ±1°C. The weather station measured global horizontal irradiance, wind speed, wind direction, ambient temperature, and humidity. A high-resolution industrial camera (DS-2DE7320IW-A) was placed perpendicular to the surface of the solar panels to capture images at regular intervals. The distance between the camera and the modules was chosen to avoid shading while providing sufficient spatial resolution.
Table 3.1 lists the key specifications of the photovoltaic modules used in the experiments.
| Parameter | Value |
|---|---|
| Maximum power (Pmax) | 260 W |
| Open-circuit voltage (Voc) | 37.7 V |
| Short-circuit current (Isc) | 8.95 A |
| Voltage at maximum power (Vmp) | 30.4 V |
| Current at maximum power (Imp) | 8.58 A |
| Module efficiency | 16.01% |
| Temperature coefficient of power | -0.408 %/K |
| Number of cells | 60 (6×10 polycrystalline) |
| Cell size | 156 mm × 156 mm |
| Maximum system voltage | 1000 V |
The data acquisition unit was based on a digital signal processor (DSP) that collected all measurement signals and transmitted them to a central control room. The central control room housed an industrial computer running a custom-designed software platform for data processing, storage, and display.
3.2 Software Design
The software was developed in C# using the Microsoft .NET framework. It provided a graphical user interface for real-time monitoring of the electrical parameters, temperature, irradiance, and camera images. The software also performed data conversion, calculation of the instantaneous power, and computation of the power loss rate using the temperature-corrected efficiency model. All data were stored in a Microsoft Access database for future retrieval and analysis. The flow of the monitoring algorithm is shown in Figure 3.2 (not included here). The main interface displayed the live data, the processed image, and the calculated values, along with alarm indicators when the power loss exceeded a preset threshold.
The online monitoring method was based on maximum power point tracking (MPPT). For a given photovoltaic array, the output voltage and current were continuously monitored. The real-time generation efficiency of the solar panels was calculated using the following equation:
$$\eta_a = \frac{P_{\text{max}}}{G \cdot S_r} \times 100\%$$
where \(\eta_a\) is the real-time generation efficiency, \(P_{\text{max}}\) is the maximum power output of the module under the given irradiance and temperature, \(G\) is the total irradiance, and \(S_r\) is the total area of the module (including the frame). Because the efficiency of solar panels varies with temperature, a temperature correction was applied. The backsheet temperature \(T_b\) of each module was measured, and the efficiency was recalculated under the nominal operating cell temperature (NOCT) reference condition (45°C). The corrected efficiency is given by
$$\eta = \frac{P_{\text{max}} – k \cdot (T_b – T_s)}{G \cdot A_m}$$
where \(k\) is the power temperature coefficient, \(T_s\) is the reference temperature, and \(A_m\) is the module area. In this way, the dust effect could be isolated from the temperature effect, because all measurements were taken at midday (11:00–13:00) when the incident angle was approximately constant.
4. Simulation of Dust Deposition and Visual Analysis
4.1 Preparation of Dust Particles
To simulate the natural accumulation of dust on solar panels, I collected dust samples from the surface of photovoltaic panels at a local power station. The samples were analyzed using a laser particle size analyzer. The particle size distribution ranged from 1 to 100 µm, with a median size of about 15 µm. The dust was dried and sieved through a 150-mesh screen to obtain particles smaller than 100 µm, ensuring uniformity in the experiments. Table 4.1 shows the main characteristics of the dust particles.
| Size fraction (µm) | Volume percentage (%) |
|---|---|
| 0–10 | 18.2 |
| 10–20 | 34.6 |
| 20–40 | 27.5 |
| 40–100 | 19.7 |
4.2 Artificial Dust Deposition on Solar Panels
Six test modules were artificially coated with dust at densities of 5, 10, 15, 20, 25, and 30 g/m², respectively. A seventh module was kept clean as a reference. The dust deposition procedure was designed to mimic natural settling as closely as possible. Each module was laid flat on the ground and covered with a sealed transparent acrylic box. A known mass of dust was blown into the box through a side inlet using a blower, creating a uniform suspension. After the air was turned off, the dust was allowed to settle for six hours in a dry environment. To improve the adhesion of particles to the glass surface, a humidifier was used to slightly moisten the surface after the initial settling. This step was necessary to prevent the dust from falling off during subsequent handling. The resulting dust layers were visually uniform and representative of real field conditions.
4.3 Image Acquisition and Grid-Line Removal
For each dust density, a high-resolution image of the corresponding solar panel was captured. The images showed the characteristic silver grid lines (fingers) that collect current from the cells. These grid lines significantly interfere with the gray-level analysis because they are much brighter than the surrounding dark silicon. Therefore, it was essential to remove the grid-line pattern from the images before calculating the average gray value. I developed a modified interpolation algorithm that combines linear interpolation and third-order interpolation based on the local variation of pixel values.
The algorithm first identifies the rows (or columns) that correspond to the grid lines. For a given pixel position \((i,j)\), let \(G(i,j)\) be its gray value. The difference between the neighboring pixels above and below the candidate grid line is computed as
$$F = G(i-1,j) – G(i+1,j)$$
and the relative error is
$$\delta = \left| \frac{F}{G(i-1,j)} \right|$$
If \(\delta\) is smaller than a preset threshold \(d\), the variation in the local background is small, and the pixel value is replaced by the linear interpolation of the two neighboring pixels:
$$G_{\text{new}}(i,j) = \frac{G(i-1,j) + G(i+1,j)}{2}$$
If \(\delta\) exceeds the threshold \(d\), the background texture changes significantly, and a more sophisticated third-order interpolation is applied. The third-order interpolation uses a 4×4 neighborhood around the pixel to preserve edge details. The interpolation kernel can be expressed as a piecewise cubic polynomial:
$$w(s) = \begin{cases} |s|^3 – 2|s|^2 + 1, & 0 \le |s| < 1 \\[6pt] -|s|^3 + 5|s|^2 – 8|s| + 4, & 1 \le |s| < 2 \\[6pt] 0, & |s| \ge 2 \end{cases}$$
where \(s\) is the normalized distance. The interpolated value is the weighted sum of the 16 neighboring pixels. Finally, a least-squares fitting is applied to replace the grid-line pixels with the fitted values. This approach effectively removes the grid lines while preserving the fine texture of the dust layer. Figure 4.1 (not shown) compares a raw image and the corresponding grid-line-removed image for several dust densities.
4.4 Calculation of the Average Gray Value
After grid-line removal, the image is transformed into an 8-bit grayscale image. Each pixel has a gray value between 0 (black) and 255 (white). The average gray value of the entire module image is calculated as
$$\overline{Gray} = \frac{1}{n} \sum_{i=1}^{n} Gray_i$$
where \(n\) is the total number of pixels in the image and \(Gray_i\) is the gray value of pixel \(i\). In practice, I converted the original RGB image to grayscale using the weighted formula:
$$Gray = 0.3 \cdot R + 0.6 \cdot G + 0.1 \cdot B$$
Table 4.2 summarizes the average gray values for the seven dust densities.
| Dust density (g/m²) | Average gray value |
|---|---|
| 0 | 50.21 |
| 5 | 76.56 |
| 10 | 87.58 |
| 15 | 94.37 |
| 20 | 98.02 |
| 25 | 104.06 |
| 30 | 112.11 |
It is evident that the average gray value increases almost linearly with the dust density. This is because the dust layer scatters light, making the surface appear brighter. The relationship can be expressed by the fitted equation
$$\overline{Gray} = 51.6 + 1.13 \rho$$
where \(\rho\) is the dust density in g/m². This simple linear model provides a convenient mapping between the visual feature and the physical amount of dust on solar panels.
5. Experimental Results and Modeling
5.1 Efficiency of Solar Panels under Different Dust Loads
Using the monitoring system, I recorded the electrical output of the seven solar panels at midday for a period of several weeks. Table 5.1 lists the measured generation efficiency of the clean module. The average efficiency was approximately 20.11%, which is close to the nominal efficiency of the module.
| Time | Irradiance (W/m²) | Voltage (V) | Current (A) | Power (W) | Backsheet temp (°C) | Efficiency (%) |
|---|---|---|---|---|---|---|
| 11:00 | 643 | 28.0 | 6.32 | 176.96 | 3.39 | 20.34 |
| 11:05 | 643 | 28.0 | 6.28 | 175.84 | 3.51 | 20.20 |
| 11:10 | 653 | 28.0 | 6.33 | 177.24 | 3.17 | 20.07 |
| 11:15 | 656 | 28.0 | 6.36 | 178.08 | 3.04 | 20.09 |
| 11:20 | 655 | 28.0 | 6.34 | 177.52 | 2.98 | 20.06 |
| 11:25 | 653 | 28.0 | 6.34 | 177.52 | 3.14 | 20.11 |
| 11:30 | 655 | 28.0 | 6.34 | 177.52 | 3.15 | 20.05 |
Using the temperature correction described in Section 3.2, I computed the efficiency of the dust-covered modules under the same reference conditions. Table 5.2 presents the average efficiency values for each dust density.
| Dust density (g/m²) | Average efficiency (%) |
|---|---|
| 0 | 20.11 |
| 5 | 19.32 |
| 10 | 17.84 |
| 15 | 15.09 |
| 20 | 14.90 |
| 25 | 12.33 |
| 30 | 10.18 |
The data clearly show that the efficiency of solar panels decreases sharply as the dust density increases. The relationship appears nearly linear in the investigated range. A similar trend is observed for the power loss rate, defined as the relative decrease in efficiency compared to the clean condition:
$$\eta_{pl} = \frac{\eta_c – \eta_d}{\eta_c} \times 100\%$$
where \(\eta_c\) is the efficiency of the clean module and \(\eta_d\) is the efficiency of the dusty module. Table 5.3 lists the power loss rates for each dust density.
| Dust density (g/m²) | Power loss rate (%) |
|---|---|
| 5 | 4.11 |
| 10 | 11.34 |
| 15 | 24.87 |
| 20 | 26.04 |
| 25 | 38.72 |
| 30 | 49.30 |
It is worth noting that the power loss rate is not perfectly linear with dust density; the rate of increase is slightly higher at larger densities, likely due to a shadowing effect that becomes more pronounced as the dust layer becomes thicker. Nevertheless, for practical purposes, a linear approximation is acceptable within a limited range.
5.2 Prediction Model Using the Average Gray Value
Since the average gray value of the images is strongly correlated with the dust density, I directly established a relationship between the gray value and the power loss rate. Using the data from Tables 4.2 and 5.3, I performed a linear regression, yielding
$$\eta_{pl} = 1.1268 \cdot (\overline{Gray} – 49.82)$$
Equivalently, the power loss rate can be expressed as
$$\eta_{pl} = 1.1268 \cdot \overline{Gray} – 56.13$$
This model allows a quick estimation of the performance degradation of solar panels solely from an image captured by a standard camera. The closeness of the fit is shown in Figure 5.1 (not included). The coefficient of determination \(R^2\) was above 0.94, indicating a strong linear relationship.
5.3 Power Output Degradation Model
In addition to the instantaneous dust-density model, I developed a time-dependent degradation model that describes the evolution of the power loss rate over the duration of dust accumulation. The accumulation of dust on solar panels is a dynamic process in which particles deposit and are removed by wind and rain. A classic asymptotic model, similar to the Kern-Seaton fouling model, can be used:
$$\eta_{pl}(t) = \eta_{pl}^* \cdot (1 – e^{-b t})$$
where \(\eta_{pl}^*\) is the asymptotic maximum power loss rate for a given site, \(b\) is the rate constant that depends on the local dust concentration and weather conditions, and \(t\) is the time since the last cleaning (in days). To determine the model parameters, I monitored the power output of a naturally soiling solar panel for several weeks. The measured power loss rates, shown in Table 5.4, were fitted to the exponential model.
| Exposure time (days) | Power loss rate (%) |
|---|---|
| 0 | 0 |
| 7 | 3.25 |
| 14 | 6.80 |
| 21 | 9.64 |
| 28 | 11.85 |
| 35 | 13.79 |
| 42 | 15.32 |
The fitted expression for the investigated region was
$$\eta_{pl}(t) = 18.9 \cdot (1 – e^{-0.043 t})$$
This asymptotic model predicts that the power loss rate approaches 18.9% after a long period without cleaning. The result is consistent with the observation that in arid regions, solar panels can lose approximately 15%–20% of their output within two months. The model is valuable because cleaning decisions must be based on the expected cost of the power loss, which increases nonlinearly with time.
6. Cleaning Cycle Optimization and Cost Evaluation
The ultimate goal of monitoring the dust status of solar panels is to determine the most economical cleaning schedule. Frequent cleaning reduces the energy loss but increases the maintenance cost. On the other hand, infrequent cleaning saves labor and water but incurs a higher loss of electricity production. I developed an optimization framework that balances these two opposing costs.
6.1 Cost Components
For a photovoltaic power plant, the total annual cost caused by dust accumulation is the sum of three components: the electricity loss cost during the cleaning interval (\(E_d\)), the cleaning and maintenance cost (\(E_c\)), and the residual dust cost immediately after cleaning (\(E_m\)). Because cleaning is never perfect, some residual dust remains, and its effect is accounted for in the residual cost.
The electricity loss cost during one cleaning interval is calculated by integrating the instantaneous power loss over the interval:
$$E_d = \int_0^{t_{ci}} e_d \, dt$$
where \(t_{ci}\) is the cleaning interval in days and \(e_d\) is the daily cost of lost electricity, which is proportional to the power loss rate:
$$e_d = P \cdot t_d \cdot C_e \cdot \eta_{pl}(t)$$
Here, \(P\) is the installed capacity (MW), \(t_d\) is the daily utilization time (hours), \(C_e\) is the grid-connected electricity price (yuan/kWh), and \(\eta_{pl}(t)\) is the instantaneous power loss rate. Similarly, the cleaning cost per cycle is the product of the specific cleaning cost and the total module area:
$$E_c = C_c \cdot F \cdot P$$
where \(C_c\) is the cleaning cost per square meter and \(F\) is the area per unit capacity (m²/MW). The residual dust cost is the cost of the power loss caused by the dust that remains after cleaning. It can be expressed as
$$E_m = e_d(t_{ci}) \cdot \frac{t_{cp}}{t_{ci}} \cdot t_{ci}$$
In practice, I adopted the following simplified expression for the residual dust cost over a cycle, based on the assumption that the residual dust density is a fixed fraction of the pre-cleaning value:
$$E_m = \frac{t_{cp}}{t_{cp} + t_{ci}} \cdot \int_0^{t_{ci}} e_d(t) \, dt$$
where \(t_{cp}\) is the cleaning time (hours). The total annual cost is then
$$E = \frac{\tau}{t_{ci} + t_{cp}} \cdot (E_d + E_c + E_m)$$
where \(\tau\) is the annual operating time in hours (e.g., 365 days). The optimal cleaning interval \(t_{ci}^*\) is the one that minimizes \(E\).
6.2 Case Study: 50 MW Photovoltaic Plant
I applied the optimization model to a 50 MWp photovoltaic power plant located in a dusty region. The relevant parameters are listed in Table 6.1.
| Parameter | Value |
|---|---|
| Installed capacity \(P\) | 50 MW |
| Annual operating time \(\tau\) | 365 days |
| Average daily utilization time \(t_d\) | 3.56 h |
| Grid electricity price \(C_e\) | 0.95 yuan/kWh |
| Cleaning cost per unit area \(C_c\) | 0.25 yuan/m² |
| Module area per unit capacity \(F\) | 50 m²/MW (approx.) |
Using the degradation model \(\eta_{pl}(t) = 18.9 \cdot (1 – e^{-0.043 t})\), I computed the total annual cost for various cleaning intervals. Figure 6.1 (not included) shows that the total cost first decreases rapidly and then increases, yielding a clear minimum. The optimal cleaning interval was found to be \(t_{ci}^* = 10.14\) days, with a cleaning time of one hour. This corresponds to approximately 36 cleaning cycles per year. At this optimal interval, the annual electricity loss cost is 0.782 million yuan, the cleaning maintenance cost is 1.608 million yuan, and the residual dust cost is 0.335 million yuan. The total minimum annual cost is 2.725 million yuan, which is equivalent to 54,500 yuan per MW per year. This amount is about 5.6% of the expected annual revenue from electricity generation, representing a substantial saving compared to a non-optimized cleaning schedule.
The corresponding power loss rate at the optimal cleaning interval is 4.45%, and the corresponding average gray value of the solar panel image is 54.89. This means that if the monitoring system captures an image and computes a gray value above approximately 55, the cleaning crew should be dispatched. Thus, the vision-based method provides a straightforward and objective threshold for maintenance decisions.
6.3 Sensitivity Analysis
To understand how the optimal cleaning interval changes with site-specific parameters, I conducted a sensitivity analysis by varying the installed capacity, the daily utilization time, the grid electricity price, and the cleaning cost per square meter. The results are summarized in Table 6.2.
| Parameter (base value) | Change | Optimal interval (days) |
|---|---|---|
| Installed capacity (50 MW) | +20% | 9.8 |
| Installed capacity (50 MW) | -20% | 10.6 |
| Daily utilization time (3.56 h) | +20% | 9.5 |
| Daily utilization time (3.56 h) | -20% | 10.9 |
| Grid electricity price (0.95 yuan/kWh) | +20% | 9.7 |
| Grid electricity price (0.95 yuan/kWh) | -20% | 10.8 |
| Cleaning cost (0.25 yuan/m²) | +50% | 12.5 |
| Cleaning cost (0.25 yuan/m²) | -50% | 8.3 |
As expected, a higher value of electricity (from larger capacity, longer utilization, or higher price) leads to a shorter optimal cleaning interval because the cost of lost energy becomes more significant relative to the cleaning cost. Conversely, a higher cleaning cost stretches the interval because cleaning becomes more expensive. This analysis can help plant operators tailor their maintenance strategies to local economic conditions.
7. Conclusion and Future Work
In this work, I proposed a comprehensive vision-based approach to monitor and evaluate the dust deposition on solar panels. The study demonstrated that the average gray value of a properly processed image is a reliable proxy for the dust density on the surface of solar panels. By combining image analysis with electrical performance measurements, I established a linear prediction model that converts the gray value into a power loss rate. Furthermore, I developed an asymptotic degradation model that describes how the power loss of solar panels increases with exposure time in the absence of cleaning. This model was integrated into an economic optimization framework that determines the most cost-effective cleaning interval for a photovoltaic plant.
The case study of a 50 MW photovoltaic plant revealed that the optimal cleaning cycle is approximately 10.14 days, which results in a minimum annual cost of 2.725 million yuan. The corresponding image gray value threshold of 54.89 offers a simple operational rule: when the average gray value of a solar panel image exceeds this threshold, cleaning should be performed. The sensitivity analysis further indicated that the optimal interval is robust to moderate changes in economic parameters, but it should be adjusted according to the local electricity price and cleaning cost.
Several limitations of the present study should be addressed in future research. First, the experiments were conducted with a single type of dust and in a specific geographical region. The chemical composition and particle size distribution of dust vary significantly from site to site, so the calibration coefficients of the model may need to be re-evaluated when applying the method to other locations. Second, the image processing algorithm was designed for a fixed camera angle and lighting condition. In practice, variations in illumination and camera perspective may affect the gray value calculation. The use of a reference color card placed next to the solar panels could help normalize the images. Third, the degradation model assumed a constant asymptotic power loss rate, but in reality, rainfall and wind events can partially clean the panels, causing the loss curve to fluctuate. An extended model that incorporates meteorological data would improve the accuracy of the cleaning-cycle optimization.
In conclusion, the vision-based monitoring system presented in this paper offers a practical and effective solution for maintaining the high performance of solar panels in dusty environments. By providing real-time information about the soiling status of solar panels, the system enables proactive maintenance that maximizes energy yield and minimizes operational costs. The successful implementation of the proposed method at the pilot site demonstrates its potential for widespread adoption in commercial photovoltaic power plants. As the global fleet of solar panels continues to grow, robust and intelligent monitoring techniques will become increasingly important, and the results of this study may contribute to the development of autonomous cleaning systems that further improve the sustainability of solar energy.