| CPC H02S 40/10 (2014.12) [B08B 3/024 (2013.01); B08B 13/00 (2013.01); G05D 1/622 (2024.01); G05D 1/6486 (2024.01); G06T 7/0002 (2013.01); G05D 1/644 (2024.01); G05D 2105/10 (2024.01); G05D 2109/20 (2024.01); G06T 2207/20081 (2013.01)] | 2 Claims |
|
1. An intelligent system for a photovoltaic cleaning robot, comprising a control terminal, an intelligent control center, a stain detection database, and an intelligent device; wherein the intelligent device comprises a camera shooting module, a motion flight module, and a cleaning module; the intelligent device and the intelligent control center are integrated on the photovoltaic cleaning robot; the intelligent control center and the control terminal are connected by a wireless communication; the intelligent control center is connected to the stain detection database through the wireless communication;
the intelligent control center receives a device start instruction issued by the control terminal: when a position of a fuselage is detected to exceed a threshold of a position range of a last sleep, then return to the control terminal to determine an inspection range; when there is no latitude change, then return to the control terminal to determine whether inspected or not, when the control terminal selects not to perform an inspection, an automatic cleaning mode is turned on;
the intelligent control center is connected to the stain detection database in real-time to facilitate continuous updating and learning of machine learning algorithms;
the intelligent control center is connected to the control terminal in real-time, to transmit back data and a picture of a cleaning operation, or receive instructions of the control terminal;
the intelligent control center sends execution instructions to the photovoltaic cleaning robot, and controls the photovoltaic cleaning robot to enter an intelligent cleaning mode;
a execution process of the intelligent cleaning mode comprises:
S100, sending the device start instruction by the control terminal, and entering the intelligent cleaning mode for the photovoltaic cleaning robot;
S200, establishing a recognition model, and automatically inspecting and collecting photovoltaic panel position information based on the recognition model;
S300, automatically planning a cleaning route and verifying learning according to the photovoltaic panel position information;
S400, storing the cleaning route to control the photovoltaic cleaning robot to perform the cleaning operation; and
S500, monitoring a status of the fuselage and an operation status in real-time during an operation according to different operating stages to judge whether to return to a supply station;
a process of establishing the recognition model in the S200 comprises:
S21, using lightweight GhostNet convolution calculation based on deep learning, a calculation formula is as follows:
Y′=F*f′
yij=Φi,j(yi),i∈[1,M],j∈[1,s];
wherein, f′∈RM×K×K×C denotes M convolution kernel of K×K size in C channels, Φi,j denotes a jth linear calculation and is configured to generate a jth Ghost feature map, F denotes an input feature map, Y denotes an output feature map processed by a convolution layer, yij denotes a jth linear operation result of an ith channel feature map in Y′, M denotes a number of channels, S denotes an internal parameter, and y; denotes the ith channel feature map in Y′; and
S22, using a CloU loss function to speed up and complete an establishment of the recognition model, and an expression is as follows:
 in the expression, Δx denotes a distance between a center point of a prediction box and a center point of a real box in an x direction, Δy denotes a distance between the center point of the prediction box and the center point of the real box in a y direction, I denotes a difference in a length of a bounding box, IOU denotes an original loss function, CIOU denotes a new improved loss function, H and W are a height and a width of a feature map respectively, α is a regularization coefficient, and σ is a standard deviation of a width and a length of the prediction box;
the step of sending the device start instruction by the control terminal, and entering the intelligent cleaning mode for the photovoltaic cleaning robot in the S100 comprises:
S110, sending the device start instruction by the control terminal;
S120, judging fuselage position information and an instruction when returning to the terminal whether to be inspected; and
S130, when the intelligent system determines a scope of the inspection, then carrying out a detection of the status of the fuselage and an electric quantity, and emptying a water tank;
the step of judging the fuselage position information and the instruction when returning to the terminal whether to be inspected in the S120 comprises:
S121, retrieving the cleaning route according to the fuselage position information when the control terminal selects not to perform the inspection;
S122, manually setting the inspection range or a default inspection range of the intelligent system when the control terminal selects to perform the inspection; and
S123, manually setting the inspection range or the default inspection range of the intelligent system when a change in the position of the fuselage is detected to be out of a preset range;
the step of automatically inspecting and collecting the photovoltaic panel position information based on the recognition model in the S200 comprises:
S210, automatically planning an inspection route according to the inspection range by the intelligent control center; and
S220, identifying a photovoltaic panel and storing the photovoltaic panel position information by a machine learning technology and mapping to an area map according to an inspection route flight;
the step of automatically planning the cleaning route and verifying the learning according to the photovoltaic panel position information in the S300 comprises:
S310, automatically planning the cleaning route and performing a no-load flight test according to the photovoltaic panel position information;
S320, avoiding obstacles and collecting obstacle information through fuselage sensors and the machine learning technology; and
S330, re-planning the cleaning route according to the obstacle information and returning results to the control terminal;
the step of storing the cleaning route to control the photovoltaic cleaning robot to perform the cleaning operation in the S400 comprises:
S410, starting the intelligent cleaning, and returning to the supply station for charging and adding a cleaning agent for the photovoltaic cleaning robot;
S420, according to position information of the cleaning route, flying to and identifying a first photovoltaic panel of the cleaning route by the intelligent control center and the machine learning technology;
S430, automatically identifying and judging whether the first photovoltaic panel needs to be cleaned according to the machine learning technology;
S431, automatically flying to a second photovoltaic panel on the cleaning route when the first photovoltaic panel does not need to be cleaned, and recording a progress of the cleaning route;
S432, spraying the cleaning agent when the second photovoltaic panel needs to be cleaned, judging a cleaning status of the second photovoltaic panel, and flying to a third photovoltaic panel on the cleaning route to continue the cleaning operation after reaching a set cleaning standard; and
S440, returning to the supply station when the cleaning operation is completed, and recording the progress of the cleaning route;
the step of judging whether to return to the supply station in the S500 comprises:
S510, in an automatic inspection stage, detecting the status of the fuselage and the electric quantity, returning to the supply station for a supply when the status of the fuselage is detected as abnormal or the electric quantity is detected as insufficient, and after completing the supply, automatically planning a shortest distance according to a progress of an operation route, and directly flying from the supply station to a next plate to be operated on the operation route in combination with an obstacle avoidance system; and
S520, in an intelligent cleaning stage, detecting the status of the fuselage, the electric quantity, and a residual quantity of the cleaning agent; returning to the supply station for the supply when the status of the fuselage is detected as abnormal or the electric quantity is detected as insufficient; after completing the supply, automatically planning the shortest distance according to the progress of the operation route, and directly flying from the supply station to the next plate to be operated on the operation route in combination with the obstacle avoidance system.
|