Intelligent control method for air conditioner air compressor

Abstract

This invention discloses an intelligent control method for air conditioning air compressors, specifically relating to the field of energy-saving control technology for air conditioning air compressors. The method includes acquiring continuous images of the air conditioning air compressor control panel, reading the page identifier area, numerical display area, status display area, and prompt display area frame by frame, extracting persistent high load, high energy consumption, or frequent switching phenomena from adjacent images, and outputting the current page, current display item, and abnormal phenomena. The method first performs page positioning and reversible click pairing to solve the abnormal phenomena in the continuous images of the control panel, then verifies whether the current abnormal phenomenon can be resolved by the current human-machine interface operation based on the return results of the trial action and the reverse click, and performs formal adjustment along the direction of the trial action after the verification is successful.

Description

Technical Field

[0001] This invention relates to the field of energy-saving control technology for air conditioning air compressors, and more specifically, to an intelligent control method for air conditioning air compressors. Background Technology

[0002] Existing energy-saving adjustment technologies for air conditioning compressors mainly focus on adjusting set parameters or operating modes based on the current operating conditions of the equipment without altering the original control system, in order to balance energy consumption reduction and operational stability. In practice, maintenance personnel often manually operate the system after viewing the pressure, temperature, load status, start / stop information, and alarm content on the control panel, or, while retaining the original touch interface, use a camera to capture the screen display results, and then have a mechanical actuator complete the corresponding clicks according to a predetermined page path. For example, in the central air conditioning room of a commercial complex, air compressors or supporting compression units that have been in operation for many years often have issues such as closed controller protocols, inaccessible internal communication links, and inability to rewrite the original control program. At the same time, automatic adjustment must be limited to the existing human-machine interface to avoid affecting the continuity of cooling due to misoperation. In this case, the existing processing method is prone to a recurring phenomenon: when the screen continuously displays high load, high energy consumption, or frequent operation switching, although the automatic system can read the screen and execute the click, it cannot distinguish whether the phenomenon is due to setpoint deviation, terminal load fluctuation, or protection logic intervention because it only relies on the external display results. As a result, the parameters have often been changed, but the equipment operating rhythm has not improved accordingly. There may even be situations where the short-term change quickly returns to the original state, and repeated clicks are still difficult to reduce energy consumption. The technical problem this application aims to solve is: under the condition that the original controller and internal communication data of the air compressor cannot be accessed, and the screen can only be read by a camera and simulated by manual touch screen operation, how to determine whether the abnormal appearance presented on the screen can indeed be resolved by the current human-machine interface operation before formal adjustment, so as to avoid ineffective adjustment. Summary of the Invention

[0003] To overcome the aforementioned deficiencies in the prior art, embodiments of the present invention provide an intelligent control method for an air compressor in an air conditioner. This method first performs page positioning and reversible click pairing to solve abnormal phenomena in the continuous image of the control screen. Then, based on the return results of the trial action and the reverse click, it verifies whether the current abnormal phenomenon can be resolved by the current human-machine interface operation. After the verification is successful, it performs formal adjustment along the direction of the trial action to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent control method for an air compressor for air conditioning, comprising: S1. Obtain continuous images of the air compressor control panel, read the page identifier area, value display area, status display area and prompt display area frame by frame, extract the continuously occurring high load, high energy consumption or frequent switching phenomena from adjacent images, and output the current page, current display item and abnormal appearance. S2. Based on the position of the clickable control on the current page, select the single-step tap pair that is on the same page as the currently displayed item and can be directly executed in reverse tap to restore within the page. Determine the forward tap in the single-step tap pair as the trial action, and output the trial action and its corresponding reverse tap. S3. Control the robotic arm to perform a single tap on the screen position corresponding to the probe action, continuously acquire the continuous images after the tap, sequentially read the direction of the tapped item change, the order of the status display area change, and the content of the prompt display area change, and output the probe response. S4. Control the robotic arm to perform a tap on the screen position corresponding to the reverse tap, continuously acquire the continuous image after the reverse tap, and determine whether the tapped item returns to the state before the tap, whether the status display area retreats in the opposite direction to the trial action, and whether the prompt display area is restored to the content before the tap. If all three conditions are met, output that the appearance can be resolved; otherwise, output that the appearance cannot be resolved.

[0005] In a preferred embodiment, it further includes: S5. When the output is a resolvable phenomenon, control the robotic arm to perform formal adjustment sequentially along the tapping direction of the trial action, and repeatedly read the trial response and confirm the phenomenon's attribution after each tap, until the abnormal phenomenon weakens or disappears, and then output the adjustment result; when the output is an unresolvable phenomenon, stop further tapping on the current page and output the adjustment prohibition result.

[0006] In a preferred embodiment, S1 includes: S1-1. Obtain continuous images of the air compressor control panel, perform on-screen positioning and segmentation of each frame image according to the fixed border, title text position and control distribution position in the image, and output the partition images corresponding to the page identification area, numerical display area, status display area and prompt display area. S1-2. Perform content comparison at the same position on each partition image in adjacent frames, determine the frame group where the page identifier area remains unchanged and the numerical display area, status display area or prompt display area changes as the valid running frame group, and read the corresponding display content according to the field position in the valid running frame group, and output the current page and the current display item. S1-3. Perform continuous merging of the current display items in the effective running frame group according to their order of appearance. Determine the display content in the numerical display area that continuously rises or stays at a high position as a high load or high energy consumption phenomenon. Determine the display content in the status display area that repeatedly alternates between two adjacent running states as a frequent switching phenomenon and output the abnormal phenomenon.

[0007] In a preferred embodiment, S2 includes: S2-1. Based on the center coordinates, border range, control characters, and display coordinates of the current display item of each clickable control on the current page, construct the control position matrix, display item position matrix, and control character code table. Perform same-page coordinate projection, relative displacement decomposition, and shortest displacement sorting on the control position matrix and display item position matrix to form the forward candidate control sequence corresponding to the current display item. S2-2. For each forward candidate control in the forward candidate control sequence, read the corresponding control character, determine the reciprocal character control group based on the pairing relationship of the control characters in the reciprocal character code table, and then perform direction vector calculation, coordinate difference decomposition and angle relationship determination on the control center coordinates in each reciprocal character control group. Determine the reciprocal character controls with opposite directions that point to the same currently displayed item as the reverse candidate controls to form the initial click pair set.

[0008] In a preferred embodiment, S2 further includes: S2-3. For each initial point pair in the initial point pair set, construct a three-point association matrix between the forward control, the currently displayed item, and the reverse control. Perform determinant evaluation, symmetry axis reconstruction, and bidirectional path length summation on the three-point association matrix. When the path length from the forward control to the currently displayed item is equal to the path length from the reverse control to the currently displayed item, the dot product of the forward direction vector and the reverse direction vector is negative, and the symmetry axis passes through the center of the currently displayed item, retain the corresponding initial point pair as an invertible candidate point pair. S2-4. Perform one-to-one matching on reversible candidate points, construct a bipartite graph of forward and reverse controls, perform minimum cost matching, conflict edge deletion and connected component splitting on the bipartite graph, and when a forward control corresponds to only one reverse control and a reverse control corresponds to only one forward control, determine the corresponding reversible candidate point pair as the target single-step point pair, and determine the forward point in the target single-step point pair as the trial action and the reverse point as the corresponding reverse point and output it.

[0009] In a preferred embodiment, S3 includes: S3-1. Control the robotic arm to perform a single tap on the screen position corresponding to the trial action, acquire the image before the tap and the continuous images after the tap, perform character splitting, position alignment and bit-by-bit difference on the tapped item area in the image before the tap and the continuous images after the tap, determine the direction of change of the tapped item according to the order of increase and decrease of the code value of each character position, and output the sequence of the direction of change of the tapped item. S3-2. Perform state segmentation and timing numbering on the state display area in the continuous image after the point is pressed. Construct a state sequence table according to the occurrence time, duration of each state segment and adjacent connection relationship. Align the state sequence table with the change direction sequence of the point-pressed item according to the same frame time and output the change order of the state display area. S3-3. Perform character extraction, prompt fragment merging, and frame position write-back verification on the prompt display area in the continuous image after tapping. Construct a prompt change table according to the appearance order, disappearance order, and duration interval of the prompt content. Combine the prompt change table with the change direction sequence of the tapped item and the change order of the status display area and write it into the trial response record. Output the trial response.

[0010] In a preferred embodiment, S4 includes: S4-1. Obtain the baseline image before the trial action is executed, the trial response record after the trial action, and the continuous image after the reverse click. Extract the character bit sequence, status bit sequence, and prompt content bit sequence of the clicked item from the baseline image and the continuous image after the reverse click, respectively. Construct the baseline bit sequence matrix, the backtrack bit sequence matrix, and the frame sequence association table according to the frame sequence number. S4-2. Perform bit-by-bit difference, sign reversal, matrix multiplication, and singular value decomposition on the point-by-item character position sequence in the reference position matrix and the backtrack position matrix. Translate the point-by-item change direction sequence in the trial response record into a backtracking vector. Then project the point-by-item character position difference vector of each frame onto the backtracking vector. When the projected sign of each frame is consistent with the sign of the corresponding position of the backtracking vector, the point-by-item character position sequence of the last frame is identical bit-by-bit to the point-by-item character position sequence in the reference image, and the first frame number that satisfies the bit-by-bit identity is unique, output the point-by-item return result.

[0011] In a preferred embodiment, S4 further includes: S4-3. Perform sequence alignment, state transition expansion, and reverse adjacency graph construction on the state bit sequences in the reference bit sequence matrix and the backtrack bit sequence matrix. Write the state display area change order in the trial response record into the target backtrack sequence in reverse order. Then perform closed path search and path coverage solution on the reverse adjacency graph. Output the state backtrack result when the solved backtrack path covers the target backtrack sequence in order, the end state of the backtrack path is the same as the first state of the reference state bit sequence, and each state in the backtrack path is retained only once. S4-4. Perform character encoding, position numbering, hash mapping, and error correction encoding expansion on the prompt content bit sequence in the base position matrix and the back position matrix. Perform bit-by-bit correspondence verification on the prompt content encoding sequence after the reverse click and the base prompt content encoding sequence. When the character encoding, position number, hash value, and codeword sequence after error correction expansion are the same, output the prompt recovery result. S4-5. Write the results of the item return, status rollback, and prompt recovery into the consistent association graph. Perform two-part matching, conflict edge deletion, connected component partitioning, and closed loop verification on the consistent association graph. When the nodes corresponding to the three types of results are located in the same closed connected component, output a resolvable representation. When the nodes corresponding to the three types of results are not located in the same closed connected component, output an unresolvable representation.

[0012] In a preferred embodiment, S5 includes: S5-1. When the appearance attribution result is a resolvable appearance, read the trial action, the corresponding reverse tap, the trial response and the current page, construct the formal tap sequence according to the tap direction of the trial action, and write the page image, tap item change, state change sequence and prompt change content after each formal tap into the adjustment record table, and output the current round of adjustment record. S5-2. For the adjustment record table, perform point difference, state sequence alignment and prompt content write-back comparison of adjacent two rounds of adjustment records. When the point direction remains unchanged, the appearance attribute result is continuously resolvable appearance, and the display item corresponding to the abnormal appearance changes continuously in the same direction of change, retain the next round of formal point and output the continued adjustment instruction. Stop adding formal points when any item is not true.

[0013] In a preferred embodiment, S5 further includes: S5-3. Construct a chronological sequence of the appearance evolution of each round of adjustment records after retention. Perform segmented merging of abnormal appearances, verification of the beginning and end of continuous segments, and solution of the disappearance position. When the high load display, high energy consumption display, or frequent switching display corresponding to the abnormal appearance changes from continuous to no longer appearing, output the adjustment result. When the abnormal appearance still appears continuously and the appearance attribution result undergoes an irresolvable transformation, output the stop adjustment result. S5-4. When the representation attribution result is an unresolvable representation, the robotic arm is prohibited from continuing to click along the direction of the trial action on the current page, and the current page, trial action, trial response and unresolvable representation are written into the prohibition record table, and the prohibition result is output.

[0014] The technical effects and advantages of this invention are as follows: 1. This solution first performs a trial action, then performs a reverse tap, and combines verification of the tapped item's return to its original position, status rollback, and prompt restoration. This allows the solution to determine whether the current abnormal phenomenon can be resolved by the current page operation before formal adjustment, thereby relatively reducing the situation of ineffective adjustment based solely on external display results. 2. Perform on-screen positioning and segmentation, content comparison at the same position, and effective running frame group filtering on continuous images of the control screen to distinguish page switching ghosting, short-term refresh, and static images from actual running changes, which helps to improve the stability of the extraction results of the current page, the current display item, and abnormal appearances; 3. By constructing a forward candidate control sequence, a group of reciprocal character controls, reversible candidate click pairs, and target single-step click pairs, the click entry points that can be verified to revert back can be solved within the current page, which helps to relatively reduce the subsequent verification deviations caused by multiple configurations of one control, misselection of controls, and irreversible clicks. 4. Record the changes in the direction of the clicked items, the order of changes in the status display area, and the content of the prompt changes in a unified frame sequence after the trial action. Incorporate the numerical changes, status changes, and prompt feedback into the same trial response record, which will help improve the consistency of subsequent return verification and appearance attribution determination. 5. Perform consistency correlation graph verification on the result of the clicked item return, status rollback, and prompt recovery after the reverse click. This expands the single field recovery into a joint verification of multiple feedbacks, which helps to relatively suppress the situation of misjudging adjustability based solely on local interface changes. Attached Figure Description

[0015] Figure 1 This is a flowchart outlining the method steps of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Refer to the instruction manual appendix Figure 1 The present invention provides an intelligent control method for an air compressor used in air conditioning systems, comprising: S1. Obtain continuous images of the air compressor control panel, read the page identifier area, value display area, status display area and prompt display area frame by frame, extract the continuously occurring high load, high energy consumption or frequent switching phenomena from adjacent images, and output the current page, current display item and abnormal appearance. In this embodiment, the purpose of S1 is to first convert the continuous images of the control screen into calculable page partitioning results, field reading results, and abnormal phenomenon results, providing a unified input for subsequent single-step click pair solving, trial response reading, and phenomenon attribution determination. The mechanism is as follows: first, the fixed borders, title text positions, and control distribution positions of the control screen remain unchanged on the same page to complete the same-screen positioning and segmentation; then, the content changes in adjacent frames at the same position are used to identify the actual running frames; finally, a merge solution is performed on the abnormal display states that continuously appear in the actual running frames, avoiding misjudging screen jitter, page switching ghosting, or short-term refreshes as adjustable phenomena. This implementation process includes the following steps: The purpose of S1-1 is to stably extract the page identifier area, numerical display area, status display area, and prompt display area from continuous images, establishing a unified coordinate reference for subsequent comparison of content at the same location. The input is the continuous image of the air conditioner compressor control panel and the camera installation and calibration results. The camera installation and calibration results include at least the pixel coordinates of the four corners of the control panel in the image, the shooting tilt direction, and the effective display range of the screen. The processing actions include: first, performing grayscale conversion and edge extraction on each frame image to extract the outer frame line segments; then, extracting the four corners of the control panel according to the intersection of the horizontal and vertical borders; after the four corners are determined, performing perspective correction on the current frame to map the effective display range of the control panel to a unified page coordinate system; subsequently, within the unified page coordinate system, reading the position band of the line where the title text is located as the page identifier area, reading the position band where controls are densely distributed and contain variable numerical characters as the numerical display area, reading the position band where status characters or status icons are continuously arranged as the status display area, and reading the position band where short prompts, alarm prompts, or operation prompts are located as the prompt display area. The fixed border is taken from the border lines that appear in ten consecutive frames and have the same position number. The title text position is taken from the character block located above the fixed border and whose character length remains unchanged in ten consecutive frames. The control distribution position is taken from the area within the border where there are multiple border blocks or character blocks and whose position number remains unchanged in ten consecutive frames. The output is the page identifier partition image, numerical display partition image, status display partition image, and prompt display partition image corresponding to each frame image. Each partition image is written into the partition image table according to the frame number for S1-2 to read. The abnormal or missing handling is as follows: when a frame fails to solve the four corners at the same time, the unified page coordinate system confirmed in the previous frame is read to perform correction on the frame; when the title text position is obscured, the position number of the previously confirmed page identifier area is maintained; when the control distribution position is shifted as a whole, the frame is re-divided using the unified page coordinate system reconstructed by the fixed border to avoid partition drift. The purpose of S1-2 is to remove invalid frames from the partitioned images and retain only valid running frame groups that truly reflect operational changes, thereby ensuring that the current page and the current displayed item originate from the actual running process rather than static images or page transition images; the input is a partitioned image table; the processing actions include: first, reading the page identifier partitioned images in adjacent frames according to frame number, performing a bit-by-bit comparison of the title characters at the same position number, and determining that the page identifier area remains unchanged when the title character encoding sequence is completely consistent; then, performing a content comparison at the same position for the numerical display partitioned image, the status display partitioned image, and the prompt display partitioned image respectively, where the numerical display area extracts the character position sequence according to the field position number and performs a bit-by-bit comparison, the status display area extracts the status code sequence according to the status position number and performs a bit-by-bit comparison, and the prompt display area extracts the character encoding sequence according to the prompt box position number and performs a bit-by-bit comparison; when the page identifier area remains unchanged and at least one of the three types of partitions has content changes, the adjacent frames are written into the same valid running frame group; Subsequently, the corresponding display content of each valid running frame group is read according to the field position. The current page is determined by the title character encoding sequence in the page identifier area, and the current display item is composed of the field position number and field character content in the numerical display area, the status bit number and status code content in the status display area, and the prompt box position number and prompt content in the prompt display area. The output is the current page and the current display item, and the frame number range, page number, field position number, field content, status code content, and prompt content of each valid running frame group are written into the running field table for S1-3 to read. The abnormal or missing handling is as follows: when the page identifier area changes between two adjacent frames, the page cut frame between the two frames is marked as a transition frame and is not written into the valid running frame group; when there is a single character recognition missing in the numerical display area, the missing character position is filled with the same character in the previous and next frames of the same field; when the status display area and the prompt display area are both empty, the frame group is only retained if the numerical display area changes, otherwise the frame group is discarded. The purpose of S1-3 is to converge continuous display changes in the effective running frame group into abnormal phenomena that can be used for subsequent testing and adjustment, avoiding miswriting single fluctuations or one-time jumps as stable anomalies; the input is the running field table; the processing actions include: first, sorting the current display items according to the page number and field position number, and then continuously merging the display content at the same field position number according to the frame number; for the same field in the numerical display area, the character position sequence of each frame is sequentially converted into a code value sequence. When the code value sequence monotonically increases according to the frame number and there is no reverse code value in the middle, it is determined as continuously rising display content; when the code value sequence remains the same in multiple consecutive effective running frame groups, and the code value corresponds to the highest level display position or full display position of the field on the current page, it is determined as continuously high-level static display content; for the same status position number in the status display area, the status code is written into the status sequence table according to the frame number. When two adjacent status codes are written alternately in sequence and form two or more complete round trips, it is determined as repeatedly alternating display content; Subsequently, continuously rising display content and continuously high-position dwell display content are respectively classified into high-load phenomena or high-energy-consumption phenomena. The field correspondence is taken from the page field description table, which is derived from the field name and position correspondence in the control panel manual. Repeatedly alternating display content is classified into frequent switching phenomena. The output is an abnormal phenomenon, and the phenomenon number, phenomenon type, corresponding page number, corresponding field position number, start frame number, end frame number, and duration of the abnormal phenomenon are written into the phenomenon record table for S2 to read. The abnormal or missing handling is as follows: when a single frame jump occurs in the continuous merging process of the same field, if the code values ​​of the two frames before and after the jump frame are the same, the jump frame is merged into the consecutive segments before and after. When a third status code is inserted for the same status bit number, the inserted status is first cut out separately according to the status sequence table, and then only the two status codes before and after the insertion are checked back and forth. When the same page shows both high-load phenomena and high-energy-consumption phenomena, two phenomenon numbers are retained respectively and written separately according to the field position number. Through the above processing, S1 first completes the unified positioning of the control screen image, then completes the filtering of the real running frame, and finally completes the formation of abnormal appearances, so that subsequent steps no longer directly face the original continuous image, but read the structured result that already has page number, field position number, display content and appearance number, thereby improving the stability of single-step click pair solution and appearance attribution determination. In practical applications: When the air compressor control panel in the central air conditioning room of a commercial complex continuously displays exhaust pressure, operating frequency, loading status, and alarm prompts, the camera first performs perspective correction on the outer frame of the control panel and cuts out the page identification area, numerical display area, status display area, and prompt display area. Then, it removes the page switching transition frames from the continuous frames, retaining only the valid operating frame groups where the title characters remain unchanged and the values ​​or statuses actually change. When the operating frequency field continuously increases in frame number in multiple consecutive valid operating frame groups, and the loading and unloading states form two or more complete round trips in the status sequence table, high load appearance and frequent switching appearance are generated and written to the appearance record table. Subsequently, the trial action and reverse click can be solved based on the appearance record table.

[0018] S2. Based on the position of the clickable control on the current page, select the single-step tap pair that is on the same page as the currently displayed item and can be directly executed in reverse tap to restore within the page. Determine the forward tap in the single-step tap pair as the trial action, and output the trial action and its corresponding reverse tap. In this embodiment, the purpose of S2 is to extract the target single-step click pair from the current page that can act on the currently displayed item and complete the backtracking verification within the same page, thereby providing a unique, reversible, and verifiable operation entry point for subsequent trial response reading and appearance attribution determination. The mechanism is as follows: first, the spatial relationship between the clickable control and the currently displayed item is fixed in the same page coordinate system; then, an initial click pair is formed based on the reciprocal relationship of the control characters and the pointing relationship from the control to the currently displayed item; subsequently, a three-point correlation matrix is ​​used to verify the backtracking symmetry between the forward control, the current displayed item, and the reverse control; finally, one-to-one matching eliminates conflicts such as one control with multiple matches, multiple controls with one match, and parallel conflicts, retaining only the unique target single-step click pair. This implementation process includes the following steps: The purpose of S2-1 is to form a sequence of forward candidate controls within the current page that can be used for subsequent pairing solutions. Its mechanism involves first mapping all clickable controls and currently displayed items on the current page to the same page coordinate system, and then filtering out the controls most likely to affect the currently displayed item based on their positional relationships. The inputs are the current page, the currently displayed item, the partition image table, and the running field table. The current page includes at least the page number and the page identifier area position number, and the currently displayed item includes at least the field position number, field content, field type, and display coordinates. The processing actions include: first reading each clickable control from the partition image corresponding to the current page... Based on the border range of the control, the average of the x-coordinates of the top-left corner and the bottom-right corner of the border is taken as the center x-coordinate, and the average of the y-coordinates of the top-left corner and the bottom-right corner of the border is taken as the center y-coordinate. Thus, the center coordinates of each clickable control are solved. Then, a control position matrix is ​​constructed with the control number as the row number and the center x-coordinate, center y-coordinate, left border, right border, top border, and bottom border as the column number. A display item position matrix is ​​constructed with the current display item number as the row number and the display center x-coordinate and display center y-coordinate as the column number. The characters read within the border range of each control are written into the control character code table in character position order. Subsequently, a same-page coordinate projection is performed on the control position matrix and the display item position matrix, ensuring that the x and y coordinates of both matrices fall within the unified position numbering system of the current page. After projection, for each clickable control, the difference between its center x-coordinate and the center x-coordinate of the current display item, and the difference between its center y-coordinate and the center y-coordinate of the current display item are calculated, and these differences are written as relative displacement pairs. Then, the controls are sorted according to the absolute values ​​of their x-coordinate differences, the absolute values ​​of their y-coordinate differences, and their position numbers, forming a sequence of forward candidate controls corresponding to the current display item. The outputs are the control position matrix and the display item position matrix. The system generates a control character code table and a sequence of forward candidate controls, and writes the sequence of forward candidate controls into a control candidate table for S2-2 to read. The handling of anomalies or missing characters is as follows: when a control character cannot be fully recognized, the center coordinates and border range of the corresponding control are still retained, and the control is marked as a characterless control, subsequently only participating in position calculation and not inverse character pairing; when overlapping borders exist on the same page, the border with the smaller border area and its center coordinates located inside the larger border is taken as the actual clickable control border; when the coordinates of the currently displayed item are missing, the display coordinates of the same field position number in the previous valid running frame group are read from the running field table and filled in. The purpose of S2-2 is to extract the corresponding reverse candidate controls from the forward candidate control sequence. Its working mechanism is to utilize the mutual inverse relationship of control characters and the spatial relationship between the controls on both sides pointing to the same currently displayed item to first form an initial click pair that can be rolled back. The input quantities are the control candidate table, the control character code table, the currently displayed item, and the mutual inverse character code table. The mutual inverse character code table comes from the control character correspondence in the control screen operation instructions and the character pairing results in the historical click records on the same page, and includes at least the character pair number, the forward character code, and the reverse character code. The processing actions include: first, reading the control character code of each forward candidate control one by one according to the forward candidate control sequence; then, searching for the reverse character code that is paired with the control character code in the mutual inverse character code table; and reading the control containing the corresponding reverse character code from all clickable controls on the current page to form a mutual inverse character control group. Subsequently, for each group of mutually inverse character controls, the direction vectors from the forward candidate control to the current displayed item and from the mutually inverse character control to the current displayed item are calculated based on the center coordinates of the controls. The direction vectors consist of the horizontal difference and the vertical difference. Then, coordinate difference decomposition is performed on the two direction vectors to obtain the horizontal component sign and the vertical component sign. Based on the angle relationship between the two vectors, it is determined whether they are located on both sides of the current displayed item. Specifically, the inner product operation is performed on the forward direction vector and the mutually inverse direction vector. When the inner product is negative and the endpoints of both vectors are the center coordinates of the current displayed item, it is determined that the two directions are opposite and both point to the same current displayed item. For mutually inverse character controls that satisfy this relationship, they are identified as reverse candidate controls, and the forward candidate controls and the corresponding reverse candidate controls are paired and written into the initial click pair set. The output consists of a set of reciprocal character controls and an initial set of click pairs. The initial set of click pairs is written to the initial set of click pairs table for S2-3 to read. The handling of exceptions or missing controls is as follows: when a forward candidate control corresponds to multiple reciprocal character controls, all are retained and added to the initial set of click pairs table, and are not resolved in this step; when there is no corresponding reverse character code in the reciprocal character code table, the forward candidate control is marked as an irreversible control and removed; when the center coordinates of a reciprocal character control coincide with the center coordinates of the currently displayed item, the reciprocal character control is judged as an obstructing control and is not included in the initial set of click pairs. The purpose of S2-3 is to perform a backtracking symmetry check on the initial point-click pairs, thereby filtering out point-click pairs that, although having a reciprocal character relationship, cannot stably return to the currently displayed item. Its mechanism is to use the spatial geometric relationship between the forward control, the currently displayed item, and the reverse control to check whether the point-click pair possesses single-step reversibility. The input quantities are the initial set of point-click pairs, the control position matrix, and the display item position matrix. The processing actions include: first, reading the center coordinates of the forward control, the center coordinates of the current displayed item, and the center coordinates of the reverse control for each initial point-click pair in the initial point-click pair set; and... Construct a three-point correlation matrix with three points as three rows and the x-coordinate, y-coordinate, and position number as three columns; then perform determinant calculation on the three-point correlation matrix to determine whether the three points are in a valid two-dimensional relationship; then, using the midpoint of the line connecting the center coordinates of the forward control and the center coordinates of the reverse control as the pivot point, construct a symmetry axis that passes through this pivot point and is perpendicular to the line connecting the forward control and the reverse control; after constructing the symmetry axis, calculate the path length from the forward control to the currently displayed item and the path length from the reverse control to the currently displayed item, respectively, and take the sum of the absolute values ​​of the horizontal and vertical differences; Next, calculate the dot product of the forward direction vector and the reverse direction vector, and determine whether the axis of symmetry passes through the center of the current displayed item. When the two path lengths are equal, the dot product is negative, and the axis of symmetry passes through the center of the current displayed item, retain the corresponding initial point pair as a reversible candidate point pair. The output is the reversible candidate point pair, and the forward control number, the current displayed item number, the reverse control number, the path length, the dot product value, and the axis of symmetry number are written into the reversible point pair table for S2-4 to read. The abnormal or missing handling is as follows: when the determinant of the three-point correlation matrix is ​​zero and the coordinates of the three point centers are collinear, it is not directly eliminated, but it continues to determine whether the axis of symmetry coincides with the center of the current displayed item. If it coincides, it is retained; if it does not coincide, it is eliminated. When the same initial point pair cannot form a complete three-point correlation matrix due to missing coordinates, the coordinates of the center of the control with the same number in the previous frame are read to fill in the missing coordinates. When the path lengths are equal but the dot product of the forward direction vector and the reverse direction vector is not negative, it is determined that the point pair only has positional symmetry and not directional backtracking and is eliminated. The purpose of S2-4 is to extract the unique target single-step click pair from the reversible candidate click pairs. Its mechanism is to eliminate one-to-one matching between forward and reverse controls, multiple matching of one control, one matching of multiple controls, and local conflicts, so that subsequent trial actions and reverse clicks have unique inputs. The input is a table of reversible click pairs. The processing actions include: first, constructing a bipartite graph with the forward control number as the left node and the reverse control number as the right node, and writing each reversible candidate click pair as a connecting edge into the bipartite graph; then calculating the cost value for each connecting edge. The cost value is composed of the path length from the forward control to the current displayed item, the path length from the reverse control to the current displayed item, and the center distance from the forward control to the reverse control, forming a cost triplet in sequence. Then, minimum cost matching is performed on the connecting edges according to the lexicographical order of the cost triples. First, the first edge corresponding to each forward control is retained, and then it is checked whether each reverse control is connected by only one retained edge. In the case where the same forward control is connected to multiple reverse controls, the remaining edges except for the first retained edge are deleted. In the case where the same reverse control is connected to multiple forward controls, the edge with the first cost triple in the order is retained and the remaining edges are deleted. After the edge deletion is completed, connected component splitting is performed on the remaining relations. When a connected component contains only one forward control node and one reverse control node, the invertible candidate point pair corresponding to the connected component is determined as the target single-step point pair, and the forward point is determined as the trial action and the reverse point is determined as the corresponding reverse point. The output consists of the target single-step click pair, the trial action, and the corresponding reverse click, and these three are written into the click solution result table for S3 to read. Abnormal or missing data is handled as follows: when there is no connected component in the bipartite graph containing only one forward control node and one reverse control node, the click pair solution for the current page is stopped, and the current page is written into the prohibited solution page table; when multiple connected components meet the conditions, the group with the first forward control position number is retained as the target single-step click pair, and the remaining groups are written into the spare click pair table; when the robotic arm reachability check finds that any control in the target single-step click pair exceeds the robotic arm's click range, the next group in the spare click pair table is retrieved for further check. Through the above processing, S2 gradually converges the clickable controls in the current page from their original position set into a unique single-step click pair. This ensures that the trial action can be applied to the currently displayed item and that the reverse click can be performed to perform back verification within the same page. This establishes a reversible, unique, and computable operational foundation for subsequent trial response reading and resolvable appearance judgment. In practical applications: When a set value field displayed on the current page of the air conditioner compressor control panel is in the position corresponding to an abnormal phenomenon, the system first calculates the center coordinates of each clickable control according to the control border, and then forms a forward candidate control sequence according to the center coordinates of the current display item. Subsequently, it reads the control group whose characters are inverses of the forward candidate controls from the reciprocal character code table, and filters out the controls located on the opposite side of the current display item and pointing to the same display coordinates as reverse candidate controls. On this basis, a three-point association matrix is ​​constructed from the forward control, the current display item, and the reverse control, retaining reversible candidate point-press pairs with equal path lengths and symmetry axes passing through the center of the current display item. Finally, through one-to-one matching of the two-part relationship graph, only the unique target single-step point-press pair is retained, and the forward point-press in the target single-step point-press pair is written as a trial action, and the reverse point-press is written as the corresponding reverse point-press. The robotic arm can then perform trial and backtracking verification accordingly.

[0019] S3. Control the robotic arm to perform a single tap on the screen position corresponding to the probe action, continuously acquire the continuous images after the tap, sequentially read the direction of the tapped item change, the order of the status display area change, and the content of the prompt display area change, and output the probe response. In this embodiment, the purpose of S3 is to transcribe the interface feedback after the probing action into a probing response that can be directly used for backtracking verification. The process is as follows: first, determine the direction of change of the clicked item; then, determine the order of change of the status display area; finally, determine the content of change of the prompt display area, and write the three into the probing response record in the same frame order. This implementation process includes the following steps: The purpose of S3-1 is to determine the actual direction of the trial action on the clicked item; the input quantities are the trial action, the image before the click, and the continuous images after the click; the processing actions are as follows: control the robotic arm to perform a click on the screen position corresponding to the trial action, extract the clicked item area according to the field border range of the currently displayed item, perform character splitting on the clicked item area in the image before the click and the continuous images after the click, perform positional alignment according to the character position number, perform bit-by-bit difference by subtracting the character code of the previous frame from the character code of the next frame, and read the first non-zero difference value from the high bit to the low bit of the character position. When the first non-zero difference is positive, it is determined to be forward increment; when the first non-zero difference is negative, it is determined to be forward decrement; when the difference of each character position is zero, it is determined to be unchanged, and a sequence of point-by-item change direction is formed according to the frame number; the output is the sequence of point-by-item change direction and is written into the point-by-item change table for subsequent reading; the abnormal or missing handling is as follows: when a character position is missing, the character code of the same character position in the previous frame and the next frame is read and written; if the two are different, the frame is recorded as an invalid frame; when a page jump occurs in the continuous images after clicking, the current test response reading is stopped and the test action is recorded as a page jump test; The purpose of S3-2 is to determine the sequential relationship of state changes after a trial action; the inputs are the continuous images after the tap and the tap change table; the processing actions are as follows: capture the images of each frame of the state display area according to the position number of the state display area, extract the state code for the same state position number, traverse the state code sequence according to the frame number, merge adjacent frame state codes into the same state segment when they are the same, close the previous state segment and open the next state segment when the state code changes, write the timing number according to the first appearance frame number of each state segment, and construct the state based on the start frame number, end frame number, duration frame number and the numbers of the adjacent state segments. The sequence list is then used as an index to align the state sequence list with the point-by-item change direction sequence, forming the state display area change order. The output is the state display area change order, and the state segment number, state code, start frame number, end frame number, duration frame number, and corresponding point-by-item change direction are written into the state sequence list for subsequent reading. Abnormal or missing status code handling is as follows: when a single frame has an isolated status code and the status codes of the preceding and following frames are the same, the single frame is merged into the preceding and following state segments. When the state display area is empty, an empty status code is written, and an empty state segment is only retained when two or more consecutive empty status codes appear. The purpose of S3-3 is to determine the prompt feedback and its duration after a trial action; the inputs are the continuous images after the tap, the tap change table, and the state sequence table; the processing actions are as follows: capture the prompt display area images of each frame according to the prompt display area position number, perform character extraction on the prompt content of each frame to form a prompt content encoding sequence, traverse the prompt content encoding sequence according to the frame number, merge adjacent frames with the same encoding sequence under the same prompt box position number into the same prompt segment, and start the next prompt segment when the encoding sequence changes; then perform a frame position write-back check on each prompt segment, and determine a complete prompt write when the frame before the first frame is empty and the frame after the last frame is empty, and the encoding sequence of the frame before the first frame is the same as the first frame of the current prompt segment. When the previous prompt segment is merged, if the encoding sequence of the last frame after the current prompt segment is the same as that of the last frame of the current prompt segment, the current prompt segment is extended; then a prompt change table is constructed according to the order of appearance, disappearance and duration of the prompt content, and the prompt change table, the change direction sequence of the clicked item and the change sequence of the status display area are written into the trial response record according to the same frame number; the output is the trial response, and the trial response record table is made available for S4 to read; the abnormal or missing handling is as follows: when the prompt content extraction fails, the prompt box position number is retained and written into the empty encoding sequence; when two adjacent frames differ by only one character position and have the same position number, they are retained as the same prompt segment and that character position is recorded as the character position to be checked; when the prompt display area is always empty, an empty prompt segment is written. Through the above processing, the test response output by S3 simultaneously includes the direction of the clicked item change, the order of state changes, and the content of the prompt change. All three share the same frame sequence and the same test action number. This can be directly used for subsequent verification of the clicked item return, state rollback, and prompt restoration after a reverse click. In practical applications: After the robotic arm performs a test action on the raised control on the current page, the system first determines the direction of the clicked item change when the set value character position changes from thirty-five to thirty-six. Then, it determines the order of state changes in standby, loading, and steady state. Subsequently, it determines the appearance frame number, disappearance frame number, and duration frame number of the parameter update prompt and writes all three into the same test response record for direct reading in subsequent return determination.

[0020] S4. Control the robotic arm to perform a tap on the screen position corresponding to the reverse tap, continuously acquire the continuous image after the reverse tap, and determine whether the tapped item returns to the state before the tap, whether the status display area retreats in the opposite direction to the trial action, and whether the prompt display area is restored to the content before the tap. If all three conditions are met, output that the appearance can be resolved; otherwise, output that the appearance cannot be resolved. In this implementation, the purpose of S4 is to perform rollback checks on each of the changes in clicked items, states, and prompts caused by the trial action, based on the continuous images after the reverse click, and to converge the three types of check results into a unique representation attribution result. The process is as follows: first, extract the unified positional object from the baseline image, the trial response record, and the continuous images after the reverse click; then, perform the clicked item rollback check, state rollback check, and prompt restoration check respectively; finally, merge the three types of check results using a consistency association graph, and output the resolvable or non-resolvable representation. This implementation process includes the following steps: The purpose of S4-1 is to establish a unified positional reference and frame sequence correspondence for the subsequent three types of rollback checks. The input quantities are the reference image before the trial action is executed, the trial response record, and the continuous images after the reverse click. The processing actions are as follows: First, according to the current display item position number in the trial response record, the clicked item area in the reference image and the continuous images after the reverse click is extracted, and the clicked item character position sequence is extracted according to the character position number. Then, the status position sequence is extracted according to the status display area position number. The status position sequence consists of the status position number and the corresponding status code. Then, the prompt content position sequence is extracted according to the prompt display area position number. The prompt content position sequence consists of the prompt box position number, the character position number, and the character code. Subsequently, according to the frame sequence number, the clicked item character position sequence, the status position sequence, and the prompt content position sequence in the reference image are written into the reference position sequence matrix. The corresponding sequence in the continuous images after the reverse click is written into the rollback position sequence matrix according to the frame number order. And a frame sequence association table is constructed using the frame number in the trial response record, the frame number of the continuous images after the reverse click, and the sequence position number. The reference position sequence matrix contains only the position sequence data of the frame before the trial action is executed, and the backtrack position sequence matrix contains the position sequence data of all consecutive frames after the reverse tap. Each row in the frame sequence association table corresponds to a frame number after the reverse tap and its corresponding position sequence position in the tapped item area, status display area, and prompt display area. The outputs are the reference position sequence matrix, the backtrack position sequence matrix, and the frame sequence association table, and are written into the backtrack verification table for S4-2 to S4-5 to read. The abnormal or missing handling is as follows: when a character position is missing in a frame of the consecutive images after the reverse tap, the character code of the same position number in the adjacent frames before and after is read and written. If the codes of the adjacent frames before and after are different, the frame is marked as a missing frame and is not used as the first backtrack frame. When the status position is missing, an empty status code is written. When the prompt content position is missing, the prompt box position number is retained and an empty character code is written. The purpose of S4-2 is to determine whether the clicked item has returned to the character state before the trial action was executed. The inputs are the reference position sequence matrix, the backtrack position sequence matrix, and the trial response record. The processing actions are as follows: First, read the clicked item character position sequence in the reference position sequence matrix and the clicked item character position sequence corresponding to each frame in the backtrack position sequence matrix. Perform bit-by-bit difference on the same character position number. The difference method is to subtract the reference frame character code from the backtrack frame character code. Then, perform sign reversal on the clicked item change direction sequence in the trial response record, converting forward increments to reverse subtractions, and converting forward subtractions to forward decrements. To achieve reverse increment, the invariant is converted to invariant and a reverse verification vector is formed according to the character position number. Then, the point-by-item character position difference vector of each back frame is multiplied by the reverse verification vector to form a frame-by-frame projection matrix. Singular value decomposition is then performed on the frame-by-frame projection matrix to extract the principal component direction of the corresponding character position change. After the principal component direction is determined, it is determined whether the projection symbol of each frame is consistent with the symbol of the corresponding position of the reverse verification vector according to the character position order. It is then further determined whether the point-by-item character position sequence of the last frame is identical to the point-by-item character position sequence in the reference image. When both conditions are met, continue searching the backtracking position matrix in ascending order of frame number for the first frame number that satisfies the condition of being identical digit by digit. If the frame number appears only once, output that the point-to-item backtracking result is true, and write the frame number as the point-to-item backtracking frame number. If there are multiple intervals with the same frame number, retain the frame with the smallest frame number as the point-to-item backtracking frame number, and write the rest as repeated backtracking frames. The output is the point-to-item backtracking result and the point-to-item backtracking frame number, and writes them into the point-to-item backtracking table for S4-5 to read. The handling of exceptions or missing characters is as follows: any character position When the difference cannot be calculated, the corresponding character position is written as an invalid character position and is not included in the principal component solution; when all character positions are invalid character positions, the result of the click item return is recorded as invalid; when no identical result appears in all consecutive frames after the reverse click, the result of the click item return is recorded as invalid; in practical applications, when the trial action changes the set value from thirty-six to thirty-seven, if the third frame in the consecutive images after the reverse click displays thirty-six again, and the third frame is the first identical frame, then the third frame is determined as the click item return frame number; The purpose of S4-3 is to determine whether the status display area has completed the rollback in the opposite direction of the trial action. The inputs are the reference bit sequence matrix, the rollback bit sequence matrix, and the trial response record. The processing actions are as follows: First, read the status bit sequence in the reference bit sequence matrix and the status bit sequence in the rollback bit sequence matrix, and perform sequence alignment for the same status bit number. The sequence alignment uses the status bit number and the frame number as dual indices. Then, perform state transition expansion according to the adjacent frame change relationship of the status bit sequence in the rollback bit sequence matrix, and write the previous status code to the next status code as a directed transition. The system constructs a reverse adjacency graph with status codes as nodes and transition edges as connections. Then, it reads the change order of the status display area in the trial response record and writes it into the target backtrack sequence in reverse chronological order. The first state in the target backtrack sequence is the last state of the trial response record, and the last state is the first state of the trial response record. After the reverse adjacency graph is constructed, a closed path search is performed starting from the node corresponding to the status code in the first frame of the backtrack position matrix. The search rule is to gradually expand the path in the direction of increasing frame number and retain only the paths that can sequentially cover the target backtrack sequence. Next, perform path coverage calculation on the retained paths. If a path can sequentially cover all the status codes in the target backtrack sequence, and the end state of the path is the same as the first state of the base state bit sequence, and only the first occurrence position of each status code in the path is retained, then the output state backtrack result is valid, and the corresponding path is written as the backtrack path; otherwise, the output state backtrack result is invalid. The output quantities are the state backtrack result and the backtrack path, and are written into the state backtrack table for S4-5 to read. The abnormal or missing handling is as follows: when a frame status code is an empty status code, the frame is not written into the target backtrack sequence coverage calculation, and only its frame number is retained for path continuity verification; when multiple state segments of the same status code appear consecutively, only the first segment is retained in the backtrack path; when there is no path in the reverse adjacency graph that can cover the target backtrack sequence, the state backtrack result is recorded as invalid. The purpose of S4-4 is to determine whether the prompt display area has been restored to the prompt state before the trial action was performed; the input quantities are the reference position sequence matrix and the backtrack position sequence matrix; the processing actions are as follows: first, read the prompt content bit sequence in the reference position sequence matrix and the prompt content bit sequence in each frame of the backtrack position sequence matrix, perform character encoding on the character positions under the same prompt box position number, and the character encoding is taken from the standard encoding in the character code table; then write the position number sequence according to the prompt box position number and the character position number; then concatenate the character encoding sequence with the position number sequence and perform hash mapping to form the reference prompt hash value and the prompt hash values ​​of each backtrack frame; Based on this, error correction encoding expansion is performed on the character encoding sequence. The error correction encoding rules are taken from the fixed codeword rules in the preset configuration, which is derived from the installation and debugging records of the prompt character display format. Then, the prompt content encoding sequence after the reverse click is checked against the baseline prompt content encoding sequence bit by bit. The check order is: the character encoding is the same, the position number is the same, the hash value is the same, and the codeword sequence after error correction expansion is the same. When all four items are true in order, the prompt recovery result is output as true, and the first true frame is written as the prompt recovery frame number. If any item is false, the prompt recovery result is recorded as false. The output is the prompt recovery result and the prompt recovery frame number, which are written to the prompt recovery table for S4-5 to read. The abnormal or missing handling is as follows: when the prompt display area is empty, the empty prompt character encoding sequence and the empty position number sequence are written together into the check object. When a single character is missing in the prompt character, the character position is restored with the codeword sequence after error correction expansion. If the restoration fails, the prompt recovery result is recorded as false. When the position of the prompt box changes, even if the character encoding is the same, it is determined that the prompt recovery result is false. The purpose of S4-5 is to unify the results of item return, status rollback, and prompt recovery into a unified representation attribution result. The inputs are the item return table, status rollback table, and prompt recovery table. The processing actions are as follows: First, the item return, status rollback, and prompt recovery results are written as three types of nodes in the consistency association graph. Each node has a trial action number, current page number, and corresponding return frame number or rollback path number. Then, the connection edges between nodes are established based on the same trial action number, the same current page number, and the corresponding frame order, forming a consistency association graph. Subsequently, a two-part matching is performed on the consistency association graph. First, the item return node and the status rollback node are matched, and then the status rollback node and the prompt recovery node are matched. After the matching is completed, conflicting edges that cannot form a two-sided correspondence are deleted. Then, the remaining nodes are partitioned into connected components, and a closed loop check is performed in each connected component. The closed loop check rule is that all three types of nodes exist and the edges between the three types of nodes are closed at both ends. When the nodes corresponding to the click-to-return result, status rollback result, and prompt recovery result are located in the same closed connected component, the output is a resolvable representation; when the nodes corresponding to the three types of results are not located in the same closed connected component, the output is an unresolvable representation; the output is the representation attribution result and is written to the representation attribution table for S5 to read; the abnormal or missing handling is as follows: if any of the three types of results is not true, the corresponding node is directly recorded as a disconnected node and participates in the connected component division; when there are multiple closed connected components in the consistent association graph, the closed connected component containing the minimum value of the click-to-return frame number is retained, and the remaining closed connected components are recorded as parallel components; when there are no closed connected components in the consistent association graph, the representation attribution result is recorded as an unresolvable representation; Through the above processing, S4 completes three types of rollback checks: item return, status rollback, and prompt restoration. It then merges the three types of results using a consistency association graph, avoiding the reliance on a single field change to determine the appearance's attribution. This improves the stability and uniqueness of the output of resolvable and non-resolvable appearances. In practical applications: when a trial action causes the set value to increase, the state sequence to change from standby to loading to steady state, and the prompt content to "parameter updated," the robotic arm performs a reverse press. The system first solves the rollback position matrix to find the first frame number of the character position sequence before the press, then solves the reverse adjacency graph to find the unique rollback path from steady state to loading to standby. Subsequently, it checks whether the prompt content returns to the empty prompt state before the press. If the three types of results are located in the same closed connected component in the consistency association graph, it is determined that the current abnormal appearance can be resolved by the current page operation; if the three types of results cannot be closed correspondingly, it is determined that the current abnormal appearance cannot be resolved by the current page operation.

[0021] S5. When outputting a resolvable phenomenon, control the robotic arm to perform formal adjustment sequentially along the direction of the trial action, and repeatedly read the trial response and confirm the phenomenon's attribution after each click, until the abnormal phenomenon weakens or disappears, and then output the adjustment result; when outputting an unresolvable phenomenon, stop further clicking on the current page and output the adjustment prohibition result. In this implementation, the purpose of S5 is to perform controlled formal adjustment on resolvable appearances and to perform adjustment prohibition processing on non-resolvable appearances after the appearance attribution result has been clarified, and to write the adjustment progress, cessation, and adjustment prohibition results into traceable records respectively; the process is as follows: first, under the condition of resolvable appearances, a formal click sequence is generated according to the direction of the trial action to form the current round of adjustment record; then, the adjustment records of the two adjacent rounds are checked for continued adjustment; subsequently, based on the adjustment records of each round, an appearance evolution sequence is constructed and the adjustment result or the adjustment cessation result is output; if the appearance attribution result is a non-resolvable appearance, then the current page is directly stopped from continuing to click and the adjustment prohibition result is output; this implementation process includes the following steps: The purpose of S5-1 is to transcribe the trial action into a continuously executable formal tap sequence when a resolvable appearance is established, and to fix the interface feedback after each round of taps as a structured adjustment record. The input quantities are appearance attribution result, trial action, corresponding reverse tap, trial response, and current page. The processing action is as follows: first, determine whether the appearance attribution result is a resolvable appearance. If so, read the control number, control center coordinates, and tap direction from the trial action, and repeatedly write them into the formal tap sequence according to the tap direction of the trial action. Each item in the formal tap sequence contains the current page number, formal tap round, control number, and tap direction. Subsequently, the robotic arm is controlled to perform one click in each round according to the formal click sequence. After each round of clicks, the page image is captured, and the click item changes, state change sequence, and prompt change content are extracted according to the reading method of S3. Then, the current page number, formal click round, page image number, click item change direction sequence, state change sequence, prompt change content, corresponding reverse click number, and appearance attribution result are written into the adjustment record table to form the current round adjustment record. The output is the current round adjustment record, and the current round adjustment record is written into the adjustment record table for S5-2 to read. The abnormal or missing handling is as follows: when the page jumps after the robotic arm clicks, the subsequent formal clicks are stopped and the round record is marked as a page jump adjustment record; when the page image is not successfully captured after the click, the first valid image in the consecutive images of the same round is read and written; when the click item change direction sequence remains unchanged after the click, the round adjustment record is still retained and marked as a no-change adjustment record. The purpose of S5-2 is to determine whether the formal clicks have the conditions to continue in the same direction. Its mechanism is to use the continuity of click item changes, state changes, and prompt changes between two adjacent rounds of adjustment records to filter out the subsequent formal clicks that can be retained. The input quantities are the adjustment record table, the appearance attribution result, and the abnormal appearance. The processing actions are as follows: First, read the adjustment records of the two adjacent rounds from the adjustment record table according to the formal click round. Perform a bit-by-bit difference on the character sequence of the click items in the two adjacent rounds of adjustment records to determine whether the direction of the click item change in the next round is consistent with the direction of the click item change in the previous round. Then, perform state segment number alignment on the state change sequence in the two adjacent rounds of adjustment records to determine whether the state change sequence in the next round continues to expand along the state change sequence of the previous round without introducing new state codes. Finally, perform position number correspondence and content write-back comparison on the prompt content in the two adjacent rounds of adjustment records to determine whether the prompt content in the next round still corresponds to the same prompt box position on the current page. After the above verification is completed, the appearance attribution results of the corresponding round are read to determine whether the appearance attribution results of two adjacent rounds are continuous and thus resolvable appearances; finally, the display item number corresponding to the abnormal appearance is read to determine whether the display item changes continuously in the same direction in the adjustment records of two adjacent rounds. Specifically, the sign of the character position difference of the display item in the two adjacent rounds is compared bit by bit, and when the sign of the first non-zero difference is consistent, it is determined to be a continuous change in the same direction; the output is a continuation adjustment command or a stop adding formal press command, and the reserved formal press of the next round is written into the continuation adjustment record table for S5-3 to read; The handling of frequent or missing events is as follows: if there is no change in the adjustment record in either of the two adjacent rounds, stop adding more formal clicks; if the appearance attribute result in either of the two adjacent rounds becomes an unresolvable appearance, stop adding more formal clicks; if the state change sequence cannot be aligned, directly output the command to stop adding formal clicks; in practical applications, if the set value changes from 36 to 37 after the first formal click, and the set value changes from 37 to 38 after the second formal click, and the state change sequence of both rounds is the steady state after loading, and the prompt content is in the same prompt box, then retain the second formal click and output the command to continue adjusting; The purpose of S5-3 is to determine whether abnormal phenomena have subsided or whether adjustment should be stopped based on the retained adjustment records of each round. Its mechanism is to expand the results of multiple rounds of adjustment into a phenomenon evolution sequence according to time, and then output the final adjustment result based on the continuous appearance and subsidence position of abnormal phenomena. The input quantities are the continuation adjustment record table, adjustment record table, abnormal phenomena, and phenomenon attribution results. The processing actions are as follows: First, read the retained adjustment records of each round according to the official point, round number, and frame number. Write the corresponding display items, state change sequence, prompt change content, and phenomenon attribution results of abnormal phenomena in each round of adjustment records into a phenomenon evolution sequence in chronological order. Then, perform abnormal phenomenon segmentation and merging on the phenomenon evolution sequence, and merge records with the same abnormal phenomenon number that appear consecutively into a phenomenon continuous segment. Subsequently, perform head-to-tail correspondence verification on each phenomenon continuous segment. The head is taken from the round and frame number of the first appearance of the phenomenon continuous segment, and the tail is taken from the round and frame number of the last appearance of the phenomenon continuous segment. It is also determined whether the abnormal phenomenon numbers between the head and tail are consistent. After the initial and final correspondences are established, the process of determining the fading position continues. Specifically, starting from the last round of adjustment records, a reverse chronological search is performed, and the first round in which the corresponding abnormal phenomenon number no longer appears is recorded as the fading position. When the high load display, high energy consumption display, or frequent switching display corresponding to the abnormal phenomenon changes from continuous occurrence to no longer occurrence, the adjustment result is output, and the fading position is written into the adjustment result table. When the abnormal phenomenon still occurs continuously and the phenomenon attribution result undergoes an irreversible transformation in subsequent rounds, the stop adjustment result is output, and the transformation round is written into the stop adjustment record. Record the data; the output is either the adjustment result or the stop adjustment result, and write it into the adjustment result table and the stop adjustment record table respectively; the handling of anomalies or missing data is as follows: when there is a skipped page adjustment record in the image evolution sequence, the skipped page adjustment record does not participate in the merging of image continuum segments; if the abnormal image number is missing in the middle round but is the same in the rounds before and after, the missing round is regarded as a transition round and is not directly judged as a regression; if the same abnormal image number always exists in the adjustment records of each round after retention, the adjustment result is not output, and the stop adjustment result is only output when the image attribution result becomes an irresolvable image; The purpose of S5-4 is to block further adjustment of the current page when an irresolvable representation is established, and to form a complete indeterminate adjustment. Its mechanism is to fix the current page, the trial action, the trial response, and the irresolvable representation into the indeterminate adjustment record to prevent the robotic arm from continuing to click in the invalid direction. The input quantities are the representation attribution result, the current page, the trial action, the trial response, and the irresolvable representation. The processing action is as follows: first, determine whether the representation attribution result is an irresolvable representation. If so, send the current page indeterminate adjustment command to the robotic arm to prevent the robotic arm from continuing to execute clicks along the trial action direction. Then, the current page number, the trial action number, the trial response record number, and the unresolvable appearance number are read and written into the prohibition record table under the same prohibition number. The prohibition result is then generated, which includes at least the prohibition number, the current page number, the trial action direction, the unresolvable appearance type, and the prohibition generation time. The output is the prohibition result, which is written into the prohibition record table for subsequent page switching or manual review. Abnormal or missing information is handled as follows: if the current page number is missing, the page number in the trial response record is read and written; if the trial response record is missing, the latest complete trial response record is read and written; if the robotic arm has received the prohibition command but a click still occurs, this click is written as an abnormal click after prohibition and appended to the prohibition record table. Through the above processing, S5 establishes a formal click-to-advance, continued adjustment verification, and result output link under the condition of resolvable appearance, and establishes a blocking link under the condition of non-resolvable appearance. This allows formal adjustment to proceed continuously and to stop in time when abnormal appearance has not subsided or appearance classification has changed. In practical application: when a high-load appearance is determined to be resolvable by S4, the robotic arm continuously performs formal clicks along the direction of the trial action, and writes the change of set value, the order of state change, and the content of prompt change after each round of clicks into the adjustment record table; if the adjustment records of two adjacent rounds maintain the same change direction and the appearance classification result continues to be a resolvable appearance, then the continued adjustment command is output; when the high-load display no longer appears in subsequent rounds, the system outputs the adjustment result; if the high-load display continues to exist and the appearance classification result in subsequent rounds changes to a non-resolvable appearance, then the stop adjustment result is output; if it is initially determined to be a non-resolvable appearance, then the robotic arm is directly prohibited from continuing to click and the blocking result is output.

[0022] Working principle: This solution does not directly treat the abnormal phenomena seen on the screen as problems that can definitely be resolved. Instead, it first clearly observes what is currently displayed on the screen, then identifies which pair of controls on the current page can form a "try and then go back" single-step click pair. A small trial is then performed, recording the changes in values, status, and prompts after the trial. A reverse click is then performed to check if the interface can return to its original state in the opposite direction. If the clicked item returns to its original position, the status reverts, and the prompts recover, it means that the current abnormal phenomenon can indeed be resolved through operations on this page. Only then will the system continue to make formal adjustments in the same direction, continuously judging whether the abnormal phenomenon has weakened or disappeared after each round of adjustment. If it cannot go back, or although a click was made but the abnormal phenomenon does not move in the direction of resolution, clicking is stopped and recorded as a prohibited result. In other words, this method does not directly adjust upon seeing an abnormality, but first verifies whether "this page can truly resolve this abnormality," and only makes formal adjustments after verification. For example, in the central air conditioning room of a commercial complex, the controller protocol of the air compressor is closed, and the system cannot directly connect to the equipment. Instead, a camera is used to view the touch screen, and a robotic arm simulates human touch. Assuming that high load and frequent switching occur continuously on the screen, the system first identifies the current page, setpoint display items, status display items, and prompts from the continuous images. Then, it finds a set of reversible controls on the current page, such as an increase control and its corresponding decrease control. The robotic arm first clicks the increase control to see if the setpoint changes, if the running status changes accordingly, and if the prompt box appears accordingly. Then, it clicks the decrease control to see if the setpoint returns to its original value, if the status reverts in reverse order, and if the prompt returns to the content before the click. If all three types of reversion are successful, it means that this type of anomaly can indeed be adjusted by the current page, and the system continues to fine-tune in the same direction until high load or frequent switching no longer occurs. If the reversion is unsuccessful, or if the anomaly persists after further adjustment, the system stops adjusting the current page to avoid blindly clicking and making the equipment more chaotic.

[0023] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent control method for an air compressor used in air conditioning, characterized in that, include: S1. Obtain continuous images of the air compressor control panel, read the page identifier area, value display area, status display area and prompt display area frame by frame, extract the continuously occurring high load, high energy consumption or frequent switching phenomena from adjacent images, and output the current page, current display item and abnormal appearance. S2. Based on the position of the clickable control on the current page, select the single-step tap pair that is on the same page as the currently displayed item and can be directly executed in reverse tap to restore within the page. Determine the forward tap in the single-step tap pair as the trial action, and output the trial action and its corresponding reverse tap. S3. Control the robotic arm to perform a single tap on the screen position corresponding to the probe action, continuously acquire the continuous images after the tap, sequentially read the direction of the tapped item change, the order of the status display area change, and the content of the prompt display area change, and output the probe response. S4. Control the robotic arm to perform a tap on the screen position corresponding to the reverse tap, continuously acquire the continuous image after the reverse tap, and determine whether the tapped item returns to the state before the tap, whether the status display area retreats in the opposite direction to the trial action, and whether the prompt display area is restored to the content before the tap. If all three conditions are met, output that the appearance can be resolved; otherwise, output that the appearance cannot be resolved.

2. The intelligent control method for an air compressor for air conditioning according to claim 1, characterized in that: Also includes: S5. When the output is a resolvable phenomenon, control the robotic arm to perform formal adjustment sequentially along the tapping direction of the trial action, and repeatedly read the trial response and confirm the phenomenon's attribution after each tap, until the abnormal phenomenon weakens or disappears, and then output the adjustment result; when the output is an unresolvable phenomenon, stop further tapping on the current page and output the adjustment prohibition result.

3. The intelligent control method for an air compressor for air conditioning according to claim 2, characterized in that: S1 includes: S1-1. Obtain continuous images of the air compressor control panel, perform on-screen positioning and segmentation of each frame image according to the fixed border, title text position and control distribution position in the image, and output the partition images corresponding to the page identification area, numerical display area, status display area and prompt display area. S1-2. Perform content comparison at the same position on each partition image in adjacent frames, determine the frame group where the page identifier area remains unchanged and the numerical display area, status display area or prompt display area changes as the valid running frame group, and read the corresponding display content according to the field position in the valid running frame group, and output the current page and the current display item. S1-3. Perform continuous merging of the current display items in the effective running frame group according to their order of appearance. Determine the display content in the numerical display area that continuously rises or stays at a high position as a high load or high energy consumption phenomenon. Determine the display content in the status display area that repeatedly alternates between two adjacent running states as a frequent switching phenomenon and output the abnormal phenomenon.

4. The intelligent control method for an air compressor for air conditioning according to claim 3, characterized in that: S2 includes: S2-1. Based on the center coordinates, border range, control characters, and display coordinates of the current display item of each clickable control on the current page, construct the control position matrix, display item position matrix, and control character code table. Perform same-page coordinate projection, relative displacement decomposition, and shortest displacement sorting on the control position matrix and display item position matrix to form the forward candidate control sequence corresponding to the current display item. S2-2. For each forward candidate control in the forward candidate control sequence, read the corresponding control character, determine the reciprocal character control group based on the pairing relationship of the control characters in the reciprocal character code table, and then perform direction vector calculation, coordinate difference decomposition and angle relationship determination on the control center coordinates in each reciprocal character control group. Determine the reciprocal character controls with opposite directions that point to the same currently displayed item as the reverse candidate controls to form the initial click pair set.

5. The intelligent control method for an air compressor for air conditioning according to claim 4, characterized in that: S2 also includes: S2-3. For each initial point pair in the initial point pair set, construct a three-point association matrix between the forward control, the currently displayed item, and the reverse control. Perform determinant evaluation, symmetry axis reconstruction, and bidirectional path length summation on the three-point association matrix. When the path length from the forward control to the currently displayed item is equal to the path length from the reverse control to the currently displayed item, the dot product of the forward direction vector and the reverse direction vector is negative, and the symmetry axis passes through the center of the currently displayed item, retain the corresponding initial point pair as an invertible candidate point pair. S2-4. Perform one-to-one matching on reversible candidate points, construct a bipartite graph of forward and reverse controls, perform minimum cost matching, conflict edge deletion and connected component splitting on the bipartite graph, and when a forward control corresponds to only one reverse control and a reverse control corresponds to only one forward control, determine the corresponding reversible candidate point pair as the target single-step point pair, and determine the forward point in the target single-step point pair as the trial action and the reverse point as the corresponding reverse point and output it.

6. The intelligent control method for an air compressor for air conditioning according to claim 5, characterized in that: S3 includes: S3-1. Control the robotic arm to perform a single tap on the screen position corresponding to the trial action, acquire the image before the tap and the continuous images after the tap, perform character splitting, position alignment and bit-by-bit difference on the tapped item area in the image before the tap and the continuous images after the tap, determine the direction of change of the tapped item according to the order of increase and decrease of the code value of each character position, and output the sequence of the direction of change of the tapped item. S3-2. Perform state segmentation and timing numbering on the state display area in the continuous image after the point is pressed. Construct a state sequence table according to the occurrence time, duration of each state segment and adjacent connection relationship. Align the state sequence table with the change direction sequence of the point-pressed item according to the same frame time and output the change order of the state display area. S3-3. Perform character extraction, prompt fragment merging, and frame position write-back verification on the prompt display area in the continuous image after tapping. Construct a prompt change table according to the appearance order, disappearance order, and duration interval of the prompt content. Combine the prompt change table with the change direction sequence of the tapped item and the change order of the status display area and write it into the trial response record. Output the trial response.

7. The intelligent control method for an air compressor for air conditioning according to claim 6, characterized in that: S4 includes: S4-1. Obtain the baseline image before the trial action is executed, the trial response record after the trial action, and the continuous image after the reverse click. Extract the character bit sequence, status bit sequence, and prompt content bit sequence of the clicked item from the baseline image and the continuous image after the reverse click, respectively. Construct the baseline bit sequence matrix, the backtrack bit sequence matrix, and the frame sequence association table according to the frame sequence number. S4-2. Perform bit-by-bit difference, sign reversal, matrix multiplication, and singular value decomposition on the point-by-item character position sequence in the reference position matrix and the backtrack position matrix. Translate the point-by-item change direction sequence in the trial response record into a backtracking vector. Then project the point-by-item character position difference vector of each frame onto the backtracking vector. When the projected sign of each frame is consistent with the sign of the corresponding position of the backtracking vector, the point-by-item character position sequence of the last frame is identical bit-by-bit to the point-by-item character position sequence in the reference image, and the first frame number that satisfies the bit-by-bit identity is unique, output the point-by-item return result.

8. The intelligent control method for an air compressor for air conditioning according to claim 7, characterized in that: S4 also includes: S4-3. Perform sequence alignment, state transition expansion, and reverse adjacency graph construction on the state bit sequences in the reference bit sequence matrix and the backtrack bit sequence matrix. Write the state display area change order in the trial response record into the target backtrack sequence in reverse order. Then perform closed path search and path coverage solution on the reverse adjacency graph. Output the state backtrack result when the solved backtrack path covers the target backtrack sequence in order, the end state of the backtrack path is the same as the first state of the reference state bit sequence, and each state in the backtrack path is retained only once. S4-4. Perform character encoding, position numbering, hash mapping, and error correction encoding expansion on the prompt content bit sequence in the base position matrix and the back position matrix. Perform bit-by-bit correspondence verification on the prompt content encoding sequence after the reverse click and the base prompt content encoding sequence. When the character encoding, position number, hash value, and codeword sequence after error correction expansion are the same, output the prompt recovery result. S4-5. Write the results of the item return, status rollback, and prompt recovery into the consistent association graph. Perform two-part matching, conflict edge deletion, connected component partitioning, and closed loop verification on the consistent association graph. When the nodes corresponding to the three types of results are located in the same closed connected component, output a resolvable representation. When the nodes corresponding to the three types of results are not located in the same closed connected component, output an unresolvable representation.

9. The intelligent control method for an air compressor for air conditioning according to claim 8, characterized in that: S5 includes: S5-1. When the appearance attribution result is a resolvable appearance, read the trial action, the corresponding reverse tap, the trial response and the current page, construct the formal tap sequence according to the tap direction of the trial action, and write the page image, tap item change, state change sequence and prompt change content after each formal tap into the adjustment record table, and output the current round of adjustment record. S5-2. For the adjustment record table, perform point difference, state sequence alignment and prompt content write-back comparison of adjacent two rounds of adjustment records. When the point direction remains unchanged, the appearance attribute result is continuously resolvable appearance, and the display item corresponding to the abnormal appearance changes continuously in the same direction of change, retain the next round of formal point and output the continued adjustment instruction. Stop adding formal points when any item is not true.

10. The intelligent control method for an air compressor for air conditioning according to claim 9, characterized in that: S5 also includes: S5-3. Construct a chronological sequence of the appearance evolution of each round of adjustment records after retention. Perform segmented merging of abnormal appearances, verification of the beginning and end of continuous segments, and solution of the disappearance position. When the high load display, high energy consumption display, or frequent switching display corresponding to the abnormal appearance changes from continuous to no longer appearing, output the adjustment result. When the abnormal appearance still appears continuously and the appearance attribution result undergoes an irresolvable transformation, output the stop adjustment result. S5-4. When the representation attribution result is an unresolvable representation, the robotic arm is prohibited from continuing to click along the direction of the trial action on the current page, and the current page, trial action, trial response and unresolvable representation are written into the prohibition record table, and the prohibition result is output.

Images