CN114332039A - Photovoltaic panel dust concentration identification network, system and method - Google Patents
Photovoltaic panel dust concentration identification network, system and method Download PDFInfo
- Publication number
- CN114332039A CN114332039A CN202111660018.0A CN202111660018A CN114332039A CN 114332039 A CN114332039 A CN 114332039A CN 202111660018 A CN202111660018 A CN 202111660018A CN 114332039 A CN114332039 A CN 114332039A
- Authority
- CN
- China
- Prior art keywords
- photovoltaic panel
- image
- dust
- module
- identification network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/50—Photovoltaic [PV] energy
Abstract
The invention provides a photovoltaic panel dust concentration identification network, a system and a method, wherein the photovoltaic panel dust concentration identification network is based on a U-Net3+ network architecture, and a multi-scale enhancement module and a light-weight attention mechanism module are added on the basis of a main body architecture. The multi-scale enhancement module can effectively perform denoising processing on an image, the light-weight attention mechanism module adopts CBAM, and comprises two independent sub-modules, a channel attention module and a space attention module which respectively perform attention mechanisms on a channel and a space. The method can identify the dust concentration of the photovoltaic panel according to the picture data transmitted on site, can provide a favorable basis for timely cleaning the photovoltaic panel, can effectively help a power grid dispatching plan to provide important reference, and has an important significance for the full grid-connected operation of optical volt-ampere.
Description
Technical Field
The invention relates to the field of solar photovoltaic power generation, in particular to a network, a system and a method for identifying the dust concentration of a photovoltaic panel.
Background
In order to solve the serious problems of global energy shortage, climate warming and the like, renewable energy has become the key point of research of countries in the world, wherein the development and utilization of solar energy also become the focus of attention of people. Solar photovoltaic power generation is taken as a power generation technology with development prospect in the field of new energy, and new vitality is released under the background of transformation and upgrading of global energy structures by virtue of the industrial characteristics of wide distribution, rich reserves and mature technology. However, the tiny particles in the air are easy to accumulate on the solar photovoltaic panel to form dust, and the low transmittance of the dust seriously reduces the power generation performance of the photovoltaic panel, thereby affecting the economy of photovoltaic power generation enterprises. The accumulated dust concentration is generally qualitatively analyzed on site according to experience of operation and maintenance personnel and a manual observation method, but the accuracy and the real-time performance are poor, and the development requirements of intelligent operation and maintenance such as real-time monitoring, cleaning, maintenance and optimization, efficient and safe grid connection and the like of the accumulated dust concentration of the photovoltaic panel are difficult to meet. Therefore, the sensing capability of improving the dust concentration of the photovoltaic panel has important significance for improving the operation reliability of the power system.
Disclosure of Invention
The invention provides a photovoltaic panel dust concentration identification network, a system and a method, and aims to solve the technical problems of insufficient timeliness and accuracy of photovoltaic panel dust concentration identification in the prior art.
Therefore, the photovoltaic panel ash deposition concentration identification network provided by the invention is based on a U-Net3+ network architecture, and a multi-scale enhancement module and a light-weight attention mechanism module are added on the basis of a main body architecture.
Further, the multi-scale enhancement module acts as a decoder of the network, performs a refinement process based on the previously estimated image by an enhancement algorithm and restores the photovoltaic panel clean image step by step, improves the signal-to-noise ratio and can refine the intermediate results of the previous iteration step by step.
Further, the lightweight attention mechanism module includes a channel attention module and a spatial attention module.
Further, the specific processing flow of the channel attention module is as follows: and finally, multiplying the final channel attention characteristic and the input characteristic graph to generate the required input characteristic.
Further, the specific processing flow of the spatial attention module is as follows: taking the feature graph output by the channel attention module as an input feature graph, firstly performing global maximum pooling and global average pooling based on a channel to obtain two feature graphs, then performing channel splicing on the two feature graphs, performing 7 multiplied by 7 convolution operation to reduce the dimension into 1 channel, performing activation operation to generate a space attention feature, and finally performing multiplication on the space attention feature and the input feature graph of the module to obtain a final feature.
Further, the photovoltaic panel soot concentration identification network relies on a photovoltaic panel soot concentration recovery model, which can be expressed as:
I(x)＝J(x)t(x)+A(1-t(x))
wherein I (x) is a dust image, J (x) is a photovoltaic panel cleaning image, A is an atmospheric light value, t (x) is a transmission matrix, and t (x) can be defined as t (x) e-βd(x)Where β is the atmospheric scattering coefficient and d (x) is the distance between the object and the camera.
The photovoltaic panel dust concentration identification system provided by the invention specifically comprises an image monitoring device and a software processing system, wherein the image monitoring device is used for collecting a dust image of the photovoltaic panel, and the software processing system can realize the photovoltaic panel dust concentration identification network.
Further, the software processing system is developed using a pytorech.
The photovoltaic panel dust concentration identification method based on the photovoltaic panel dust concentration identification network provided by the invention specifically comprises the following steps:
s1, subjecting the collected photovoltaic panel dust image to perspective transformation to obtain a corrected photovoltaic panel image;
s2, removing silver grid information to obtain a photovoltaic panel surface image without grid lines;
s3, inputting the photovoltaic panel dust deposition image into a photovoltaic panel dust deposition concentration identification network, and outputting a dust deposition result by the photovoltaic panel dust deposition concentration identification network;
and S4, when the dust accumulation result is larger than the threshold value, giving an alarm to remind a worker to clean.
The computer-readable storage medium provided by the invention stores a computer program capable of being executed by a processor, and the processor can realize the method for identifying the dust concentration of the photovoltaic panel by executing the computer program.
Compared with the prior art, the invention has the following beneficial effects:
according to the invention, the accurate identification of the dust concentration of the photovoltaic panel can be realized through the photovoltaic panel dust image monitored on the photovoltaic site, and the safety and the economy of a photovoltaic system are favorably improved.
Drawings
FIG. 1 is a structural diagram of a grid concentration identification network of a photovoltaic panel according to an embodiment of the present invention;
FIG. 2 is a block diagram of a multi-scale enhancement module according to an embodiment of the present invention;
FIG. 3 is a block diagram of a CBAM attention mechanism module according to an embodiment of the present invention;
FIG. 4 is a flow chart illustrating a process of a channel attention module according to an embodiment of the invention;
FIG. 5 is a flowchart illustrating a process of the spatial attention module according to an embodiment of the invention.
Detailed Description
In order to more clearly understand the technical features, objects, and effects of the present invention, embodiments of the present invention will now be described with reference to the accompanying drawings.
As shown in fig. 1, the photovoltaic panel dust deposition concentration identification network is based on a U-Net3+ network architecture, and a multi-scale enhancement module and a light-weight attention mechanism module are added on the basis of a main body architecture, so that the denoising processing of the photovoltaic panel dust deposition image is enhanced, the feature extraction capability of a channel and a space is improved, important features such as color and texture of the photovoltaic panel dust deposition image are effectively extracted, the classification performance and the detection performance of the proposed network are improved, and the dust deposition concentration of the on-site photovoltaic panel is effectively identified in real time. The encoder of the photovoltaic panel dust concentration identification network has 5 layers, a single photovoltaic panel dust image is used as input, and a corresponding decoder also has 5 layers.
The U-Net3+ network architecture utilizes full-scale skip connection and depth supervision to combine high-level semantics with low-level semantics in different-scale feature maps extracted from photovoltaic panel gray images and learn hierarchical representations of gray images from multi-scale aggregated feature maps. The U-Net3+ network architecture utilizes visual transformers from the backbone encoder to the task decoder. The connection relationship of the U-Net3+ network architecture can be expressed by the following formula:
to be provided with
By way of example, by
And
for inputting, the merged features are enhanced by a multi-scale enhancement module, and then a feature map is reserved by an attention mechanism moduleThe space and channel feature extraction capability.
The multi-scale enhancement module can effectively perform denoising processing on an image, an SOS (Signal-to-Noise Ratio) enhancement algorithm performs thinning processing on the basis of a previously estimated image and gradually recovers a photovoltaic panel clean image, improves a Signal-to-Noise Ratio and can gradually thin an intermediate result of previous iteration, the image denoising processing method has been proved to be effective for image denoising, and the Signal-to-Noise Ratio (SNR) has been proved to be improved, and the better result can be obtained on the basis of the denoising method in the aspect of SNR on the image of the same scene with less Noise points. For identifying the dust concentration of the photovoltaic panel, the calculation mode of the SOS enhancement algorithm is as follows:
Jn+1＝g(I+Jn)-Jn (2)
in the photovoltaic panel dust concentration identification network, a decoder is regarded as a photovoltaic panel clean image recovery module, in order to gradually perfect features from the feature recovery module, an SOS enhancement algorithm is introduced into the decoder of the proposed network, as shown in FIG. 2, the structure of a multi-scale enhancement module is formulated as follows:
the lightweight attention mechanism module sequentially deduces attention weight along two dimensions of a space and a channel by acquiring a U-Net3+ framework middle feature map, then multiplies the feature with an original feature map to perform self-adaptive adjustment on the feature to improve the feature extraction capability of the channel and the space, retains complete channel information and space information and reasonably utilizes non-adjacent features in the feature map, so that the classification performance and the detection performance of the network are improved. The lightweight attention mechanism module employs CBAM. There are many limitations in view of the U-Net3+ network architecture, such as: spatial information is lacking during down-sampling of the encoder and features between non-adjacent layers lack sufficient connectivity. To correct missing spatial information in higher level features and fully exploit features of non-adjacent levels, a straightforward approach is to first resample all features to the same scale and then fuse them together with bottleneck (connection and convolution layers) as nodes in the DenseNet. However, simply using concatenation is less effective for feature fusion because features from different levels have different proportions and sizes, and therefore CBAM attention mechanism modules are added. As can be seen from fig. 3, the CBAM includes 2 independent sub-modules, namely a Channel Attention Module (CAM) and a Spatial Attention Module (SAM), for performing Channel and spatial Attention, respectively. This not only saves parameters and computing power, but also ensures that it can be integrated into existing network architectures as a plug-and-play module.
As shown in fig. 4, the specific processing flow of the channel attention module is as follows: the input feature graph F (H multiplied by W multiplied by C) is respectively subjected to global max pooling and global average pooling based on width and height to obtain two 1 multiplied by C feature graphs, and then the two feature graphs are respectively sent into a two-layer neural network (MLP), the number of neurons in a first layer is C/r (r is a reduction rate), an activation function is Relu, the number of neurons in a second layer is C, and the two-layer neural network is shared. And then, carrying out addition operation based on element-wise on the characteristics of the MLP output, and then carrying out sigmoid activation operation to generate a final channel attribute feature, namely M _ c. And finally, performing element-wise multiplication operation on the M _ c and the input feature diagram F to generate the input features required by the Spatial attribute module.
As shown in fig. 5, the specific processing flow of the spatial attention module is as follows: and taking the feature map output by the Channel attribute module as an input feature map of the module. Firstly, making a channel-based global max and global average potential to obtain two H multiplied by W multiplied by 1 feature maps, and then making concat operation (channel splicing) on the basis of the channel for the 2 feature maps. Then, after a 7 × 7 convolution operation (7 × 7 is better than 3 × 3), the dimensionality reduction is 1 channel, i.e., hxwx 1. And generating a spatial attribute feature by the sigmoid. And finally, multiplying the feature by the input feature of the module to obtain the finally generated feature.
The photovoltaic panel dust concentration identification network depends on a photovoltaic panel dust concentration recovery model. The atmospheric scattering model has been a classical method for describing the generation of a haze image, and since the haze image is similar to the image of a photovoltaic panel, the model can be expressed as:
I(x)＝J(x)t(x)+A(1-t(x)) (4)
wherein I (x) is a dust image, and J (x) is a photovoltaic panel cleaning image. A is the atmospheric light value. t (x) is a transmission matrix, which can be defined as
t(x)＝e-βd(x) (5)
Where β is the atmospheric scattering coefficient and d (x) is the distance between the object and the camera.
To reduce the reconstruction error of the model and the eligible dust accumulation, the reconstruction error of q (x) needs to be minimized. Therefore, the photovoltaic panel soot concentration recovery model can be written as:
J(x)＝Q(x)I(x)-Q(x)+c (6)
wherein the content of the first and second substances,
where t (x) and A incorporate Q (x), c defaults to a constant offset of 1. Therefore, our goal is to build an input adaptive depth model with parameters that vary with the input image.
P(J)＝(1-Q)A/J (8)
Wherein P (J) is the deposition concentration of the photovoltaic panel, and the deposition concentration of the photovoltaic panel is determined by utilizing the inverse relation between the deposition image of the photovoltaic panel and Q.
Based on the photovoltaic panel dust concentration identification network, the invention provides a photovoltaic panel dust concentration identification system which specifically comprises an image monitoring device and a software processing system. The image monitoring device collects the dust deposition picture of the photovoltaic panel, the selection of the image monitoring points follows the law of comprehensive coverage, reasonable resource utilization and cost benefit consideration, and the picture covers the panel surface of the whole photovoltaic panel. The software processing system is developed by using a Pythrch, can realize a photovoltaic panel dust concentration identification network, collects the dust images of the photovoltaic panel at the time interval of 24 hours, processes the images, identifies the dust concentration through the photovoltaic panel dust concentration identification network, stores data and provides a historical data query function.
Based on the photovoltaic panel dust concentration identification network, the invention provides a photovoltaic panel dust concentration identification method, which specifically comprises the following steps:
s1, subjecting the collected photovoltaic panel dust image to perspective transformation to obtain a corrected photovoltaic panel image;
s2, removing silver grid information by adopting an FMM method to obtain a photovoltaic panel surface image without grid lines;
s3, inputting the photovoltaic panel dust deposition image into a photovoltaic panel dust deposition concentration identification network, and outputting a dust deposition result by the photovoltaic panel dust deposition concentration identification network;
s4, setting a threshold value of the dust concentration of the photovoltaic panel, and giving an alarm to remind workers to clean when the detection result is larger than the threshold value.
According to the photovoltaic panel dust deposition concentration identification network provided by the invention, the relationship between the photovoltaic panel dust deposition concentration recovery model and the dust deposition is utilized, and the recovery model is utilized to directly identify the dust deposition concentration of the photovoltaic panel from the photovoltaic panel dust deposition image. The method has the advantages that the encoder of the network has 5 layers, a single photovoltaic panel dust deposition image is used as input, the clean image of the photovoltaic panel is recovered by using the 5 layers of the decoder, the dust deposition concentration of the photovoltaic panel is estimated by using the intermediate weight parameter from the dust deposition image to the clean image mapping, the method has the advantages of high accuracy, high calculating speed and the like, the dust deposition concentration of the photovoltaic panel can be monitored in real time, and reliable data are provided for an operation and maintenance center.
The above disclosure is only for the purpose of illustrating the preferred embodiments of the present invention, and it should not be understood that the scope of the present invention is limited thereby. It should be noted that those skilled in the art should recognize that they may make equivalent variations to the embodiments of the present invention without departing from the spirit and scope of the present invention.
Claims (10)
1. The photovoltaic panel ash deposition concentration identification network is characterized in that the photovoltaic panel ash deposition concentration identification network is based on a U-Net3+ network architecture, and a multi-scale enhancement module and a light-weight attention mechanism module are added on the basis of a main body architecture.
2. The photovoltaic panel ash concentration identification network of claim 1, wherein the multi-scale enhancement module acts as a decoder of the network, performs a refinement process based on a previously estimated image through an enhancement algorithm and gradually restores the photovoltaic panel clean image, improves the signal-to-noise ratio and can gradually refine the intermediate results of the previous iteration.
3. The photovoltaic panel ash concentration identification network of claim 1, wherein the lightweight attention mechanism module comprises a channel attention module and a spatial attention module.
4. The photovoltaic panel ash deposition concentration identification network according to claim 3, wherein the specific processing flow of the channel attention module is as follows: and finally, multiplying the final channel attention characteristic and the input characteristic graph to generate the required input characteristic.
5. The photovoltaic panel ash deposition concentration identification network according to claim 3, wherein the specific processing flow of the spatial attention module is as follows: taking the feature graph output by the channel attention module as an input feature graph, firstly performing global maximum pooling and global average pooling based on a channel to obtain two feature graphs, then performing channel splicing on the two feature graphs, performing 7 multiplied by 7 convolution operation to reduce the dimension into 1 channel, performing activation operation to generate a space attention feature, and finally performing multiplication on the space attention feature and the input feature graph of the module to obtain a final feature.
6. The photovoltaic panel ash deposition concentration recognition network according to any one of claims 1 to 5, wherein the photovoltaic panel ash deposition concentration recognition network relies on a photovoltaic panel ash deposition concentration recovery model, which can be expressed as:
I(x)＝J(x)t(x)+A(1-t(x))
wherein I (x) is a dust image, J (x) is a photovoltaic panel cleaning image, A is an atmospheric light value, t (x) is a transmission matrix, and t (x) can be defined as t (x) e-βd(x)Where β is the atmospheric scattering coefficient and d (x) is the distance between the object and the camera.
7. The photovoltaic panel dust concentration identification system is characterized by specifically comprising an image monitoring device and a software processing system, wherein the image monitoring device is used for collecting a dust image of a photovoltaic panel, and the software processing system can be used for realizing the photovoltaic panel dust concentration identification network of any one of claims 1 to 6.
8. The photovoltaic panel ash deposition concentration identification network of claim 7 wherein the software processing system is developed using a Pythrch.
9. The method for identifying the dust concentration of the photovoltaic panel based on the network for identifying the dust concentration of the photovoltaic panel as claimed in any one of claims 1 to 6 is characterized by comprising the following steps:
s1, subjecting the collected photovoltaic panel dust image to perspective transformation to obtain a corrected photovoltaic panel image;
s2, removing silver grid information to obtain a photovoltaic panel surface image without grid lines;
s3, inputting the photovoltaic panel dust deposition image into a photovoltaic panel dust deposition concentration identification network, and outputting a dust deposition result by the photovoltaic panel dust deposition concentration identification network;
and S4, when the dust accumulation result is larger than the threshold value, giving an alarm to remind a worker to clean.
10. A computer-readable storage medium, characterized in that a computer program executable by a processor is stored, and the processor is capable of implementing the method for identifying the deposition concentration of a photovoltaic panel according to claim 9 by executing the computer program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111660018.0A CN114332039A (en) | 2021-12-30 | 2021-12-30 | Photovoltaic panel dust concentration identification network, system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111660018.0A CN114332039A (en) | 2021-12-30 | 2021-12-30 | Photovoltaic panel dust concentration identification network, system and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114332039A true CN114332039A (en) | 2022-04-12 |
Family
ID=81018892
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111660018.0A Pending CN114332039A (en) | 2021-12-30 | 2021-12-30 | Photovoltaic panel dust concentration identification network, system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114332039A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115937567A (en) * | 2022-09-07 | 2023-04-07 | 北京交通大学 | Image classification method based on wavelet scattering network and ViT |
-
2021
- 2021-12-30 CN CN202111660018.0A patent/CN114332039A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115937567A (en) * | 2022-09-07 | 2023-04-07 | 北京交通大学 | Image classification method based on wavelet scattering network and ViT |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111563508B (en) | Semantic segmentation method based on spatial information fusion | |
| Si et al. | Hybrid solar forecasting method using satellite visible images and modified convolutional neural networks | |
| Liu et al. | Spatial air quality index prediction model based on decomposition, adaptive boosting, and three-stage feature selection: A case study in China | |
| CN112871938B (en) | Photovoltaic power plant cleans control integration management system | |
| CN111914815A (en) | Machine vision intelligent recognition system and method for garbage target | |
| CN112365468A (en) | AA-gate-Unet-based offshore wind power tower coating defect detection method | |
| CN112163447B (en) | Multi-task real-time gesture detection and recognition method based on Attention and Squeezenet | |
| CN114332039A (en) | Photovoltaic panel dust concentration identification network, system and method | |
| CN113496310A (en) | Atmospheric pollutant prediction method and system based on deep learning model | |
| CN115294550A (en) | Automatic driving automobile road scene understanding method based on multi-task learning | |
| CN113870160A (en) | Point cloud data processing method based on converter neural network | |
| CN114358246A (en) | Graph convolution neural network module of attention mechanism of three-dimensional point cloud scene | |
| CN111488974B (en) | Ocean wind energy downscaling method based on deep learning neural network | |
| CN116128168A (en) | Weather prediction method based on causal expansion convolution and Autoformer | |
| CN115984263A (en) | Bolt looseness detection algorithm and detection system based on improved twin neural network | |
| CN115880691A (en) | Roof photovoltaic potential estimation method based on computer vision | |
| CN113421222B (en) | Lightweight coal gangue target detection method | |
| CN115146842A (en) | Multivariate time series trend prediction method and system based on deep learning | |
| Cao et al. | IDS-Net: Integrated Network for Identifying Dust State of Photovoltaic Panels | |
| Ullah et al. | Real-time video dehazing for industrial image processing | |
| Zhang | Deep learning-based hybrid short-term solar forecast using sky images and meteorological data | |
| Cong et al. | Object Detection and Image Segmentation for Autonomous Vehicles | |
| CN116680624B (en) | Classification method, system and storage medium for metadata of power system | |
| CN117878910A (en) | Wind power prediction method based on edge feature reconstruction and federal learning | |
| Yuan et al. | Ultra-short-term Photovoltaic Power Prediction Based on STCN Model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |