1. Introduction
The global energy landscape is undergoing a profound transformation driven by the twin imperatives of mitigating climate change and ensuring long-term energy security. As fossil fuel reserves dwindle and the environmental consequences of their combustion become increasingly apparent, the world has turned its collective attention toward renewable energy sources. Among these, solar energy stands out as one of the most abundant and accessible options available to humanity. Photovoltaic (PV) technology, which converts sunlight directly into electricity, has experienced remarkable growth over the past few decades, evolving from a niche laboratory curiosity into a mainstream power generation technology deployed across the globe.
The fundamental appeal of solar panel technology lies in its environmental compatibility and its capacity for distributed deployment. Unlike conventional thermal power plants that require centralized infrastructure and produce significant pollution, solar panels can be installed on rooftops, in desert expanses, in agricultural settings, and even on water surfaces. This versatility has driven the exponential growth in global photovoltaic capacity. By the end of June 2020, China alone had achieved a cumulative installed capacity of 216 million kilowatts of photovoltaic power generation, with new installations primarily distributed across the central and western regions of the country. This massive deployment represents not only a significant investment in clean energy infrastructure but also a substantial engineering challenge in terms of operation and maintenance.
However, the expansion of solar panel installations has been accompanied by a concerning trend: the increasing frequency of photovoltaic module fires and performance degradation incidents. Statistical data from the photovoltaic industry reveals that at least 45 typical fire accidents occurred in the three-year period preceding this study. These incidents range from large-scale fires on solar panel-covered slopes in Shanxi Province to devastating facility fires in industrial installations. The economic losses associated with these events are substantial, with some individual incidents resulting in direct economic losses exceeding three million yuan. The underlying causes of these fires are multifaceted and include: electrical leakage resulting from panel perforation; severe heating of wiring due to poor heat dissipation; hot spot effects caused by shading, soot accumulation, or foreign object coverage that create localized temperature elevations; and external fire or heat sources compromising the integrity of the photovoltaic modules.
The relationship between temperature and photovoltaic performance is well-established in the scientific literature. Solar panel operating temperature has a direct and significant influence on both power output and long-term module reliability. When the temperature of a photovoltaic cell exceeds its designed operating range, the photoelectric conversion efficiency degrades measurably. More critically, localized overheating—whether caused by partial shading, cell mismatch, or cell damage—can induce a phenomenon known as the hot spot effect. A hot spot occurs when a shaded or damaged cell within a series string operates in reverse bias, dissipating power as heat rather than generating electricity. Under sustained irradiance, the temperature in the vicinity of this defective cell can escalate dramatically, leading to secondary breakdown of the cell, melting of solder joints, delamination of encapsulant materials, and ultimately, catastrophic fire.
Beyond the immediate safety concerns, temperature monitoring of solar panels serves an equally important function in optimizing performance. Extensive research has demonstrated that solar panel conversion efficiency exhibits a temperature coefficient typically ranging from -0.3%/°C to -0.5%/°C for crystalline silicon cells. This means that a solar panel operating at 60°C rather than 25°C can suffer an efficiency loss of 10% to 17% relative to its rated performance. The ability to monitor and respond to temperature variations in real time therefore offers a pathway toward both increased energy yield and extended system lifetime. When the monitored temperature of a solar panel exceeds a predetermined threshold, active cooling measures can be initiated promptly to restore the operating temperature to an optimal range.
The imperative for effective solar panel temperature monitoring has driven the development of various sensing technologies. However, each of the currently available approaches presents inherent limitations that restrict their applicability to large-scale photovoltaic installations. The present work addresses this gap by proposing and demonstrating a distributed fiber-optic Raman temperature sensing system specifically tailored for solar panel temperature monitoring, coupled with an intelligent early warning algorithm based on fuzzy logic theory.
This thesis is organized as follows: Chapter 2 presents the comprehensive design of the distributed Raman temperature sensing system, encompassing both software and hardware implementations. Chapter 3 details the solar panel temperature monitoring methodology and presents the fuzzy temperature difference threshold early warning model, followed by experimental results and analysis. Chapter 4 provides concluding remarks and outlines directions for future research.
2. Design of the Distributed Fiber-Optic Raman Temperature Sensing System
2.1 Theoretical Foundations of Raman Scattering
When a laser pulse propagates through an optical fiber, it interacts with the fiber medium through several distinct scattering processes. These processes are classified according to whether energy exchange occurs between the incident photons and the scattering medium. The three primary scattering mechanisms are Rayleigh scattering, Brillouin scattering, and Raman scattering. Each of these mechanisms carries specific information about the propagation medium and its environmental conditions.
Rayleigh scattering is an elastic scattering process in which the scattered photons retain the same energy (and therefore the same wavelength) as the incident photons. This scattering mechanism arises from microscopic density fluctuations in the fiber material and forms the basis for optical time-domain reflectometry (OTDR) techniques used in fiber-optic fault location. However, since Rayleigh scattering involves no energy exchange, it exhibits virtually no temperature sensitivity.
Raman scattering, in contrast, is an inelastic scattering process resulting from the interaction between the incident photons and the vibrational modes of the atoms or molecules in the fiber core material. The energy exchange between photons and matter causes the scattered photons to be shifted in frequency relative to the incident light. This frequency shift is characteristic of the material and is known as the Raman shift. The scattered light is split into two components: the Stokes component, which appears at a longer wavelength (lower frequency) than the incident light, and the anti-Stokes component, which appears at a shorter wavelength (higher frequency).
The critical observation for temperature sensing is that the relative intensities of the Stokes and anti-Stokes scattering components are temperature-dependent. This phenomenon can be understood through quantum theory, which describes the Raman scattering process in terms of energy level transitions. In a material at thermal equilibrium, the population density of molecules at different vibrational energy levels follows the Boltzmann distribution. When an incident photon interacts with a molecule in the lower vibrational energy state, it can absorb the molecule’s thermal energy and scatter at a higher frequency (anti-Stokes). Conversely, when a photon interacts with a molecule already in an excited vibrational state, the scattering process transfers energy to the photon, resulting in a lower frequency (Stokes).
The intensity of Rayleigh backscattered light can be expressed as:
$$P_R = K_R v_0^4 S_b P_0 \exp(-2\alpha_0 L)$$
Similarly, the Stokes and anti-Stokes backscattered intensities are given by:
$$P_S = K_S v_S^4 S_b P_0 \exp[-(\alpha_0 + \alpha_s)L] R_S(T)$$
$$P_{AS} = K_{AS} v_{AS}^4 S_b P_0 \exp[-(\alpha_0 + \alpha_{AS})L] R_{AS}(T)$$
where $K_R$, $K_S$, and $K_{AS}$ are the cross-section coefficients, $v_0$, $v_S$, and $v_{AS}$ are the frequencies of the incident, Stokes, and anti-Stokes photons respectively, $P_0$ is the injected laser power, $S_b$ is the backscatter capture coefficient, $\alpha_0$, $\alpha_S$, and $\alpha_{AS}$ are the attenuation coefficients, and $L$ is the distance along the fiber. The temperature-dependent coefficients $R_S(T)$ and $R_{AS}(T)$ are defined as:
$$R_S(T) = \frac{1}{\exp(h\Delta v/kT) – 1}$$
$$R_{AS}(T) = \frac{1}{1 – \exp(1 – h\Delta v/kT)^{-1}}$$
Here, $h$ is Planck’s constant, $k$ is Boltzmann’s constant, $\Delta v$ is the Raman frequency shift, and $T$ is the absolute temperature at the measurement point. Since $R_{AS}(T)$ exhibits a much stronger temperature dependence than $R_S(T)$, the anti-Stokes component serves as the signal channel, while the Stokes component provides an ideal reference channel in dual-channel demodulation schemes.
2.2 Temperature Demodulation Methods
Three principal demodulation methods have been developed for distributed Raman temperature sensing systems. The first is the single-channel anti-Stokes demodulation method, which utilizes the temperature sensitivity of the anti-Stokes backscattered light alone. In this approach, a reference measurement at a known temperature $T_0$ is stored, and subsequent temperature values are calculated using the ratio of the measured anti-Stokes intensities:
$$\frac{P_{AS}(T)}{P_{AS}(T_0)} = \frac{1/\exp(h\Delta v/kT_0) – 1}{1/\exp(h\Delta v/kT) – 1}$$
While this method offers a simple configuration and low computational complexity, it suffers from susceptibility to pump power fluctuations and component noise.
The second method, which has become the industry standard, is the dual-channel Stokes and anti-Stokes ratio demodulation method. This approach exploits the fact that both the Stokes and anti-Stokes signals are affected similarly by common-mode effects such as fiber attenuation and source power variations. By taking the ratio of anti-Stokes to Stokes intensity, these common-mode effects are suppressed:
$$\frac{P_{AS}(T)/P_S(T)}{P_{AS}(T_0)/P_S(T_0)} = \frac{\exp(h\Delta v/kT_0) – 1}{\exp(h\Delta v/kT) – 1}$$
Solving for the temperature $T$ at the measurement location yields:
$$\frac{1}{T} = \frac{1}{T_0} – \frac{k}{h\Delta v} \ln\left( \frac{P_{AS}(T)/P_S(T)}{P_{AS}(T_0)/P_S(T_0)} \right)$$
The dual-channel method significantly improves measurement accuracy by eliminating the effects of local fiber bends and inhomogeneous attenuation. However, it is important to note that the attenuation coefficients $\alpha_S$ and $\alpha_{AS}$ differ between the Stokes and anti-Stokes channels. This residual differential attenuation can introduce errors in long-distance measurements.
The third method, the loop demodulation approach, overcomes this limitation by acquiring anti-Stokes light intensity data from both the forward and backward directions. The geometric mean of the forward-scattered and backward-scattered intensities eliminates the position-dependent attenuation factors. The effective signal becomes:
$$P_{AS,eff}(Z) = \sqrt{P_{AS,forward}(Z) \cdot P_{AS,backward}(Z)}$$
This loop configuration ensures that the attenuation is integrated over the entire fiber length, effectively canceling position-dependent losses due to bends, splices, and other localized perturbations.
2.3 Positioning Principles: Optical Time-Domain Reflectometry
The distributed measurement capability of the Raman temperature sensing system relies on optical time-domain reflectometry (OTDR) principles. When a short laser pulse is launched into the sensing fiber, the backscattered light arriving at the detector at time $t$ originates from a location at distance $L$ from the launch point. The distance can be calculated as:
$$L = \frac{v \cdot t}{2}$$
where $v = c/n$ is the group velocity of light in the fiber, $c$ is the speed of light in vacuum, and $n$ is the effective refractive index of the fiber core. The factor of 2 accounts for the round-trip path of the light pulse. By continuously sampling the backscattered signal, the temperature along the entire fiber axis can be resolved with a spatial resolution determined primarily by the pulse width and the detector response characteristics.
To avoid ambiguity between successive pulses, the pulse repetition period must exceed the transit time of light through the entire fiber length. For a fiber of length $L_{max}$, the minimum repetition period is:
$$\Delta T_{rep} = \frac{2nL_{max}}{c}$$
2.4 Software Design of the Distributed Raman Temperature Sensing System
The software architecture for the distributed Raman temperature sensing system was developed using the LabVIEW graphical programming environment. This platform was selected for its powerful data acquisition capabilities, extensive hardware driver support, and excellent visualization options. The software is organized into several interconnected modules, each serving a distinct function within the overall system architecture.
2.4.1 Signal Acquisition and Parameter Configuration Module
The signal acquisition module manages the high-speed digitizer interface, configuring sampling rate, accumulation averaging counts, and measurement length parameters. The module continuously acquires backscattered signals from both the Stokes and anti-Stokes channels and performs cumulative averaging to enhance the signal-to-noise ratio. The averaging process is critical because the Raman backscattered light is extremely weak—typically at the nanowatts level—and is accompanied by substantial noise from the detectors and electronic circuits.
The software implementation utilizes a flat sequence structure to ensure deterministic execution order, with each frame of processing completing before data is passed to the next stage. The program first initializes the digitizer hardware, configures the acquisition parameters, and verifies that all settings are within valid ranges. Once the averaging count reaches the configured value, the processed signals are transferred to the demodulation module for temperature extraction.
2.4.2 Temperature Demodulation Module
The temperature demodulation module implements the mathematical algorithms described above for converting raw intensity data into temperature measurements. In the dual-channel approach, the measured Stokes and anti-Stokes intensity arrays are stored as one-dimensional numerical arrays. The program reads the calibration data acquired at a reference temperature, computes the average intensity over a specified reference fiber segment to minimize noise effects, and applies the ratio-based demodulation formula.
The program incorporates a conditional structure that adapts the processing path based on whether the system is operating in calibration or measurement mode. During calibration, the system stores the reference intensity data in association with a precisely measured reference temperature. During normal operation, the stored calibration constants are retrieved and used with the live acquired signals to compute real-time temperature profiles.
Additionally, the module applies a wavelet transform modulus maximum denoising algorithm to the demodulated temperature curves. Wavelet denoising is particularly effective in this application because it preserves rapid temperature transitions while eliminating high-frequency noise components, resulting in smoother and more accurate temperature profiles.
2.4.3 Data Storage Module
Reliable data persistence is essential for both real-time monitoring and post-incident analysis. The data storage module automatically generates directory structures based on acquisition dates, creating paths in the format YYYYMMDD. The module checks for file name conflicts and implements an automatic incrementing mechanism to ensure unique file names for each measurement session. Both the raw intensity data and the demodulated temperature profiles are saved in structured formats suitable for subsequent analysis and database queries.
The storage logic includes conditional structures to distinguish between first-time daily runs and subsequent sessions, ensuring efficient file organization and preventing data loss from file overwrites. The LabVIEW implementation utilizes relative and absolute path functions to maintain portability across different computer systems while ensuring access to the intended storage locations.
2.4.4 Database Query Module
Historical data retrieval is implemented through the database query module, which allows operators to specify a date range and retrieves the corresponding temperature data for review. The module incorporates string manipulation functions to construct file paths from user inputs, reads the stored data from electronic spreadsheet formats, and converts the raw data into a dynamic data type compatible with LabVIEW’s analysis functions. Statistical metrics including the mean, maximum, minimum temperatures and the locations of these extrema are computed and displayed on the front panel.
2.5 Hardware Design of the Distributed Raman Temperature Sensing System
The hardware implementation of the distributed Raman temperature sensing system integrates several key components, each selected for its specific contribution to overall system performance. The complete experimental configuration is illustrated schematically in the system device diagram. The primary components include:
High-Speed Pulse Laser: The laser source is a critical determinant of system performance. The selected laser is a high-speed pulsed fiber laser operating at a center wavelength of 1550.1 nm. This wavelength sits in the low-loss transmission window of standard silica optical fibers, balancing the competing requirements of minimized attenuation (favoring longer wavelengths) and maximized Raman scattering efficiency (favoring shorter wavelengths). The laser produces pulses with a width of 10 ns, achieving a spatial resolution of approximately 1 m. The peak optical power is adjustable up to 4 W, with the constraint that the power must remain below the threshold for stimulated Raman scattering to preserve the temperature-intensity relationship characteristic of spontaneous Raman scattering. The laser incorporates an external trigger port synchronized with the data acquisition system to enable precise time-of-flight measurements.
Wavelength Division Multiplexer (WDM): The WDM serves as the optical front-end, coupling the incident laser pulses into the sensing fiber and separating the backscattered Stokes and anti-Stokes components. The 1×3 WDM configuration connects to the laser (1550 nm port), the sensing fiber (common port), the anti-Stokes detector (1450 nm port), and the Stokes detector (1650 nm port). The WDM provides an isolation of 30 dB between channels, suppressing crosstalk from the strong Rayleigh backscattered light that could otherwise contaminate the weak Raman signals. The insertion loss is specified at a maximum of 0.2 dB, and all fiber connections use FC/APC angle-polished connectors to minimize back reflections.
Dual-Channel Avalanche Photodiode (APD) System: The optical signals exiting the WDM are extremely weak, particularly the anti-Stokes component which carries the primary temperature information. Each channel employs an avalanche photodiode detector operating at a reverse bias near the breakdown voltage to achieve internal gain. The APD characteristics are summarized in Table 2-1:
| Parameter | Condition | Minimum | Typical | Maximum | Unit |
|---|---|---|---|---|---|
| Spectral Response Range | — | 900 | — | 1700 | nm |
| Responsivity | — | 8.5 | — | 20 | A/W |
| Reverse Breakdown Voltage | — | 40 | — | 60 | V |
| Dark Current | M = 10 | — | 2 | 10 | nA |
| Total Capacitance | M = 10 | — | 0.5 | — | pF |
| Operating Temperature Coefficient | −40 to 85°C | 0.1 | — | 0.15 | V/°C |
| Power Supply (+5V) | — | 4.9 | 5 | 5.5 | V |
| Power Supply (−5V) | — | −4.9 | −5 | −5.5 | V |
| Power Supply (+48V) | — | 45 | 46 | 48.5 | V |
Signal Amplifiers: The electrical signals from the APD detectors are too weak for direct digitization. Two wideband operational amplifiers with a bandwidth of 100 MHz amplify the signals before submission to the data acquisition system. The amplifiers introduce a gain that is calibrated as part of the system initialization procedure.
High-Speed Data Acquisition Card: The digitizer is a PCI104-interface card with dual channels, each operating at a sampling rate of 100 megasamples per second with 12-bit resolution. The onboard FPGA implements real-time cumulative averaging of up to 32768 acquisitions, improving the signal-to-noise ratio by a factor of up to 180. The card supports both internal and external trigger modes with a trigger synchronization jitter of 5 ps RMS between channels. This precise timing is essential for accurately correlating the time-of-flight information with spatial position along the sensing fiber.
2.6 System Integration and Performance Characterization
Following the selection and bench testing of individual components, the distributed Raman temperature sensing system was assembled into an integrated instrument. The integration process involved several stages: mechanical design, optical circuit alignment, electrical interconnection, and thermal management.
The instrument chassis utilizes a two-layer construction with a top layer for external connectors and a bottom layer for the core sensing components. Carbon fiber composite material was selected for the chassis construction based on its excellent strength-to-weight ratio, dimensional stability, and electromagnetic shielding properties. The thermal management system incorporates four air intake vents and two exhaust outlets, establishing directed airflow pathways that dissipate heat from the main power-consuming components including the computer motherboard and data acquisition card.
The system performance was evaluated using a test configuration consisting of an 11 km multi-mode sensing fiber with fiber under test (FUT) sections located at approximately 2.3 km and 10.36 km distances. The FUT sections were placed in an oven at controlled temperatures of 35.0 °C, 50.0 °C, 65.0 °C, 80.0 °C, and 95.0 °C while the remaining fiber remained at ambient temperature. The results of the performance tests are summarized in Table 2-2:
| Parameter | Measured Value |
|---|---|
| Sensing Distance | 11.0 km |
| Response Time | 2 s |
| Temperature Measurement Accuracy | ±1.00 °C |
| Temperature Resolution | 0.50 °C |
| Spatial Resolution | 1.0 m |
| Laser Wavelength | 1550 nm |
| Fiber Type | Multi-mode fiber |
The temperature measurement accuracy was quantified by comparing the RDTS measurements with a precision temperature reference at both FUT locations. At the 2.27 km location, the maximum measurement error was 0.8 °C with a mean error of 0.6 °C, while at the 10.36 km location, the maximum error was 1.24 °C with a mean error of 1.0 °C. This distance-dependent error degradation is attributable to the reduction in backscattered signal intensity with increasing fiber length, which lowers the signal-to-noise ratio and consequently broadens the variance of the demodulated temperature.
The spatial resolution was determined by measuring the distance over which the temperature transition from 10% to 90% of step response occurred. At 2.27 km, the spatial resolution was 2.0 m, while at 10.36 km it worsened to 4.6 m. The degradation in spatial resolution at longer distances results from inter-modal dispersion in the multi-mode fiber, which causes temporal broadening of the optical pulses as they propagate through the fiber.
3. Solar Panel Temperature Monitoring and Early Warning System
3.1 Monitoring Methodology
The established RDTS system was applied to the distributed temperature monitoring of solar panels through the deployment configuration illustrated in the monitoring arrangement diagram. The experimental setup comprised seven photovoltaic modules organized into three distinct groups to facilitate controlled comparison of temperature-related effects. The PV modules used in these experiments were rated at a peak power output of 270 W, with maximum power voltage of 30.2 V and maximum power current of 8.13 A.
The first group (P1, P2, P3) consisted of cleaned panel surfaces installed at a tilt angle of 60°. The second group (P4, P5) maintained the same tilt angle with dusty surfaces, which had accumulated approximately three months of environmental contamination. The third group (P6, P7) featured cleaned surfaces installed at a reduced tilt angle of 30°. This experimental design allowed independent assessment of the effects of dust accumulation and tilt angle on solar panel temperature distribution.
For the temperature sensing deployment, the optical fiber was routed along the back surface of each solar panel using an epoxy-based thermally conductive adhesive to ensure efficient heat transfer from the panel to the fiber. The fiber was arranged in a serpentine pattern across the panel surface, maximizing the coverage area while following horizontal paths to capture the expected vertical temperature gradient. The temperature measurement system operated with a temporal resolution of 1 minute, providing a continuous record of solar panel thermal behaviour throughout the day.
3.2 Experimental Results of Solar Panel Temperature Monitoring
The distributed temperature measurements revealed significant spatial and temporal variations in solar panel temperature profiles. Quantitative results at different times of day are presented in Table 3-1:
| Time | T1 | T2 | T3 | T4 | T5 |
|---|---|---|---|---|---|
| Group 1 (Clean, 60°) | 39.30 | 36.47 | 34.02 | 27.81 | 18.72 |
| Group 2 (Dusty, 60°) | 42.31 | 38.95 | 35.49 | 28.65 | 19.24 |
| Group 3 (Clean, 30°) | 35.31 | 33.12 | 31.05 | 25.43 | 16.98 |
The experimental results demonstrate a consistent decrease in solar panel temperature from noon (T1) through the afternoon (T5), corresponding to the gradual reduction in solar irradiance. The maximum temperature of 39.30 °C was recorded at 12:00 PM for the clean panel at 60° inclination, while the minimum afternoon temperature of 18.72 °C occurred at 4:00 PM.
The effect of dust accumulation on solar panel temperature was quantified by comparing Group 1 and Group 2. Solar panels with dust-covered surfaces exhibited consistently higher temperatures than their clean counterparts, with the maximum temperature difference of 3.01 °C occurring at the T1 time point. This temperature elevation is attributed to partial shading of the cells caused by accumulated dust particles, which creates localized hot spots where the shaded cells operate in reverse bias mode. During the system temperature measurement period, the formation of scorched darker areas was observed on panels from Group 2, consistent with high-temperature stress damage to the cell encapsulants.
Comparing Group 1 with Group 3 revealed the influence of inclination angle on solar panel temperature. The panels installed at the steeper 60° angle exhibited higher temperatures than those at the shallower 30° angle, with a maximum temperature difference of 3.99 °C at the T1 time point. This difference reflects the varying energy incidence angle and consequent heat absorption characteristics associated with different panel orientations.
The combined influence of dust accumulation and inclination angle was evaluated by comparing Group 2 with Group 3, yielding a maximum temperature difference of 6.24 °C. This substantially greater difference highlights the compounding effect of both factors, demonstrating that solar panel installations in dusty environments require careful consideration of tilt angle selection to optimize thermal performance.
Beyond the inter-group comparisons, measurements within individual solar panels revealed substantial temperature gradients across the panel surface. The serpentine fiber layout enabled identification of localized hot spots within a single solar panel, a capability that distinguishes distributed fiber sensing from conventional point-based sensing approaches. The temperature at the hottest point of a panel exceeded the panel mean temperature by up to several degrees Celsius, emphasizing the importance of spatially resolved monitoring for early hot spot detection.
3.3 Relationship between Solar Panel Temperature and Output Power
To establish the practical significance of solar panel temperature monitoring, a six-hour continuous experiment was conducted correlating real-time panel temperature with electrical output power. The results revealed a strong positive correlation between these parameters over short time scales, with both simultaneously reaching their maximum values of 157 W and 39.60 °C at 12:40 PM. This observation is consistent with the understanding that higher irradiance simultaneously drives both increased power generation and elevated panel temperature.
During the experiment, both temperature and power exhibited a rapid simultaneous decrease at approximately 1:50 PM, followed by recovery to normal levels around 3:00 PM. This synchronized behavior was traced to transient cloud cover that temporarily reduced solar irradiance. The coordinated response of temperature and output power validates the predictive utility of temperature monitoring: temperature serves as a reliable indicator of power generation performance, enabling detection of operational anomalies that might otherwise require direct electrical measurements to reveal.
3.4 Fuzzy Temperature Difference Threshold Method (FTDTM) for Early Warning
While the RDTS system provides comprehensive real-time temperature information for solar panels, the value of this information is substantially enhanced when the system can anticipate temperature changes before they occur. Early warning of impending overheating enables proactive intervention rather than reactive response. To address this requirement, the present research developed a predictive early warning model based on fuzzy time series theory.
Fuzzy time series theory, originally proposed by Song and Chissom, provides a framework for modeling dynamic processes using linguistic values rather than crisp numerical values. This approach is particularly well-suited to temperature forecasting in outdoor environments where the system behavior is influenced by numerous interacting factors including solar irradiance, wind speed, ambient temperature, and cloud cover—factors that are difficult to model precisely through conventional analytical or numerical methods.
The fuzzy temperature difference threshold method (FTDTM) developed in this work operates on the principle that future temperature changes can be inferred from recent temperature trends. The thermal dynamics of solar panels exhibit inertia, implying that temperature changes in the immediate past provide meaningful information about imminent future changes. The FTDTM algorithm comprises six steps:
Step 1: Partitioning of the Universe of Discourse
Based on historical temperature data from the solar panel monitoring system, the maximum temperature difference $\Delta T_{max}$ and minimum temperature difference $\Delta T_{min}$ between consecutive measurement instants are computed. The universe of discourse $U$ is defined as:
$$U = [\Delta T_{min} – T_1, \Delta T_{max} + T_2]$$
where $T_1$ and $T_2$ are positive margin values chosen based on the distribution characteristics of the temperature difference data. The interval $U$ is divided into six equal-length subintervals to define the fuzzy linguistic categories.
Step 2: Establishing Fuzzy Relations
For each subinterval $u_i$, a fuzzy set $A_i$ is defined using membership functions. The fuzzy sets are computed using the finite-set representation:
$$A_i = u_A(u_1)/u_1 + u_A(u_2)/u_2 + \cdots + u_A(u_6)/u_6$$
Each temperature difference $\Delta T$ is mapped to its corresponding fuzzy linguistic value based on the membership degree $u_A(u_i)$, where $u_A(u_i)$ achieves its maximum value within the subinterval $u_i$ containing $\Delta T$.
Step 3: Computing the Fuzzy Relation Matrix
With the prediction window base designated as $w$, the standard matrix $C(t)$ for the most recent temperature difference is:
$$C(t) = F(t-1) = [C_1, C_2, \dots, C_m]$$
and the operating matrix $O_w(t)$ incorporates the temperature differences from the preceding $w$ time intervals:
$$O_w(t) = \begin{bmatrix} F(t-2) \\ F(t-3) \\ \vdots \\ F(t-w-1) \end{bmatrix} = \begin{bmatrix} O_{11} & O_{12} & \cdots & O_{1m} \\ O_{21} & O_{22} & \cdots & O_{2m} \\ \vdots & \vdots & \ddots & \vdots \\ O_{w1} & O_{w2} & \cdots & O_{wm} \end{bmatrix}$$
The fuzzy relation matrix $R(t)$ is computed as the product of the operating matrix and the transposed standard matrix:
$$R(t)[w, m] = O_w(t) \odot C^T(t)$$
where the symbol indicates element-wise multiplication. The fuzzy time series $F(t)$ is derived from $R(t)$ by extracting the column maximum values.
Step 4: Computing the Predicted Temperature Difference
If the fuzzy time series $F(t)$ contains multiple equal maximum values, the predicted temperature difference $m_i$ is calculated as the arithmetic mean of the subinterval midpoints:
$$m_i = \frac{m_1 + m_2 + \cdots + m_k}{k}$$
where $m_1, m_2, \dots, m_k$ represent the midpoints of the corresponding subintervals.
Step 5: Calculating the Predicted Temperature
The predicted temperature $T_t$ at time instant $t$ is computed by adding the predicted temperature difference to the observed temperature at the previous time instant:
$$T_t = T_{t-1} + m_i$$
Boundary conditions are applied to clip the prediction to a physically plausible range defined by $T_{max}$ and $T_{min}$ derived from the historical temperature data:
$$T_t = \begin{cases} T_{max}, & \text{if } T_{t-1} + m_i > T_{max} \\ T_{min}, & \text{if } T_{t-1} + m_i < T_{min} \\ T_{t-1} + m_i, & \text{otherwise} \end{cases}$$
This clipping operation constrains the maximum possible prediction error to $\max(|T_{max}|, |T_{min}|)$.
Step 6: Temperature Threshold Alarm Setting
An alarm threshold $T_h$ is configured for the solar panel monitoring application. When the predicted temperature exceeds this threshold, the system initiates the cooling mechanism, compensating for the inherent start-up latency of active cooling systems by triggering slightly before the temperature reaches the critical limit.
3.5 Experimental Verification of the Prediction Model
The FTDTM prediction model was validated using the real-time temperature data acquired from the solar panel monitoring experiments. The RDTS system collected temperature measurements at one-minute intervals over the six-hour test period, with the FTDTM predicting the temperature one minute ahead based on historical trend information.
Figure 3-8 in the original thesis presents the comparison between real-time measured temperatures and the corresponding predictions. The analysis of prediction errors revealed that the FTDTM achieved a mean absolute error of 1.08 °C, with prediction error fluctuations confined to the range of ±3.70 °C. The error magnitude exhibited a correlation with the rate of temperature change: larger prediction errors occurred during rapid temperature transitions, consistent with the underlying principle that the method extrapolates recent trends.
To optimize the prediction performance, experiments were conducted using different window base values $w$ ranging from 2 to 8. The mean relative prediction errors for each window base are summarized in Table 3-2:
| 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 |
The experimental results demonstrate a clear monotonic relationship between the window base and prediction error: smaller window bases consistently produced more accurate predictions. This behavior emerges from the statistical property that temperature observations in the immediate past hold the greatest predictive information for the near future. As the window base increases, the influence of more distant observations dilutes the predictive signal, increasing the error.
The optimal performance was achieved with a window base of $w=2$, yielding a mean relative error of 3.16%. This configuration was therefore selected as the default setting for the solar panel temperature early warning system.
4. Discussion and Future Directions
The research presented in this thesis demonstrates that the combination of distributed Raman optical fiber sensing with fuzzy logic-based predictive analytics offers a robust and practical solution for solar panel temperature monitoring and early warning. The RDTS system provides several distinctive advantages that address the limitations of existing solar panel temperature monitoring technologies:
First, the distributed measurement capability of the RDTS system enables complete temperature profiling of every solar panel within a large installation, rather than relying on sparse point measurements. As demonstrated in the experimental results, significant temperature variations exist within individual panels, and these localized hot spots represent the primary fire risk. Point-based sensing approaches, whether using electrical sensors or fiber Bragg gratings, risk missing exactly the critical hot spots that the monitoring system is intended to detect.
Second, the all-optical sensing medium provides immunity to electromagnetic interference, which is particularly important in photovoltaic installations where high DC currents and the associated magnetic fields can degrade the accuracy of electrical sensors. The optical fiber also offers inherent corrosion resistance, making it suitable for the harsh outdoor environments in which solar panels typically operate.
Third, the fiber-optic approach enables long-distance monitoring without intermediate electronics. A single RDTS system can interrogate sensing fibers deployed over distances of 11 km or more, accommodating large-scale solar panel installations with minimal distributed electronics. This configuration reduces system cost, simplifies maintenance, and enhances overall reliability.
The fuzzy temperature difference threshold method provides a computational framework for early warning that is both effective and computationally efficient. A key advantage of the FTDTM over alternative prediction approaches is its minimal data requirements. Unlike energy balance models or neural network approaches that require comprehensive metrological inputs (irradiance, wind speed, ambient temperature, cloud cover), the FTDTM operates exclusively on the temperature history itself. This self-contained nature makes the method robust to sensor failures and applicable to situations where supplementary environmental monitoring infrastructure is unavailable.
Several limitations of the current approach warrant acknowledgment. The prediction accuracy of 1.08 °C mean absolute error and ±3.70 °C error range, while adequate for early warning applications, may be insufficient for tasks requiring precise temperature knowledge. The error is expected to increase during periods of rapidly fluctuating irradiance, such as partly cloudy conditions when moving clouds create rapidly varying shadow patterns across solar panel arrays.
Future research directions for improving the solar panel temperature monitoring system include:
(1) Enhancing system response speed: The present system prioritizes measurement accuracy and spatial resolution, resulting in a response time of 2 seconds per complete temperature profile. For applications requiring faster detection of fast-developing hot spots, the acquisition averaging, signal processing, and demodulation algorithms could be streamlined or parallelized to reduce this latency.
(2) Improving Rayleigh scattering isolation: The WDM provides 30 dB isolation between the Raman channels and the Rayleigh wavelength. Higher isolation values would further suppress crosstalk and improve signal-to-noise ratio in the Raman channels. Alternative filtering approaches, such as tandem fiber Bragg grating filters, could be evaluated for this purpose.
(3) Optimizing sensor fiber deployment: The current serpentine layout on each solar panel is simple and effective but does not capture all spatial temperature variations. A higher density fiber layout or deployment of fiber loops around the panel perimeter and junction box regions could reveal additional thermal phenomena.
(4) Refining the universe of discourse partitioning: In the FTDTM, the selection of margins $T_1$ and $T_2$ and the number of subintervals directly affects the model’s prediction precision. A dynamic partitioning scheme that adapts to the observed temperature difference statistics could reduce prediction error. Alternatively, optimization algorithms could search for the optimal partition parameters using historical data.
(5) Developing multi-factor prediction models: The current FTDTM uses only temperature history as the predictive input. Incorporating additional measured quantities such as output current, output voltage, or ambient conditions available from other sensors in the photovoltaic installation could increase prediction accuracy. Multi-factor fuzzy time series models would extend the single-factor approach presented here to exploit these additional information sources.
(6) Expanding field deployment and long-term validation: The experimental validation in this thesis was conducted over a single day’s measurement period. Long-term deployment at an actual photovoltaic installation would provide valuable data on system stability, prediction model performance across seasonal variations, and the practical effectiveness of the early warning system in preventing panel damage.
5. Conclusion
This thesis has addressed the critical need for effective temperature monitoring and early warning systems in photovoltaic installations. The research contributions can be summarized as follows:
First, a complete distributed fiber-optic Raman temperature sensing system was designed, developed, and characterized. The hardware integration included selection and testing of individual components, assembly into a portable instrument, and comprehensive performance validation. The software architecture implemented through LabVIEW provides a user-friendly interface for real-time temperature visualization, data persistence, and historical data retrieval. The system achieves a sensing distance of 11 km, a response time of 2 s, temperature measurement accuracy of ±1.00 °C, temperature resolution of 0.50 °C, and spatial resolution of 1.0 m.
Second, the RDTS system was deployed for distributed temperature monitoring of solar panels under varying conditions of surface contamination, inclination angle, and solar irradiance. The experimental results revealed that solar panel temperature increases with dust layer thickness, increases with radiation intensity, and decreases with increasing inclination angle within the tested range. The temperature differences attributable to these three factors reached 3.01 °C, 9.58 °C, and 3.99 °C, respectively. The experiments also confirmed a positive correlation between solar panel temperature and output power under normal operating conditions.
Third, a fuzzy temperature difference threshold method was developed and validated for early warning of solar panel overheating. The FTDTM operates on historical temperature data alone, making it robust to missing environmental data and computationally efficient enough for real-time deployment. Systematic evaluation of different window base values established that $w=2$ provides optimal prediction accuracy, with a mean absolute error of 1.08 °C and prediction error range of ±3.70 °C. The method successfully identifies temperature trends that could precede hot spot formation, enabling proactive cooling or maintenance intervention.
In conclusion, the combination of distributed Raman optical fiber sensing and fuzzy logic-based prediction provides an effective, reliable, and scalable solution for solar panel temperature monitoring and early warning. The system addresses the limitations of existing point-based and laboratory-oriented approaches, offering the distributed coverage, environmental robustness, and predictive capability required for modern large-scale photovoltaic installations. As solar energy continues to expand its role as a primary pillar of the global energy mix, distributed sensing technologies of the type developed in this research will play an increasingly important role in safeguarding the performance and safety of solar panel energy infrastructure.