Crawler-type photovoltaic cleaning robot control method based on fuzzy logic and proportional differential integral corrector

Abstract

The invention provides a crawler-type photovoltaic cleaning robot control method based on fuzzy logic and a proportional differential integral corrector, and relates to the field of robot control. A current course straight line is obtained through an image acquisition system in cooperation with an image algorithm, a current yaw angle error and a current transverse error are obtained, fuzzy reasoning is carried out on the yaw angle error and the transverse error to obtain a coefficient, and a control signal is obtained by combining two proportional differential integral correctors; and the control signal is subjected to hard constraint to obtain a correction value, and a motor control signal is generated to control a motor of the crawler-type photovoltaic cleaning robot so as to enable the crawler-type photovoltaic cleaning robot to run along the texture of the photovoltaic panel. Motion information can be obtained only through an image acquisition system, and the hardware structure is simple; the pose data obtained through an image algorithm is small in fluctuation, the crawler-type photovoltaic cleaning robot can be better controlled, and the cleaning effect on the photovoltaic panel is improved.

Description

Technical Field

[0001] The invention relates to the field of robot control, and in particular to a control method for a crawler-type photovoltaic cleaning robot. Background Art

[0002] With the growing demand for energy in human society and considerations for environmental protection and sustainable development, more and more clean energy and renewable energy have entered people's lives. Solar energy, as a renewable clean energy, has become an important force in the current energy transformation. However, due to the relatively harsh environment in the normal deployment area of ​​​​photovoltaic panels, dust and dirt are easily accumulated on the surface of photovoltaic panels. According to scientific calculations, the power generation of photovoltaic panels that have not been scientifically cleaned can drop by up to 40% to 60%, and the power generation can be attenuated by 20% to 30%. Therefore, cleaning robots for photovoltaic panels have incomparable significance.

[0003] The existing tracked photovoltaic cleaning robots will gradually deviate from the original driving direction during driving due to the characteristics of their tracks, the inclination of the photovoltaic panels, the rain and snow conditions on site and other conditions. Without automatic control, they are very likely to fall due to excessive deviation from the original driving direction, causing economic losses.

[0004] The existing error acquisition methods of crawler photovoltaic cleaning robots are mainly divided into four types: through machine vision, through inertial navigation unit, through GPS positioning, and multi-sensor data fusion. Among them, the acquisition through inertial navigation unit will have the problem of error accumulation, and this problem is positively correlated with the running time. It is difficult to achieve more accurate error acquisition in long-term and high-complexity cleaning tasks; the method through GPS positioning requires a satellite communication device to be placed on the crawler photovoltaic cleaning robot, and the signal requirements are high. It is very easy to lose positioning in a complex environment, resulting in an open loop of the system; the method through multi-sensor data fusion has higher accuracy, but relies on data from multiple sensors, which will significantly increase the system complexity and computing time.

[0005] Patent CN119002507A proposes a photovoltaic cleaning robot control method based on a proportional-integral-differential controller. This method has low computational complexity and good system stability. It can control the robot to travel in a straight line along the grain on the photovoltaic panel. However, due to the limitations of the proportional-integral-differential controller, its control system has poor flexibility and cannot control the yaw angle error and lateral error at the same time. Patent CN118838344A proposes a control method based on tilt parameters, which uses inertial navigation to obtain the tilt parameters. However, due to the error accumulation problem of inertial navigation, it performs poorly in long-term, high-complexity cleaning tasks. Patent CN117666593A proposes a robot control optimization method based on neural networks, but due to the high complexity of neural networks and high computing power requirements, it is not suitable for deployment in robots, an embedded system.

[0006] In summary, it is of great significance to design a control method with high flexibility, high system stability, low computational complexity, and good performance in long-term and high-complexity cleaning tasks. It can realize the automatic control of the tracked photovoltaic cleaning robot and reduce the economic losses caused by yaw and fall. Summary of the invention

[0007] In order to overcome the above problems and realize automatic control of a tracked photovoltaic cleaning robot in a complex environment, the present invention proposes a control method for a tracked photovoltaic cleaning robot based on fuzzy control and a proportional differential integral corrector, which realizes the control of the yaw angle while maintaining the control of the lateral error; the present invention also uses a discrimination method based on the error change rate, which further reduces the misjudgment of the system error and improves the stability of the system.

[0008] In order to achieve the above invention objectives, the technical solutions adopted are as follows:

[0009] The texture image of the photovoltaic panel surface is obtained through the image acquisition system, and the center line of the photovoltaic panel texture is set as the target yaw angle of the robot. The current heading line of the robot is extracted from the image using an image algorithm. The current yaw angle error and lateral error of the robot are determined by comparing the heading line with the center line of the photovoltaic panel texture. The yaw angle error and lateral error are simultaneously fuzzy inferred by fuzzy logic to obtain coefficients. The coefficients are combined with two proportional integral differential correctors to obtain a control signal. The control signal is restricted by hard constraints to obtain a correction value. The correction value generates a motor control signal for controlling the motor of the crawler photovoltaic cleaning robot, so that the crawler photovoltaic cleaning robot can stably travel along the photovoltaic panel texture. The specific steps are as follows:

[0010] Step 1: Obtaining the rough error from the image: The image of the texture of the photovoltaic panel where the crawler photovoltaic cleaning robot is located is obtained through the image acquisition system, the image center line is obtained through double threshold image binarization, the noise is removed and the path features are highlighted through image erosion and image expansion, the straight line is extracted through Huffman transformation, and the rough error is obtained by comparing with the image center line;

[0011] Step 2: Error discrimination: By analyzing the rate of change of the error, if there is a sudden change in the rate of change, it is considered that a path jump has occurred, and the previous error is used to replace the current error to achieve continuous target path planning;

[0012] Step 3: Fuzzification of input errors: A dual-input fuzzy inference system is used to classify the input yaw angle error and lateral error into three states through the triangular input membership function;

[0013] Step 4: Fuzzy reasoning: Through 9 fuzzy reasoning rules, the input yaw angle error and lateral error are inferred into three output states;

[0014] Step 5: Defuzzification: Defuzzification is performed by using the isosceles triangle barycentric projection method. The input yaw angle error and the lateral error are integrated and added. The addition result is used as the area of ​​an isosceles triangle with a base length of 1. The barycentric of the isosceles triangle is calculated, and the height corresponding to the barycentric is the fuzzy reasoning output.

[0015] Step 6: Correct the output: The yaw angle error and the lateral error are calculated by the yaw angle proportional differential integral corrector and the lateral proportional differential integral corrector respectively to obtain the correction of each error. The correction of each error is added after the gain of the fuzzy inference output. The final result is the total output correction of the crawler photovoltaic controller.

[0016] Step 7: Hard constraint to prevent loss of control: limit the maximum angular velocity and linear velocity of the crawler photovoltaic cleaning robot to prevent the controller from losing control and causing the crawler photovoltaic cleaning robot to lose control;

[0017] Step 8: Motor control: The angular velocity and linear velocity obtained in the above steps are calculated through kinematic change matrix to obtain the motor control signal, thereby realizing the automatic control of the crawler-type photovoltaic cleaning robot traveling along the texture of the photovoltaic panel.

[0018] Furthermore, the three states are S (small), M (medium), and B (large), wherein for S (small), the normalized error falls between 0 and 0.4; for M (medium), the normalized error falls between 0.3 and 0.7; and for B (large), the normalized error falls between 0.6 and 1.

[0019] Furthermore, the three states are R (yaw angle-oriented), N (normal), and H (lateral-oriented).

[0020] Furthermore, the 9 fuzzy reasoning rules are:

[0021] Rule 1: If the yaw error is small and the lateral error is small, the output state is normal;

[0022] Rule 2: If the yaw error is small and the lateral error is medium, the output state is lateral-oriented;

[0023] Rule 3: If the yaw error is small and the lateral error is large, the output state is lateral-oriented;

[0024] Rule 4: If the yaw error is moderate and the lateral error is small, the output state is yaw-focused;

[0025] Rule 5: If the yaw error is medium and the lateral error is medium, the output status is normal;

[0026] Rule 6: If the yaw error is moderate and the lateral error is large, the output state is lateral-oriented;

[0027] Rule 7: If the yaw error is large and the lateral error is small, the output state is yaw-oriented;

[0028] Rule 8: If the yaw error is large and the lateral error is medium, the output state is yaw-focused;

[0029] Rule 9: If the yaw angle error is large and the lateral error is large, the output status is normal.

[0030] Furthermore, the image acquisition system includes an optical camera, a surrounding light shield and a fill light.

[0031] Furthermore, the control method is deployed on an Ubuntu 18.04LTS system development board and is written in Python.

[0032] The present invention proposes a control method for a crawler-type photovoltaic cleaning robot based on fuzzy control and proportional differential integral corrector, which has the following significant advantages:

[0033] (1) By adopting fuzzy control, the yaw angle and lateral error can be controlled simultaneously with only a slight increase in computational complexity;

[0034] (2) The proportional differential integral corrector control algorithm with fixed parameters is superimposed to reduce the complexity of fuzzy reasoning and ensure the stability of the system to a great extent;

[0035] (3) The detection equipment only requires one optical camera, which reduces the complexity of the system and avoids the problem of error accumulation of the inertial navigation unit;

[0036] (4) The error change rate-based discrimination method is used to avoid the problem of the camera outputting erroneous data at the connection of the photovoltaic panel, thus achieving long-distance stable driving of the crawler photovoltaic cleaning robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The control system block diagram of the method of the present invention.

[0038] Figure 2 This is a hardware installation diagram.

[0039] Figure 3 is a diagram of the membership function of the fuzzy controller: (a) input membership; (b) output membership.

[0040] Figure 4 It is the fuzzy reasoning rule table.

[0041] Figure 5 This is a simulation diagram of the control effect. DETAILED DESCRIPTION

[0042] The following is a detailed description of the control method implementation scheme of the crawler-type photovoltaic cleaning robot based on fuzzy control and proportional differential integral corrector of the present invention in conjunction with the accompanying drawings of the specification, but the following examples are only descriptive and not restrictive.

[0043] The present invention uses machine vision to obtain the error. The control block diagram is shown in the attached figure. Figure 1 As shown in the figure, when the tracked photovoltaic cleaning robot is driving on the photovoltaic panel, the image is acquired through the image acquisition system, the current heading line and photovoltaic panel texture of the tracked photovoltaic cleaning robot are acquired through the image algorithm, and the current yaw angle error and the current lateral error are acquired by comparing the photovoltaic panel texture and the heading line; the yaw angle error and the lateral error are simultaneously fuzzy inferred through fuzzy logic, and the motor of the tracked photovoltaic cleaning robot is controlled through two proportional integral differential correctors combined with the coefficients obtained by fuzzy inference to control the tracked photovoltaic cleaning robot to drive along the photovoltaic panel texture. The image acquisition system is shown in the attached figure. Figure 2 As shown, the direction indicated by the arrow is the movement direction of the crawler photovoltaic cleaning robot. The image acquisition system, including an optical camera, a surrounding light shield and a fill light, is installed at the front of the crawler photovoltaic cleaning robot. When the system is turned on, the image acquisition system starts to work, and real-time images of the texture where the crawler photovoltaic cleaning robot is located are collected and transmitted to the controller through the USB interface; the control method is deployed on the Ubuntu18.04LTS system development board and is written in Python language.

[0044] In each control cycle, the following steps are followed:

[0045] (1) Obtaining gross errors from images: In order to reduce the real-time computing workload of the system, the original color image is grayed out, meaningless RGB information is discarded, and then a median filter is used to remove small noise points caused by image quality and screen jitter. In order to highlight the features of the thick white lines in the picture, the double threshold method is used to turn the dark background and the brighter reflective areas into black, leaving only the white lines in the picture; since the white lines of the photovoltaic panels are thick and thin, with more thin lines and fewer thick lines, this solution uses image dilation and image erosion operations. Image dilation is used to fill the rough edges of the white lines, which is more conducive to subsequent straight line recognition. Image erosion is used to filter out all the thin lines, leaving only the thick lines for tracking; the Canny operator is used for edge detection, and the Hough transform is used to identify all the straight lines appearing in the figure; if the maximum allowable interval (break) of two line segments in the same direction determined as one line segment exceeds the set value, the two line segments are regarded as one line segment. The larger the value, the larger the break on the line segment is allowed, and the more likely it is to detect potential straight line segments. In the application scenario of the present invention, it is expected that there is only one straight line, so the constraint of the maximum allowable interval is very loose; find the straight line closest to the center of the picture as the target path, take the center line of the picture as the heading line, and take the deviation between the heading line and the target path as the output yaw angle error and lateral error.

[0046] (2) Error discrimination: Since the spacing of the lines on different photovoltaic panels is different, and multiple straight lines may appear in the camera's field of view during operation (especially in the initial state), if the camera's output error is directly used as the final error, when the crawler photovoltaic cleaning robot shakes for some reason, the determined straight line may jump, causing the system to fall into a positive feedback loop. Therefore, before the lateral error is input into the controller, it is discriminated. The present invention uses the change in the lateral error change rate to determine whether a target trajectory jump has occurred; if a target trajectory jump has occurred, the lateral error will suddenly change, and the lateral error change rate will suddenly increase. If this increase is detected, it can be considered that a jump has occurred. At this time, the error discriminator abandons the lateral error measured this time and uses the last lateral error for output. Through the above method, sensor errors caused by environmental interference can be avoided to a large extent.

[0047] (3) Input error fuzzification: Through the above steps, the controller believes that the true values ​​of the yaw angle error and the lateral error have been obtained, and the heading angle error and the lateral error are fuzzified. The fuzzification method adopted by the present invention is to normalize the yaw angle error and the lateral error for discrimination, and classify them into three states through the triangular input membership function, S (small), M (medium), and B (large). These states correspond to different error ranges, S (small), the normalized error falls between 0 and 0.4; M (medium), the normalized error falls between 0.3 and 0.7; B (large), the normalized error falls between 0.6 and 1; for state S (small), the left endpoint is 0 and the vertex is 0.4; for state M (medium), the left endpoint is 0.3 and the vertex is 0.7; for state B (large), the left endpoint is 0.6 and the vertex is 1; Figure 3 (a) shows the image of the input membership function, in which each state corresponds to a triangular curve. These curves describe how the error value is mapped to the membership value.

[0048] (4) Fuzzy reasoning: According to the attached Figure 4 The fuzzy inference rules shown in the figure can obtain the state of fuzzy inference output, which is divided into three states: N (normal), H (lateral emphasis), and R (yaw emphasis). There are 9 fuzzy inference rules in total, which are:

[0049] Rule 1: If the yaw error is small and the lateral error is small, the output state is normal;

[0050] Rule 2: If the yaw error is small and the lateral error is medium, the output state is lateral-oriented;

[0051] Rule 3: If the yaw error is small and the lateral error is large, the output state is lateral-oriented;

[0052] Rule 4: If the yaw error is moderate and the lateral error is small, the output state is yaw-focused;

[0053] Rule 5: If the yaw error is medium and the lateral error is medium, the output status is normal;

[0054] Rule 6: If the yaw error is moderate and the lateral error is large, the output state is lateral-oriented;

[0055] Rule 7: If the yaw error is large and the lateral error is small, the output state is yaw-oriented;

[0056] Rule 8: If the yaw error is large and the lateral error is medium, the output state is yaw-focused;

[0057] Rule 9: If the yaw angle error is large and the lateral error is large, the output status is normal.

[0058] (5) Defuzzification: According to the above steps, the output state can be obtained. According to the output membership function shown in Figure 3(b), it is a triangular curve. The projection of the obtained output state will fall within a triangle. The present invention proposes a defuzzification method based on the isosceles triangle centroid projection method, which is specifically: adding the input membership of the yaw angle error and the lateral error to the three states, and taking it as the projection area of ​​the state corresponding to the output membership. According to the projection area, the centroid position of an isosceles triangle with a base length of 1 can be calculated. The output membership function value corresponding to the centroid is the output q of this fuzzy reasoning.

[0059] (6) Correction of output: The controller commonly used in the industry is a proportional differential integral corrector, which can be divided into a positional proportional differential integral corrector and an incremental proportional differential integral corrector. Since the output of the control system of the present invention is speed, and in order to reduce memory usage, the present invention uses an incremental proportional differential integral corrector, and its formula is:

[0060] u(t)=Kp*(err(t)-err(t-1))+Ki*err(t)+Kd*(err(t)-2err(t-1)+err(t-2))

[0061] Where Kp is the proportional coefficient, Ki is the integral coefficient, Kd is the differential coefficient, err(t) is the current error, err(t-1) is the previous error, err(t-2) is the previous quadratic error, and u(t) is the output correction value; the yaw angle error correction value and the lateral error correction value are multiplied by the output q and 1-q of the above fuzzy reasoning respectively, and then superimposed. The result is the final output correction value. This correction value is a change quantity, not a position quantity. Therefore, the previous speed plus this correction value is the current speed output value.

[0062] (7) Hard constraints to prevent loss of control: Due to the mechanical structure limitations of the crawler-type photovoltaic cleaning robot, its output speed needs to be limited within a certain range. Otherwise, it will cause damage to the mechanical structure or excessive output current to burn electronic components or controller loss of control, causing the crawler-type photovoltaic cleaning robot to lose control.

[0063] (8) Motor control: The angular velocity and linear velocity obtained in the above steps are subjected to kinematic change matrix operations to obtain a motor control signal, which is then output to the motor driver, thereby achieving automatic control of the crawler-type photovoltaic cleaning robot moving along the texture of the photovoltaic panel.

[0064] After the above steps, a control calculation is completed, and the calculated speed and other commands are packaged and sent to the motion control board through the USB interface. After receiving the command, the motion control board parses it, and the parsed speed is calculated through the kinematic change matrix to obtain the motor control signal, thereby realizing the automatic control of the crawler photovoltaic cleaning robot; Figure 5 The figure shows the simulation result of the tracked photovoltaic cleaning robot in Simulink. The lateral error reaches the control target in 4 seconds and has no obvious overshoot. There is no obvious oscillation after entering the steady state. The yaw angle error reaches the control target in 6 seconds and the overshoot is extremely small. There is no obvious oscillation after entering the steady state. It can be seen that in the control process, the tracked photovoltaic robot effectively controls the yaw angle error and the lateral error at the same time, which shows that the method proposed in the present invention has a control effect of fast response and good stability. The actual driving distance of the tracked photovoltaic cleaning robot is 200 meters without route deviation. The results show that the method proposed in the present invention can realize the control of the tracked photovoltaic cleaning robot with good effect.

Claims

1. A control method for a crawler photovoltaic cleaning robot based on fuzzy logic and proportional differential integral corrector, characterized in that: include: The texture image of the photovoltaic panel surface is obtained through the image acquisition system, and the center line of the photovoltaic panel texture is set as the target yaw angle of the robot. The current heading line of the robot is extracted from the image using an image algorithm. The current yaw angle error and lateral error of the robot are determined by comparing the heading line with the center line of the photovoltaic panel texture. The yaw angle error and lateral error are simultaneously fuzzy inferred by fuzzy logic to obtain coefficients. The coefficients are combined with two proportional integral differential correctors to obtain a control signal. The control signal is restricted by hard constraints to obtain a correction value. The correction value generates a motor control signal for controlling the motor of the crawler photovoltaic cleaning robot, so that the crawler photovoltaic cleaning robot can stably travel along the photovoltaic panel texture. The specific steps are as follows: Step 1: Obtaining the rough error from the image: The image of the texture of the photovoltaic panel where the crawler photovoltaic cleaning robot is located is obtained through the image acquisition system, the image center line is obtained through double threshold image binarization, the noise is removed and the path features are highlighted through image erosion and image expansion, the straight line is extracted through Huffman transformation, and the rough error is obtained by comparing with the image center line; Step 2: Error discrimination: By analyzing the rate of change of the error, if there is a sudden change in the rate of change, it is considered that a path jump has occurred, and the previous error is used to replace the current error to achieve continuous target path planning; Step 3: Fuzzification of input errors: A dual-input fuzzy inference system is used to classify the input yaw angle error and lateral error into three states through the triangular input membership function; Step 4: Fuzzy reasoning: Through 9 fuzzy reasoning rules, the input yaw angle error and lateral error are inferred into three output states; Step 5: Defuzzification: Defuzzification is performed by using the isosceles triangle barycentric projection method. The input yaw angle error and the lateral error are integrated and added. The addition result is used as the area of ​​an isosceles triangle with a base length of 1. The barycentric of the isosceles triangle is calculated, and the height corresponding to the barycentric is the fuzzy reasoning output. Step 6: Correct the output: The yaw angle error and the lateral error are calculated by the yaw angle proportional differential integral corrector and the lateral proportional differential integral corrector respectively to obtain the correction of each error. The correction of each error is added after the gain of the fuzzy inference output. The final result is the total output correction of the crawler photovoltaic controller. Step 7: Hard constraint to prevent loss of control: limit the maximum angular velocity and linear velocity of the crawler photovoltaic cleaning robot to prevent the controller from losing control and causing the crawler photovoltaic cleaning robot to lose control; Step 8: Motor control: The angular velocity and linear velocity obtained in the above steps are calculated through kinematic change matrix to obtain the motor control signal, thereby realizing the automatic control of the crawler photovoltaic cleaning robot traveling along the texture of the photovoltaic panel.

2. The control method of a crawler photovoltaic cleaning robot based on fuzzy logic and proportional differential integral corrector according to claim 1 is characterized in that: The three states are S (small), M (medium), and B (large), where S (small) has a normalized error between 0 and 0.4; M (medium), the normalized error falls between 0.3 and 0.7; B (large), the normalized error falls between 0.6 and 1.

3. The control method of a crawler photovoltaic cleaning robot based on fuzzy logic and proportional differential integral corrector according to claim 1 is characterized in that: The three states are R (yaw angle-oriented), N (normal), and H (lateral-oriented).

4. The control method of a crawler photovoltaic cleaning robot based on fuzzy logic and proportional differential integral corrector according to claim 1, characterized in that: The 9 fuzzy inference rules are: Rule 1: If the yaw error is small and the lateral error is small, the output state is normal; Rule 2: If the yaw error is small and the lateral error is medium, the output state is lateral-oriented; Rule 3: If the yaw error is small and the lateral error is large, the output state is lateral-oriented; Rule 4: If the yaw error is moderate and the lateral error is small, the output state is yaw-focused; Rule 5: If the yaw error is medium and the lateral error is medium, the output status is normal; Rule 6: If the yaw error is moderate and the lateral error is large, the output state is lateral-oriented; Rule 7: If the yaw error is large and the lateral error is small, the output state is yaw-oriented; Rule 8: If the yaw error is large and the lateral error is medium, the output state is yaw-focused; Rule 9: If the yaw angle error is large and the lateral error is large, the output status is normal.

5. The control method of a crawler photovoltaic cleaning robot based on fuzzy logic and proportional differential integral corrector according to claim 1, characterized in that: The image acquisition system includes an optical camera, a surrounding light shield and a fill light.

6. The control method of a crawler photovoltaic cleaning robot based on fuzzy logic and proportional differential integral corrector according to claim 1, characterized in that: The control method is deployed on the Ubuntu 18.04LTS system development board and is written in Python.

Images