CN112495978A - Photovoltaic cleaning robot cleaning rate adjusting method based on visual perception - Google Patents
Photovoltaic cleaning robot cleaning rate adjusting method based on visual perception Download PDFInfo
- Publication number
- CN112495978A CN112495978A CN202011254267.5A CN202011254267A CN112495978A CN 112495978 A CN112495978 A CN 112495978A CN 202011254267 A CN202011254267 A CN 202011254267A CN 112495978 A CN112495978 A CN 112495978A
- Authority
- CN
- China
- Prior art keywords
- robot
- cleaning
- panel
- dust
- grade
- 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.)
- Withdrawn
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B08—CLEANING
- B08B—CLEANING IN GENERAL; PREVENTION OF FOULING IN GENERAL
- B08B13/00—Accessories or details of general applicability for machines or apparatus for cleaning
Landscapes
- Electric Vacuum Cleaner (AREA)
Abstract
The invention discloses a photovoltaic cleaning robot cleaning rate adjusting method based on visual perception, which solves the problem that the photovoltaic cleaning robot cleaning rate adjustment needs to be intelligentized. The invention comprises the following steps: 1. acquiring the dirt degree of each photovoltaic cell panel on each row of cell panels in advance by using an unmanned aerial vehicle; 2. the cleaning robot acquires an image of the surface of the battery board in the current field by using the camera to obtain the dust grade and the stubborn stain grade; 3. establishing a cleaning rate adjustment model according to the dust level and the level of stubborn stains, and obtaining the cleaning rate of the robot by using the model; 4. detecting the change of the output power of the battery panel in real time in the cleaning process of the robot to obtain the influence of the shielding of the robot on the power; 5. and correcting the rate according to the influence of the shielding of the robot on the output power to obtain a corrected cleaning rate adjustment model. The robot that this technique finally obtained cleans speed more accurate reasonable, can not influence the output of panel because of sheltering from.
Description
Technical Field
The invention relates to the technical field of machine vision image processing, in particular to a photovoltaic cleaning robot cleaning rate adjusting method based on visual perception.
Background
When a photovoltaic cell panel of a photovoltaic power station works outdoors for a long time, the surface of the photovoltaic cell panel can be covered by dust and can be shielded by other stains, so that the output power of the cell panel is reduced, hot spots are generated on the surface of the cell panel, and the cell panel can be damaged seriously. In order to reduce the interference and erosion of dust and dirt on the battery panel, a cleaning robot is generally adopted to clean the surface of the battery panel, so that the surface of the battery panel becomes clean and tidy, and the power generation power is improved. When the robot cleans the battery plates with different dirt degrees, the cleaning speed is different, and when the dirt degree is high, the cleaning speed is slower, so that the cleaning effect is good; when the dirt degree is low, the cleaning speed is higher, and the cleaning efficiency is improved. However, when the cleaning speed is too slow, the robot stays at a certain position of the solar panel for a long time, and the cleaning robot shields the solar panel, so that the output power of the solar panel fluctuates and decreases, which is unfavorable for the photovoltaic panel, and therefore, an operation and control method which has less influence on the solar panel due to shielding is required to be designed.
Disclosure of Invention
The invention overcomes the problem that the cleaning rate adjustment of the photovoltaic cleaning robot in the prior art needs to be intelligentized, and provides the photovoltaic cleaning robot cleaning rate adjustment method based on visual perception, which has a good processing effect.
The technical scheme of the invention is to provide a photovoltaic cleaning robot cleaning rate adjusting method based on visual perception, which comprises the following steps: comprises the following steps:
step 1: before the cleaning robot cleans the solar panels, an unmanned aerial vehicle is used for acquiring the dirt degree of each photovoltaic cell panel on each row of cell panels in advance;
step 2: the cleaning robot acquires an image of the surface of the battery plate in the current visual field by using a camera on the cleaning robot to obtain the dust grade and the grade of stubborn stains;
and step 3: establishing a cleaning rate adjustment model according to the dust level and the level of stubborn stains, and obtaining the cleaning rate of the robot by using the model;
and 4, step 4: detecting the change of the output power of the battery panel in real time in the cleaning process of the robot to obtain the influence of the shielding of the robot on the power;
and 5: and correcting the rate according to the influence of the shielding of the robot on the output power to obtain a corrected cleaning rate adjustment model.
Preferably, the step 1 comprises the following steps:
step 1.1, the unmanned aerial vehicle flies in low-altitude, the camera is downward in overlook, the visual field at least comprises a row of photovoltaic cell panels, and RGB images collected by the camera are obtained;
step 1.2, inputting the RGB image into a photovoltaic module positioning network to obtain a boundary frame of each row of photovoltaic modules and obtain the position of each row of photovoltaic modules; acquiring a boundary frame of each photovoltaic cell panel on each row of photovoltaic modules by using a photovoltaic cell panel positioning network, and acquiring the position of each photovoltaic cell panel on one row of photovoltaic modules; finally, acquiring the stain degree of each photovoltaic cell panel by using a stain degree judging network;
step 1.3, adopting a DNN network structure of an encoder-decoder to obtain a boundary box by the photovoltaic module positioning network and the photovoltaic panel positioning network, and obtaining the boundary box by an SSD or a YOLOV4 network structure;
step 1.4, when the battery panel is overlooked in the air, the ratio of the dust distribution area on the surface of the battery panel to the area of the battery panel is large, the larger the dirty degree is, the larger the dust area ratio is, and the dirty grade is divided into ten grade degrees: 0,1,2, …, 9; the larger the dirt degree is, the more the dust on the surface of the battery panel is distributed, the dirt degree is 0, and the dirt degree judging network acquires the dirt degree by adopting a resnet50 network structure.
Preferably, the step 2 comprises the following steps:
step 2.1, installing an RGB camera on the cleaning robot, wherein the camera is obliquely downward in a top view and collects RGB texture image data of the surface of the battery panel to be cleaned in real time;
2.2, obtaining the dust grade on the image by using a dust grade detection network;
and 2.3, acquiring stubborn stain grades on the image by using a stubborn stain grade detection network, and dividing the stubborn stain grades into ten grades: 0,1,2, …, 9; the stubborn stain grade is 0, which indicates that no stubborn stain grade exists on the surface of the battery plate, and the stubborn stain grade is 9, which indicates that most areas of the surface of the battery plate have stubborn stain grade distribution; the dust grade detection network and the stubborn stain grade detection network both adopt a resnet50 network structure.
Preferably, the step 3 comprises the following steps:
step 3.1: when the cleaning robot cleans the battery panel, the dust grade and stubborn stain grade on the battery panel to be cleaned are set to be L1 and L2, L1 and L2 are normalized, and the normalization method comprises the following steps:
the symbol "means that the result on the right side of the symbol is assigned to a variable of the symbol coordinate;
step 3.2: setting the running speed of the cleaning robot on the clean battery plate surface as V0,V0The robot is set when leaving the factory, and the speed V when the robot is cleaned to clean the battery panel is as follows:
V＝V0(1-α)
the above formula is a speed regulation model of the cleaning robot, the parameter α is a speed attenuation factor, and when the values of L1 and L2 are larger, α is larger, the robot needs to slow down speed for cleaning, that is, V is smaller; since stubborn stains on the panels are difficult to clean, the effect of L2 on speed is greater, i.e. alpha is more responsive to L2, such that:
α＝L1 exp(L2)(aL1+bL2)
wherein a and b are undetermined coefficients, and the method for acquiring the undetermined coefficients comprises the following steps:
step 3.2.1: in a laboratory environment, a cleaning robot is enabled to clean battery panels with different dust levels and stubborn stain levels at different rates, the cleaning rate V, the dust level L1 before cleaning, the stubborn stain level L2 before cleaning, the dust level L1_1 after cleaning and the stubborn stain level L2_1 after cleaning of the robot are recorded in each cleaning, the five parameters (V, L1, L2, L1_1 and L2_1) are called as sample data, and a plurality of sample data are obtained through a plurality of tests;
step 3.2.2: sample data with parameters L1_1 and L2_1 being 0 are screened from all the sample data, and the sample data show that the cleaning robot can clean the battery panel with the dust grade of L1 and the stubborn stain grade of L2 at the speed V, and the sample data are effective samples;
step 3.2.3: and fitting the cleaning robot speed adjusting model by using the effective sample, and fitting undetermined coefficients a and b by using a least square method.
Preferably, the step 4 comprises the following steps:
step 4.1: the robot acquires the output power of the battery plate row in real time in the cleaning process, sets the current time t, and acquires the output P of the battery plate row from the time t-k to the time t ═ P { (P)t-k,Pt-k+1,Pt-k+2,…,PtK is a hyper-parameter, and the value of k is 30 seconds;
step 4.2: let L3 be the degree of influence of robot occlusion on output power, let:
max (P) represents the maximum value of the sequence P, P0Indicating that the panel has no dirt, namely the output power, wherein gamma is a scaling coefficient and is a hyper-parameter, and the gamma is made to be 0.5; max (P) -PtRepresenting the difference between the maximum power and the power at the current moment, the value of L3 is smaller if the power shown in sequence P is gradually increased with a smaller fluctuation amplitude; when the robot blocks the battery plate at a certain position for a long time due to too slow movement, the power change of the sequence P presents the characteristic that the power change is gradually increased with small fluctuation firstly and then is increased to the maximum value and then is suddenly reduced, and max (P) -P at the momenttThe larger the value of L3, the larger the power attenuation amplitude of the battery panel output, the larger the influence of the robot shielding on the battery panel output power.
Preferably, the step 5 comprises the steps of:
step 5.1: when the influence of the shielding of the robot on the output power of the battery panel is large, the cleaning speed of the cleaning robot is increased, and the speed of the robot is increased by combining the amount of the dust to be cleaned of the robot;
step 5.2: the robot knows how many blocks of panels have been cleaned according to the vision odometer that itself carried to and how many remaining panels need clear up, according to the dirty degree of every panel that unmanned aerial vehicle provided, learns the dirty degree on the panel of not clearing up, the average value of the dirty degree of these panels is marked as L4, what L4 characterized is exactly that treats the dust degree of clear panel, carries out the normalization to L4 and handles: l4: l4/10;
step 5.3: correcting a cleaning rate adjustment model of the robot according to the influence degree L3 of the shielding of the robot on the output power and the dust degree L4 to be cleaned; setting the adjusted rate to
Order:
wherein:
V＝V0(1-α)
α＝L1 exp(L2)(aL1+bL2)
beta is a rate correction factor, when L3 is larger, the influence of the shielding of the robot on the output power of the battery panel is larger, the rate of the cleaning robot needs to be increased, namely the beta is larger; the larger the dust degree L4 of the panel to be cleaned is, the more the other panels are in urgent need of cleaning, and at this time, the speed of the robot is also increased, to sum up, the order:
β＝exp(L3)(θL4+1)
θ is a scaling factor, which indicates the weight of L4, and is a hyperparameter, where θ is 0.2.
Compared with the prior art, the photovoltaic cleaning robot cleaning rate adjusting method based on visual perception has the following advantages: and adjusting the cleaning speed of the robot according to the dust grade and the stubborn stain grade of the photovoltaic cell panel, and further correcting the speed through the change characteristic of the output power of the cell panel and the number of stains to be cleaned to finally obtain a corrected speed adjustment model of the cleaning robot. The cleaning speed of the robot is controlled by detecting the amount of dust and stains on the surface of the battery panel, so that the battery panel can be cleaned quickly and efficiently; meanwhile, the cleaning speed is corrected and adjusted according to the influence of the shielding of the robot on the battery panel and the quantity of the dust to be cleaned.
The distribution situation of dust and stubborn stains on the panel that the cleaning robot speed adjustment model that finally obtains was synthesized and is considered to shelter from factors such as can change the output of panel for a long time of considering the panel makes the robot that finally obtains clean speed more accurate, reasonable, and the high efficiency clearance panel can not influence the output of panel because of sheltering from again.
Drawings
FIG. 1 is a schematic workflow diagram of the present invention.
Detailed Description
The method for adjusting the cleaning rate of the photovoltaic cleaning robot based on visual perception of the invention is further described with reference to the accompanying drawings and the specific implementation mode: as shown in the figure, the specific implementation process of this embodiment is as follows:
1. use unmanned aerial vehicle high altitude flight earlier before photovoltaic cleaning machines people clearance panel, utilize the camera that unmanned aerial vehicle carried on to gather every row of photovoltaic module image information, acquire every row of dirty degree of photovoltaic cell panel on the photovoltaic cell panel according to the image of gathering, the purpose is that can learn the roughly area of the spot of treating the clearance when cleaning machines people and cleaning the panel. The degree of soiling of the panels is reflected by: when the battery panel is overlooked in the air, the ratio of the dust distribution area on the surface of the battery panel to the area of the battery panel is large, and the larger the dirt degree is, the larger the dust area ratio is.
2. The method for acquiring the stain degree of each battery panel on each row of battery panels by the unmanned aerial vehicle comprises the following steps:
1) unmanned aerial vehicle low-altitude flight, the camera overlooks downwards, and the field of vision includes one row of photovoltaic cell board at least, acquires the RGB image that the camera was gathered.
2) Inputting the RGB image into a photovoltaic module positioning network to obtain a boundary frame of each row of photovoltaic modules and obtain the position of each row of photovoltaic modules; acquiring a boundary frame of each photovoltaic cell panel on each row of photovoltaic modules by using a photovoltaic cell panel positioning network, and acquiring the position of each photovoltaic cell panel on one row of photovoltaic modules; and finally, acquiring the stain degree of each photovoltaic cell panel by using a stain degree judging network.
3) The photovoltaic module positioning network and the photovoltaic panel positioning network adopt DNN network structures of encoders and decoders to obtain the boundary frame, the network structures are known, such as SSD, YOLOV4 and the like, and the boundary frame is obtained by adopting a YOLOV4 network structure, so that the photovoltaic module positioning network and the photovoltaic panel positioning network have the advantages of higher precision and higher rapid speed. .
4) The dirty degree judging network is used for acquiring the dirty grade of each battery panel, and the dirty grade reflects that: when the battery panel is overlooked in the air, the ratio of the dust distribution area on the surface of the battery panel to the area of the battery panel is large, and the larger the dirt degree is, the larger the dust area ratio is. The invention divides the dirt grade into ten grade degrees: 0,1,2, …, 9. The larger the degree of contamination, the more the dust distribution on the surface of the battery panel, the degree of contamination was 0, which indicated that there was no dust on the surface of the battery panel, and the degree of contamination was 9, which indicated that the surface of the battery panel was almost completely covered with dust. The dirty degree judging network acquires the dirty degree by adopting a network structure of resnet50, quantitatively expresses the dirty condition of the battery panel and is beneficial to subsequent calculation and analysis of the dirty condition.
5) So far, utilize unmanned aerial vehicle to obtain every panel on every row of photovoltaic module at dirty degree. The beneficial effects of the method are two, firstly, the method can know which row of the battery panel is large in dirt degree and large in dust coverage, and is used for deciding whether to clean the battery panel or not; and secondly, when the cleaning robot cleans the battery plate, the cleaning robot can know the approximate area and distribution of the dust to be cleaned. The invention utilizes the latter to decide the cleaning rate of the cleaning robot.
3. The invention aims to control the cleaning rate of the cleaning robot according to the dirt condition of a battery plate when the cleaning robot cleans the battery plate. Above-mentioned unmanned aerial vehicle is owing to be high altitude flight, can only obtain dirty degree according to the coverage volume of dust, can't be meticulous to the thickness of dust and the distribution quantity of stubborn spot. In order to realize more accurate control, the cleaning robot needs to acquire the dust grade and the stubborn stain grade of the battery panel to be cleaned in real time. The dust grade reflects the thickness and the area of dust on the surface of the battery, and the thicker the dust, the larger the area, the higher the dust grade; the stubborn stains refer to blocky stains adhered to the surface of the battery plate, such as bird droppings and the like; the stubborn stain rating reflects the number and distribution area of stubborn stains.
4. The method for acquiring the dust grade and stubborn stain grade on the surface of the battery panel by the cleaning robot comprises the following steps:
1) an RGB camera is installed on the cleaning robot, the camera looks down obliquely and collects RGB texture image data of the surface of a battery panel to be cleaned in real time
2) The dust grade detection network is used for obtaining the dust grade on the image, and the dust grade is divided into ten grades: 0,1,2, …, 9. A dust rating of 0 indicates that there is no dust on the surface of the panel, and a dust rating of 9 indicates that there is dust distributed over most of the area of the panel surface, with the dust covering almost the entire panel area.
3) The method comprises the following steps of obtaining stubborn stain grades on an image by using a stubborn stain grade detection network, and dividing the stubborn stain grades into ten grades: 0,1,2, …, 9. The stubborn stain grade is 0, which indicates that no stubborn stain is on the surface of the battery plate, and the stubborn stain grade is 9, which indicates that most areas of the surface of the battery plate have stubborn stain grade distribution.
4) The invention relates to a dust grade detection network and a stubborn stain grade detection network resnet50 network structure. The beneficial effects of respectively obtaining the dust grade and the stubborn stain grade are that the distribution conditions of the dust and the stubborn stains on the battery panel are quantified, and the dust and the stubborn stains on the battery panel are respectively analyzed, so that the subsequent calculation result of the cleaning speed of the robot is reliable.
5. When the cleaning robot cleans the battery panel, the dust grade and stubborn stain grade on the battery panel to be cleaned are set to be L1 and L2, L1 and L2 are normalized, and the normalization method comprises the following steps:
the symbol ": means" a variable in which the result to the right of the symbol is assigned to the symbol coordinate. The beneficial effect of normalization is that the values of L1 and L2 are not more than 1.0, so that subsequent calculation cannot generate data with overlarge orders of magnitude.
6. Setting the running speed of the cleaning robot on the clean battery plate surface as V0The speed is determined by the robot when leaving the factory, and the speed V when the robot is cleaned to clean the battery plate is as follows:
V＝V0(1-α)
the above equation is a speed adjustment model of the cleaning robot, the parameter α is a speed attenuation factor, and when the values of L1 and L2 are larger, α is larger, the robot needs to slow down speed cleaning, i.e. V is smaller. Using the formula V ═ V0(1-alpha) the beneficial effect of calculating the panel velocity is through the method of attenuation coefficient, at V0On the basis of increasing or decreasing by an alphav0Such that the calculated V is at V0And the calculated V is reasonable and accurate due to the change of nearby fluctuation.
7. Since stubborn stains on the panels are difficult to clean, the effect of L2 on speed is greater, i.e. alpha is more responsive to L2, such that:
α＝L1 exp(L2)(aL1+bL2)
exp (L2) is a mapping to L2, so that alpha is more sensitive to the change of L2, namely, if L2 changes a little bit, alpha has a larger response, so that the result is more accurate and is in accordance with the reality.
8. Wherein a and b are undetermined coefficients, and the method for acquiring the undetermined coefficients comprises the following steps:
1) in a laboratory environment, a cleaning robot is enabled to clean battery panels with different dust levels and stubborn stain levels at different rates, the cleaning rate V, the dust level L1 before cleaning, the stubborn stain level L2 before cleaning, the dust level L1_1 after cleaning and the stubborn stain level L2_1 after cleaning of the robot are recorded in each cleaning, the five parameters (V, L1, L2, L1_1 and L2_1) are called as sample data, and a plurality of sample data are obtained through a plurality of tests.
2) Sample data with parameters L1_1 and L2_1 of 0 are screened from all sample data, the sample data indicate that the cleaning robot can clean the battery panel with the dust grade of L1 and the stubborn stain grade of L2 at the speed V, and the sample data are called effective samples.
3) And fitting the cleaning robot speed adjusting model by using effective samples to obtain undetermined coefficients a and b. The invention uses least square method to fit the coefficient to be determined. The implementer can also use other methods to fit the undetermined coefficient, and the method for fitting the undetermined coefficient by using the least square method has small calculation amount and is relatively simple and convenient.
9. When the cleaning robot cleans the surface of the battery plate, the robot can shield the surface of the battery plate; when the robot cleans fast, the influence of the shielding on the battery panel can be ignored, but when the surface of the battery panel is stained more and the cleaning speed of the robot is slower, the robot can stay at a certain position of the battery panel for a longer time, and the shielding at the moment can seriously influence the output power of the battery panel. Therefore, the invention obtains the influence of the shielding of the robot on the output power by detecting the output power of the battery plate, and the specific method comprises the following steps:
1) when the robot cleans the battery board, the output power of the battery board row is acquired in real time. If the cleaning robot cannot shield the battery panel, the output probability of the battery panel is gradually increased when the robot cleans the battery panel, because the dirt on the surface of the battery panel is cleaned, the output power is increased; however, in consideration of the blocking of the battery panel by the robot, the output power gradually increases with fluctuation. When the robot speed is too slow and the shielding is severe, the power is greatly reduced.
2) Setting the current time t, and acquiring the output P ═ P of the battery plate array from the time t-k to the time tt-k,Pt-k+1,Pt-k+2,…,PtK is a hyper-parameter, and the value of k is 30 seconds.
3) Let L3 be the degree of influence of robot occlusion on output power, let:
max (P) represents the maximum value of the sequence P. P0The power output is that the battery plate has no dirt, gamma is a scaling coefficient and is a super parameter, and the gamma is made to be 0.5
4)max(P)-PtRepresenting the difference between the maximum power and the power at the current moment. If the power shown in sequence P is gradually increased with a smaller fluctuation amplitude, the value of L3 is smaller; when the robot blocks the battery plate at a certain position for a long time due to too slow movement, the power change of the sequence P presents the characteristic that the power change is gradually increased with small fluctuation firstly and then is increased to the maximum value and then is suddenly reduced, and max (P) -P at the momenttThe value of (a) is large.
5) The larger L3 indicates that the power attenuation amplitude of the battery panel output is larger, and the influence of the robot shielding on the battery panel output power is larger. The beneficial effects of obtaining L3 are: the influence of the shielding of the cleaning robot on the output power of the battery panel can be intuitively obtained through the size of L3, the influence is an important index for correcting the cleaning speed of the robot, and the influence has important significance on subsequent calculation.
10. When the influence of the robot shielding on the output power of the battery panel is large, the cleaning speed of the cleaning robot should be increased, and in addition, in combination with the amount of dust to be cleaned by the robot, if the cleaning belt cleans a large amount of dust, the speed of the robot should be increased a little bit.
11. The robot can learn how many blocks of panels have been cleaned according to the vision odometer that itself carried to and how many remaining panels need clear up, according to the dirty degree of every panel that unmanned aerial vehicle provided, can learn the dirty degree on the panel of not clearing up, the average value of the dirty degree of these panels is recorded as L4, what L4 characterized is exactly the dust degree of treating clear panel. Note that the normalization process is performed on L4: l4: l4/10
12. According to the shielding pair of the robotThe degree of influence of the output power L3 and the degree of dust to be cleaned L4 modify the cleaning rate adjustment model of the robot. Setting the adjusted rate to
Order:
wherein:
V＝V0(1-α)
α＝L1 exp(L2)(aL1+bL2)
utilizing type
To obtain a corrected rate
Has the advantages that: increasing a beta V value on the basis of the original sweeping speed V so that
The change is made on the basis of V so that the result of the correction is relatively accurate.
13. Beta is a rate correction factor, and when L3 is larger, the influence of the shielding of the robot on the output power of the battery panel is larger, so that the cleaning robot needs to have larger rate increase, namely the beta is larger; on the other hand, the larger the dust degree L4 of the battery board to be cleaned is, which indicates that other battery boards are in urgent need of cleaning, and at the moment, the speed of the robot also influences the lifting greatly. In summary, let:
β＝exp(L3)(θL4+1)
θ is a scaling factor, which indicates the weight of L4, and is a super parameter, and θ is 0.2 in the present invention.
The beneficial effects of the rate correction factor beta are: determining the size of beta according to the influence degree L3 of the shielding of the robot on the output power of the battery panel and the dust deposition degree L4 to be cleaned, and comprehensively considering the influence machineFactor of the cleaning rate of the robot, the corrected robot rate thus obtained
Is more accurate and complete.
14. In summary, the corrected speed adjustment model of the cleaning robot
The speed can be adjusted according to the dirty condition on the surface of the battery panel, and the battery panel can be prevented from being influenced by the long-term shielding of the robot.
15. Thus, the present invention has been completed.
Claims (6)
1. A photovoltaic cleaning robot cleaning rate adjusting method based on visual perception is characterized by comprising the following steps: comprises the following steps:
step 1: before the cleaning robot cleans the solar panels, an unmanned aerial vehicle is used for acquiring the dirt degree of each photovoltaic cell panel on each row of cell panels in advance;
step 2: the cleaning robot acquires an image of the surface of the battery plate in the current visual field by using a camera on the cleaning robot to obtain the dust grade and the grade of stubborn stains;
and step 3: establishing a cleaning rate adjustment model according to the dust level and the level of stubborn stains, and obtaining the cleaning rate of the robot by using the model;
and 4, step 4: detecting the change of the output power of the battery panel in real time in the cleaning process of the robot to obtain the influence of the shielding of the robot on the power;
and 5: and correcting the rate according to the influence of the shielding of the robot on the output power to obtain a corrected cleaning rate adjustment model.
2. The photovoltaic cleaning robot cleaning rate adjusting method based on visual perception according to claim 1, characterized in that: the step 1 comprises the following steps:
step 1.1, the unmanned aerial vehicle flies in low-altitude, the camera is downward in overlook, the visual field at least comprises a row of photovoltaic cell panels, and RGB images collected by the camera are obtained;
step 1.2, inputting the RGB image into a photovoltaic module positioning network to obtain a boundary frame of each row of photovoltaic modules and obtain the position of each row of photovoltaic modules; acquiring a boundary frame of each photovoltaic cell panel on each row of photovoltaic modules by using a photovoltaic cell panel positioning network, and acquiring the position of each photovoltaic cell panel on one row of photovoltaic modules; finally, acquiring the stain degree of each photovoltaic cell panel by using a stain degree judging network;
step 1.3, adopting a DNN network structure of an encoder-decoder to obtain a boundary box by the photovoltaic module positioning network and the photovoltaic panel positioning network, and obtaining the boundary box by an SSD or a YOLOV4 network structure;
step 1.4, when the battery panel is overlooked in the air, the ratio of the dust distribution area on the surface of the battery panel to the area of the battery panel is large, the larger the dirty degree is, the larger the dust area ratio is, and the dirty grade is divided into ten grade degrees: 0,1,2, …, 9; the larger the dirt degree is, the more the dust on the surface of the battery panel is distributed, the dirt degree is 0, and the dirt degree judging network acquires the dirt degree by adopting a resnet50 network structure.
3. The photovoltaic cleaning robot cleaning rate adjusting method based on visual perception according to claim 1, characterized in that: the step 2 comprises the following steps:
step 2.1, installing an RGB camera on the cleaning robot, wherein the camera is obliquely downward in a top view and collects RGB texture image data of the surface of the battery panel to be cleaned in real time;
2.2, obtaining the dust grade on the image by using a dust grade detection network;
and 2.3, acquiring stubborn stain grades on the image by using a stubborn stain grade detection network, and dividing the stubborn stain grades into ten grades: 0,1,2, …, 9; the stubborn stain grade is 0, which indicates that no stubborn stain grade exists on the surface of the battery plate, and the stubborn stain grade is 9, which indicates that most areas of the surface of the battery plate have stubborn stain grade distribution; the dust grade detection network and the stubborn stain grade detection network both adopt a resnet50 network structure.
4. The photovoltaic cleaning robot cleaning rate adjusting method based on visual perception according to claim 1, characterized in that: the step 3 comprises the following steps:
step 3.1: when the cleaning robot cleans the battery panel, the dust grade and stubborn stain grade on the battery panel to be cleaned are set to be L1 and L2, L1 and L2 are normalized, and the normalization method comprises the following steps:
symbol ": "denotes a variable that assigns the result to the right of the symbol to the symbol coordinate;
step 3.2: setting the running speed of the cleaning robot on the clean battery plate surface as V0，V0The robot is set when leaving the factory, and the speed V when the robot is cleaned to clean the battery panel is as follows:
V＝V0(1-α)
the above formula is a speed regulation model of the cleaning robot, the parameter α is a speed attenuation factor, and when the values of L1 and L2 are larger, α is larger, the robot needs to slow down speed for cleaning, that is, V is smaller; since stubborn stains on the panels are difficult to clean, the effect of L2 on speed is greater, i.e. alpha is more responsive to L2, such that:
α＝L1exp(L2)(aL1+bL2)
wherein a and b are undetermined coefficients, and the method for acquiring the undetermined coefficients comprises the following steps:
step 3.2.1: in a laboratory environment, a cleaning robot is enabled to clean battery panels with different dust levels and stubborn stain levels at different rates, the cleaning rate V, the dust level L1 before cleaning, the stubborn stain level L2 before cleaning, the dust level L1_1 after cleaning and the stubborn stain level L2_1 after cleaning of the robot are recorded in each cleaning, the five parameters (V, L1, L2, L1_1 and L2_1) are called as sample data, and a plurality of sample data are obtained through a plurality of tests;
step 3.2.2: sample data with parameters L1_1 and L2_1 being 0 are screened from all the sample data, and the sample data show that the cleaning robot can clean the battery panel with the dust grade of L1 and the stubborn stain grade of L2 at the speed V, and the sample data are effective samples;
step 3.2.3: and fitting the cleaning robot speed adjusting model by using the effective sample, and fitting undetermined coefficients a and b by using a least square method.
5. The photovoltaic cleaning robot cleaning rate adjusting method based on visual perception according to claim 1, characterized in that: the step 4 comprises the following steps:
step 4.1: the robot acquires the output power of the battery plate row in real time in the cleaning process, sets the current time t, and acquires the output P of the battery plate row from the time t-k to the time t ═ P { (P)t-k，Pt-k+1，Pt-k+2，..，PtK is a hyper-parameter, and the value of k is 30 seconds;
step 4.2: let L3 be the degree of influence of robot occlusion on output power, let:
max (P) represents the maximum value of the sequence P, P0Indicating that the panel has no dirt, namely the output power, wherein gamma is a scaling coefficient and is a hyper-parameter, and the gamma is made to be 0.5; max (P) -PtRepresenting the difference between the maximum power and the power at the current moment, the value of L3 is smaller if the power shown in sequence P is gradually increased with a smaller fluctuation amplitude; when the robot blocks the battery plate at a certain position for a long time due to too slow movement, the power change of the sequence P presents the characteristic that the power change is gradually increased with small fluctuation firstly and then is increased to the maximum value and then is suddenly reduced, and max (P) -P at the momenttThe larger the value of L3, the larger the power attenuation amplitude of the output of the battery panel, the larger the value of L3, the more the machine is representedThe greater the impact of human occlusion on the output power of the panel.
6. The photovoltaic cleaning robot cleaning rate adjusting method based on visual perception according to claim 1, characterized in that: the step 5 comprises the following steps:
step 5.1: when the influence of the shielding of the robot on the output power of the battery panel is large, the cleaning speed of the cleaning robot is increased, and the speed of the robot is increased by combining the amount of the dust to be cleaned of the robot;
step 5.2: the robot knows how many blocks of panels have been cleaned according to the vision odometer that itself carried to and how many remaining panels need clear up, according to the dirty degree of every panel that unmanned aerial vehicle provided, learns the dirty degree on the panel of not clearing up, the average value of the dirty degree of these panels is marked as L4, what L4 characterized is exactly that treats the dust degree of clear panel, carries out the normalization to L4 and handles: l4: l4/10;
step 5.3: correcting a cleaning rate adjustment model of the robot according to the influence degree L3 of the shielding of the robot on the output power and the dust degree L4 to be cleaned; setting the adjusted rate to
Order:
wherein:
V＝V0(1-α)
α＝L1exp(L2)(aL1+bL2)
beta is a rate correction factor, when L3 is larger, the influence of the shielding of the robot on the output power of the battery panel is larger, the rate of the cleaning robot needs to be increased, namely the beta is larger; the larger the dust degree L4 of the panel to be cleaned is, the more the other panels are in urgent need of cleaning, and at this time, the speed of the robot is also increased, to sum up, the order:
β＝exp(L3)(θL4+1)
θ is a scaling factor, which indicates the weight of L4, and is a hyperparameter, where θ is 0.2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011254267.5A CN112495978A (en) | 2020-11-11 | 2020-11-11 | Photovoltaic cleaning robot cleaning rate adjusting method based on visual perception |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011254267.5A CN112495978A (en) | 2020-11-11 | 2020-11-11 | Photovoltaic cleaning robot cleaning rate adjusting method based on visual perception |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112495978A true CN112495978A (en) | 2021-03-16 |
Family
ID=74956297
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011254267.5A Withdrawn CN112495978A (en) | 2020-11-11 | 2020-11-11 | Photovoltaic cleaning robot cleaning rate adjusting method based on visual perception |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112495978A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115283360A (en) * | 2022-10-08 | 2022-11-04 | 天津盛安机械设备有限公司 | Automatic visual point cloud path planning system and method based on intelligent subway purging |
| CN117021105A (en) * | 2023-08-30 | 2023-11-10 | 北京瑞科同创能源科技有限公司 | Control method, device and equipment of photovoltaic cleaning robot |
-
2020
- 2020-11-11 CN CN202011254267.5A patent/CN112495978A/en not_active Withdrawn
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115283360A (en) * | 2022-10-08 | 2022-11-04 | 天津盛安机械设备有限公司 | Automatic visual point cloud path planning system and method based on intelligent subway purging |
| CN115283360B (en) * | 2022-10-08 | 2022-12-27 | 天津盛安机械设备有限公司 | Automatic visual point cloud path planning system and method based on intelligent subway purging |
| CN117021105A (en) * | 2023-08-30 | 2023-11-10 | 北京瑞科同创能源科技有限公司 | Control method, device and equipment of photovoltaic cleaning robot |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Aghaei et al. | Innovative automated control system for PV fields inspection and remote control | |
| CN109615629B (en) | Convolutional neural network-based photovoltaic panel dust accumulation state image recognition system and analysis, regulation and control method thereof | |
| CN112495978A (en) | Photovoltaic cleaning robot cleaning rate adjusting method based on visual perception | |
| CN113538391B (en) | Photovoltaic defect detection method based on Yolov4 and thermal infrared image | |
| CN107179768B (en) | Obstacle identification method and device | |
| CN109799442B (en) | Insulator pollution flashover prediction method and system based on airborne hyperspectrum | |
| US9659237B2 (en) | Imaging through aerosol obscurants | |
| CN104517125B (en) | The image method for real time tracking and system of high-speed object | |
| CN116690613B (en) | Control method and system of photovoltaic intelligent cleaning robot | |
| US20170140077A1 (en) | Method for assessing efficiency of power generation systems | |
| WO2020144618A1 (en) | Device and method for measuring effect of soiling on photovoltaic device | |
| CN114342910A (en) | Laser bird repelling method and related device | |
| Yang et al. | Soiling quantification using an image-based method: Effects of imaging conditions | |
| CN115018872B (en) | Intelligent control method of dust collection equipment for municipal construction | |
| CN116542383A (en) | Distributed photovoltaic system output prediction method based on small fluctuation weather satellite cloud image | |
| CN116797944A (en) | Detection method and system for identifying cleanliness of photovoltaic panel based on unmanned aerial vehicle image | |
| CN115457001A (en) | Photovoltaic panel foreign matter detection method, system, device and medium based on VGG network | |
| CN112801413A (en) | Photovoltaic power station generated power prediction method and device | |
| CN115392494A (en) | Intelligent photovoltaic ash removal method and system | |
| CN116029160B (en) | Method and system for constructing mapping model of defects and power generation efficiency loss of photovoltaic module | |
| KR102579901B1 (en) | System and method for diagnosing photovoltaic power generation using i-v curve | |
| JP2013113797A (en) | Solar radiation sunlight loss evaluation system in photovoltaic power generation system | |
| CN115166866A (en) | Citrus disease and insect pest occurrence forecasting method and system based on lattice point meteorological data | |
| US10615745B1 (en) | Device and method for measuring effect of soiling on photovoltaic device | |
| CN114563771A (en) | Double-threshold laser radar cloud layer detection algorithm based on cluster analysis |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210316 |