Photovoltaic panel bimodal dynamic detection method and system based on lightweight edge intelligent box

Abstract

The invention discloses a photovoltaic panel bimodal dynamic detection method and system based on a lightweight edge intelligent box, and belongs to the technical field of photovoltaic power station intelligent operation and maintenance and edge computing crossing. The system comprises a Raspberry Pi Compute Module 5 core processor, a Raspberry Pi core processor, a Raspberry Pi core processor, a Raspberry Pi core processor and a Raspberry Pi core processor; a double-model dynamic architecture; and the security collaboration mechanism has the advantages of realizing collaborative analysis, dynamic switching and security processing of visible light and infrared thermal imaging, and solves the problems of hardware limitation, algorithm defects and system splitting in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of intersection of intelligent operation and maintenance of photovoltaic power stations and edge computing, and in particular relates to a dual-modal dynamic detection method and system for photovoltaic panels based on a lightweight edge intelligent box. Background Art

[0002] The patent number disclosed in the comparative document 1 is: CN113706458A. The core technical defect is that the problem of low-altitude small target recognition has not been solved. The patent number disclosed in the comparative document 2 is: CN114519587A. The core technical defect is that the infrared detection has not optimized the low-resolution sensor. The patent number disclosed in the comparative document 3 is: CN120029306A. The core technical defect is that the model has not been optimized for photovoltaic scenarios.

[0003] Pain points of existing technologies:

[0004] Hardware limitations: The power consumption of the transmission industrial computer solution is greater than 60W, which cannot meet the drone's endurance requirements. The dual-model inference latency of devices such as Jetson Nano is greater than 200ms, causing detection lag.

[0005] Algorithm flaws: General object detection models (such as YOLOv5) have a false detection rate of 35% for photovoltaic reflective scenes. Infrared hot spot detection directly upsamples to 416x416 resolution, resulting in a detail loss rate exceeding 70%.

[0006] System fragmentation: Visible light and infrared detection require separate flights, which is inefficient (two flights / power station). There is a lack of a spatiotemporal alignment mechanism, making it impossible to locate the hot spot corresponding to the photovoltaic panel. Summary of the Invention

[0007] In response to the problems existing in the existing technology, the present invention provides a dual-modal dynamic detection method and system for photovoltaic panels based on a lightweight edge intelligent box, which has the advantages of realizing collaborative analysis, dynamic switching and safe processing of visible light and infrared thermal imaging, and solves the problems of existing hardware limitations, algorithm defects and system fragmentation.

[0008] The present invention is implemented as follows: a photovoltaic panel dual-modal dynamic detection method and system based on a lightweight edge intelligent box, comprising:

[0009] Raspberry Pi Compute Module 5-core processor;

[0010] Dual-model dynamic architecture;

[0011] Security coordination mechanism.

[0012] As preferred embodiment of the present invention, the Raspberry Pi Compute Module 5 core processor includes a Raspberry Pi Compute Module 5 core board, and the Raspberry Pi Compute Module 5 core board includes an AI acceleration expansion board, a TEE security encryption chip and a multi-protocol drone interface. The AI ​​acceleration expansion board supports TensorRT INT8 quantization, the TEE security encryption chip is used for model weight encryption storage, and the multi-protocol drone interface includes DJI / XAG / Parrot automatic adaptation.

[0013] As a preferred embodiment of the present invention, the dual-model dynamic architecture includes a photovoltaic visible light detection model Binomial-PV1 and an infrared hot spot detection model Binomial-ThermoScannerV1, the photovoltaic visible light detection model Binomial-PV1 includes a PV-LEA module, and the infrared hot spot detection model Binomial-ThermoScannerV1 includes a TGAM module.

[0014] As a preferred embodiment of the present invention, the PV-LEA module includes: a shadow feature extraction convolution layer, a photovoltaic reflection suppression attention mechanism and a feature weighted output unit;

[0015] PV-LEA module:

[0016] Class PV-LEA(nn.Module):

[0017] Def_init_(self):

[0018] Self.shadow_net=nn.sequential(

[0019] nn.Conv2d(256, 128, 3),

[0020] nn.ReLU(),

[0021] SolarAttention()#Photovoltaic reflection suppression mechanism )

[0023] def forward(self, x):

[0024] Return x*self.shadow_net(x) #Shadow feature weighting.

[0025] The dual-mode dynamic architecture also includes a dual-mode dynamic switching engine:

[0026] Workflow:

[0027] Drone → Edge Box: Real-time altitude data;

[0028] Edge box → Decision engine: if height ≤ 4m: start fine mode;

[0029] Decision Engine → Binomial-PV1: Set the resolution to 640×640;

[0030] Decision Engine → Binomial-ThermoScannerV1: Preloaded model;

[0031] Binomial-PV1 → edge box: bird droppings detected (confidence > 0.7);

[0032] Edge box → Drone: Hover command + descend to 3.5m;

[0033] Edge box → Binomial-ThermoScannerV1: Start hot spot scanning;

[0034] Edge box → cloud: Encrypted transmission of associated data.

[0035] As preferred embodiment of the present invention, the security collaboration mechanism includes a data encryption process: sensor data → TEE encryption → edge computing → encryption result → secure transmission → cloud decryption → spatiotemporal correlation matrix;

[0036] The spatiotemporal correlation matrix is ​​as follows:

[0037]

[0038] A photovoltaic panel dual-modal dynamic detection method based on a lightweight edge intelligent box includes the following steps:

[0039] Step S1, dynamically switching the detection resolution based on the UAV's flight altitude;

[0040] Step S2: After the visible light model identifies a specific defect, infrared detection is triggered;

[0041] Step S3, fusing bimodal data through spatiotemporal correlation matrix;

[0042] Step S4: Use TEE to encrypt and transmit the detection results.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The Raspberry Pi Compute Module 5+ expansion board enables real-time detection of 10W power consumption, dual-model inference latency of less than 35ms, meeting real-time requirements, and operates in a wide temperature range of -40°C to 85°C, making it suitable for photovoltaic field environments.

[0045] Dual-model dynamic architecture: Binomial-PV1's PV-LEA module solves the industry's reflection pain point, and Binomial-ThermoScannerV1's TGAM module improves the micro-hot spot recognition rate by 71%.

[0046] Security collaboration mechanism: TEE encryption + space-time correlation matrix realizes end-to-end security diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a diagram of the Raspberry Pi Compute Module 5 edge smart box system. DETAILED DESCRIPTION

[0048] In order to further understand the content, features and effects of the present invention, the following embodiments are given as examples and described in detail with reference to the accompanying drawings.

[0049] The structure of the present invention is described in detail below with reference to the accompanying drawings.

[0050] like Figure 1 As shown, the embodiment of the present invention provides a dual-modal dynamic detection method and system for photovoltaic panels based on a lightweight edge intelligent box, including:

[0051] Raspberry Pi Compute Module 5-core processor;

[0052] Dual-model dynamic architecture;

[0053] Security coordination mechanism.

[0054] The Raspberry Pi Compute Module 5 core processor includes a Raspberry Pi Compute Module 5 core board, which includes an AI acceleration expansion board, a TEE security encryption chip, and a multi-protocol drone interface. The AI ​​acceleration expansion board supports TensorRT INT8 quantization, the TEE security encryption chip is used for encrypted storage of model weights, and the multi-protocol drone interface includes automatic adaptation for DJI / XAG / Parrot.

[0055] The dual-model dynamic architecture includes a photovoltaic visible light detection model Binomial-PV1 and an infrared hot spot detection model Binomial-ThermoScannerV1. The photovoltaic visible light detection model Binomial-PV1 includes a PV-LEA module, and the infrared hot spot detection model Binomial-ThermoScannerV1 includes a TGAM module.

[0056] The PV-LEA module includes: a shadow feature extraction convolution layer, a photovoltaic reflection suppression attention mechanism and a feature weighted output unit;

[0057] PV-LEA module:

[0058] Class PV-LEA(nn.Module):

[0059] Def_init_(self):

[0060] Self.shadow_net=nn.sequential(

[0061] nn.Conv2d(256, 128, 3),

[0062] nn.ReLU(),

[0063] SolarAttention()#Photovoltaic reflection suppression mechanism )

[0065] def forward(self, x):

[0066] Return x*self.shadow_net(x) #Shadow feature weighting.

[0067] The dual-mode dynamic architecture also includes a dual-mode dynamic switching engine:

[0068] Workflow:

[0069] Drone → Edge Box: Real-time altitude data;

[0070] Edge box → Decision engine: if height ≤ 4m: start fine mode;

[0071] Decision Engine → Binomial-PV1: Set the resolution to 640×640;

[0072] Decision Engine → Binomial-ThermoScannerV1: Preloaded model;

[0073] Binomial-PV1 → edge box: bird droppings detected (confidence > 0.7);

[0074] Edge box → Drone: Hover command + descend to 3.5m;

[0075] Edge box → Binomial-ThermoScannerV1: Start hot spot scanning;

[0076] Edge box → cloud: Encrypted transmission of associated data.

[0077] The security collaboration mechanism includes the data encryption process: sensor data → TEE encryption → edge computing → encryption result → secure transmission → cloud decryption → spatiotemporal correlation matrix;

[0078] The spatiotemporal correlation matrix is ​​as follows:

[0079]

[0080] A photovoltaic panel dual-modal dynamic detection method based on a lightweight edge intelligent box includes the following steps:

[0081] Step S1, dynamically switching the detection resolution based on the UAV's flight altitude;

[0082] Step S2: After the visible light model identifies a specific defect, infrared detection is triggered;

[0083] Step S3, fusing bimodal data through spatiotemporal correlation matrix;

[0084] Step S4: Use TEE to encrypt and transmit the detection results.

[0085] The specific implementation is as follows: Taking the detection of 10MW distributed photovoltaic power station as an example

[0086] 1. Hardware deployment

[0087] Edge smart box configuration

[0088] Hardware={

[0089] "core":"Raspberry Pi CM5",

[0090] "accelerator":AI Extensin Board(4TOPS)",

[0091] "security":"TEE Chip(ARM TrustZone)",

[0092] "sensors":["MLX90640","GPS-RTK"]

[0093] }

[0094] 2. Detection task execution

[0095] Step 1: The drone takes off to a height of 6m

[0096] Automatically enters fast mode (PV-Detector V1@416×416)

[0097] Flight speed 3m / s, coverage area 15 acres / hour

[0098] Step 2: Identify component failure

[0099] ```json

[0100] / / PV-Detector V1 output

[0101] {

[0102] "panel_id":"bird-drop",

[0103] "confidence":0.86,

[0104] "position":[1280,720] / / Image coordinates

[0105] }

[0106] Step 3: Triggering Infrared Detection

[0107] Hover and descend to 3.5m

[0108] Start the ThermoScanner V1 scan

[0109] Detected associated hot spot ΔT = 22°C

[0110] Step 4: Secure data transmission

[0111] encrypted_data=tee_encrypt({

[0112] "visble_result":bird_drop_data,

[0113] "location": [28.6812, 115.8563] / / GPS coordinates

[0114] })

[0115] Send_to_cioud(encrypted_data).

[0116] The performance comparison table with traditional solutions is as follows:

[0117]

[0118] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0119] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A dual-modal dynamic detection system for photovoltaic panels based on a lightweight edge intelligent box, characterized by: include: Raspberry Pi Compute Module 5-core processor; Dual-model dynamic architecture; Security coordination mechanism.

2. The photovoltaic panel dual-modal dynamic detection system based on a lightweight edge intelligent box according to claim 1, characterized in that: The Raspberry Pi Compute Module 5 core processor includes a Raspberry Pi Compute Module 5 core board, which includes an AI acceleration expansion board, a TEE security encryption chip, and a multi-protocol drone interface. The AI ​​acceleration expansion board supports TensorRT INT8 quantization, the TEE security encryption chip is used for encrypted storage of model weights, and the multi-protocol drone interface includes automatic adaptation for DJI / XAG / Parrot.

3. The photovoltaic panel dual-modal dynamic detection system based on a lightweight edge intelligent box according to claim 1, characterized in that: The dual-model dynamic architecture includes a photovoltaic visible light detection model Binomial-PV1 and an infrared hot spot detection model Binomial-ThermoScannerV1. The photovoltaic visible light detection model Binomial-PV1 includes a PV-LEA module, and the infrared hot spot detection model Binomial-ThermoScannerV1 includes a TGAM module.

4. The photovoltaic panel dual-modal dynamic detection system based on a lightweight edge intelligent box according to claim 3 is characterized by: The PV-LEA module includes: a shadow feature extraction convolutional layer, a photovoltaic reflection suppression attention mechanism and a feature weighted output unit.

5. The photovoltaic panel dual-modal dynamic detection system based on a lightweight edge intelligent box according to claim 1, characterized in that: The security collaboration mechanism includes the data encryption process: sensor data → TEE encryption → edge computing → encryption result → secure transmission → cloud decryption → spatiotemporal correlation matrix; The spatiotemporal correlation matrix is ​​as follows:

6. The photovoltaic panel dual-modal dynamic detection method based on a lightweight edge intelligent box according to any one of claims 1 to 5, characterized in that: The photovoltaic panel dual-mode dynamic detection method comprises the following steps: Step S1, dynamically switching the detection resolution based on the UAV's flight altitude; Step S2: After the visible light model identifies a specific defect, infrared detection is triggered; Step S3, fusing bimodal data through spatiotemporal correlation matrix; Step S4: Use TEE to encrypt and transmit the detection results.

Images