Photovoltaic module fault early warning method and system based on time sequence and space double flow model

Abstract

The present disclosure provides a photovoltaic module fault early warning method and system based on a time and space double-flow model, relates to the technical field of photovoltaic module prediction and early warning, and comprises the following steps: obtaining multi-source heterogeneous data through an improved PTPv2 time synchronization algorithm; inputting the preprocessed multi-source heterogeneous data into a space-time feature joint model, extracting electric parameter flow features in the multi-source heterogeneous data by using an LSTM network in the space-time feature joint model, extracting thermal image features in the multi-source heterogeneous data by using a ViT network in the space-time feature joint model, modeling a space relationship by using a self-attention mechanism, correlating cross-frame hot spots, and obtaining thermal imaging flow features; introducing a health index and a fusion weight controlled by real-time irradiance, dynamically fusing the thermal imaging flow features and the electric parameter flow features, and obtaining double-modality joint features; inputting the double-modality joint features into a federal model for decision-making, and outputting fault types and early warnings.

Description

Technical Field

[0001] This disclosure relates to the field of photovoltaic module prediction and early warning technology, specifically to a photovoltaic module fault early warning method and system based on a time-series and spatial dual-flow model. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] A photovoltaic module, also known as a solar panel, is a device that converts solar energy into electrical energy. It is composed of multiple photovoltaic cells (solar cells) encapsulated in series and parallel. It is the core component of a photovoltaic power generation system. The core components of a photovoltaic module are: photovoltaic cells, encapsulation materials and frames, and junction boxes.

[0004] Photovoltaic module fault prediction and early warning refers to a technical system that uses real-time monitoring, data analysis, and intelligent algorithms to identify potential fault risks in photovoltaic modules in advance and trigger a tiered response mechanism. Its core objective is to achieve a shift from "passive maintenance" to "proactive prevention."

[0005] Traditional photovoltaic module fault prediction and early warning methods suffer from low data synchronization accuracy, leading to spatiotemporal misalignment of electrical and thermal data and affecting fault correlation analysis. When current and voltage data are affected by environmental interference (such as electromagnetic noise), conventional filtering methods are prone to signal distortion. Centralized training makes it difficult for models to adapt to the environmental differences of different power plants. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes a photovoltaic module fault early warning method and system based on a temporal and spatial dual-stream model. It achieves μs-level synchronization through an improved PTPv2 algorithm, enhances data quality by combining VMD-ADAN joint noise reduction and 3D nonlocal mean filtering, and introduces HI and G through dual-stream modeling of LSTM (temporal) and ViT (spatial). t The fusion weight of regulation enhances fault sensitivity.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] A photovoltaic module fault early warning method based on a time-series and spatial dual-flow model includes:

[0009] μs-level time synchronization of multi-source heterogeneous data is achieved through an improved PTPv2 time synchronization algorithm.

[0010] Preprocess the acquired multi-source heterogeneous data;

[0011] The preprocessed multi-source heterogeneous data is input into the spatiotemporal feature joint model. The LSTM network in the spatiotemporal feature joint model is used to extract the electrical parameter flow features in the multi-source heterogeneous data. The ViT network in the spatiotemporal feature joint model is used to extract the thermal image features in the multi-source heterogeneous data. The self-attention mechanism is used to model the spatial relationship and associate cross-frame hotspots to obtain thermal imaging flow features.

[0012] By introducing the fusion weights of health index and real-time irradiance regulation, thermal imaging flow characteristics and electrical parameter flow characteristics are dynamically fused to obtain dual-modal joint characteristics;

[0013] The joint features of the two modes are input into the federated model for decision-making, and the output is the fault type and warning.

[0014] According to some embodiments, the present disclosure adopts the following technical solutions:

[0015] A photovoltaic module fault early warning system based on a temporal and spatial dual-flow model includes:

[0016] The data acquisition module is used to acquire multi-source heterogeneous data at the μs level using an improved PTPv2 time synchronization algorithm;

[0017] The preprocessing module is used to preprocess the acquired multi-source heterogeneous data;

[0018] The feature extraction module is used to input the preprocessed multi-source heterogeneous data into the spatiotemporal feature joint model, use the LSTM network in the spatiotemporal feature joint model to extract the electrical parameter flow features in the multi-source heterogeneous data, use the ViT network in the spatiotemporal feature joint model to extract the thermal image features in the multi-source heterogeneous data, and use the self-attention mechanism to model spatial relationships, associate cross-frame hotspots, and obtain thermal imaging flow features.

[0019] The joint characterization module is used to introduce the fusion weights of health index and real-time irradiance regulation, and to dynamically fuse thermal imaging flow features with electrical parameter flow features to obtain dual-modal joint features;

[0020] The fault prediction and early warning module is used to input the joint features of the two modes into the federated model for decision-making, output the fault type and issue an early warning.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] A computer program product includes a computer program that, when executed by a processor, implements the photovoltaic module fault early warning method based on a time-series and spatial dual-flow model.

[0023] According to some embodiments, the present disclosure adopts the following technical solutions:

[0024] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the photovoltaic module fault early warning method based on a time-series and spatial dual-stream model.

[0025] According to some embodiments, the present disclosure adopts the following technical solutions:

[0026] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the photovoltaic module fault early warning method based on the time-series and spatial dual-flow model.

[0027] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0028] This disclosure presents a photovoltaic module fault early warning method based on a temporal and spatial dual-stream model. During data acquisition, an improved PTPv2 algorithm achieves μs-level synchronization between the infrared thermal imager and the IV scanner. In preprocessing, VMD-ADAN joint noise reduction and 3D nonlocal mean filtering are combined to improve data quality. A joint spatiotemporal feature model is constructed, incorporating LSTM (temporal) and ViT (spatial) dual-stream modeling, introducing HI and G... t The fusion weight σ is adjusted to enhance fault sensitivity and achieve accurate feature extraction and high-quality, effective fusion.

[0029] This disclosed photovoltaic module fault early warning method based on a temporal and spatial dual-flow model, through the fusion and collaborative detection of multimodal data and the joint decision-making of electrical parameters and thermal imaging flow characteristics, covers the entire scenario from microcracks to system-level faults. It has the advantages of more efficient and accurate fault prediction and early warning, and solves the problems of low data synchronization accuracy, spatiotemporal misalignment of electrical and thermal data, and impaired fault correlation analysis in traditional photovoltaic module fault prediction and early warning methods. Attached Figure Description

[0030] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0031] Figure 1 This is a flowchart illustrating the overall architecture of the photovoltaic module fault early warning method based on a temporal and spatial dual-flow model, as described in this embodiment of the disclosure.

[0032] Figure 2 This is a schematic diagram of the photovoltaic module fault early warning method based on a time-series and spatial dual-flow model according to an embodiment of the present disclosure. Detailed Implementation

[0033] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0034] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0035] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0036] Example 1

[0037] One embodiment of this disclosure provides a photovoltaic module fault early warning method based on a time-series and spatial dual-flow model, the method comprising:

[0038] Step 1: Acquire μs-level time synchronization of multi-source heterogeneous data using the improved PTPv2 time synchronization algorithm;

[0039] Step 2: Preprocess the acquired multi-source heterogeneous data;

[0040] Step 3: Input the preprocessed multi-source heterogeneous data into the spatiotemporal feature joint model, use the LSTM network in the spatiotemporal feature joint model to extract the electrical parameter flow features in the multi-source heterogeneous data, use the ViT network in the spatiotemporal feature joint model to extract the thermal image features in the multi-source heterogeneous data, and use the self-attention mechanism to model the spatial relationship, associate cross-frame hotspots, and obtain thermal imaging flow features.

[0041] Step 4: Introduce the fusion weights of health index and real-time irradiance regulation, and dynamically fuse thermal imaging flow features with electrical parameter flow features to obtain dual-modal joint features;

[0042] Step 5: Input the dual-modal joint features into the federated model for decision-making, output the fault type and issue a warning.

[0043] As one embodiment, this disclosure presents a photovoltaic module fault early warning method based on a temporal and spatial dual-stream model. It achieves μs-level synchronization through an improved PTPv2 algorithm, improves data quality by combining VMD-ADAN joint noise reduction and 3D nonlocal mean filtering, and enhances fault sensitivity by introducing fusion weights controlled by HI and Gt through dual-stream modeling of LSTM (temporal) and ViT (spatial). The specific implementation process of the method is as follows:

[0044] Step 1: Collect data and perform μs-level time synchronization using the improved PTPv2 time synchronization algorithm;

[0045] Specifically, infrared thermal imagers and IV scanners are deployed to collect multi-source heterogeneous data, and μs-level time synchronization is achieved through an improved PTPv2 time synchronization algorithm;

[0046] The multi-source heterogeneous data includes current and voltage data as well as thermal imaging data. An infrared thermal imager acquires thermal imaging data, while an IV scanner collects current and voltage data. An improved PTPv2 algorithm is used, incorporating temperature coefficients, proportional-integral coefficients, and crystal oscillator temperature drift compensation terms. Path weights are applied to the data transmission and reception processes to achieve μs-level time synchronization between the infrared thermal imager and the IV scanner. The framework formula of the PTPv2 time synchronization algorithm is as follows:

[0047]

[0048] Where t1 and t4 are the timestamps of the master clock sending and receiving messages, respectively (master clock time field); t2 and t3 are the timestamps of the slave clock receiving and sending messages, respectively (slave clock time field). Crystal oscillator temperature drift compensation term, where α is the temperature coefficient and ΔT is the temperature difference; ResidenceTime i W is the message dwell time for the i-th transparent clock (TC); i K represents the path weight. p ,K i This is the proportional-integral coefficient, used for servo control.

[0049] This disclosure achieves precise synchronization (error ≤ 1μs) between an infrared thermal imager and an IV scanner by using an improved PTPv2 algorithm (including crystal oscillator temperature drift compensation), thus solving the spatiotemporal misalignment problem of traditional methods.

[0050] Step 2: Preprocess the acquired multi-source heterogeneous data; this includes VMD-ADAN joint noise reduction for current and voltage data, and 3D nonlocal mean filtering for thermal imaging data. Details are as follows:

[0051] Step 21: Perform VMD-ADAN joint noise reduction on the current and voltage data, which also includes the following steps:

[0052] Step 211: First, perform variational mode decomposition (VMD) preprocessing, preset the number of IMF components (K value) and bandwidth limit parameters, decompose the original signal into K quasi-orthogonal mode components, and suppress mode aliasing through Hilbert transform and frequency mixing optimization to ensure that the physical meaning of each IMF is clear.

[0053] Step 212: Adaptive Denoising (ADAN) processing, calculating the signal-to-noise ratio (SNR) for each IMF component, and dynamically selecting wavelet thresholding or total variation (TV) regularization method:

[0054] (1) High-frequency noise dominant component: The improved Stein unbiased risk estimation (SURE) threshold is adopted.

[0055] (2) Low-frequency useful signal components: edge features are preserved by applying a non-convex penalty term.

[0056] Step 213: Signal reconstruction and verification. Weighted fusion and noise reduction of the IMF components, calculate the root mean square error (RMSE) and signal-to-noise ratio improvement (ΔSNR), with a target ΔSNR ≥ 10dB.

[0057] Step 22: Perform 3D nonlocal mean filtering on the thermal imaging data, which also includes the following steps:

[0058] Step 221: Spatiotemporal noise modeling, analyze noise types such as pot lid effect, cold reflection, etc., confirm additive / multiplicative noise characteristics, construct a spatiotemporal cube for continuous frame sequences, and mark cloud reflection areas as high noise areas;

[0059] Step 222: Perform 3D nonlocal similarity calculation, extend traditional NLM to three-dimensional space, search for similar blocks in the spatiotemporal cube, and set the similarity window to: 7×7 pixels + 5 frames;

[0060] Then, weights are calculated: Gaussian weighting based on the Euclidean distance between blocks;

[0061] Step 223: Dynamic filtering and compensation, enhancing the weight attenuation coefficient of the cloud reflection area to suppress high-frequency reflection noise.

[0062] As one embodiment, the key parameters are shown in Table 1.

[0063] Table 1 Key Parameter Reference

[0064]

[0065] This disclosure employs a combined VMD-ADAN noise reduction method for electrical parameters (ΔSNR ≥ 10dB), using a SURE threshold for high-frequency noise and preserving edge features for low-frequency signals. Furthermore, 3D nonlocal mean filtering is applied to thermal imaging data (resulting in a high-frequency detail loss of <3%), effectively suppressing dynamic noise such as cloud reflections. This significantly improves the data signal-to-noise ratio, providing high-fidelity input for subsequent feature extraction.

[0066] Step 3: Input the preprocessed multi-source heterogeneous data into the spatiotemporal feature joint model, use the LSTM network in the spatiotemporal feature joint model to extract the electrical parameter flow features in the multi-source heterogeneous data, use the ViT network in the spatiotemporal feature joint model to extract the thermal image features in the multi-source heterogeneous data, and use the self-attention mechanism to model the spatial relationship, associate cross-frame hotspots, and obtain thermal imaging flow features.

[0067] Specifically, a joint spatiotemporal feature model including an LSTM network and a ViT network is constructed to process voltage-current time series data and thermal imaging image data, respectively.

[0068] Step 31: Standardize the IV curve time series data (voltage-current sequence), align the length to a fixed step size T, and use a 256-element bidirectional LSTM layer to extract the electrical parameter flow characteristics of the time series data:

[0069] h t =LSTM(x t ,h t-1 ),t∈[1,T]

[0070] Where, x t The electrical parameter input at time t, h t It is in a hidden state;

[0071] Furthermore, a local attention mechanism is employed to dynamically assign weights to voltage anomaly intervals, where the attention mechanism focuses on the 0.5Vmp-0.8Vmp voltage interval, and the weight assignment coefficient α∈[0.6,1.0]:

[0072]

[0073] Among them, the weight matrix Optimize through backpropagation.

[0074] Step 32: ViT network spatial modeling, dividing the thermal image into 16×16 blocks, linearly projecting it into a D-dimensional vector:

[0075]

[0076] in, For embedding matrix, For position encoding.

[0077] Further, cross-frame hotspot correlations are captured using multi-head self-attention (MSA):

[0078] MSA(Z) = concat(head1, …, head k W O head i = Attention(ZW i Q ZW i K ZW i V )

[0079] Finally, the output feature thermal imaging flow feature Z is processed by LayerNorm and MLP.

[0080] Step 33: Compare the LSTM output h^ (temporal features) with the ViT's CLS tokenclass z class (Spatial features) are mapped to the same dimension d;

[0081] Step 34: Adaptive fusion coefficients, composed of Health Index (HI) and real-time irradiance (G). t Adjusting the fusion weight σ:

[0082]

[0083] in, These are learnable parameters.

[0084] Step 35: Adjust the fusion weights to perform dynamic fusion and obtain the bimodal joint features. The weighted dynamic fusion yields the following bimodal joint features:

[0085]

[0086] in, It is a dual-modal joint feature. To integrate weights, For electrical parameters and current characteristics, This is for thermal imaging flow characteristics. Final output. Used for degradation detection or fault classification. Key parameter configurations are shown in Table 2.

[0087] Table 2 Key Parameter Configuration

[0088]

[0089] This disclosure extracts temporal features through dual-stream modeling and dynamic fusion of spatiotemporal features: 256-unit BiLSTM + local attention mechanism (focusing on the 0.5Vmp-0.8Vmp anomaly interval).

[0090] Spatial features: The ViT model (16×16 blocks) captures cross-frame hotspot associations.

[0091] Furthermore, through adaptive weighting: the fusion coefficient σ is composed of the health index (HI) and real-time irradiance (G). t Dynamic calculations enhance fault sensitivity, improving the early detection rate of latent defects such as microcracks and PID effects by more than 30%.

[0092] Step 4: Input the dual-modal joint features into the federated model for decision-making, output the fault type and issue a warning.

[0093] Specifically, the federated model is trained using the MobileViT-S local model architecture. In the MobileViT-S local model architecture, the weighted cross-entropy loss function is used to handle the class imbalance in photovoltaic module fault detection. The federated model outputs the fault probability, obtains the fault type, and triggers graded early warning actions.

[0094] The formula for the weighted cross-entropy loss function is as follows:

[0095]

[0096] Where N is the number of samples; y i p is the true label (one-hot encoded) of the i-th sample; i w represents the model's predicted probability for the i-th sample. yi For category y i The weights are used to balance class imbalance.

[0097] In the MobileViT-Transformer hybrid architecture, this loss function is often used to handle class imbalance in photovoltaic module fault detection, such as when there are far more normal samples than abnormal samples. Adjusting wyi can improve the recall rate of the minority class.

[0098] The local model adopts the MobileViT-S architecture and uses incremental federated learning for training. The number of optimized parameters is ≤2M. Gradients are aggregated monthly using the DP-FedAvg algorithm, and feature distillation is used to retain knowledge from the old model. The global model aggregation cycle is 24±4 hours, and the differential privacy noise ε=0.5.

[0099] As one embodiment, the tiered early warning system of this disclosure is a three-level early warning system, with different measures taken for different levels, specifically including:

[0100] L1: HI < 0.85 and lasting for 10 minutes;

[0101] L2: Single component ΔT > 10℃ or IV curve distortion rate > 15%;

[0102] L3: Insulation resistance <0.5MΩ or arc energy >20J, millisecond-level disconnection of string power supply.

[0103] This disclosure includes four types of faults, as follows:

[0104] (1) Cell-level fault

[0105] Microcracks: Microcracks that cause the current path to break;

[0106] Fragment: Visible damage caused by mechanical stress;

[0107] Mixed-cell performance: Inconsistent cell efficiency parameters cause hot spots;

[0108] PID effect: Potential-induced decay leads to a sharp drop in power;

[0109] Lightning pattern: A dendritic defect formed by the propagation of hidden cracks;

[0110] Electrode corrosion: Moisture penetration leads to grid line oxidation.

[0111] (2) Welding and connection failures

[0112] Poor soldering: The solder strip does not make good contact with the battery cell;

[0113] Over-soldering: High-temperature damage to the internal structure of the battery cell;

[0114] Solder strip misalignment: Deviation in welding position reduces the conductive area;

[0115] Busbar fracture: caused by fatigue stress or improper installation;

[0116] Connector failure: plug oxidation or seal aging.

[0117] (3) Packaging material failure

[0118] EVA delamination: Insufficient cross-linking leads to component delamination;

[0119] Yellowing of the back panel: UV aging leads to a decrease in insulation performance;

[0120] Glass shattering: Damage caused by thermal stress or hail impact;

[0121] Sealant cracking: Moisture intrusion accelerates corrosion;

[0122] (4) System-level failure

[0123] Hot spot effect: overheating due to localized shading or battery defects;

[0124] MPPT failure: Inverter tracking algorithm malfunction;

[0125] DC arc: High-temperature discharge caused by poor contact;

[0126] Insulation failure: A humid environment causes excessive leakage current;

[0127] Shading: Vegetation or dust accumulation causes power loss;

[0128] Support deformation: Strong winds or snow accumulation can cause structural deformation;

[0129] Communication interruption: Data transmission module failure.

[0130] As one embodiment, the workflow of the photovoltaic module fault early warning method based on a time-series and spatial dual-flow model disclosed herein includes:

[0131] Input layer: Multi-source sensor data (IV curves, thermal images);

[0132] Feature layer: LSTM / ViT extracts features, and dynamic weights are fused;

[0133] Decision-making level: The federated model outputs the failure probability, triggering tiered actions;

[0134] As one example, virtual-real combined verification is performed at the feedback layer, and model iteration is driven by false alarm cases of digital twins. Specifically, 22 fault modes are injected into the digital twin platform for closed-loop testing, and false alarm cases are fed back to the local model for fine-tuning.

[0135] Example 2

[0136] One embodiment of this disclosure provides a photovoltaic module fault early warning system based on a temporal and spatial dual-flow model, including:

[0137] The data acquisition module is used to acquire multi-source heterogeneous data at the μs level using an improved PTPv2 time synchronization algorithm;

[0138] The preprocessing module is used to preprocess the acquired multi-source heterogeneous data;

[0139] The feature extraction module is used to input the preprocessed multi-source heterogeneous data into the spatiotemporal feature joint model, use the LSTM network in the spatiotemporal feature joint model to extract the electrical parameter flow features in the multi-source heterogeneous data, use the ViT network in the spatiotemporal feature joint model to extract the thermal image features in the multi-source heterogeneous data, and use the self-attention mechanism to model spatial relationships, associate cross-frame hotspots, and obtain thermal imaging flow features.

[0140] The joint characterization module is used to introduce the fusion weights of health index and real-time irradiance regulation, and to dynamically fuse thermal imaging flow features with electrical parameter flow features to obtain dual-modal joint features;

[0141] The fault prediction and early warning module is used to input the joint features of the two modes into the federated model for decision-making, output the fault type and issue an early warning.

[0142] Example 3

[0143] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the photovoltaic module fault early warning method based on a time-series and spatial dual-stream model.

[0144] Example 4

[0145] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the photovoltaic module fault early warning method based on a time-series and spatial dual-stream model.

[0146] Example 5

[0147] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the photovoltaic module fault early warning method based on the time-series and spatial dual-flow model.

[0148] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0149] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0150] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A photovoltaic module fault early warning method based on a time sequence and space double-flow model, characterized in that, The method comprises the following steps: Multi-source heterogeneous data μs-level time synchronization acquisition is performed through an improved PTPv2 time synchronization algorithm; The multi-source heterogeneous data includes current-voltage data and thermal imaging data, the thermal imaging data is acquired by using an infrared thermal imager, the current-voltage data is acquired by using an IV scanner, the improved PTPv2 algorithm is used, a temperature coefficient, a proportional-integral coefficient and a crystal oscillator temperature drift compensation term are introduced, and the process of data sending and data receiving is weighted through path weight, so that the infrared thermal imager and the IV scanner are μs-level time synchronized; the framework formula of the PTPv2 time synchronization algorithm is as follows: The acquired multi-source heterogeneous data is preprocessed; Wherein, t1, t4 are respectively the time stamp of the master clock sending and receiving the message; t2, t3 are respectively the time stamp of the slave clock receiving and sending the message, : crystal oscillator temperature drift compensation term, α is the temperature coefficient, ΔT is the temperature difference; ResidenceTime i ResidenceTime is the message residence time of the ith transparent clock; W i W is the path weight; K p K i K is the proportional-integral coefficient, used for servo control; The preprocessed multi-source heterogeneous data is input into a space-time feature joint model, the electric parameter flow features in the multi-source heterogeneous data are extracted by using an LSTM network in the space-time feature joint model, the thermal image features in the multi-source heterogeneous data are extracted by using a ViT network in the space-time feature joint model, and the space relationship is modeled by using a self-attention mechanism, so that the cross-frame hot spots are associated, and the thermal imaging flow features are obtained; The dual-mode joint features are input into a federal model for decision-making, and the fault type is output and an early warning is given. The health index and the real-time irradiance regulated fusion weight are introduced, the thermal imaging flow features and the electrical parameter flow features are dynamically fused, and a dual-modality combined feature is obtained; the electrical parameter flow features output by the LSTM network and the thermal imaging flow features output by the ViT network are mapped to the same dimension, and the health index HI and the real-time irradiance G t The health index and the real-time irradiance regulated fusion weight are introduced, the thermal imaging flow features and the electrical parameter flow features are dynamically fused, and a dual-modality combined feature is obtained; the electrical parameter flow features output by the LSTM network and the thermal imaging flow features output by the ViT network are mapped to the same dimension, and the health index HI and the real-time irradiance G wherein, is a bimodal joint feature, is a fusion weight, is an electrical parameter flow feature, is a thermal imaging flow feature; Controlling fusion weight by health index and real-time irradiance : wherein, are learnable parameters, HI is a health index, G t is the real-time irradiance; The current-voltage data preprocessing includes VMD-ADAN joint denoising, first, the original signal is decomposed into K quasi-orthogonal modal components by presetting the IMF component number and the bandwidth limit parameter, the Hilbert transform and the frequency mixing optimization are used to suppress the modal aliasing phenomenon, then the adaptive denoising is performed, the signal-to-noise ratio of each IMF component is calculated, and the wavelet threshold or the total variation regularization method is dynamically selected for denoising; the thermal imaging data is subjected to 3D non-local mean filtering.

2. The photovoltaic module failure early warning method based on time and space dual-flow model according to claim 1, wherein, The space-time feature joint model including the LSTM network and the ViT network is constructed, the voltage-current time series data and the thermal imaging image data are processed respectively, the length of the preprocessed voltage-current time series is aligned to a fixed step, the 256-unit bidirectional LSTM layer is used to extract the electric parameter flow features of the time series, and the weight is dynamically allocated to the voltage abnormal interval; 3.The photovoltaic module fault early warning method based on time and space double-flow model according to claim 1, wherein, The preprocessed thermal imaging image data is input into the ViT network, the spatial features of the thermal imaging image data are extracted, the cross-frame hot spot association is captured through the multi-head self-attention, and the thermal imaging flow features are obtained. The federal model is trained by using the MobileViT-S local model architecture, in the MobileViT-S local model architecture, the weighted cross-entropy loss function is used to process the class imbalance in the photovoltaic component fault detection, the federal model outputs the fault probability, the fault type is obtained, and the hierarchical early warning action is triggered. 4.The photovoltaic module fault early warning method based on time and space double-flow model according to claim 1, wherein, The method comprises the following steps:

5. A photovoltaic module failure warning system based on time series and spatial dual-flow model, characterized in that, The data acquisition module is used for acquiring multi-source heterogeneous data through an improved PTPv2 time synchronization algorithm; ​ The infrared thermal imager acquires thermal imaging data, the IV scanner collects current-voltage data, the improved PTPv2 algorithm is adopted, the temperature coefficient, the proportional-integral coefficient and the crystal oscillator temperature drift compensation term are introduced, the process of data sending and data receiving is weighted through path weight, and the time synchronization of the infrared thermal imager and the IV scanner is realized in the order of microseconds; the framework formula of the PTPv2 time synchronization algorithm is: Wherein, t1, t4 are respectively the time stamp of the master clock sending and receiving the message; t2, t3 are respectively the time stamp of the slave clock receiving and sending the message, : crystal oscillator temperature drift compensation term, α is the temperature coefficient, ΔT is the temperature difference; ResidenceTime i is the message residence time of the i th transparent clock; W i is the path weight; K p ,K i is the proportional-integral coefficient, used for servo control; The preprocessing module is used for preprocessing the obtained multi-source heterogeneous data. The feature extraction module is used for inputting the preprocessed multi-source heterogeneous data into the spatio-temporal feature joint model, extracting the electrical parameter flow features in the multi-source heterogeneous data by using the LSTM network in the spatio-temporal feature joint model, extracting the thermal image features in the multi-source heterogeneous data by using the ViT network in the spatio-temporal feature joint model, and modeling the spatial relationship by using the self-attention mechanism, correlating the cross-frame hot spots, and obtaining the thermal imaging flow features. The joint representation module is used for introducing a fusion weight of a health index and real-time irradiance regulation, dynamically fusing thermal imaging flow features and electrical parameter flow features, and obtaining a dual-modality joint feature; the electrical parameter flow features output by the LSTM network and the thermal imaging flow features output by the ViT network are mapped to the same dimension by the health index HI and real-time irradiance G t regulation, and dynamically fusing to obtain a dual-modality joint feature, wherein the dual-modality joint feature obtained by the weighted dynamic fusion is: wherein, is a bimodal joint feature, is a fusion weight, is an electrical parameter flow feature, is a thermal imaging flow feature; Controlling fusion weight by health index and real-time irradiance : wherein, are learnable parameters, HI is a health index, G t is the real-time irradiance; The fault prediction and early warning module is used for inputting the dual-modal joint features into the federal model for decision-making, and outputting the fault type and early warning.

6. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to realize the photovoltaic module fault early warning method based on the time sequence and space double-flow model in any one of claims 1-4.

7. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is used for storing computer instructions, and the computer instructions are executed by the processor to realize the photovoltaic module fault early warning method based on the time sequence and space double-flow model in any one of claims 1-4.

8. An electronic device, comprising: It comprises: A processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the photovoltaic module fault early warning method based on the time sequence and space double-flow model in any one of claims 1-4.

Images