Photovoltaic (PV) power generation is one of the most promising approaches to harvest solar energy. However, PV modules are subject to overheating caused by factors such as mechanical damage, partial shading, cell mismatch, and soiling. This overheating not only degrades the photoelectric conversion efficiency but may also induce secondary breakdown of the cells or even catastrophic fire accidents. Therefore, real-time temperature monitoring of solar panels is essential to prevent high-temperature damage and to improve the overall energy yield. Moreover, if the module temperature can be predicted before a critical threshold is exceeded, proactive cooling measures can be activated, which further enhances the operational stability of the PV plant.
Traditional temperature monitoring techniques for solar panels include infrared thermography, fiber Bragg grating (FBG) sensors, and electrical sensors. Each of these approaches has its own limitations. Infrared thermography suffers from low accuracy due to surface emissivity variations and atmospheric disturbances, and it cannot provide a fully distributed measurement along the panel array. FBG sensors can achieve quasi-distributed sensing but leave blind spots and require a complex demodulation process. Electrical sensors are vulnerable to electromagnetic interference, corrosion, and are not suitable for large-scale deployment. To overcome these limitations, this work presents a distributed optical fiber Raman temperature sensing (RDTS) system specifically designed for solar panel temperature monitoring and early warning. The proposed system leverages the Raman scattering effect in optical fibers to deliver continuous, real-time temperature profiles over kilometers of sensing fiber.
In this paper, I describe the complete development of the RDTS system, including software and hardware design, system integration, performance evaluation, and a novel fuzzy temperature difference threshold method (FTDTM) for predictive alarm generation. Experimental results demonstrate that the system achieves a sensing range of 11 km, a response time of 2 s, a temperature accuracy of ±1.00 °C, a temperature resolution of 0.50 °C, and a spatial resolution of 1.0 m. The proposed FTDTM enables one-minute-ahead prediction of module temperature with an average absolute error of 1.08 °C when the window base is set to 2. The work provides a viable solution for large-scale distributed temperature monitoring and early warning for solar panels.
2. Theoretical Background of Raman Distributed Temperature Sensing
Raman scattering is an inelastic scattering process that occurs when an incident photon interacts with molecular vibrations in the optical fiber. The scattered spectrum comprises the Rayleigh peak (elastic), the Stokes component (lower frequency), and the anti-Stokes component (higher frequency). The intensity of the anti-Stokes light is strongly temperature-dependent, whereas the Stokes light is only weakly dependent on temperature. Thus, the ratio of anti-Stokes to Stokes intensity can be used to determine the local temperature.
For a laser pulse injected into the fiber, the backscattered Stokes and anti-Stokes powers at a distance \(L\) from the source can be expressed as:
$$P_S(L,T) = K_S \nu_S^4 S_b R_S(T) P_0 \exp[-(a_0+a_S)L]$$
$$P_A(L,T) = K_A \nu_A^4 S_b R_A(T) P_0 \exp[-(a_0+a_A)L]$$
where \(K_S\) and \(K_A\) are constants related to the scattering cross-sections, \(\nu_S\) and \(\nu_A\) are the frequencies of the Stokes and anti-Stokes photons, \(S_b\) is the backscatter capture coefficient, \(P_0\) is the incident pulse power, \(a_0\), \(a_S\), and \(a_A\) are the attenuation coefficients at the corresponding wavelengths, and \(R_S(T)\) and \(R_A(T)\) are temperature-dependent population factors:
$$R_S(T) = [1-\exp(-h\Delta\nu/kT)]^{-1}$$
$$R_A(T) = [\exp(h\Delta\nu/kT)-1]^{-1}$$
Here, \(h\) is Planck’s constant, \(k\) is Boltzmann’s constant, \(\Delta\nu\) is the Raman frequency shift, and \(T\) is the absolute temperature in Kelvin.
Using the ratio of anti-Stokes to Stokes intensity at a reference temperature \(T_0\) and at an unknown temperature \(T\), the temperature can be retrieved as:
$$T = \left[ \frac{1}{T_0} – \frac{k}{h\Delta\nu} \ln\left( \frac{F(T)}{F(T_0)} \right) \right]^{-1}$$
where \(F(T) = P_A(T)/P_S(T)\). This dual-channel demodulation method eliminates the dependency on source power fluctuations and other common-mode disturbances. Alternatively, a loop configuration can be adopted to compensate for position-dependent attenuation and bending losses.
For location determination, the system utilizes optical time-domain reflectometry (OTDR). The distance \(L\) to a scattering point is calculated by:
$$L = \frac{v t}{2}$$
where \(v = c/n\) is the speed of light in the fiber, \(c\) is the speed of light in vacuum, \(n\) is the effective refractive index of the fiber core, and \(t\) is the round-trip time of the optical pulse.
3. System Software Design
The host computer software was developed using LabVIEW, a graphical programming environment that enables rapid development of data acquisition, processing, and display modules. The software architecture comprises four main modules: signal acquisition and parameter setting, temperature demodulation, data storage, and database query. The overall workflow is illustrated in the following sequence:
- Check the hardware connections of the data acquisition card and laser source.
- Set the acquisition parameters, such as sampling rate, number of accumulations, and sensing distance.
- Perform a calibration procedure at a known reference temperature.
- Acquire the Stokes and anti-Stokes backscattered light intensity data.
- Apply the demodulation algorithm to compute the temperature profile along the fiber.
- Display the real-time temperature waveform and store the data in a user-selected file path.
- Compare the temperature values with a preset alarm threshold and trigger an alarm if necessary.
The temperature demodulation module implements both the dual-channel ratio method and the loop demodulation method. The dual-channel method uses the real-time ratio of anti-Stokes to Stokes intensities, while the loop method averages forward and backward propagation measurements to cancel the asymmetric loss effects. In LabVIEW, the program uses a flat sequence structure to ensure the correct order of execution. The front panel includes:
- Waveform display area for raw light intensity and temperature profiles.
- Alarm information area showing the exact position and temperature of abnormal points.
- Parameter setting area for sensing distance, average number, and filtering method.
- Additional function area for historical data retrieval and real-time temperature logging.
The data storage module automatically creates a new folder based on the current date and saves the data with incremental file names. The database query module allows the user to input a specific date and time, then retrieves the corresponding temperature data and displays statistical information such as maximum, minimum, and average temperatures.
4. System Hardware Design and Integration
The hardware of the RDTS system consists of the following main components:
- High-speed pulse laser module: A 1550.1 nm wavelength distributed-feedback laser with adjustable pulse width (10 ns) and peak power up to 4 W. The repetition rate is set according to the required sensing range.
- Wavelength division multiplexer (WDM): A 1×3 WDM that couples the incident pulse into the sensing fiber and separates the backscattered anti-Stokes (1450 nm) and Stokes (1650 nm) wavelengths. The isolation is 30 dB to suppress Rayleigh crosstalk.
- Two avalanche photodiode (APD) detectors: These convert the weak optical signals into electrical signals. The APDs have a bandwidth of 100 MHz, a dark current of 2 nA, and a responsivity of 8.5 A/W.
- Two electrical signal amplifiers: They amplify the APD output to a level suitable for the data acquisition card.
- High-speed data acquisition card (DAQ): A PCI104 card with a sampling rate of 100 MS/s, 12-bit resolution, and on-board FPGA for real-time cumulative averaging. The maximum number of accumulations is 32768, which improves the signal-to-noise ratio by a factor of 180.
- Industrial computer motherboard: A 3.5-inch mainboard with an Intel i5 processor, running at 1.86 GHz, embedded in a rugged chassis.
- Portable chassis: The system is packaged in a carbon-fiber enclosure with four air inlets and two outlets for efficient heat dissipation. The outer panel includes optical fiber flange connectors and openings for display and USB connections.
The system layout is shown in the experimental setup: the pulse laser output is connected to the WDM input. The WDM common port is connected to the sensing fiber. The two output ports are fed into the APDs. The APD output signals are sent to the two input channels of the DAQ. The DAQ is triggered by the laser sync signal. The computer receives the digitized data and performs the demodulation and visualization.
5. Performance Evaluation
The integrated RDTS system was tested with a 11 km multimode fiber. An 80 m reference fiber placed in a high-precision thermostat was used for calibration. Two sections of fiber (each 20 m long) were placed in a temperature-controlled chamber at 35.0 °C, 50.0 °C, 65.0 °C, 80.0 °C, and 95.0 °C, while the rest of the fiber remained at room temperature. Figure below (not shown) illustrates the temperature profiles obtained at different distances. The measured temperature accuracy at 2.27 km and 10.36 km are summarized in the following table.
| Location | Maximum error (°C) | Average error (°C) | Spatial resolution (m) |
|---|---|---|---|
| 2.27 km | 0.80 | 0.60 | 2.0 |
| 10.36 km | 1.24 | 1.00 | 4.6 |
The degradation of accuracy and spatial resolution with distance is attributed to the increasing attenuation and intermodal dispersion. Nevertheless, the system meets the designed specifications for a sensing distance of 11 km, a response time of 2 s, a temperature accuracy of ±1.00 °C, a temperature resolution of 0.50 °C, and a spatial resolution of 1.0 m at the near end. The overall system parameters are listed in the table below.
| Parameter | Value |
|---|---|
| Temperature accuracy | ±1 °C |
| Positioning accuracy | ±1 m |
| Measurement distance | 11 km |
| Response time | 3 s |
| Temperature resolution | 0.5 °C |
| Laser wavelength | 1550 nm |
| Fiber type | Multimode (62.5/125 µm) |
6. Solar Panel Temperature Monitoring Scheme
The RDTS system was deployed for the distributed monitoring of multiple solar panels. The sensing fiber was attached to the backside of the PV modules using a thermally conductive epoxy adhesive. The fiber was routed horizontally across the rear surface, enabling the measurement of the temperature gradient along each module. The monitoring layout is illustrated in the schematic diagram (not reproduced here). The experimental setup involved seven PV modules divided into three groups:
- Group 1 (P1, P2, P3): cleaned surface, tilt angle 60°.
- Group 2 (P4, P5): dusty surface, tilt angle 60°.
- Group 3 (P6, P7): cleaned surface, tilt angle 30°.
The distributed temperature profiles were recorded at different times of the day. The highest local temperature reached 39.30 °C at noon, while at 16:00 the temperature dropped to 18.72 °C. The temperature variation follows the solar radiation intensity. A comparison of the three groups reveals the following effects:
- The cleaned panels at 60° (Group 1) showed the same temperature distribution within the group, confirming the consistency of the RDTS measurements.
- The maximum temperature difference due to the change in radiation intensity (from T1 to T5) was 9.58 °C.
- The panels with dust (Group 2) were hotter than the cleaned panels (Group 1) by up to 3.01 °C at T1, because the dust layer blocked heat dissipation and caused localized hot spots.
- The panels at 30° (Group 3) had a lower temperature than those at 60° by up to 3.99 °C, meaning that the tilt angle affects the thermal balance of solar panels.
- The combined effect of dust and tilt angle resulted in a maximum temperature difference of 6.24 °C between Group 2 and Group 3.
These results indicate that radiation intensity, soiling, and tilt angle are critical factors that influence the temperature distribution of solar panels.
7. Correlation between Module Temperature and Output Power
During the monitoring experiment, I also recorded the real-time output power of the PV modules. It was observed that the output power and the module temperature reached their peaks simultaneously at 12:40 PM, with values of 157 W and 39.60 °C, respectively. A sudden drop in both output power and temperature occurred at 13:50 due to cloud cover, and they recovered by 15:00. This strong correlation suggests that temperature monitoring can be used as a proxy to estimate the power generation status and to detect anomalies such as partial shading or cell failure.
8. Predictive Alarm Model Based on Fuzzy Temperature Difference Threshold Method
To enable early warning before the temperature reaches a dangerous level, I developed a predictive alarm algorithm called the Fuzzy Temperature Difference Threshold Method (FTDTM). This method is based on fuzzy time series theory and is designed to handle the large amount of distributed data from the RDTS system. The algorithm predicts the temperature at the next time step by analyzing the historical temperature differences within a sliding window. The FTDTM consists of the following six steps:
Step 1: Domain partition. Let \(w\) be the prediction window base. The temperature at each minute is averaged to reduce noise. From the historical data, the maximum temperature difference \(\Delta T_{\mathrm{max}}\) and minimum difference \(\Delta T_{\mathrm{min}}\) are calculated. The universe of discourse \(U\) is defined as:
$$U = [\Delta T_{\mathrm{min}} – T_1, \Delta T_{\mathrm{max}} + T_2]$$
where \(T_1\) and \(T_2\) are positive constants. The domain is divided into six equal intervals.
Step 2: Fuzzy relation establishment. Each temperature difference is fuzzified into one of the linguistic fuzzy sets \(A_i\). The membership function is chosen such that the difference \(\Delta T\) is converted into a fuzzy set \(A_i\) if it belongs to interval \(u_i\) with maximum membership.
Step 3: Calculation of relation matrix. The standard matrix at time \(t\) is defined as:
$$C(t) = F(t-1) = [C_1, C_2, \ldots, C_m]$$
and the operation matrix for the window base \(w\) is:
$$O_w(t) = \begin{bmatrix} F(t-2) \\ F(t-3) \\ \vdots \\ F(t-w-1) \end{bmatrix}$$
The relation matrix \(R(t, w, m)\) is obtained by multiplying each row of \(O_w(t)\) with the corresponding column of \(C^T(t)\):
$$R(t, w, m) = O_w(t) \boxtimes C^T(t)$$
where \(\boxtimes\) denotes element-wise multiplication. The fuzzy time series \(F(t)\) is the maximum value of each column in \(R\).
Step 4: Prediction of temperature difference. If all elements of \(F(t)\) are zero, no significant trend is detected. If the maximum appears only once, the predicted difference \(m_i\) is the midpoint of the corresponding interval \(u_i\). If the maximum appears \(k\) times, the predicted difference is the average of the midpoints:
$$m_i = \frac{m_1 + m_2 + \cdots + m_k}{k}$$
Step 5: Prediction of temperature. The predicted temperature at time \(t\) is \(T_t = T_{t-1} + m_i\). The predicted value is clamped to the historical minimum and maximum bounds to prevent large overshoots.
Step 6: Alarm threshold. A temperature alarm threshold \(T_h\) is set. When the predicted temperature exceeds \(T_h\), a cooling system is activated slightly before the actual threshold is reached.
9. Experimental Results of the Predictive Model
I conducted a six-hour continuous monitoring experiment on a solar panel. The RDTS system collected temperature data every minute. The FTDTM was then applied to predict the temperature one minute ahead. The real-time measured temperature and the predicted temperature are plotted as curves (not shown). The results indicate that the predicted temperature follows the measured trend closely except when abrupt changes occur due to sudden cloud cover.
The temperature error between the prediction and the actual measurement is shown in a separate plot (not reproduced). The average absolute error over the six-hour period was 1.08 °C, and the error fluctuation range was ±3.70 °C.
To evaluate the effect of the window base, I tested values of \(w\) from 2 to 8. The average relative prediction errors are summarized in the table below.
| Window base (w) | w=2 | w=3 | w=4 | w=5 | w=6 | w=7 | w=8 |
|---|---|---|---|---|---|---|---|
| Error (%) | 3.16 | 3.29 | 3.73 | 3.76 | 3.83 | 3.88 | 3.92 |
It is clear that a smaller window base yields better prediction accuracy because the most recent temperature difference has the highest correlation with the near-future trend. Therefore, the optimal window base is \(w=2\), providing an average absolute error of 1.08 °C.
10. Discussion on the Impact of Environmental Factors
The experiments demonstrated that solar panels exhibit complex temperature distribution patterns due to varying solar radiation, soiling levels, and tilt angles. The RDTS system successfully captured these distributed temperature profiles, which would be impossible with single-point sensors. The prediction capability of FTDTM further enhances the usability of the system by allowing proactive interventions. However, the current model uses only temperature differences as the input variable. In practice, the temperature of solar panels is also affected by wind speed, ambient temperature, humidity, and load current. Future work will focus on developing a multi-factor FTDTM that incorporates these additional parameters to improve prediction accuracy.
11. Conclusion
In this thesis, I have presented the complete development and validation of a distributed optical fiber Raman temperature sensing (RDTS) system for solar panel temperature monitoring and early warning. The system software, designed in LabVIEW, provides a user-friendly interface for real-time temperature visualization, data storage, and historical query. The hardware integration achieved a portable and robust instrument with a sensing distance of 11 km, response time of 2 s, temperature accuracy of ±1.00 °C, temperature resolution of 0.50 °C, and spatial resolution of 1.0 m. Field experiments on solar panels revealed that radiation intensity, dust coverage, and tilt angle significantly influence the temperature distribution. The maximum temperature differences due to these factors were 9.58 °C, 3.01 °C, and 3.99 °C, respectively. The proposed fuzzy temperature difference threshold method (FTDTM) enables accurate one-minute-ahead prediction of the module temperature. With a window base of 2, the average absolute prediction error was 1.08 °C, and the error fluctuation range was ±3.70 °C. These results confirm that the RDTS system with FTDTM is a promising solution for large-scale distributed monitoring and risk prevention in photovoltaic power plants.
12. Future Work
There are several avenues for further improvement:
- Optimizing the software execution to achieve parallel processing, thus reducing the overall computational time.
- Enhancing the WDM isolation to further suppress Rayleigh crosstalk and improve the signal-to-noise ratio.
- Refining the fiber routing on each solar panel to capture more detailed thermal gradients, using multiple loops per module.
- Investigating adaptive domain partition strategies for FTDTM to balance computational complexity and prediction accuracy.
- Extending the FTDTM to a multi-factor version that incorporates environmental variables such as irradiance, wind speed, and ambient temperature for higher precision.
In summary, the combination of RDTS and FTDTM offers a robust, scalable, and predictive monitoring framework for solar panels, contributing to safer and more efficient photovoltaic energy production.