A laboratory intelligent control system

By combining control feedback acquisition and step token generation in the laboratory automation control system, the completion of experimental steps can be accurately determined, solving the problem of experimental process misalignment caused by delays in the device interface status. This enables accurate advancement of the experimental process and effective utilization of device feedback results.

CN123114248APending Publication Date: 2026-11-17HUBEI SUQING TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611087967.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing laboratory automation control systems, delays or anomalies in the status of the equipment interface can lead to errors in the determination of experimental steps, resulting in misalignment of sample batches and equipment usage periods. Existing technologies cannot accurately advance the experimental process based on feedback results from the equipment.

Method used

By linking control operations, device feedback collection, and step token generation, a basis for node continuation is formed. The completion of experimental steps is determined by feedback items such as pixel difference, edit box text, static text, and status bar text. The target control is determined by combining control fingerprints, adjacent text, and rectangle intersection-union ratio, and step tokens are generated to ensure accurate process advancement.

Benefits of technology

It reduces the number of cases where a successful control action is mistakenly identified as the completion of an experimental step, reduces the misalignment between process issuance and status determination, ensures that the target control can still be accurately located after the device software restarts or the interface shifts, improves the status basis for node transitions, and reduces the re-execution or skipping of invalid states during interruption recovery.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114248A_ABST
    Figure CN123114248A_ABST
Patent Text Reader

Abstract

This invention discloses a laboratory intelligent control system, specifically relating to the field of laboratory automation control. It includes: a task module for receiving experimental procedures, current nodes, node edges, device connection records, and the previous token; parsing the current node into control identifiers, action content, feedback items, and candidate successor nodes; generating node tasks based on device connection records, including node sequence numbers, control identifiers, action content, feedback items, and candidate successor nodes; and outputting the node tasks. A control-seeking module receives node tasks, candidate controls, and historical control records, recording each candidate control as a control-seeking path. A feedback module converts pixel differences, edit box text, static text, and status bar text into complete item sets, and a token module generates step tokens, enabling node continuation to advance based on device feedback results, thus relatively reducing the possibility of misjudging successful control actions as completed experimental steps.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of laboratory automation control technology, and more specifically, to a laboratory intelligent control system. Background Technology

[0002] In the automated operation of scientific research laboratories, existing control platforms mainly revolve around experimental process modeling and equipment software operation. Experimental steps are configured as actions, waiting, judgment and loop nodes. The server saves the process and sends tasks to the device client. The device uses edge computing capabilities to complete Windows control recognition, click, text reading and partial screen detection. In continuous experiments such as cell culture, material aging, and sample testing, a single sample batch needs to complete parameter setting, startup, status confirmation, and result reading in sequence according to the status of the equipment software interface. The next step cannot be entered until the previous step is actually completed, otherwise it will cause misalignment between sample batches, equipment occupancy time, and process data. Existing processing methods often use control clicks to return, script execution completion, countdown completion, or single text changes as the criteria for node termination. However, the device interface exhibits issues such as initialization delays, brief status bar refreshes, pop-up window obstructions, and delayed result box writing. On-site observations reveal phenomena such as buttons being triggered but the device not yet entering the running state, waiting time ending but no results being generated, text briefly changing and then reverting, and the client returning success but the device interface remaining on the error message page. These issues cause subsequent nodes to be issued even when the experimental state is not yet established. The technical problem this application aims to solve is: how to ensure that the success of the experimental control process is based on the step completion judgment results formed by the feedback from the device end, and avoid mistakenly taking the successful control action or the end of the waiting as the completion of the experimental step. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a laboratory intelligent control system. This system associates the control operations of the current node, the feedback collection from the device, the determination of the validity of feedback items, and the generation of step tokens as the basis for node continuation. This enables subsequent node issuance, conditional branch selection, and loop continuation to proceed based on the device feedback results, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a laboratory intelligent control system, comprising: The task module is used to receive the experimental process, the current node, node edges, device connection records and the previous token, parse the current node into control identifier, action content, feedback item and candidate successor node, generate node tasks containing node sequence number, control identifier, action content, feedback item and candidate successor node according to the device connection record, and output the node task. The control search module receives node tasks, candidate controls, and historical control records. It records each candidate control as a control search path and calculates the fingerprint similarity value, title similarity value, class name similarity value, rectangle intersection-union ratio, and adjacent text similarity value between the candidate control and the control identifier. It sums these five items to obtain the path matching value. It divides the number of valid items in the previous token by the sum of the number of valid items and invalid items in the previous token to obtain the pheromone increment. It multiplies the path matching value by the pheromone increment to obtain the path weight and writes the candidate control with the highest path weight into the target control record. The feedback module is used to receive target control records and node tasks, perform actions on the target control, collect pixel differences, edit box text, static text or status bar text in the area where the target control is located, compare the collected results with the feedback items one by one, record the feedback items that match as true items, record the feedback items that do not match as false items, and output the node feedback record.

[0005] In a preferred embodiment, it further includes: The token module is used to receive node feedback records, current node, candidate successor nodes, and historical node records. It adds the number of established items in the historical node records to the number of established items in the current round to obtain the posterior established number, adds the number of invalid items in the historical node records to the number of invalid items in the current round to obtain the posterior invalid number, divides the posterior established number by the sum of the posterior established number and the posterior invalid number to obtain the node posterior value, calculates the number of hits between the feedback items and the established items in the current round for each candidate successor node, writes the candidate successor node with the highest number of hits into the token destination, and outputs the step token. The continuation module receives interrupted nodes, last feedback items, unread step tokens, and candidate successor nodes. After the device client reconnects, it collects the current status items, compares the current status items with the last feedback items item by item to obtain the number of successful continuation items and the number of unsuccessful continuation items. It adds the number of successful continuation items to the a posteriori number in the unread step token to obtain the number of successful breakpoints, and adds the number of unsuccessful continuation items to the a posteriori number in the unread step token to obtain the number of unsuccessful breakpoints. Finally, it outputs the continuation task based on the number of hits between the candidate successor node, the current node, and the interrupt handling node and the current status items.

[0006] In a preferred embodiment, the execution of the task module includes: Read the current node, node edge, device connection record and previous token in the experimental process. Extract the node number, control identifier, action content and feedback item from the current node. Read the starting node and ending node of each node edge one by one. Record the node edge whose starting node is equal to the current node as the current flow edge. Write the ending node of the current flow edge into the candidate successor node set. Read the token destination from the previous token and compare it with each candidate successor node in the candidate successor node set in turn. If the token destination is equal to any candidate successor node, write the candidate successor node into the priority successor node. If the token destination is not equal to any candidate successor node, read the endpoint node of the first current flow edge according to the storage order of the node edges in the experimental process and write the endpoint node into the priority successor node. Read the device terminal number, connection mark, and last heartbeat time from the device connection record. Compare the device terminal number bound to the current node with the device terminal number in the device connection record item by item, and write the device connection records with the same device terminal number and the connection mark as online into the target connection record. Based on the target connection record, write the node sequence number, control identifier, action content, feedback item, candidate successor node set, priority successor node, and last heartbeat time into the node task, and output the node task.

[0007] In a preferred embodiment, the execution of the control module includes: Receive node tasks, candidate control sets, historical control records, and the previous token. Read control identifiers from the node tasks, and read candidate controls one by one from the candidate control sets. Record 1 if the control fingerprint of a candidate control is the same as the historical control fingerprint corresponding to the control identifier, and 0 if they are different. Record 1 if the window title of a candidate control is the same as the historical window title corresponding to the control identifier, and 0 if they are different. Record 1 if the class name of a candidate control is the same as the historical class name corresponding to the control identifier, and 0 if they are different. Record 1 if the adjacent text of a candidate control is the same as the historical adjacent text corresponding to the control identifier, and 0 if they are different. Divide the intersection area of ​​the control rectangle of a candidate control and the union area of ​​the control rectangle corresponding to the control identifier to obtain the rectangle intersection-union ratio.

[0008] In a preferred embodiment, the execution of the control module further includes: The path matching value of a candidate control is obtained by adding the fingerprint identical value, title identical value, class name identical value, adjacent text identical value, and rectangle intersection-union ratio. The number of valid and invalid items in the previous token is read, the number of valid items is incremented by 1 to obtain the valid correction number, the number of invalid items is incremented by 1 to obtain the invalid correction number, the valid correction number is divided by the sum of the valid correction number and the invalid correction number to obtain the pheromone increment, and the path matching value is multiplied by the pheromone increment to obtain the path weight of the candidate control. Repeatedly calculate the path weight for all candidate controls in the candidate control set, and arrange the candidate controls in descending order of path weight. If the path weights are the same, arrange them in descending order of fingerprint same value, adjacent text same value, rectangle intersection-union ratio, title same value, and class name same value. Write the first-ranked candidate control, its path matching value, pheromone increment, and path weight into the target control record, and output the target control record.

[0009] In a preferred embodiment, the execution of the feedback module includes: Receive target control records and node tasks, read the target control, control rectangle and window title from the target control record, read the action content and feedback items from the node task, determine the collection area in the window corresponding to the window title according to the control rectangle, execute the action content on the target control, and collect the area image, edit box text, static text and status bar text of the collection area before and after the action is executed. Subtract the region image before the action is executed from the region image after the action is executed point by point according to the same pixel coordinates and take the absolute value. Summ up all the absolute values ​​to get the pixel difference. Write the edit box text, static text and status bar text after the action is executed into the collection text item respectively. Combine the pixel difference and the collection text item to form the collection result. Read the feedback fields in each feedback item one by one. Compare the feedback items with the pixel field to the pixel difference. Compare the feedback items with the text field to the collected text item word by word. Write the feedback items that are the same as the collected result into the true item. Write the feedback items that are different from the collected result into the false item. Write the target control, action content, collected result, true item and false item into the node feedback record.

[0010] In a preferred embodiment, the execution of the token module includes: Receive node feedback records, current node, candidate successor node set, node edges and historical node records. Read the set of valid items, the set of invalid items and the target control from the node feedback records. Count the number of items in the set of valid items to get the number of valid items in this round. Count the number of items in the set of invalid items to get the number of invalid items in this round. Read the number of valid items and the number of invalid items in the past from the historical node records where the node number is equal to the current node. The posterior valid number is obtained by adding the number of valid items in the past to the number of valid items in the current round. The posterior invalid number is obtained by adding the number of invalid items in the past to the number of invalid items in the current round. The posterior valid number is divided by the sum of the posterior valid number and the posterior invalid number to obtain the posterior value of the node. The current node, the posterior valid number, the posterior invalid number, and the posterior value of the node are written into the node posterior record.

[0011] In a preferred embodiment, the execution of the token module further includes: Read the candidate successor nodes in the candidate successor node set one by one. Read the flow edges from the node edges where the starting node is equal to the current node and the ending node is equal to the candidate successor node. Read the edge feedback items bound to the flow edge. Compare each edge feedback item with each valid item in the valid item set. If the edge feedback item is equal to any valid item, it is counted as a hit. Accumulate the hit items to obtain the number of hit items for the candidate successor node. Arrange the candidate successor nodes in descending order of the number of hits. For nodes with the same number of hits, arrange them according to the storage order of the corresponding flow edges in the node edges. Write the candidate successor node ranked first into the token destination, and write the current node, target control, established itemset, invalid itemset, posterior established number, posterior invalid number, node posterior value, and token destination into the step token.

[0012] In a preferred embodiment, the execution of the connection continuation module includes: When the device client disconnects or the node task stops responding, read the node task being executed, the step tokens that have been generated but not yet read by the next node, and the node edges. Write the current node in the node task to the interrupted node, write the established itemset in the step token to the last feedback itemset, write the step token to the unread step token, and read the flow edges whose starting node is equal to the interrupted node from the node edges. Write the ending node of the flow edge to the candidate successor node set. After the device client reconnects, read the target control and action content in the unread step token, re-collect the pixel difference, edit box text, static text and status bar text of the area where the target control is located, and write the pixel difference, edit box text, static text and status bar text into the current status item set according to the collected fields; Read each current state item in the current state item set one by one. Compare each current state item with the last feedback item in the last feedback item set. If the current state item is equal to any last feedback item, write the current state item into the continuation valid itemset. If the current state item is not equal to any last feedback item, write the current state item into the continuation invalid itemset. Count the number of items in the continuation valid itemset to get the continuation valid itemset count. Count the number of items in the continuation invalid itemset to get the continuation invalid itemset count. Add the posterior valid count in the unread step tokens to the continuation valid itemset count to get the breakpoint valid count. Add the posterior invalid count in the unread step tokens to the continuation invalid itemset count to get the breakpoint invalid count. Divide the breakpoint valid count by the sum of the breakpoint valid count and the breakpoint invalid count to get the breakpoint posterior value.

[0013] In a preferred embodiment, the execution of the connection continuation module further includes: Read the following from the node edges: flow edges where the starting node is equal to the interruption node and the ending node belongs to the candidate successor node set; current node feedback items where the ending node is equal to the current node; and handling feedback items where the ending node is equal to the interruption handling node. Compare each flow edge's bound edge feedback item, current node feedback item, and handling feedback item with the current state item set item by item. A feedback item equal to any current state item is counted as a hit. Accumulate the number of hit items for candidate successor nodes, current nodes, and interruption handling nodes. Sort the three types of hit item counts from largest to smallest. For nodes with the same number of hit items, sort them in the order of candidate successor node set, current node, and interruption handling node. Write the node ranked first into the continuation destination. Write the interruption node, current state item set, number of successful continuation items, number of unsuccessful continuation items, number of successful breakpoints, number of unsuccessful breakpoints, breakpoint post-test value, and continuation destination into the continuation task.

[0014] The technical effects and advantages of this invention are as follows: 1. The feedback module converts pixel differences, edit box text, static text and status bar text into a set of items, and the token module generates a step token, so that the node continues to advance based on the device feedback results, which reduces the situation of misjudging the successful control action as the completion of the experimental step. 2. By writing the current node, node edges, feedback items, and candidate successor nodes into the node task through the task module, the execution content, feedback judgment objects, and subsequent flow scope of the device end are all carried out in the same task, which helps to reduce the misalignment between process issuance and status judgment. 3. By combining the control fingerprint, adjacent text, rectangle intersection-union ratio and pheromone increment, the target control is determined by the control search module, so that the target control can still be repositioned after the device software restarts or the interface shifts, which can alleviate the misoperation caused by handle changes and coordinate shifts within a certain range. 4. By incorporating the historical number of established items and the number of established items in the current round into the node's posterior value through the token module, and determining the token's destination based on the number of items hit by the edge feedback item, the conditional branches and loops continue to select based on the feedback establishment results, thereby relatively improving the state basis for node flow. 5. By reading the unread step token through the continuation module and collecting the current status item after reconnection, the breakpoint post-test value and the number of hit items are used to generate the continuation task, so as to reduce the situation of repeatedly executing effective actions or skipping the invalid state when the interruption is restored. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the system modules 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 a laboratory intelligent control system, comprising: The task module is used to receive the experimental process, the current node, node edges, device connection records and the previous token, parse the current node into control identifier, action content, feedback item and candidate successor node, generate node tasks containing node sequence number, control identifier, action content, feedback item and candidate successor node according to the device connection record, and output the node task. The task module is used to transform the current node to be executed in the experimental process into a node task that can be read by the device client, so that the subsequent control module, feedback module, and token module can read the same set of node fields. The experimental process includes a node table and node edges. The node table records the node sequence number, node type, control identifier, action content, feedback item, and bound device terminal number. The node edges record the starting node, ending node, and edge feedback item. The previous token is the step token generated by the previous node. If the first node to be executed does not have a previous token, the token destination in the previous token is written as an empty field. The processing procedure of this implementation is as follows. Read the current node, node edges, device connection records, and previous token from the experimental process. Read the node sequence number, control identifier, action content, feedback item, and bound device number from the current node. Read the start node and end node of each node edge one by one. Record the node edge whose start node is equal to the current node as the current flow edge. Write the end node of each current flow edge into the candidate successor node set according to the storage order of the node edges in the experimental process. The candidate successor node set is used for subsequent generation of priority successor nodes and step tokens. If the current node does not have a current flow edge, write the candidate successor node set as an empty set and write the node sequence number, control identifier, action content, feedback item, and bound device number of the current node into the task pending verification record. Read the token destination from the previous token, and compare the token destination with the node sequence number of each candidate successor node in the candidate successor node set in turn. If the token destination is equal to any candidate successor node, write the candidate successor node into the priority successor node. If the token destination is not equal to any candidate successor node in the candidate successor node set, read the endpoint node of the first current flow edge according to the storage order of the node edges in the experimental process, and write the endpoint node into the priority successor node. When the previous token is an empty field, do not perform token destination comparison, directly read the endpoint node of the first current flow edge and write it into the priority successor node. When the candidate successor node set is empty, write the priority successor node as an empty field, and write the current node into the no successor node state record. Read the device terminal number, connection flag, and last heartbeat time from the device connection record. Compare the device terminal number bound to the current node with the device terminal numbers in the device connection record item by item, and write the device connection records with the same device terminal number and the connection flag as online into the target connection record. Based on the target connection record, write the node sequence number, control identifier, action content, feedback item, candidate successor node set, priority successor node, device terminal number from the target connection record, and last heartbeat time into the node task, and output the node task to the corresponding device client. If no device connection record with the same device terminal number and the connection flag as online is read, do not output the node task, and write the node sequence number, the bound device terminal number, and the connection missing result into the task waiting record for the device client to read after reconnecting. After the above processing, the task module converts the node data, flow edges, and device connection status in the experimental process into node tasks. The node tasks simultaneously retain control identifiers, action content, feedback items, and a set of candidate successor nodes, so that the device client does not need to re-parse the experimental process in subsequent executions. In practical applications, the current node of the cell culture process is to start the incubator. This current node is bound to device terminal number C01, the action content is to click the start button, the feedback item is the status bar text "Running", and the node edges include two current flow edges for entering the temperature stability detection node and entering the abnormal handling node. After reading the previous token, if the token is destined for the temperature stability detection node, the task module writes the temperature stability detection node into the priority successor node, and reads the target connection record with device terminal number C01 and connection mark as online in the device connection record. Then, it generates a node task containing the start button control identifier, click action, status bar text feedback item, and a set of candidate successor nodes, and outputs it to device client C01.

[0018] The control search module receives node tasks, candidate controls, and historical control records. It records each candidate control as a control search path and calculates the fingerprint similarity value, title similarity value, class name similarity value, rectangle intersection-union ratio, and adjacent text similarity value between the candidate control and the control identifier. It sums these five items to obtain the path matching value. It divides the number of valid items in the previous token by the sum of the number of valid items and invalid items in the previous token to obtain the pheromone increment. It multiplies the path matching value by the pheromone increment to obtain the path weight and writes the candidate control with the highest path weight into the target control record. The control search module addresses issues arising from device software restarts, window title changes, control handle refreshes, and resolution adjustments where target controls cannot be directly accessed using their original handles. Instead of using coordinate clicks as the basis for target control location, the device client first generates a candidate control set within the current device interface. Then, it compares each candidate control with its corresponding historical control record, converting the feedback from the previous token into a reinforcement value for the current control search, allowing results from the previous experimental state to participate in the next control location step. The candidate control set is enumerated by the device client within the current window. Each candidate control records at least its control fingerprint, window title, class name, control rectangle, and adjacent text. Historical control records record at least its control identifier, historical control fingerprint, historical window title, historical class name, historical control rectangle, historical adjacent text, and historical acquisition time. When multiple historical control records exist for the same control identifier, the latest historical acquisition time is used for calculation. After receiving the node task, candidate control set, historical control record, and previous token, the control search module first reads the control identifier from the node task and then reads the corresponding historical control record from the historical control record. If no corresponding historical control record is found, the path weight calculation is not performed, and the control identifier, node number, and the result of no historical control record are written to the control search failure record. After reading the historical control record, the module reads the candidate controls in the candidate control set one by one, marking 1 if the control fingerprint of the candidate control is the same as the historical control fingerprint, and 0 if they are different. The module then matches the window title of the candidate control with the historical control fingerprint. A 1 is recorded if the window title is the same as the historical title, and a 0 is recorded if they are different. A 1 is recorded if the class name of the candidate control is the same as the historical class name, and a 0 is recorded if they are different. A 1 is recorded if the adjacent text of the candidate control is the same as the adjacent text of the historical control, and a 0 is recorded if they are different. The control fingerprint is obtained by hashing the process name, window title, class name, control text, control rectangle width and height, parent control class name, and sibling number in a fixed order. When a field is empty, it is marked as an empty field in the concatenation. This process is used to avoid the loss of control association due to changes in the window handle alone, and also to prevent different buttons in the same window from being mistakenly selected simply because their coordinates are close. Adjacent text is used to provide semantic reference for the interface when multiple candidate controls have similar appearances. The control search module reads the nearest text control in the left, top, right, and bottom order of the candidate control rectangle and concatenates the reading results into adjacent text. When multiple text controls exist in a single direction, the distance from the boundary of the text control rectangle to the boundary of the candidate control rectangle is calculated, and the text control with the smallest distance value is selected. When no text control exists in a single direction, an empty field is marked. Then, the intersection area and union area of ​​the candidate control rectangle and the historical control rectangle are calculated. The intersection area is the area of ​​the overlapping area of ​​the two rectangles, and the union area is the area of ​​the candidate control rectangle plus the area of ​​the historical control rectangle minus the intersection area. The rectangle intersection-union ratio is the intersection area divided by the union area. When the union area is 0, the rectangle intersection-union ratio is recorded as 0. This rectangle intersection-union ratio does not determine the target control alone, but is only used as a component of the path matching value to handle the positioning relationship after slight control offset, window scaling, or interface layout changes. For the same candidate control, the path matching value is obtained by adding the fingerprint similarity value, title similarity value, class name similarity value, adjacent text similarity value, and rectangle intersection / union ratio. The path matching value represents the static similarity between the candidate control and the historical target control. Then, the number of true and false items in the previous token is read. The number of true items is incremented by 1 to obtain the true correction number, and the number of false items is incremented by 1 to obtain the false correction number. The true correction number is divided by the sum of the true and false correction numbers to obtain the pheromone increment. Finally, the path matching value is multiplied by the pheromone increment to obtain the path matching value. The path weight to the candidate control; when the first node does not have a previous token, both the number of valid and invalid items are read as 0, and the pheromone increment obtained after being corrected by adding 1 is used in the calculation; this addition of 1 is used to avoid the denominator being 0 and to ensure that the first node can still complete the control search according to the control characteristics; the role of the pheromone increment is that the more valid items the previous node reports, the more likely the current control search will continue to locate along the experimental state that has been verified in the previous round; the more invalid items the previous node reports, the more the path weight of the current control search will be reduced accordingly, to avoid continuing to advance along the high-weight path after abnormal feedback; After calculating the path weights of all candidate controls in the candidate control set, the control search module sorts only within the candidate control set corresponding to the same node task, without comparing across devices or nodes. During sorting, candidate controls are first arranged from largest to smallest path weight. If the path weights are the same, they are arranged from largest to smallest fingerprint value, adjacent text value, rectangle intersection-union ratio, title value, and class name value. If the above five items are still the same, they are arranged according to the enumeration order of the candidate controls in the candidate control set. After sorting, the first-ranked candidate control and its control fingerprint, window title, class name, control rectangle, adjacent text, path matching value, pheromone increment, and path weight are written into the target control record and output to the feedback module. When the candidate control set is empty, the target control record is not output, and the control identifier, node number, and empty candidate result are written into the control search failure record for the task module to wait for the device client to re-enumerate the window controls. Through the above processing, the control search module merges the historical features of the control, the features of the current candidate control, and the feedback status of the previous node into the same control search calculation, so that the target control no longer depends on a fixed handle or fixed screen coordinates. In practical applications, after the incubator control software restarts, the handle of the start button changes and the window position shifts. The device client enumerates multiple button controls such as start, stop, and reset. After the control search module reads the historical control record corresponding to the start button, it calculates the fingerprint same value, adjacent text same value, and rectangle intersection-union ratio of each candidate button. Then, it writes the number of established items that the device is ready to complete in the previous token into the pheromone increment. Finally, it writes the start button with the highest path weight into the target control record. The feedback module then executes the click and collects the status bar text accordingly.

[0019] The feedback module is used to receive target control records and node tasks, perform actions on the target control, collect pixel differences, edit box text, static text or status bar text in the area where the target control is located, compare the collected results with the feedback items one by one, record the feedback items that match as valid items, record the feedback items that do not match as invalid items, and output the node feedback record. The feedback module converts the operation results of the target control into node feedback records, so that the subsequent token module does not use the issuance of an action as the basis for node completion, but determines whether the feedback item is valid based on the collection results of pixel fields, edit box text fields, static text fields, or status bar text fields. A feedback item includes a feedback field, a feedback target value, and a feedback source. The feedback fields include pixel fields, edit box text fields, static text fields, and status bar text fields. When a feedback item is matched, it is written to the "valid" field; when it is not matched, it is written to the "invalid" field, and the collection results are separately written to the node feedback record. The processing procedure in this implementation is as follows: After receiving the target control record and node task, the feedback module determines the action execution object and feedback collection range. During processing, it reads the target control, control rectangle, and window title from the target control record, and the action content and feedback items from the node task. Within the window corresponding to the window title, it determines the collection area based on the control rectangle. When the action content is a click, it performs a mouse click based on the center coordinates of the control rectangle. When the action content is text input, it first activates the target control and then writes the input text in the node task. When the action content is a selection, it performs the selection operation according to the option number in the node task. Before the action is executed, it collects the area image, edit box text, static text, and status bar text of the collection area. After the action is executed, it collects the area image, edit box text, static text, and status bar text of the same collection area again. If the window corresponding to the window title does not exist, or the width or height of the control rectangle is 0, the action content is not executed, the target control, action content, and collection failure field are written to the node feedback record, and all feedback items are written to the invalid item. The feedback module performs difference and text processing on the data collected before and after the action to form the collection results required for subsequent item-by-item comparison. During processing, the region image before and after the action is converted into grayscale images. If the collected region image is a color image, the red, green, and blue channel values ​​for the same pixel coordinate are averaged to obtain the grayscale value of that pixel coordinate. Then, the grayscale image before and after the action is subtracted point by point for the same pixel coordinate and the absolute value is taken. The absolute values ​​corresponding to all pixel coordinates are accumulated to obtain the pixel difference. If the size of the region image before and after the action is different, it is re-collected according to the control rectangle. If the size is still different after re-collection, the pixel difference is written as a collection failure field. The text in the edit box after the action is executed is written into the edit box text item, the static text after the action is executed is written into the static text item, and the status bar text after the action is executed is written into the status bar text item. The pixel difference, edit box text item, static text item, and status bar text item are combined to form the collection result. The feedback module reads each feedback item and compares it with the collected results to convert device interface changes into valid and invalid items. During processing, it reads the feedback field and target value from each feedback item. If the feedback field is a pixel field, it reads the pixel difference from the collected results and compares it with the target value. If they are equal, the feedback item is written as a valid item; otherwise, it is written as an invalid item if they are not equal or the pixel difference is a failed collection field. If the feedback field is a text field in an edit box, it compares the target value with the text item character by character. If the number of characters and their order are the same, the feedback item is written as a valid item; otherwise, it is written as an invalid item. If the feedback field is a static text field, it compares the target value with the static text item character by character. If the feedback field is a status bar text field, it compares the target value with the status bar text item character by character. After all feedback items are compared, the target control, action content, collected results, valid items, and invalid items are written into the node feedback record and output to the token module. After the above processing, the feedback module breaks down the interface changes after the control action into pixel differences and text items, and writes the feedback item itself into the true or false items, so that the token module can directly read the true items for subsequent node hit calculation. In practical applications, after the incubator start button is clicked, the feedback module collects the button area image, edit box text, static text, and status bar text before and after the action in the button window. If the status bar text item is identical to the feedback target value word by word during operation, the feedback item corresponding to the status bar text field is written into the true items. If the pixel difference is not equal to the feedback target value corresponding to the pixel field, the feedback item corresponding to the pixel field is written into the false items. Then, the collection results and the two types of feedback items are written into the node feedback record for the token module to calculate the number of true items and the number of false items in this round.

[0020] The token module is used to receive node feedback records, current node, candidate successor nodes, and historical node records. It adds the number of established items in the historical node records to the number of established items in the current round to obtain the posterior established number, adds the number of invalid items in the historical node records to the number of invalid items in the current round to obtain the posterior invalid number, divides the posterior established number by the sum of the posterior established number and the posterior invalid number to obtain the node posterior value, calculates the number of hits between the feedback items and the established items in the current round for each candidate successor node, writes the candidate successor node with the highest number of hits into the token destination, and outputs the step token. The token module transforms the sets of valid and invalid items in the node feedback records into step tokens. This ensures that whether the current node can enter a successor node is no longer determined by the action completion flag, but by the validity of the feedback items, historical node records, and node edges. Historical node records store the number of valid items, the number of invalid items, and the node's posterior value, categorized by node number. Node edges store the starting node, ending node, and edge feedback items. Edge feedback items are formed by the feedback items referenced in the entry conditions of the corresponding flow edges. The edge feedback items for sequential edges are taken from all feedback items of the current node. The token module's processing procedure is as follows: After receiving node feedback records, the current node, the candidate successor node set, node edges, and historical node records, the token module first reads the set of valid items, the set of invalid items, and the target control from the node feedback records to determine the contribution of this round of feedback to the current node. During processing, it counts the number of feedback items in the set of valid items to obtain the number of valid items in this round, and counts the number of feedback items in the set of invalid items to obtain the number of invalid items in this round. Then, it reads the records with node numbers equal to the current node from the historical node records, and uses the historical number of valid items and the historical number of invalid items in these records as the historical quantity for this update. If no historical node record corresponding to the current node is read, the historical number of valid items and the historical number of invalid items are both recorded as 0. If the feedback items of the current node are empty, it does not enter the posterior calculation, and writes the current node into the record of no feedback items to avoid generating a push token for nodes without feedback basis. The token module incorporates the feedback results of the current round into the historical node record to form the feedback posterior result of the current node after multiple executions. During processing, the number of historical true items is added to the number of true items in the current round to obtain the posterior true number, and the number of historical false items is added to the number of false items in the current round to obtain the posterior false number. The posterior true number is then divided by the sum of the posterior true number and the posterior false number to obtain the node posterior value. When the sum of the posterior true number and the posterior false number is 0, the node posterior value is not calculated, and the current node is written to the posterior no input record. After obtaining the node posterior value, the current node, the posterior true number, the posterior false number, and the node posterior value are written to the node posterior record, and the historical node record is updated with the node number as the index, so that the next time the same current node is executed, the posterior true number and the posterior false number written in this operation can be read. After completing the node post-update, the token module reads the candidate successor node set to determine the entry criteria for each candidate successor node based on the overlap between the edge feedback items and the current round's established item set. During processing, each candidate successor node in the candidate successor node set is read one by one. Flow edges whose starting node is equal to the current node and whose ending node is equal to the candidate successor node are read from the node edges, and the edge feedback items bound to these flow edges are read. Each edge feedback item is compared sequentially with each established item in the established item set. If the edge feedback item has the same feedback field, feedback target value, and feedback source as any established item, it is counted as a hit, and the number of hit items for the candidate successor node is accumulated. If no corresponding flow edge is read, the number of hit items for the candidate successor node is recorded as 0. When a flow edge is not bound to an edge feedback item, all feedback items of the current node are used as edge feedback items for comparison with this flow edge. The token module generates a token destination based on the number of hits for each candidate successor node, which is used to transform the feedback result into the basis for reading the next node. During processing, candidate successor nodes are arranged in descending order of the number of hits. If the number of hits is the same, they are arranged according to the storage order of the corresponding flow edge in the node edge. When there is at least one candidate successor node with a non-zero number of hits, the candidate successor node ranked first is written into the token destination. When the number of hits for all candidate successor nodes is zero, the current node is written into the token destination, and the token type is set as supplementary sampling feedback token, so that the subsequent task module does not directly advance to the successor node. Then, the current node, target control, established itemset, invalid itemset, posterior established number, posterior invalid number, node posterior value, token destination, and token type are written into the step token. When generating the step token, the read flag is set to unread, so that the task module can read it when generating the next node task. After the task module reads the step token, it changes the read flag to read. Through the above processing, the token module transforms the feedback result after a control action into a step token with a node posterior value and a token destination. This step token is then used as input for subsequent task modules and the control-seeking module, ensuring that the advancement of the experimental process is based on the results of the feedback items being valid, rather than the results returned by the control actions. In practical applications, after the incubator startup node is executed, the set of valid items in the node feedback record includes feedback items where the status bar text field indicates "running," while the set of invalid items includes feedback items where the pixel field has not reached the target value. After reading the historical number of valid items and the historical number of invalid items for this node, the token module calculates the posterior valid number, the posterior invalid number, and the node posterior value. It then reads the two flow edges from the current node to the temperature stability detection node and the anomaly handling node, compares the edge feedback items of the two flow edges with the set of valid items, and if the edge feedback item corresponding to the temperature stability detection node has the highest number of hits, then the temperature stability detection node is written into the token destination, and a step token is generated for the task module to read.

[0021] The continuation module receives interrupted nodes, last feedback items, unread step tokens, and candidate successor nodes. After the device client reconnects, it collects the current status items, compares the current status items with the last feedback items one by one to obtain the number of successful continuation items and the number of unsuccessful continuation items. It adds the number of successful continuation items to the a posteriori number in the unread step token to obtain the number of successful breakpoints, and adds the number of unsuccessful continuation items to the a posteriori number in the unread step token to obtain the number of unsuccessful breakpoints. Finally, it outputs the continuation task based on the number of hits between the candidate successor node, the current node, and the interrupt handling node and the current status items. The continuation module is used to determine whether the experimental process should enter a candidate successor node, remain at the interrupted node for supplementary feedback, or enter an interruption handling node after the device client disconnects, the device software restarts, or the node task stops responding, based on the step tokens generated before the interruption and the device status re-collected after reconnection. The step tokens include the current node, target control, established itemset, invalid itemset, posterior established number, posterior invalid number, node posterior value, token destination, and read flag. The read flag is written as unread when the step token is generated and rewritten as read when the task module generates the next node task. Node edges include the starting node, ending node, and edge feedback items. The interruption handling node is written to the node table when the experimental process is loaded. The feedback items of the interruption handling node are used to record pop-up prompts, error text, device stop status, or abnormal status after reconnection. The processing procedure of this implementation is as follows: When a device client disconnects or a node task stops responding, the continuation module is used to fix the process position at the time of interruption, preventing the device from continuing execution based solely on the node number after recovery. During processing, it reads the currently executing node task, reads the step tokens marked as unread, and node edges; writes the current node in the node task to the interrupted node, writes the established item set in the step token to the last feedback item set, writes the step token to the unread step token, and reads the starting node and ending node one by one from the node edges. The node edge whose starting node is equal to the interrupted node is recorded as an interrupted flow edge, and the ending node of the interrupted flow edge is written to the candidate successor node set according to the node edge storage order. The device client disconnects when the connection mark in the device connection record changes from online to offline, and the node task stops responding when the currently executing node task has not written to the node feedback record and the device connection record no longer updates the last heartbeat time. When a step token marked as unread is not read, the feedback items in the currently executing node task are written to the last feedback item set, and the unread step token is written as an empty token record, for subsequent re-formation of breakpoint data based on the current status items. After the device client reconnects, the continuation module is used to reacquire the device interface state corresponding to the interruption location, so that the continuation judgment is based on the real state item after reconnection. During processing, the target control and action content in the unread step token are read, and the collection area is re-determined according to the window title and control rectangle in the target control record. Without repeating the action content, the area image, edit box text, static text and status bar text of the target control area are collected. The area image is processed according to the pixel difference calculation rules in the feedback module to obtain the pixel difference, and the pixel difference, edit box text, static text and status bar text are written into the current state item set according to the collection fields. When the unread step token is an empty token record, the target control record corresponding to the interruption node is read to determine the collection area. When the window corresponding to the window title does not exist or the width and height of the control rectangle are 0, the missing window field or abnormal control rectangle field is written into the current state item set, and the field is used for comparison with the processing feedback item of the interruption handling node. The continuation module compares the current status item set with the last feedback item set item by item to determine whether the device status after reconnection is still in the same experimental stage as the feedback result established before the interruption. During processing, it reads the current status item in the current status item set item by item and compares each current status item with the last feedback item in the last feedback item set in turn. If the collected field of the current status item is the same as the feedback field of the last feedback item, and the collected value of the current status item is the same as the feedback target value of the last feedback item, the current status item is written into the continuation established item set. If the current status item does not satisfy the condition that both the field and value are the same as any of the last feedback items, the current status item is written into the continuation established item set. Write the itemset to the set of items that are not valid for continuation; then count the number of items in the set of items that are valid for continuation to get the number of valid items for continuation, count the number of items in the set of items that are not valid for continuation to get the number of invalid items for continuation, add the number of valid items for continuation to the posterior number of valid items in the unread step token to get the number of valid breakpoints, add the number of invalid items for continuation to the posterior number of invalid items in the unread step token to get the number of invalid breakpoints, and divide the number of valid breakpoints by the sum of the number of valid breakpoints and the number of invalid breakpoints to get the posterior value of the breakpoint; when the sum of the number of valid breakpoints and the number of invalid breakpoints is 0, do not calculate the posterior value of the breakpoint, and write the interrupted node and the current state itemset to the breakpoint no input record; The continuation module calculates the hit counts for candidate successor nodes, interruption nodes, and interruption handling nodes based on the current state item set to determine the direction and content of the continuation task. During processing, it reads interruption flow edges from the node edges where the starting node equals the interruption node and the ending node belongs to the candidate successor node set, and reads the edge feedback items bound to each interruption flow edge. It reads feedback items from the node table where the node number equals the interruption node as interruption node feedback items; it reads feedback items from the node table where the node number equals the interruption handling node as handling feedback items. Each edge feedback item, interruption node feedback item, and handling feedback item is compared item by item with the current state item set. If the feedback field of a feedback item is the same as the collected field of any current state item and the feedback target value is the same as the collected value of that current state item, it is counted as a hit. These are then accumulated to obtain the number of candidate successor node hits. The system calculates the number of hits for interrupt nodes and interrupt handling nodes. It then sorts the three types of hit counts from largest to smallest. For nodes with the same number of hits, it arranges them in the order of reading the candidate successor node set, the interrupt node, and the interrupt handling node. The node ranked first is written to the continuation destination. When the continuation destination is a candidate successor node, the continuation task writes the corresponding control identifier, action content, and feedback item for that candidate successor node. When the continuation destination is an interrupt node, the continuation task writes a supplementary feedback task. The supplementary feedback task does not repeat the action content; it only re-collects the feedback item of the interrupt node. When the continuation destination is an interrupt handling node, the continuation task writes the control identifier, action content, and feedback item for the interrupt handling node, and writes the interrupt node, current status item set, number of successful continuation items, number of unsuccessful continuation items, number of successful breakpoints, number of unsuccessful breakpoints, breakpoint post-test value, and continuation destination to the continuation task. After the above processing, the continuation module does not use the manually selected continuation position as the basis for recovery. Instead, it re-compares the last feedback item in the step token before the interruption, the current status item after reconnection, and the feedback items in the node edges to generate a continuation task that can be read by the task module. In actual application, the incubator startup node has already generated a step token after execution, but the device client disconnects before entering the temperature stability detection node. After reconnection, the continuation module reads the target control in the unread step token, re-collects the status bar text and the pixel difference of the button area. If the status bar text in the current status item is still "running", then the current status item is written into the continuation establishment item set and matches the edge feedback item corresponding to the temperature stability detection node. The continuation destination is written into the temperature stability detection node. If the current status item contains communication error text, then the current status item matches the handling feedback item of the interruption handling node. The continuation destination is written into the interruption handling node, and the task module generates an interruption handling task accordingly.

[0022] Working principle: The core of this solution is to make the experimental process advance based on the feedback results from the device, rather than based on whether the control action is issued. The system first uses a task module to read the experimental process, current node, node edges, and device connection records to generate a node task containing control identifiers, action content, feedback items, and candidate successor nodes. The device client, leveraging edge computing capabilities, performs interface acquisition and control enumeration locally. The control search module combines historical control records, current candidate control characteristics, and feedback validity from the previous token to calculate path weights and determine the target control. After the feedback module executes an action on the target control, it collects pixel differences, edit box text, static text, and status bar text, comparing the collected results with feedback items to generate valid and invalid item sets. The token module calculates the node's posterior value based on the valid and invalid item sets and historical node records, and generates a step token based on the edge feedback item hit status of the candidate successor node. This step token serves as the basis for entering the next node, selecting conditional branches, or continuing the loop. When the device client is interrupted, the reconnection module re-determines the reconnection direction based on the unread step token and the current status item after reconnection, reducing the re-execution of already effective experimental actions. In experiments involving the continuous operation of incubators, centrifuges, and detectors, users pre-program nodes for startup, waiting, detection, and anomaly handling. Taking incubator startup as an example, after the system sends a task to the device client to click the startup button, it doesn't directly consider a successful click as startup completion. Instead, the device client uses edge computing to locally read the control fingerprint, window title, class name, control rectangle, and adjacent text to determine the startup button on the current interface. After clicking, the device client continues to collect status bar text and button area image changes. Only when the collected results match the corresponding feedback item during operation does the system generate a step token pointing to the temperature stability detection node. If the device client disconnects after clicking, the system reconnects and re-collects the incubator interface status using edge computing. If the status still corresponds to "running," it continues to the temperature stability detection node. If a communication error or a stopped state appears on the interface, it proceeds to the interruption handling node, thereby reducing the possibility of the experimental process being mistakenly advanced or repeatedly executed. The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A laboratory intelligent control system, characterized in that, include: The task module is used to receive the experimental process, the current node, node edges, device connection records and the previous token, parse the current node into control identifier, action content, feedback item and candidate successor node, generate node tasks containing node sequence number, control identifier, action content, feedback item and candidate successor node according to the device connection record, and output the node task. The control search module receives node tasks, candidate controls, and historical control records. It records each candidate control as a control search path and calculates the fingerprint similarity value, title similarity value, class name similarity value, rectangle intersection-union ratio, and adjacent text similarity value between the candidate control and the control identifier. It sums these five items to obtain the path matching value. It divides the number of valid items in the previous token by the sum of the number of valid items and invalid items in the previous token to obtain the pheromone increment. It multiplies the path matching value by the pheromone increment to obtain the path weight and writes the candidate control with the highest path weight into the target control record. The feedback module is used to receive target control records and node tasks, perform actions on the target control, collect pixel differences, edit box text, static text or status bar text in the area where the target control is located, compare the collected results with the feedback items one by one, record the feedback items that match as true items, record the feedback items that do not match as false items, and output the node feedback record.

2. The intelligent control system for a laboratory according to claim 1, characterized in that: Also includes: The token module is used to receive node feedback records, current node, candidate successor nodes, and historical node records. It adds the number of established items in the historical node records to the number of established items in the current round to obtain the posterior established number, adds the number of invalid items in the historical node records to the number of invalid items in the current round to obtain the posterior invalid number, divides the posterior established number by the sum of the posterior established number and the posterior invalid number to obtain the node posterior value, calculates the number of hits between the feedback items and the established items in the current round for each candidate successor node, writes the candidate successor node with the highest number of hits into the token destination, and outputs the step token. The continuation module receives interrupted nodes, last feedback items, unread step tokens, and candidate successor nodes. After the device client reconnects, it collects the current status items, compares the current status items with the last feedback items item by item to obtain the number of successful continuation items and the number of unsuccessful continuation items. It adds the number of successful continuation items to the a posteriori number in the unread step token to obtain the number of successful breakpoints, and adds the number of unsuccessful continuation items to the a posteriori number in the unread step token to obtain the number of unsuccessful breakpoints. Finally, it outputs the continuation task based on the number of hits between the candidate successor node, the current node, and the interrupt handling node and the current status items.

3. The intelligent control system for a laboratory according to claim 2, characterized in that: The execution of the task module includes: Read the current node, node edge, device connection record and previous token in the experimental process. Extract the node number, control identifier, action content and feedback item from the current node. Read the starting node and ending node of each node edge one by one. Record the node edge whose starting node is equal to the current node as the current flow edge. Write the ending node of the current flow edge into the candidate successor node set. Read the token destination from the previous token and compare it with each candidate successor node in the candidate successor node set in turn. If the token destination is equal to any candidate successor node, write the candidate successor node into the priority successor node. If the token destination is not equal to any candidate successor node, read the endpoint node of the first current flow edge according to the storage order of the node edges in the experimental process and write the endpoint node into the priority successor node. Read the device terminal number, connection mark, and last heartbeat time from the device connection record. Compare the device terminal number bound to the current node with the device terminal number in the device connection record item by item, and write the device connection records with the same device terminal number and the connection mark as online into the target connection record. Based on the target connection record, write the node sequence number, control identifier, action content, feedback item, candidate successor node set, priority successor node, and last heartbeat time into the node task, and output the node task.

4. The intelligent control system for a laboratory according to claim 3, characterized in that: The execution of the control module includes: Receive node tasks, candidate control sets, historical control records, and the previous token. Read control identifiers from the node tasks, and read candidate controls one by one from the candidate control sets. Record 1 if the control fingerprint of a candidate control is the same as the historical control fingerprint corresponding to the control identifier, and 0 if they are different. Record 1 if the window title of a candidate control is the same as the historical window title corresponding to the control identifier, and 0 if they are different. Record 1 if the class name of a candidate control is the same as the historical class name corresponding to the control identifier, and 0 if they are different. Record 1 if the adjacent text of a candidate control is the same as the historical adjacent text corresponding to the control identifier, and 0 if they are different. Divide the intersection area of ​​the control rectangle of a candidate control and the union area of ​​the control rectangle corresponding to the control identifier to obtain the rectangle intersection-union ratio.

5. A laboratory intelligent control system according to claim 4, characterized in that: The execution of the search and control module also includes: The path matching value of a candidate control is obtained by adding the fingerprint identical value, title identical value, class name identical value, adjacent text identical value, and rectangle intersection-union ratio. The number of valid and invalid items in the previous token is read, the number of valid items is incremented by 1 to obtain the valid correction number, the number of invalid items is incremented by 1 to obtain the invalid correction number, the valid correction number is divided by the sum of the valid correction number and the invalid correction number to obtain the pheromone increment, and the path matching value is multiplied by the pheromone increment to obtain the path weight of the candidate control. Repeatedly calculate the path weight for all candidate controls in the candidate control set, and arrange the candidate controls in descending order of path weight. If the path weights are the same, arrange them in descending order of fingerprint same value, adjacent text same value, rectangle intersection-union ratio, title same value, and class name same value. Write the first-ranked candidate control, its path matching value, pheromone increment, and path weight into the target control record, and output the target control record.

6. A laboratory intelligent control system according to claim 5, characterized in that: The execution of the feedback module includes: Receive target control records and node tasks, read the target control, control rectangle and window title from the target control record, read the action content and feedback items from the node task, determine the collection area in the window corresponding to the window title according to the control rectangle, execute the action content on the target control, and collect the area image, edit box text, static text and status bar text of the collection area before and after the action is executed. Subtract the region image before the action is executed from the region image after the action is executed point by point according to the same pixel coordinates and take the absolute value. Summ up all the absolute values ​​to get the pixel difference. Write the edit box text, static text and status bar text after the action is executed into the collection text item respectively. Combine the pixel difference and the collection text item to form the collection result. Read the feedback fields in each feedback item one by one. Compare the feedback items with the pixel field to the pixel difference. Compare the feedback items with the text field to the collected text item word by word. Write the feedback items that are the same as the collected result into the true item. Write the feedback items that are different from the collected result into the false item. Write the target control, action content, collected result, true item and false item into the node feedback record.

7. A laboratory intelligent control system according to claim 6, characterized in that: The execution of the token module includes: Receive node feedback records, current node, candidate successor node set, node edges and historical node records. Read the set of valid items, the set of invalid items and the target control from the node feedback records. Count the number of items in the set of valid items to get the number of valid items in this round. Count the number of items in the set of invalid items to get the number of invalid items in this round. Read the number of valid items and the number of invalid items in the past from the historical node records where the node number is equal to the current node. The posterior valid number is obtained by adding the number of valid items in the past to the number of valid items in the current round. The posterior invalid number is obtained by adding the number of invalid items in the past to the number of invalid items in the current round. The posterior valid number is divided by the sum of the posterior valid number and the posterior invalid number to obtain the posterior value of the node. The current node, the posterior valid number, the posterior invalid number, and the posterior value of the node are written into the node posterior record.

8. A laboratory intelligent control system according to claim 7, characterized in that: The execution of the token module also includes: Read the candidate successor nodes in the candidate successor node set one by one. Read the flow edges from the node edges where the starting node is equal to the current node and the ending node is equal to the candidate successor node. Read the edge feedback items bound to the flow edge. Compare each edge feedback item with each valid item in the valid item set. If the edge feedback item is equal to any valid item, it is counted as a hit. Accumulate the hit items to obtain the number of hit items for the candidate successor node. Arrange the candidate successor nodes in descending order of the number of hits. For nodes with the same number of hits, arrange them according to the storage order of the corresponding flow edges in the node edges. Write the candidate successor node ranked first into the token destination, and write the current node, target control, established itemset, invalid itemset, posterior established number, posterior invalid number, node posterior value, and token destination into the step token.

9. A laboratory intelligent control system according to claim 8, characterized in that: The execution of the continuation module includes: When the device client disconnects or the node task stops responding, read the node task being executed, the step tokens that have been generated but not yet read by the next node, and the node edges. Write the current node in the node task to the interrupted node, write the established itemset in the step token to the last feedback itemset, write the step token to the unread step token, and read the flow edges whose starting node is equal to the interrupted node from the node edges. Write the ending node of the flow edge to the candidate successor node set. After the device client reconnects, read the target control and action content in the unread step token, re-collect the pixel difference, edit box text, static text and status bar text of the area where the target control is located, and write the pixel difference, edit box text, static text and status bar text into the current status item set according to the collected fields; Read each current state item in the current state item set one by one. Compare each current state item with the last feedback item in the last feedback item set. If the current state item is equal to any last feedback item, write the current state item into the continuation valid itemset. If the current state item is not equal to any last feedback item, write the current state item into the continuation invalid itemset. Count the number of items in the continuation valid itemset to get the continuation valid itemset count. Count the number of items in the continuation invalid itemset to get the continuation invalid itemset count. Add the posterior valid count in the unread step tokens to the continuation valid itemset count to get the breakpoint valid count. Add the posterior invalid count in the unread step tokens to the continuation invalid itemset count to get the breakpoint invalid count. Divide the breakpoint valid count by the sum of the breakpoint valid count and the breakpoint invalid count to get the breakpoint posterior value.

10. A laboratory intelligent control system according to claim 9, characterized in that: The execution of the continuation module also includes: Read the following from the node edges: flow edges where the starting node is equal to the interruption node and the ending node belongs to the candidate successor node set; current node feedback items where the ending node is equal to the current node; and handling feedback items where the ending node is equal to the interruption handling node. Compare each flow edge's bound edge feedback item, current node feedback item, and handling feedback item with the current state item set item by item. A feedback item equal to any current state item is counted as a hit. Accumulate the number of hit items for candidate successor nodes, current nodes, and interruption handling nodes. Sort the three types of hit item counts from largest to smallest. For nodes with the same number of hit items, sort them in the order of candidate successor node set, current node, and interruption handling node. Write the node ranked first into the continuation destination. Write the interruption node, current state item set, number of successful continuation items, number of unsuccessful continuation items, number of successful breakpoints, number of unsuccessful breakpoints, breakpoint post-test value, and continuation destination into the continuation task.