Photovoltaic inverter fault rapid detection method based on infrared sensor and saliency feature extraction

Abstract

The invention provides a photovoltaic inverter fault rapid detection method based on an infrared sensor and saliency feature extraction. The method comprises the following steps: firstly, acquiring an infrared image of an inverter in a photovoltaic power station by using the infrared sensor; smooth denoising processing of the image is completed by using Gaussian filtering; and finally, performing region detection and segmentation on the fault in the image by using a saliency feature extraction algorithm to obtain a target region. According to the photovoltaic inverter rapid detection method based on the infrared sensor and saliency feature extraction, fault identification of the photovoltaic inverter in a photovoltaic power station can be rapidly completed by using a visual saliency principle, so that fault work of the photovoltaic inverter in the photovoltaic power station is more efficient and reliable.

Description

Technical Field

[0001] The invention relates to the technical field of primary equipment fault detection in photovoltaic power stations, and in particular to a photovoltaic inverter fault rapid detection method based on infrared sensors and significant feature extraction. Background Art

[0002] In response to the crisis of traditional energy, my country has made great efforts to support and build new energy. Among them, photovoltaic power generation has become the largest sector with its low cost and convenient installation, and has been rapidly developed. However, since this type of device needs to make the most of sunlight, photovoltaic power stations are usually built in vast areas such as mountains or deserts. Even for small-scale installations, there are generally areas such as water surfaces and tops of buildings. At the same time, this geographical environment brings great difficulties to the safety detection of primary equipment in photovoltaic power stations, especially photovoltaic inverter fault detection. Since photovoltaic inverter faults are relatively hidden, it is difficult to complete this task by manual detection alone.

[0003] In order to solve this problem, the traditional method of inverter fault detection in photovoltaic modules is to use manual patrols on site to complete inspections when the fault is too large and causes a system alarm. This method has a long cycle and a large lag. Later, with the rapid development of information technology, image processing and deep learning artificial intelligence algorithms were introduced to this field. However, due to the hidden nature of this type of fault, ordinary camera technology cannot directly capture the characteristics of this type of fault, resulting in a decrease in the accuracy of this method. In addition, this type of algorithm has another obvious disadvantage that it requires upgrading existing hardware, increasing additional investment, and the real-time performance is not very high. Summary of the invention

[0004] In order to meet the requirements of real-time accuracy without increasing the cost budget, the present invention proposes an algorithm for extracting significant features based on infrared sensors to complete inverter fault detection in photovoltaic modules. This method requires very little computational cost, and the infrared sensor can accurately capture the hot spot features of such faults. At the same time, it has the rapidity of the significant feature extraction algorithm in the visual field, and can thus efficiently and accurately complete the rapid detection of inverter faults in photovoltaic power stations.

[0005] In order to realize the above technical features, the object of the present invention is achieved as follows: a photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction, comprising the following steps: Step 1: Use an infrared sensor to obtain an infrared image of the inverter in the photovoltaic power station; Step 2: Use Gaussian filtering to complete the infrared image smoothing and denoising process; Step 3: Use the histogram equalization algorithm to perform image enhancement processing on the infrared image; Step 4: Use the salient feature extraction algorithm to detect the fault area in the image, and then segment it to obtain the target fault area.

[0006] Preferably, in step 1, the infrared image of the inverter in the photovoltaic power station is obtained by using an infrared sensor, wherein the infrared sensor is an infrared camera, and the process is to fix the infrared camera around the inverter in the photovoltaic power station to collect infrared images of the inverter in real time.

[0007] Preferably, in step 2, Gaussian filtering is used to complete the infrared image smoothing and denoising processing, specifically, a second-order Gaussian filtering is used to perform the infrared image smoothing and denoising processing, and the processing process is as follows: Step 2.1, convert the colors in the image Model is brightness model, L is the brightness of the brightness model, a and b are the two color channels of the brightness model; Step 2.2, set the two-dimensional Gaussian function, the formula is as follows: ; In the formula, is a two-dimensional Gaussian function; are two-dimensional Gaussian variables respectively; is the standard deviation in Gaussian filtering, which is a constant; Step 2.3, Schema Definition The coordinates in the image are A brightness model at a certain point The attribute is set to Gaussian filter template. , the matrix generated by the high-speed function is convolved with the matrices of the three attributes to obtain new attribute values, update the original image, and complete the infrared image smoothing and denoising processing.

[0008] Preferably, step 3 uses a histogram equalization algorithm to perform image enhancement processing on the infrared image, and the process is as follows: Step 3.1: For the brightness of the infrared image that has been denoised in step 2 The three model attributes of brightness L, color channels a and b in the model are respectively generated by cumulative distribution functions for histogram equalization, where the cumulative distribution function formula is as follows: ; In the formula, It is a value at a certain level in the attribute. is the probability of this level value appearing in the entire image, is the number of occurrences of this level value in the entire image, is the total number of pixels in the entire image, is the maximum value of the attribute; Step 3.2, then the equalization of the original image is completed through the transformation function, and its formula is as follows: ; In the formula, yes It is the new value of a certain brightness value in the attribute.

[0009] Preferably, step 4 uses a significant feature extraction algorithm to detect the fault area in the image, and then segment to obtain the target fault area. The processing process of this step is as follows: Step 4.1: extract the infrared image that has been denoised and enhanced in step 3, and adjust the brightness of the image. The brightness attribute L in the model is subjected to Harris node solving operation to obtain the fault significance seed node of the image; Step 4.2, perform convex hull operation on the Harris nodes obtained in step 4.1, further expand the area of ​​the fault significance seed node, and complete the characteristic information of the target area; Step 4.3, propagate and diffuse the significant feature information of the seed node area in step 4.2, obtain the complete photovoltaic inverter fault area, and complete the rapid detection of photovoltaic inverter faults in photovoltaic power stations.

[0010] Preferably, the step 4.2 specifically includes: Step 4.2.1, first calculate and define the convex hull area containing the Harris nodes. At this time, it is assumed that there is nodes in the convex hull area and The brightness of the two areas The model attribute vectors are and Two sets are conducted; Step 4.2.2, then use Represents an image node To the brightness of other nodes in the image object The distance of the model attribute vector is solved as follows: ; In the formula, for arrive The Euclidean gap; Step 4.2.3, finally, the critical value of the fault saliency seed node of the image is calculated by the distance: ; In the formula, is the critical value; According to the critical value, the area defined by the nodes in the defined convex hull is Nodes that meet the conditions are selected as fault saliency diffusion seed nodes.

[0011] Preferably, the step 4.3 specifically includes: Step 4.3.1, first set the fault significance diffusion seed node feature vector of the photovoltaic power station photovoltaic inverter: ; In the formula, is the above seed node, that is, the total number of pixels; Depend on Derived a feature undirected graph , using the adjacency matrix To verify the undirected graph ,use Used to represent undirected graphs Midpoint Distance and The weight of is the adjacency matrix An element in , so the maximum weight Means node and Spread by similar salient features; Step 4.3.2, then use the local feature information to reconstruct the HGF algorithm to propagate the information of the significant features to diffuse the final fault area, the process is as follows: The formula for propagating significant features using the HGF algorithm is as follows: ,satisfy , ; In the formula, is the saliency diffusion seed node described in Step 4.2.3, , and for and exist The node element of the position, In the original image, The calculated Laplace operator is W is the adjacency matrix of the current image, is a diagonal matrix whose diagonal elements are ; Step 4.3.3, finally complete the rapid fault detection based on the propagation results.

[0012] Beneficial effects of the present invention: The photovoltaic inverter rapid detection method based on infrared sensors and significant feature extraction of the present invention utilizes the principle of visual significance and can quickly complete the photovoltaic inverter fault identification in the photovoltaic power station, making the photovoltaic inverter fault work in the photovoltaic power station more efficient and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0014] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION

[0015] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0016] Embodiment 1: See also Figure 1 A photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction comprises the following steps: Step 1: Use an infrared sensor to obtain an infrared image of the inverter in the photovoltaic power station; The infrared sensor is an infrared camera, and the process is to fix the infrared camera around the inverter in the photovoltaic power station to collect infrared images of the inverter in real time.

[0017] Step 2: Use Gaussian filtering to complete the infrared image smoothing and denoising process; Specifically, a second-order Gaussian filter is used to smooth and denoise the infrared image. The processing process is as follows: Step 2.1, convert the colors in the image Model is brightness model, L is the brightness of the brightness model, a and b are the two color channels of the brightness model; Step 2.2, set the two-dimensional Gaussian function, the formula is as follows: ; In the formula, is a two-dimensional Gaussian function; are two-dimensional Gaussian variables respectively; is the standard deviation in Gaussian filtering, which is a constant value. .

[0018] Step 2.3, Schema Definition The coordinates in the image are A brightness model at a certain point The attribute is set to Gaussian filter template. , the matrix generated by the high-speed function is convolved with the matrices of the three attributes to obtain new attribute values, update the original image, and complete the infrared image smoothing and denoising processing.

[0019] Step 3: Use the histogram equalization algorithm to perform image enhancement processing on the infrared image; Step 3.1: For the brightness of the infrared image that has been denoised in step 2 The three model attributes of brightness L, color channels a and b in the model are respectively generated by cumulative distribution functions for histogram equalization, where the cumulative distribution function formula is as follows: ; In the formula, It is a value at a certain level in the attribute. is the probability of this level value appearing in the entire image, is the number of occurrences of this level value in the entire image, is the total number of pixels in the entire image, is the maximum value of the attribute; Step 3.2, then the equalization of the original image is completed through the transformation function, and its formula is as follows: ; In the formula, yes It is the new value of a certain brightness value in the attribute.

[0020] Step 4: Use the salient feature extraction algorithm to detect the fault area in the image, and then segment the target fault area; Step 4.1: extract the infrared image that has been denoised and enhanced in step 3, and adjust the brightness of the image. The brightness attribute L in the model is subjected to Harris node solution operation to obtain the fault significance seed node of the image; Step 4.2, perform convex hull operation on the Harris nodes obtained in step 4.1, further expand the region of the fault significance seed node, and complete the feature information of the target region; the process is as follows: Step 4.2.1, first calculate and define the convex hull area containing the Harris node. At this time, it is assumed that there is nodes in the convex hull area and The brightness of the two areas The model attribute vectors are and Two sets are conducted; Step 4.2.2, then use Represents an image node To the brightness of other nodes in the image object The distance of the model attribute vector is solved as follows: ; In the formula, for arrive The Euclidean gap; Step 4.2.3, finally, the critical value of the fault saliency seed node of the image is calculated by the distance: ; In the formula, is the critical value; According to the critical value, the area defined by the nodes in the defined convex hull is Nodes that meet the conditions are selected as fault saliency diffusion seed nodes.

[0021] Step 4.3, propagate and diffuse the significant feature information of the seed node area in step 4.2, obtain the complete photovoltaic inverter fault area, and complete the rapid detection of photovoltaic inverter faults in photovoltaic power stations. The process is as follows: Step 4.3.1, first set the fault significance diffusion seed node feature vector of the photovoltaic power station photovoltaic inverter: ; In the formula, is the above seed node, that is, the total number of pixels; Depend on Derived a feature undirected graph , using the adjacency matrix To verify the undirected graph ,use Used to represent undirected graphs Midpoint Distance and The weight of is the adjacency matrix An element in , so the maximum weight Means node and Spread by similar salient features; Step 4.3.2, then use the local feature information to reconstruct the HGF algorithm to propagate the information of the significant features to diffuse the final fault area, the process is as follows: The formula for propagating significant features using the HGF algorithm is as follows: ,satisfy , ; In the formula, is the saliency diffusion seed node described in Step 4.2.3, , and for and exist The node element of the position, In the original image, The calculated Laplace operator is W is the adjacency matrix of the current image, is a diagonal matrix whose diagonal elements are ; Step 4.3.3, finally complete the rapid fault detection based on the propagation results.

[0022] Embodiment 2: In the implementation process, infrared images collected from a photovoltaic power station in September 2020 were collected as the test data source. Through the above four steps, the results show that the fault detection success rate of this method is above 95%.

[0023] The above description is only a specific example of the present invention and does not constitute any limitation to the present invention. It is obvious that for professionals in this field, after understanding the content and principle of the present invention, various modifications and changes in form and details may be made without departing from the principle and structure of the present invention, but these modifications and changes based on the idea of ​​the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction, characterized in that: The following steps are involved: Step 1: Use an infrared sensor to obtain an infrared image of the inverter in the photovoltaic power station; Step 2: Use Gaussian filtering to complete the infrared image smoothing and denoising process; Step 3: Use the histogram equalization algorithm to perform image enhancement processing on the infrared image; Step 4: Use the salient feature extraction algorithm to detect the fault area in the image, and then segment it to obtain the target fault area.

2. A photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction according to claim 1, characterized in that: In step 1, an infrared image of the inverter in the photovoltaic power station is obtained by using an infrared sensor, wherein the infrared sensor is an infrared camera, and the process is to fix the infrared camera around the inverter in the photovoltaic power station to collect infrared images of the inverter in real time.

3. A photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction according to claim 2, characterized in that: In step 2, Gaussian filtering is used to complete the infrared image smoothing and denoising process. Specifically, second-order Gaussian filtering is used to perform infrared image smoothing and denoising. The processing process is as follows: Step 2.1, convert the colors in the image Model is brightness model, L is the brightness of the brightness model, a and b are the two color channels of the brightness model; Step 2.2, set the two-dimensional Gaussian function, the formula is as follows: ； In the formula, is a two-dimensional Gaussian function; are two-dimensional Gaussian variables respectively; is the standard deviation in Gaussian filtering, which is a constant; Step 2.3, Schema Definition The coordinates in the image are A brightness model at a certain point The attribute is set to Gaussian filter template. , the matrix generated by the high-speed function is convolved with the matrices of the three attributes to obtain new attribute values, update the original image, and complete the infrared image smoothing and denoising processing.

4. A photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction according to claim 3, characterized in that: Step 3 uses the histogram equalization algorithm to perform image enhancement processing on the infrared image. The process is as follows: Step 3.1: For the brightness of the infrared image that has been denoised in step 2 The three model attributes of brightness L, color channels a and b in the model are respectively generated by cumulative distribution functions for histogram equalization, where the cumulative distribution function formula is as follows: ； In the formula, It is a value at a certain level in the attribute. is the probability of this level value appearing in the entire image, is the number of occurrences of this level value in the entire image, is the total number of pixels in the entire image, is the maximum value of the attribute; Step 3.2, then the equalization of the original image is completed through the transformation function, and its formula is as follows: ； In the formula, yes It is the new value of a certain brightness value in the attribute.

5. A photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction according to claim 4, characterized in that: Step 4 uses a significant feature extraction algorithm to detect the fault area in the image, and then segment the target fault area. The processing process of this step is as follows: Step 4.1: extract the infrared image that has been denoised and enhanced in step 3, and adjust the brightness of the image. The brightness attribute L in the model is subjected to Harris node solving operation to obtain the fault significance seed node of the image; Step 4.2, perform convex hull operation on the Harris nodes obtained in step 4.1, further expand the area of ​​the fault significance seed node, and complete the characteristic information of the target area; Step 4.3, propagate and diffuse the significant feature information of the seed node area in step 4.2, obtain the complete photovoltaic inverter fault area, and complete the rapid detection of photovoltaic inverter faults in photovoltaic power stations.

6. A photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction according to claim 5, characterized in that: The step 4.2 specifically includes: Step 4.2.1, first calculate and define the convex hull area containing the Harris nodes. At this time, it is assumed that there is nodes in the convex hull area and The brightness of the two areas The model attribute vectors are and Two sets are conducted; Step 4.2.2, then use Represents an image node To the brightness of other nodes in the image object The distance of the model attribute vector is solved as follows: ； In the formula, for arrive The Euclidean gap; Step 4.2.3, finally, the critical value of the fault saliency seed node of the image is calculated by the distance: ； In the formula, is the critical value; According to the critical value, the area defined by the nodes in the defined convex hull is Nodes that meet the conditions are selected as fault saliency diffusion seed nodes.

7. A photovoltaic inverter fault rapid detection method based on infrared sensor and significant feature extraction according to claim 6, characterized in that: The step 4.3 specifically includes: Step 4.3.1, first set the fault significance diffusion seed node feature vector of the photovoltaic power station photovoltaic inverter: ； In the formula, is the above seed node, that is, the total number of pixels; Depend on Derived a feature undirected graph , using the adjacency matrix To verify the undirected graph ,use Used to represent undirected graphs Midpoint Distance and The weight of is the adjacency matrix An element in , so the maximum weight Means node and Spread by similar salient features; Step 4.3.2, then use the local feature information to reconstruct the HGF algorithm to propagate the information of the significant features to diffuse the final fault area, the process is as follows: The formula for propagating significant features using the HGF algorithm is as follows: ,satisfy , ; In the formula, is the saliency diffusion seed node described in Step 4.2.3, , and for and exist The node element of the position, In the original image, The calculated Laplace operator is W is the adjacency matrix of the current image, is a diagonal matrix whose diagonal elements are ; Step 4.3.3, finally complete the rapid fault detection based on the propagation results.

Images