Photovoltaic module early fault warning system based on multi-modal fusion

Abstract

The application relates to the technical field of early fault warning, and discloses a photovoltaic module early fault warning system based on multi-modal fusion, which comprises a driving excitation and synchronous control module, a double-mode high-speed acquisition module, a phase-sensitive response extraction and alignment module and a fault diagnosis module based on a physical model which are sequentially signal-connected; the driving excitation and synchronous control module receives a fault suspected area coordinate instruction issued by an upper-layer system, drives a semiconductor laser with adjustable power and spot size, and applies a sinusoidal low-frequency micro-thermal excitation to a target area. The photovoltaic module early fault warning system based on multi-modal fusion breaks through the time-space matching bottleneck of traditional multi-modal technology by actively applying controllable micro-thermal excitation to construct an explicit cause-and-effect relationship between excitation and response, efficiently suppresses environmental noise by using a phase-sensitive detection technology, realizes high-sensitivity extraction of weak fault features, and effectively improves power station power generation efficiency and safe operation level.

Description

Technical Field

[0001] This invention relates to the field of early fault warning technology, and in particular to an early fault warning system for photovoltaic modules based on multimodal fusion. Background Technology

[0002] During long-term operation in complex outdoor environments, photovoltaic (PV) modules are susceptible to various faults such as true hot spots, pseudo hot spots caused by shading, early microcracks, diode failure, and solder ribbon detachment. These faults not only lead to a significant decrease in module power generation efficiency and economic losses, but in severe cases, they can also cause module burnout, fires, and other safety accidents, directly threatening the operational safety and stability of PV power plants. To address these issues, existing PV module fault diagnosis technologies mainly fall into three categories: first, infrared thermal imaging detection technology, which identifies faults by capturing differences in the surface temperature field of the module, and has become a common method for large-scale inspections due to its visualization characteristics; second, electrical performance testing technology, which judges the overall operating status of the module by monitoring global electrical parameters such as current, voltage, and power of the PV string; and third, simple multimodal fusion technology, which attempts to integrate infrared thermal images with electrical parameter data to combine the detection advantages of both modes.

[0003] However, existing infrared thermal imaging technology can only capture the thermal manifestations of faults, has a slow response speed, is easily affected by environmental thermal noise such as atmospheric radiation and cloud drift, cannot effectively distinguish between true hot spots and spurious hot spots caused by shading, and has extremely low sensitivity to faults without obvious thermal characteristics, such as early microcracks. Electrical performance testing technology focuses on string-level global monitoring, lacks sensitivity to early weak faults in local areas, makes it difficult to achieve accurate fault location, and is also easily affected by background electrical noise such as power grid fluctuations and changes in illumination. Simple multimodal fusion technology has not solved the essential difference between the global fast response of electrical signals and the local slow response of thermal images, and only stays at the data level of superposition and splicing, without forming an effective spatiotemporal matching mechanism, and lacks the core means to extract weak fault features from strong noise environments, resulting in a high false alarm rate in fault diagnosis and insufficient early warning capability, which cannot meet the actual needs of accurate identification of early faults in photovoltaic modules. Summary of the Invention

[0004] The technical problem to be solved by this invention is that in existing photovoltaic module fault diagnosis technologies, infrared thermal imaging is difficult to distinguish between real and false hot spots and has low sensitivity to early microcracks, electrical performance testing cannot accurately locate local weak faults, and simple multimodal fusion technology has defects such as spatiotemporal matching bottleneck and insufficient noise suppression, resulting in high false alarm rate and lack of early warning capability in fault diagnosis. To address this, we propose a photovoltaic module fault early warning system based on multimodal fusion.

[0005] To achieve the above objectives, this application adopts the following technical solution: a photovoltaic module fault early warning system based on multimodal fusion, comprising an active excitation and synchronization control module, a dual-mode high-speed acquisition module, a phase-sensitive response extraction and alignment module, and a fault diagnosis module based on a physical model, which are connected in sequence. The active excitation and synchronization control module receives the coordinate command of the suspected fault area issued by the upper-level system, drives a semiconductor laser with adjustable power and spot size to apply a sinusoidal low-frequency micro-thermal excitation to the target area, and simultaneously generates a frequency reference signal and a global timestamp signal synchronized with the excitation waveform, which are distributed to subsequent modules to provide a unified time reference. The dual-mode high-speed acquisition module responds to the synchronization signal and acquires the instantaneous data of the string through current and voltage sensors. Current and voltage data are simultaneously captured by a high-speed infrared thermal imager at a frame rate higher than the excitation frequency to form an infrared thermal image sequence. All raw data packets are marked with a unified high-precision timestamp before transmission. The phase-sensitive response extraction and alignment module uses a digital phase-locked amplification algorithm to receive electrical signals from the dual-mode high-speed acquisition module, extract current and voltage perturbation features, and generate phase-locked thermal amplitude and phase maps from the infrared image sequence using a phase-locked thermal imaging algorithm. Based on the timestamp and pre-calibrated coordinate transformation relationship, the dual-mode features are spatiotemporally aligned to generate fused feature pairs. The physical model-based fault diagnosis module calls a pre-stored fault response feature database to normalize and calculate the similarity of the fused feature pairs, and outputs the diagnostic results.

[0006] Preferably, the active excitation and synchronization control module is also used to calibrate the coordinates of the target area based on the real-time image of the infrared thermal imager before driving the semiconductor laser.

[0007] Preferably, the dual-modal high-speed acquisition module is also used to remove abnormal data points in the electrical signal and invalid frames in the thermal image in real time during the acquisition process.

[0008] Preferably, the electrical response extraction unit in the phase-sensitive response extraction and alignment module is used to preprocess the raw current and voltage data, including DC component stripping and baseline calibration.

[0009] Preferably, the thermal response extraction unit in the phase-sensitive response extraction and alignment module is used to perform image quality screening on the infrared thermal image sequence and remove image frames with a signal-to-noise ratio lower than a preset threshold.

[0010] Preferably, when the feature alignment and fusion unit in the phase-sensitive response extraction and alignment module performs time-domain correlation based on a unified timestamp, if the difference between the timestamp of the electrical feature data and the timestamp of the thermal feature data exceeds a preset tolerance, it triggers resynchronization or performs data compensation.

[0011] Preferably, the physical model-based fault diagnosis module normalizes the fused feature pairs and calculates their similarity to each fault feature model. First, it determines whether there is an anomaly based on the comparison with the normal state model. If there is an anomaly, it identifies the fault type based on the principle of highest similarity and preset logical rules.

[0012] Preferably, the preset logic rules include: if the fusion feature pair satisfies significant thermal features and negative and out-of-phase electrical perturbation, it is determined to be a true hot spot; if the thermal features are significant but the electrical perturbation is not significantly different from the normal state, it is determined to be a shading pseudo hot spot; if the electrical perturbation phase shows characteristic hysteresis but the thermal features are not significantly abnormal, it is determined to be an early microcrack.

[0013] Preferably, after determining the fault type, the fault diagnosis module based on the physical model quantifies and classifies the severity of the fault according to the feature parameter values ​​in the fused feature pair.

[0014] Preferably, the physical model-based fault diagnosis module performs consistency verification on the diagnosis results of multiple consecutive excitation cycles and outputs a structured diagnostic report containing the fault type, location, and severity level.

[0015] The technical effects and advantages of this invention are as follows: Addressing the core pain points of existing technologies, this invention establishes a clear causal relationship between excitation and response by actively applying controllable micro-thermal excitation, overcoming the spatiotemporal matching bottleneck of traditional multimodal technologies. It utilizes phase-sensitive detection technology to efficiently suppress environmental noise, achieving high-sensitivity extraction of weak fault characteristics. Combined with the physical mechanisms of photovoltaic modules, a fault model library is constructed, enabling deep fusion and intelligent diagnosis of electrical and thermal dual-modal information. This invention can accurately identify easily confused faults such as true hot spots, shading-induced false hot spots, and early microcracks, achieving early fault detection, precise location, and severity quantification. Furthermore, its safe and controllable excitation design ensures a non-destructive testing process, adapting to various photovoltaic modules such as monocrystalline silicon, polycrystalline silicon, and thin-film photovoltaic modules, as well as centralized and distributed photovoltaic power station scenarios. This provides technical support for the refined operation and maintenance of photovoltaic power stations, effectively improving power generation efficiency and safe operation levels. Attached Figure Description

[0016] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0017] Figure 1 This is a diagram illustrating the overall system architecture and workflow of the present invention. Figure 2 This is a block diagram of the internal processing of the phase-sensitive response extraction and alignment module of the present invention. Detailed Implementation

[0018] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0019] Reference Figure 1-2 As shown, this invention provides a technical solution: a photovoltaic module fault early warning system based on multimodal fusion, comprising: an active excitation and synchronization control module, a dual-modal high-speed acquisition module, a phase-sensitive response extraction and alignment module, and a fault diagnosis module based on a physical model; the active excitation and synchronization control module is used to generate and apply a controllable, safe, and separable physical excitation from the module under test, and to provide a unique time reference for all subsequent data acquisition and processing actions; the active excitation and synchronization control module includes an excitation control unit, a micro-thermal excitation source, and a synchronization signal generator; the excitation control unit receives the coordinate instructions of the suspected fault area from the upper-level system and identifies the target excitation area; according to the instructions, the excitation control unit drives the micro-thermal excitation source, which is a semiconductor laser with adjustable power and spot size, and the control unit adjusts the output of the laser to a low-frequency periodic waveform of a sine wave in the range of 0.1-1Hz, controlling the laser to irradiate the target coordinate area, and key parameters ensure that the instantaneous temperature rise generated by the excitation is limited to 3-5℃; the synchronization signal generator generates a reference clock signal that is strictly synchronized with the excitation waveform in real time, and the synchronized reference clock signal includes a frequency reference signal and a global timestamp signal. The reference clock signal is distributed in real time to the dual-mode high-speed acquisition module and the phase-sensitive response extraction module.

[0020] The active excitation and synchronization control module generates and applies a controllable, safe, and separable physical excitation from the component under test, providing a unique time reference for all subsequent data acquisition and processing actions. The excitation control unit receives coordinate instructions for suspected fault areas from the upper-level system. These instructions, in JSON format, include the component number, the pixel coordinates of the area's corners, and a minimum area threshold of ≥0.5. After CRC32 verification, the target excitation area is clearly identified. Upon receiving the data, coordinate calibration is performed using real-time images from an infrared thermal imager, with errors controlled within ≤1 pixel. The excitation control unit drives a micro-thermal excitation source, a semiconductor laser with adjustable power and spot size (wavelength 808nm-980nm, continuously adjustable output power 1W-10W, circular spot size 0.5cm-2cm adjusted in 0.1cm increments, uniformity ≥90%). The control unit adjusts the laser output to a low-frequency sinusoidal waveform within the 0.1-1Hz range, ensuring precise illumination of the target coordinate area. Simultaneously, a miniature contact temperature sensor provides real-time temperature feedback, and PID closed-loop control ensures the instantaneous temperature rise from the excitation is limited to within a certain range. 3-5℃; The synchronization signal generator generates a reference clock signal that is strictly synchronized with the excitation waveform in real time. The synchronized reference clock signal includes a frequency reference signal in the form of a TTL level square wave and a global timestamp signal with 64-bit UTC and nanosecond offset. The reference clock signal is distributed to the dual-mode high-speed acquisition module and the phase-sensitive response extraction module in real time through an anti-interference differential line. It also has a built-in synchronization verification mechanism that immediately triggers resynchronization when a synchronization error exceeds 5μs. The module adopts a dual-core architecture of FPGA and ARM. The FPGA is responsible for real-time excitation and synchronization signal distribution, while the ARM is responsible for instruction parsing and status monitoring. The micro-thermal excitation source is equipped with an air-cooled heat dissipation module. At the same time, it uploads parameters such as excitation status, power, and temperature rise to the monitoring center every 100ms through an Ethernet or RS485 interface.

[0021] The dual-mode high-speed acquisition module includes an electrical signal acquisition channel and a thermal signal acquisition channel. The electrical signal acquisition channel first securely connects the current and voltage sensors through a pre-defined interface specification to ensure electrical matching with the photovoltaic string circuit. Upon receiving the synchronization trigger signal from the active excitation and synchronization control module, it initiates continuous acquisition at a sampling rate of no less than 1kHz. During acquisition, baseline calibration and preliminary outlier detection are performed on the raw signals in real time, continuously capturing the raw time-domain waveforms of the instantaneous operating current and voltage to ground of the entire string. The thermal signal acquisition channel first completes the setup, positioning, and parameter calibration of the high-speed infrared thermal imager, ensuring the lens is focused on the target excitation area and the field of view completely covers it. A frame rate much higher than the micro-thermal excitation frequency is determined and set. Upon arrival of the synchronization trigger signal, continuous shooting begins, sequentially capturing temperature field images containing the target excitation area to form an infrared thermal image sequence. ,in Here, t represents pixel space coordinates and time. During the shooting process, invalid frames such as blurry or overexposed frames are filtered and removed in real time. After acquiring each set of raw data, the electrical signal acquisition channel and the thermal signal acquisition channel immediately receive a unified high-precision timestamp from the synchronization control module and complete the marking. The marking process is executed synchronously with the acquisition action to ensure that each data point corresponds to a unique time reference. At the same time, the marked raw data packets are temporarily stored in the module's built-in cache in chronological order. When the cached data reaches the preset batch threshold or a transmission command is received from the underlying layer, the dual-modal high-speed acquisition module automatically determines the current network status and data type, selects parallel or serial transmission mode, and packages and sends the raw electrical signal data and thermal image sequence data stream with timestamps through the preset data interface. During the transmission process, the data integrity is checked in real time. If a transmission interruption or data loss is detected, a retransmission mechanism is immediately initiated to ensure that the data is accurately delivered to the subsequent phase-sensitive response extraction and alignment module.

[0022] The phase-sensitive response extraction and alignment module includes: a point response extraction unit, a thermal response extraction unit, and a feature alignment and fusion unit. The electrical response extraction unit receives raw current and voltage data, employs a digital phase-locked loop amplification algorithm, and uses the excitation frequency signal provided by the synchronization control module as the reference frequency to demodulate the raw signal, separating the signal components with the same frequency and phase as the excitation from the broadband background electrical noise. DC component stripping is performed on the raw current and voltage, and the data standard deviation is calculated. If the standard deviation > 0.1A current or > 1V voltage, it is determined to be abnormal data, triggering a re-reception. The reference signal is generated using the active excitation frequency... Based on this, a digital reference signal is generated: ; ;in For in-phase reference signal, This is an orthogonal reference signal; it will be compared with the processed current signal. Voltage signal Multiplying each signal by the two reference signals and performing quadrature demodulation yields: ; ; ; ;in and It is the quadrature demodulation product of the current signal. and The product signal is the quadrature demodulation product of the voltage signal; the product signal is then passed through an 8th-order IIR low-pass filter to remove high-frequency noise and unexcited frequency components, yielding the DC component. , and , ,in for The filtered DC component, for The filtered DC component, for The filtered DC component, for The DC component is then used to calculate the amplitude of the current perturbation. With phase The amplitude of voltage perturbation and phase : ; ; ; ;in The amplitude of the current perturbation. For the phase of the current perturbation, the range is within , The voltage perturbation amplitude, For voltage perturbation phase, the range is within , This represents the average steady-state operating current of the photovoltaic module under test under conditions without micro-thermal excitation. The average steady-state operating voltage is used, and the relative value of the current perturbation is also calculated: Relative value of voltage perturbation ;like and If no valid electrical response is found, the data is marked as invalid.

[0023] Receive infrared thermal image sequence The locked-in thermal imaging processing algorithm is used, also taking the excitation frequency as a reference, to process each pixel in the image sequence. Phase-locked analysis was performed on the brightness (temperature) curve over time, and this process generated a phase-locked thermal amplitude map. and lock-in thermal phase diagram .

[0024] Pixel-level phase-locked thermal imaging analysis is performed on the infrared thermal image sequence. The specific process is as follows: For each frame of the infrared image, bad pixel repair, non-uniformity correction, and thermal noise suppression are performed. The signal-to-noise ratio (SNR) of a single frame is calculated as: SNR = signal mean / noise standard deviation. If SNR < 10 dB, the frame is discarded and re-acquisition is triggered. Based on the synchronization timestamp, a subset of images strictly synchronized with the micro-thermal excitation cycle is selected, retaining N frames per cycle, where N ≥ 10, to ensure that the temperature changes in each excitation cycle are completely captured. For each pixel... Extracting temperature time series , Represents pixels At any moment Temperature value; independently execute the same quadrature demodulation process as the electrical signal respectively with , After multiplication and low-pass filtering, orthogonal components are obtained. , ,in For in-phase components, For orthogonal components, the branch-cutting method is used for phase unwrapping; then, phase-locked thermal maps are generated, and the phase-locked thermal amplitude map of the entire field is calculated. and phase diagram : ; ;in For pixels The phase-locked thermal amplitude is linearly mapped to a 16-bit grayscale value, with a mapping range of 0-1℃. For pixels The phase-locked thermal phase, ranging from 0° to 360°, is excluded. Furthermore, it eliminates the need for phase-invalid pixels to avoid background noise interference.

[0025] The feature alignment and fusion unit first correlates the global electrical perturbation parameters calculated by the electrical response extraction unit with the phase-locked thermal image generated by the thermal response extraction unit in time, based on a unified timestamp. Second, it maps the physical spatial location of the applied micro-thermal excitation to a specific pixel region of the phase-locked thermal image using a pre-calibrated coordinate mapping relationship. Finally, one or more fused feature pairs are formed.

[0026] Calculate the time difference between electrical and thermal characteristics based on a unified timestamp ;like Directly related; if After supplementing the electrical response data using linear interpolation, correlation is performed; if The system is determined to be out of sync with time, triggering the active excitation and synchronization control module to resynchronize. It then invokes the coordinate transformation relationship obtained beforehand by the system using a 1cm×1cm checkerboard calibration plate to adjust the physical coordinates of the micro-thermal excitation. Mapping to a specific pixel region of the phase-locked thermal image, mapping error Extract pixel regions Internal thermal characteristic statistics: ; ;in pixel area The number of effective pixels within the area; combining thermal feature statistics with global electrical perturbation features to generate a fusion feature F representing location and electrothermal response: The fusion feature pair F serves as the input data for the physical model-based fault diagnosis module.

[0027] The physical model-based fault diagnosis module is the intelligent decision-making center of the system. It achieves accurate fault diagnosis based on the physical mechanism and fusion characteristics of photovoltaic modules. It includes a fault response feature database and a diagnostic inference algorithm processing unit. Based on the physical mechanism of photovoltaic modules, it establishes and pre-stores quantitative feature models of various typical faults, as follows: It contains more than 1,000 samples covering different module types, environmental conditions and fault severity. Each sample contains basic information, namely module model, fusion feature parameters and fault label. It supports offline import via USB interface and incremental update mechanism of online distribution from monitoring center. Every 500 new samples are accumulated, the feature mean and variance are re-optimized.

[0028] A fault response characteristic database was used to establish a true hot spot model, a shading pseudo hot spot model, and an early microcrack model. The average phase-locked thermal amplitude of the target area; The amplitude of the current perturbation; This represents the relative value of the current perturbation. The phase of the current perturbation; The phase of the current perturbation in the normal region; The physical area of ​​the fault region; the true hot spot model is... : The series resistance at the fault point is extremely high, and micro-thermal excitation further increases it, leading to a decrease in the loop current; a shading pseudo-hotspot model. : The obstruction only alters the local thermal equilibrium and does not damage the intrinsic electrical properties of the solar cell, therefore there is no abnormal electrical response; early microcrack model. : Microcracks alter carrier recombination dynamics, resulting in a phase delay in the electrical response relative to thermal excitation.

[0029] The input fused feature pairs are matched and compared with the database, and similarity calculations are performed. The fault diagnosis module based on the physical model outputs a structured diagnostic conclusion, which includes at least: whether the fault exists, the type of fault, the precise location of the fault on the component, and the severity level of the fault. This result is uploaded to the monitoring center or triggers an early warning signal.

[0030] The parameters in F of the fusion feature are subjected to Min-Max normalization to eliminate baseline differences between the environment and operating conditions. The normalization formula is as follows: Where xg is the feature parameter to be normalized. This is the minimum value of this feature in the fault response feature database. This represents the maximum value of this feature in the database. The normalized feature values ​​range from 0 to 1; in the similarity calculation, a weighted Euclidean distance metric is used to fuse feature pairs F with each fault model. The similarity is calculated as follows: the smaller the distance, the higher the similarity. ;in To fuse features F with the i-th type of fault model The weighted Euclidean distance, , , , , The weights for the remaining feature parameters, For the i-th type of fault model The standard value, For the i-th type of fault model The standard value; if F is different from the normal template Weighted Euclidean distance If it is determined to be fault-free; Enter fault type identification; if F and distance Minimum, and satisfying: If it is determined to be a true hot spot; If it is determined to be a false hot spot that is blocked; If the condition is determined to be an early-stage microcrack, and multiple fault determination conditions are met simultaneously, the type with the highest percentage of fault labels is selected. If the percentage difference is ≤10%, a manual review prompt is triggered.

[0031] The severity of true hot spots is based on the relative value of current perturbations. Average phase-locked thermal amplitude of the target area Physical area of ​​the fault zone The three indicators are used to classify the levels as follows: the relative value of the current disturbance is between -0.05% and -0.03%, the average phase-locked thermal amplitude of the target area is between 0.5℃ and 0.8℃, and the physical area of ​​the fault region is less than 1. The fault is mild; the relative value of the current disturbance is between -0.1% and -0.05%, the average phase-locked thermal amplitude of the target area is between 0.8℃ and 1.2℃, and the physical area of ​​the fault region is within 1... A fault area of ​​3 cm² or less is considered moderate; the relative value of the current disturbance is not higher than -0.1%, the average phase-locked thermal amplitude of the target area is not lower than 1.2℃, and the physical area of ​​the fault region is not less than 3 cm². It is severe.

[0032] The severity of early-stage microcracks is classified based on two indicators: the hysteresis of the current perturbation phase relative to the normal region and the physical area of ​​the fault region. Specific levels are as follows: the hysteresis of the current perturbation phase relative to the normal region is between 30° and 45°, and the physical area of ​​the fault region is less than 0.3. -0.5 The range is classified as Level 1; the phase hysteresis of the current perturbation relative to the normal region is >45°-60°, and the physical area of ​​the fault region is >0.5. -1 The interval is between level two; the phase hysteresis of the current perturbation relative to the normal region is between >60° and 90°, and the physical area of ​​the fault region is >1. It is classified as Level 3.

[0033] The severity of pseudo-hot spots caused by shading is classified based on two indicators: the physical area of ​​the fault region and the percentage of component power degradation. The specific levels are as follows: Fault region physical area less than 2... It is a minor fault; the physical area of ​​the fault region is within 2. Up to 5 The condition is considered moderate if the component power degradation percentage does not exceed 5% and the physical area of ​​the fault zone is not less than 5 square meters. Furthermore, a component power attenuation percentage exceeding 5% is considered severe.

[0034] Finally, after verifying the consistency of diagnostic results over three consecutive excitation cycles, a structured diagnostic report is output, including diagnostic timestamp, component number, presence or absence of fault, specific type, precise physical coordinates, severity level, power attenuation estimate, handling suggestions, and confidence level. This report is then uploaded to the monitoring center via the MQTT protocol, and corresponding alerts are triggered based on the severity level.

[0035] Working principle: This multimodal photovoltaic module fault early warning system achieves accurate early warning of photovoltaic module faults through the execution process of active excitation, synchronous data acquisition, phase-sensitive extraction and model diagnosis.

[0036] First, the upper-level system identifies suspected fault areas in the photovoltaic modules through initial screening and sends the physical coordinates of these areas to the active excitation and synchronization control module. Upon receiving the command, this module drives a semiconductor laser with adjustable power and spot size to apply a sinusoidal micro-thermal excitation with a frequency of 0.1-1Hz to the target area. Simultaneously, through PID closed-loop control combined with temperature sensor feedback, the instantaneous temperature rise of the target area is strictly limited to a safe range of 3-5℃ to avoid damaging the modules. At the same time, the module's built-in synchronization signal generator generates a reference clock signal that is strictly synchronized with the excitation waveform and distributes it to subsequent modules through a dedicated line or high-precision network protocol to ensure that the time synchronization accuracy of the entire system is better than 1μs.

[0037] Next, the dual-mode high-speed acquisition module responds to the synchronization signal and simultaneously starts the acquisition of the electrical and thermal channels. The electrical signal acquisition channel continuously acquires the instantaneous current and voltage raw data of the photovoltaic string at a sampling rate of not less than 1kHz through high-precision current and voltage sensors. The thermal signal acquisition channel continuously captures temperature field images of the target area through a high-speed infrared thermal imager to form an infrared thermal image sequence.

[0038] All acquisition actions on both channels are triggered by a synchronization signal. The acquired raw data packets are marked with a unified nanosecond-level timestamp in real time and then transmitted to the phase-sensitive response extraction and alignment module through a high-speed interface.

[0039] In the phase-sensitive response extraction and alignment module, the received raw electrical signal is first preprocessed to remove the DC component and calibrate the baseline. After removing abnormal data, an in-phase and quadrature reference signal with the same excitation frequency is generated. The preprocessed current and voltage signals are multiplied with the reference signal respectively, and then noise is filtered out by low-pass filtering to obtain the perturbation characteristics of current and voltage. At the same time, the received infrared thermal image sequence is preprocessed to repair bad pixels, correct non-uniformity and suppress thermal noise. After removing low signal-to-noise ratio images, a subset of images synchronized with the excitation period is selected. Phase-locked analysis is performed independently on the temperature time series of each pixel to generate a phase-locked thermal amplitude map and phase map that only reflect the thermal characteristics of the excitation frequency.

[0040] The module then performs spatiotemporal alignment, associating the electrical perturbation features with the phase-locked thermal image in time based on a unified timestamp. It calls the coordinate transformation relationship obtained through pre-calibration to map the physical coordinates of the target area to a specific pixel area of ​​the phase-locked thermal image, extracts the average thermal amplitude and phase within that area, and finally fuses these thermal features with the global electrical perturbation features to form a fused feature pair containing location, electrical features, and thermal features.

[0041] Subsequently, the fused feature pairs are transmitted to the physical model-based fault diagnosis module. This module first calls the pre-stored fault response feature database, which includes quantitative fault models based on physical mechanisms, such as true hot spots, shading pseudo hot spots, and early microcracks. The input fused feature pairs are preprocessed to eliminate environmental baseline differences. Then, the similarity between the feature pairs and each fault model is calculated. The module first determines whether the similarity between the feature pairs and the normal model exceeds a threshold. If it does, an anomaly is identified, and the fault type identification is initiated: if the feature pair meets the following conditions, such as high thermal amplitude, negative current perturbation, and phase out of phase with the excitation, it is identified as a true hot spot; if the thermal amplitude is high but the current perturbation is not significantly different from the normal area, it is identified as a shading pseudo hot spot; if the thermal amplitude is normal but the current perturbation phase shows a 30-90° lag, it is identified as an early microcrack; if multiple fault conditions are met simultaneously, the type with the highest fault label ratio is selected. If the ratio difference is too small, manual review is triggered.

[0042] After identifying the fault type, the module classifies the severity according to the corresponding indicators: true hot spots are classified into light, moderate, and severe based on the relative value of current perturbation, thermal amplitude, and fault area; early microcracks are classified into level one, level two, and level three based on phase hysteresis and fault area; and shading pseudo hot spots are classified into light, moderate, and severe based on fault area and power attenuation. Subsequently, the module performs consistency verification on the diagnostic results of three consecutive excitation cycles. If the results are consistent, a structured report containing the diagnostic timestamp, component number, fault type, precise location, severity level, and handling suggestions is output and uploaded to the monitoring center via the communication protocol. At the same time, corresponding early warnings are triggered according to the severity, thus completing the entire fault early warning process.

[0043] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A photovoltaic module fault early warning system based on multimodal fusion, characterized in that, The system comprises an active excitation and synchronization control module, a dual-mode high-speed acquisition module, a phase-sensitive response extraction and alignment module, and a physical model-based fault diagnosis module, all connected sequentially by signals. The active excitation and synchronization control module receives coordinate instructions for the suspected fault area from the upper-level system, drives a semiconductor laser with adjustable power and spot size to apply a sinusoidal low-frequency micro-thermal excitation to the target area, and simultaneously generates a frequency reference signal and a global timestamp signal synchronized with the excitation waveform, distributing them to subsequent modules to provide a unified time reference. The dual-mode high-speed acquisition module responds to the synchronization signal by sensing current and voltage... The device collects instantaneous current and voltage data of the string, and simultaneously captures temperature field images of the target area at a frame rate higher than the excitation frequency using a high-speed infrared thermal imager to form an infrared thermal image sequence. All raw data packets are marked with a unified high-precision timestamp before transmission. The phase-sensitive response extraction and alignment module adopts a digital phase-locked amplification algorithm to receive electrical signals from the dual-mode high-speed acquisition module, extract current and voltage perturbation features, and generate phase-locked thermal amplitude and phase maps from the infrared image sequence using a phase-locked thermal imaging algorithm. Based on the timestamp and pre-calibrated coordinate transformation relationship, the dual-mode features are spatiotemporally aligned to generate fused feature pairs. The physical model-based fault diagnosis module calls the pre-stored fault response feature database, performs normalization and similarity calculation on the fused feature pairs, and outputs the diagnosis results.

2. The photovoltaic module fault early warning system based on multimodal fusion according to claim 1, characterized in that: The active excitation and synchronization control module is also used to calibrate the coordinates of the target area based on the real-time image of the infrared thermal imager before driving the semiconductor laser.

3. The photovoltaic module fault early warning system based on multimodal fusion according to claim 1, characterized in that: The dual-modal high-speed acquisition module is also used to remove abnormal data points in the electrical signal and invalid frames in the thermal image in real time during the acquisition process.

4. The photovoltaic module fault early warning system based on multimodal fusion according to claim 1, characterized in that: The electrical response extraction unit in the phase-sensitive response extraction and alignment module is used to preprocess the raw current and voltage data, including DC component stripping and baseline calibration.

5. The photovoltaic module fault early warning system based on multimodal fusion according to claim 1, characterized in that: The thermal response extraction unit in the phase-sensitive response extraction and alignment module is used to perform image quality screening on the infrared thermal image sequence and remove image frames with a signal-to-noise ratio lower than a preset threshold.

6. The photovoltaic module fault early warning system based on multimodal fusion according to claim 1, characterized in that: When the feature alignment and fusion unit in the phase-sensitive response extraction and alignment module performs time-domain correlation based on a unified timestamp, if the difference between the timestamp of the electrical feature data and the timestamp of the thermal feature data exceeds a preset tolerance, it will trigger resynchronization or data compensation.

7. The photovoltaic module fault early warning system based on multimodal fusion according to claim 1, characterized in that: The physical model-based fault diagnosis module normalizes the fused feature pairs and calculates their similarity to each fault feature model. First, it determines whether there is an anomaly based on the comparison with the normal state model. If there is an anomaly, it identifies the fault type based on the principle of highest similarity and preset logical rules.

8. The photovoltaic module fault early warning system based on multimodal fusion according to claim 7, characterized in that: The preset logical rules include: if the fusion feature satisfies significant thermal characteristics and negative and out-of-phase electrical perturbation, it is determined to be a true hot spot; if the thermal characteristics are significant but the electrical perturbation is not significantly different from the normal state, it is determined to be a shading pseudo hot spot; if the electrical perturbation phase shows characteristic hysteresis but the thermal characteristics are not significantly abnormal, it is determined to be an early microcrack.

9. The photovoltaic module fault early warning system based on multimodal fusion according to claim 8, characterized in that: After determining the fault type, the physical model-based fault diagnosis module quantifies and classifies the severity of the fault based on the feature parameter values ​​in the fused feature pair.

10. The photovoltaic module fault early warning system based on multimodal fusion according to claim 9, characterized in that: The physical model-based fault diagnosis module performs consistency verification on the diagnostic results of multiple consecutive excitation cycles and outputs a structured diagnostic report containing the fault type, location, and severity level.

Images