Job feedback-based machine artificial intelligence scheduling method and system

Abstract

The present application relates to the technical field of robot operation scheduling and operation data processing, and particularly relates to a robot artificial intelligence scheduling method and system based on operation feedback. The method comprises the following steps: obtaining historical operation samples, grouping according to crop types, slope levels and wind speed levels, calculating double historical deviation rates and generating scene verification rule records; receiving operation requests, generating operation work orders bound with rule versions, splitting sub-tasks and generating task assignment records according to robot resource states; generating operation coverage areas and actual pesticide consumption according to motion trajectories and pesticide metering data of spraying start states, calculating double deviation rates, matching double verification thresholds and forming verification results; generating effective feedback samples according to the verification results, recalculating preset quantiles when update triggering conditions are met, generating new rule versions and recording version migration relationships. The present application effectively improves the consistency of verification bases and the traceability of rule versions.

Description

Technical Field

[0001] This invention relates to the field of robot job scheduling and job data processing technology, and in particular to a robot artificial intelligence scheduling method and system based on job feedback. Background Technology

[0002] In the field of robot operation scheduling technology, existing solutions for agricultural robot operations typically receive operation requests from a plot of land, break down the requests into multiple sub-tasks, and assign these sub-tasks based on the robot's working status, remaining battery power, remaining pesticide dosage, and location. After completing the operation, the robot sends back motion trajectory data, spraying status data, pesticide dosage data, and task completion status to the scheduling system, which then records the operation results or adjusts subsequent task assignments accordingly. Some solutions also verify the operation completion status based on the declared operation area, the area covered by the trajectory, or the amount of pesticide consumed.

[0003] Existing solutions often rely on fixed verification thresholds or manually configured rules to process operational feedback. Different crop types, plot slopes, and wind speeds during operation periods result in variations in robot movement, spraying coverage, and pesticide output. Using uniform verification boundaries makes it difficult to consistently correlate the operational coverage area and actual pesticide consumption with specific operational scenarios. Furthermore, motion trajectory data and pesticide metering data are typically stored separately, with some solutions using only one type of data as the basis for operation completion. This can easily lead to discontinuities in the correlation between task assignment records, execution feedback data, and verification status.

[0004] Existing solutions, when adjusting subsequent rules based on historical operation feedback, typically write completed operation data directly into the history record, lacking unified verification of data integrity, pesticide dosage status, and duplicate feedback status. When abnormal feedback or single operation feedback enters the rule update process, it easily leads to the mixing of historical operation samples from different scenarios. After the rule update, there may still be a lack of clear version correspondence between existing work orders and the updated rules, resulting in different judgment criteria being used when the same work order is re-verified. Therefore, it is necessary to address the data connection between agricultural robot work orders, execution feedback data, and scenario verification rules, and to solve the problem of how to form a double-bias verification basis for scenario grouping based on operation feedback, while maintaining the original rule version of the work order when valid feedback samples are updated. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a robot artificial intelligence scheduling method based on job feedback, comprising:

[0006] S100. Obtain historical operation samples and form scene groups according to crop type, slope level and wind speed level; calculate the first historical deviation rate and the second historical deviation rate, and generate scene cancellation rule records containing the first cancellation threshold, the second cancellation threshold and the rule version identifier according to the preset quantile;

[0007] S200: Receive job requests based on the scenario verification rule record, and generate a job order bound to the rule version identifier; split the job order and generate task assignment records and subtask scheduling execution instructions based on robot resource status parameters;

[0008] S300: Based on the work order, receive execution feedback data, calculate the operation coverage area using the motion trajectory data of the spraying in the on state, and convert the actual pesticide consumption according to the pesticide metering data; calculate the first deviation rate and the second deviation rate, match the first verification threshold and the second verification threshold from the scene verification rule record bound to the work order according to crop type, slope level, and actual wind speed level, generate verification result record and update the verification status;

[0009] S400. Generate valid feedback samples based on the reconciliation result record, data integrity status, and drug dosage status; write the valid feedback samples into the corresponding scenario group; recalculate the preset quantile when the number of newly added valid samples meets the update trigger condition, generate a new rule version, and record the version migration relationship.

[0010] Furthermore, the historical operation samples are derived from completed and verified work orders, and include crop type, plot slope, wind speed data, declared operation area, operation coverage area, theoretical pesticide consumption, actual pesticide consumption, verification status, data integrity status, and pesticide metering status.

[0011] Perform field integrity verification, duplicate feedback verification, and reagent dosage status verification on the historical operation samples, and set the historical operation samples that pass the verification to the sample valid status;

[0012] The plot slope is mapped to a slope grade, and the wind speed data is mapped to a wind speed grade. Scene grouping identifiers are generated based on crop type, slope grade, and wind speed grade.

[0013] Furthermore, the first historical deviation rate and the first deviation rate are the absolute value of the difference between the operating coverage area and the declared operating area divided by the declared operating area;

[0014] The second historical deviation rate and the second deviation rate are the absolute value of the difference between the actual drug consumption and the theoretical drug consumption, divided by the theoretical drug consumption;

[0015] A first preset quantile is calculated on the first historical deviation rate set to obtain a first write-off threshold, and a second preset quantile is calculated on the second historical deviation rate set to obtain a second write-off threshold.

[0016] When the number of samples in a scenario group is less than the minimum sample size, the scenario cancellation rule record of the higher-level scenario group is matched according to the small sample rollback rule.

[0017] Furthermore, based on the effective status of the rules for the planned work period, the version identifier of the rules that are in effect is written into the work order;

[0018] The rule version identifier remains unchanged after the work order is generated;

[0019] The new rule version is linked to the work orders generated after its effective date;

[0020] When re-verifying the work order, the rule version identifier originally bound to the work order is read.

[0021] Furthermore, the work order is divided into sub-tasks according to the work area, and the theoretical pesticide consumption is calculated based on the target work area and unit area application parameters of the sub-tasks.

[0022] The robot resource status parameters include robot identifier, robot working status, remaining battery power, remaining medicine, robot position, and data acquisition time.

[0023] A candidate robot set is generated by screening robots whose working status is executable, whose remaining power meets the power constraint, whose remaining drug quantity meets the drug quantity constraint, and whose data acquisition time meets the data validity condition.

[0024] A task assignment record is generated based on the set of candidate robots.

[0025] Furthermore, the execution feedback data includes work order identifier, subtask identifier, robot identifier, execution round, motion trajectory data, and spraying status data;

[0026] Aggregate execution feedback data according to the work order identifier, subtask identifier, robot identifier, and execution round;

[0027] Extract the motion trajectory data corresponding to the spraying activation status during the operation period to generate valid trajectory data;

[0028] Based on the effective trajectory data and the robot's spray pattern, a coverage area is formed. The overlapping parts of multiple coverage areas are deduplicated to generate a coverage area calculation record and the operational coverage area.

[0029] Furthermore, the drug metering data includes drug consumption pulse count, pulse equivalent, calibration version, and spraying status data;

[0030] Extract the pesticide consumption pulse count corresponding to the spraying activation state during the operation period, and generate the single pesticide consumption amount based on the product of the pesticide consumption pulse count and the pulse equivalent.

[0031] The single-time drug consumption is aggregated according to the work order identifier, sub-task identifier, and execution round to generate a drug consumption conversion record and the actual drug consumption.

[0032] Furthermore, when the first deviation rate is not greater than the first reconciliation threshold and the second deviation rate is not greater than the second reconciliation threshold, a reconciliation pass status is generated;

[0033] When the first deviation rate is greater than the first write-off threshold and the second deviation rate is not greater than the second write-off threshold, a first deviation anomaly is generated.

[0034] When the first deviation rate is not greater than the first write-off threshold and the second deviation rate is greater than the second write-off threshold, a second deviation anomaly is generated.

[0035] When both the first deviation rate and the second deviation rate are greater than the corresponding write-off threshold, a double deviation anomaly is generated.

[0036] When the data integrity status changes to the data incomplete status, a supplementary entry waiting status and a supplementary entry instruction are generated.

[0037] Write the verification status, first deviation anomaly, second deviation anomaly, double deviation anomaly, or supplementary entry waiting status into the verification result record.

[0038] Furthermore, the valid feedback samples satisfy the following sample admission criteria:

[0039] The data is complete, not incomplete; the dosage status is not abnormal; the repeated feedback verification result is verified; and the cancellation status is either cancelled or completed.

[0040] Accumulate the number of newly valid samples since the current rule version took effect, and generate a threshold update task when the number of newly valid samples reaches the update trigger value;

[0041] Recalculate the first preset quantile and the second preset quantile to generate candidate first write-off threshold and candidate second write-off threshold;

[0042] When the change in the candidate write-off threshold relative to the current write-off threshold is greater than the maximum single change, a new write-off threshold is generated based on the maximum single change.

[0043] The new write-off threshold, updated source sample range, and effective time will be written into the new rule version.

[0044] Furthermore, the robot AI scheduling system based on job feedback includes: a scenario rule generation module, a job order scheduling module, a job feedback verification module, and a rule feedback update module; the system is used to implement the method described in any one of the above.

[0045] The following are its main beneficial effects:

[0046] (1) To address the problem that it is difficult to match the actual operation status when using a unified verification boundary for different crop types, plot slopes and operation wind speeds, the coverage area verification boundary and the pesticide consumption verification boundary are formed by scenario grouping and double deviation rate quantiles, so that the scenario verification rule record is associated with the historical operation sample of the corresponding scenario.

[0047] (2) To address the issue of discontinuous connection between motion trajectory data, pesticide metering data and work order verification status, a dual-source verification basis is formed by filtering spraying start status, aggregating and deduplicating coverage areas, converting pesticide metering, and comparing double deviation rates. When a work order is re-verified, the original bound rule version is still called.

[0048] (3) To address the issues of incomplete data, abnormal drug dosage, or duplicate feedback directly participating in rule updates, candidate feedback samples that do not meet the conditions are excluded through sample admission criteria, and new valid samples are added in groups according to scenarios; combined with update triggering conditions, threshold change limits, and version migration records, rule updates have clear sample sources, triggering times, and version applicability relationships. Attached Figure Description

[0049] Figure 1 A flowchart illustrating the robot artificial intelligence scheduling method based on job feedback provided in this application embodiment;

[0050] Figure 2 This is a structural block diagram of a robot artificial intelligence scheduling system based on job feedback provided in an embodiment of this application. Detailed Implementation

[0051] Example 1: Refer to Figure 1 This is a flowchart illustrating the robot artificial intelligence scheduling method based on job feedback provided in an embodiment of the present invention. The process may include at least steps S100-S400:

[0052] S100. Obtain historical operation samples and form scene groups according to crop type, slope level and wind speed level; calculate the first historical deviation rate and the second historical deviation rate, and generate scene cancellation rule records containing the first cancellation threshold, the second cancellation threshold and the rule version identifier according to the preset quantile;

[0053] S200: Receive job requests based on the scenario verification rule record, and generate a job order bound to the rule version identifier; split the job order and generate task assignment records and subtask scheduling execution instructions based on robot resource status parameters;

[0054] S300: Based on the work order, receive execution feedback data, calculate the operation coverage area using the motion trajectory data of the spraying in the on state, and convert the actual pesticide consumption according to the pesticide metering data; calculate the first deviation rate and the second deviation rate, match the first verification threshold and the second verification threshold from the scene verification rule record bound to the work order according to crop type, slope level, and actual wind speed level, generate verification result record and update the verification status;

[0055] S400. Generate valid feedback samples based on the reconciliation result record, data integrity status, and drug dosage status; write the valid feedback samples into the corresponding scenario group; recalculate the preset quantile when the number of newly added valid samples meets the update trigger condition, generate a new rule version, and record the version migration relationship.

[0056] S100. Obtain historical operation samples and form scene groups according to crop type, slope level and wind speed level; calculate the first historical deviation rate and the second historical deviation rate, and generate scene cancellation rule records containing the first cancellation threshold, the second cancellation threshold and the rule version identifier according to the preset quantile.

[0057] This step is executed by the scene rule generation module in the robot artificial intelligence scheduling system. The scene rule generation module is deployed on the job scheduling server, which interacts with the job order database, plot archive database, robot execution record database, pesticide dosage record database, and meteorological data record database. The historical job samples are derived from completed and verified job orders; each historical job sample is associated with a job order identifier and includes crop type, plot slope, wind speed data, declared job area, job coverage area, theoretical pesticide consumption, actual pesticide consumption, verification status, data integrity status, and pesticide dosage status. The declared job area is the plot job area written into the job order when the job request is submitted; the job coverage area is the coverage area calculated based on the robot's movement trajectory data when spraying is in progress; the theoretical pesticide consumption is the product of the declared job area or sub-task target job area and the unit area application parameters; the actual pesticide consumption is the pesticide output amount calculated based on pesticide dosage data and aggregated according to the job order.

[0058] After receiving a rule initialization command or threshold update task, the scenario rule generation module reads historical operation samples according to the work order identifier. The rule initialization command is generated when the system first establishes a scenario verification rule record; the threshold update task is generated by S400 when the number of newly added valid samples in the corresponding scenario group reaches the update trigger value. The scenario rule generation module first performs field integrity verification; the field integrity verification includes reading the existence status, data type, and value status of each field, and setting historical operation samples lacking crop type, declared operation area, operation coverage area, theoretical pesticide consumption, or actual pesticide consumption to a data incomplete state. When the declared operation area is 0, the first historical deviation rate is not calculated for the historical operation sample; when the theoretical pesticide consumption is 0, the second historical deviation rate is not calculated for the historical operation sample, and the corresponding abnormal reason is written into the abnormal reason record.

[0059] After the field integrity verification is completed, the scenario rule generation module performs duplicate feedback verification. Specifically, it reads the job order identifier, subtask identifier, robot identifier, execution round, and feedback sequence number associated with the historical job samples. When multiple historical job samples have the same job order identifier, subtask identifier, robot identifier, execution round, and feedback sequence number, the historical job sample with the later data collection time and completed verification is retained, and the remaining historical job samples are set to duplicate feedback status. For job orders with supplementary records, the scenario rule generation module reads the status fields of the original feedback version and the supplementary version. If the supplementary version has been verified, it is set as the current valid version, and the original feedback version is retained in the execution log but does not participate in this quantile calculation.

[0060] The scenario rule generation module also performs pesticide metering status verification. The pesticide metering status is generated based on pesticide consumption pulse counts, pulse equivalents, calibration versions, and spraying status data. When pesticide consumption pulse counts are missing, the calibration version is not active, or continuous pesticide consumption pulses exist while spraying is off, the corresponding historical operation sample is set to an abnormal pesticide metering state. Historical operation samples that pass field integrity verification, duplicate feedback verification, and pesticide metering status verification are set to a valid sample state. Historical operation samples in incomplete data state, duplicate feedback state, or abnormal pesticide metering state retain the original fields and anomaly reason records, but are not included in the deviation rate set for this scenario grouping.

[0061] For historical work samples set as valid samples, the scene rule generation module reads the plot slope and wind speed data and performs grade mapping. The plot slope is derived from the slope record in the plot file; in one embodiment, the slope record is calculated based on the elevation difference between the highest and lowest points in the plot elevation data and the corresponding horizontal distance. The wind speed data is derived from the robot's meteorological sensor records or the plot's meteorological station records during the work period; when a work order corresponds to multiple wind speed sampling values, the work period is formed according to the start time to the end time of the work, the wind speed sampling values ​​within the work period are extracted, and wind speed values ​​for grade mapping are formed according to preset statistical rules. In one embodiment, the arithmetic mean of the wind speed sampling values ​​within the work period is used; in another embodiment, the value corresponding to the highest wind speed interval that lasts for a preset duration within the work period is used.

[0062] Slope and wind speed grades are generated from pre-configured grade mapping records. These records include continuous value ranges, range boundaries, and corresponding grade identifiers. When a plot's slope falls within a certain slope range, the corresponding slope grade is written; similarly, when a wind speed falls within a certain wind speed range, the corresponding wind speed grade is written. Crop types use standard type identifiers from crop archives. When the same crop has different names, the scene rule generation module maps the names to the same crop type based on the standardized crop type record. Subsequently, scene group identifiers are generated using a combination of crop type, slope grade, and wind speed grade. Samples with the same crop type, slope grade, and wind speed grade and their valid historical operation status are written into the same scene group.

[0063] The scenario rule generation module calculates a first historical deviation rate and a second historical deviation rate for historical operation samples in each scenario group. The first historical deviation rate is the absolute value of the difference between the operation coverage area and the declared operation area, divided by the declared operation area; the second historical deviation rate is the absolute value of the difference between the actual reagent consumption and the theoretical reagent consumption, divided by the theoretical reagent consumption. Each historical operation sample corresponds to one first historical deviation rate record and one second historical deviation rate record; the deviation rate record includes the historical operation sample identifier, scenario group identifier, deviation rate value, calculation time, and data source version. The scenario rule generation module aggregates multiple first historical deviation rate records according to the scenario group identifier to generate a first historical deviation rate set; and aggregates multiple second historical deviation rate records according to the same scenario group identifier to generate a second historical deviation rate set.

[0064] The scenario rule generation module reads the minimum sample size record and compares the number of samples in each scenario group with the minimum sample size. When the number of samples in a scenario group is not less than the minimum sample size, a first preset quantile is calculated for the first historical deviation rate set to obtain a first write-off threshold; a second preset quantile is calculated for the second historical deviation rate set to obtain a second write-off threshold. The first and second preset quantiles are stored in a quantile configuration record, which can use the same quantile or be configured separately. In one implementation, the deviation rate set is first arranged in ascending order of value, and then the quantile position is calculated based on the sample size and quantile parameters; when the quantile position falls between two adjacent deviation rate records, a linear calculation is performed based on the adjacent deviation rate values, and the result is written into the corresponding write-off threshold field.

[0065] When the number of samples in a scene group is less than the minimum sample size, the scene rule generation module matches the scene cancellation rule record of the superior scene group according to the small sample rollback rule. The small sample rollback rule includes the scene field rollback order and the rule source record; in one implementation, the crop type and slope level are kept unchanged, and the scene cancellation rule record of the adjacent wind speed level is matched; if no rule in effect is matched, the crop type is kept unchanged and the scene cancellation rule record that does not distinguish between slope levels is matched; if no rule is matched, the basic scene cancellation rule record is called. When calling the superior scene cancellation rule record, the scene group identifier, rule version identifier, and rollback level to be called are written into the rule source field.

[0066] After the first and second reversal thresholds are generated, the scenario rule generation module associates the scenario group identifier, the first reversal threshold, the second reversal threshold, the sample range, the rule version identifier, the rule effective status, the effective time, and the rule source to generate a scenario reversal rule record. The sample range includes a set of historical job sample identifiers involved in quantile calculation, the sample start time, and the sample end time; the rule version identifier is used to distinguish scenario reversal rule records from different generation batches. The same rule version may include scenario reversal rule records corresponding to multiple scenario groups; once a rule version enters the effective state, it is available for S200 to read when generating a new job order. Old rule versions retain their original rule effective status, expiration time, and sample range, allowing job orders already bound to this old rule version to be reversed or re-reversed in S300.

[0067] In one implementation, the job scheduling server reads 36 historical job samples of the same crop type from historical job orders that have been completed and verified. Four samples are excluded due to abnormal pesticide dosage status, two samples are duplicated due to duplicate feedback sequence numbers, and the remaining 30 historical job samples are divided into multiple scenario groups according to slope and wind speed levels. Each scenario group contains 12 valid sample status records, with the sample size not less than the configured minimum sample size. The scenario rule generation module calculates 12 first historical deviation rates and 12 second historical deviation rates, and generates first and second verification thresholds according to the corresponding preset quantiles. The scenario group identifier, the dual verification thresholds, the identifiers of the 12 historical job samples, and the rule version identifier are encapsulated into a single scenario verification rule record.

[0068] The scenario verification rule record generated in this step serves as the input for S200. The rule version identifier is used to write to the work order, the scenario group identifier and dual verification threshold are used by S300 for matching within the work order's bound version, and the sample range and rule source are used by S400 to record the update source when generating a new rule version. The scenario rule generation module also writes the rule generation time, the number of scenario groups involved in the calculation, the identifiers of scenario groups that did not meet the minimum sample size, and the source of their fallback rules to the execution log.

[0069] S200: Receive job requests based on the scenario verification rule record, and generate a job order bound to the rule version identifier; split the job order and generate task assignment records and subtask scheduling execution instructions based on robot resource status parameters.

[0070] This step is executed by the work order scheduling module. The work order scheduling module receives the scenario verification rule record and rule version identifier generated by S100, and receives work requests sent by the external work application system. The work request includes a work request identifier, plot identifier, plot boundary, declared work area, crop type, pest and disease level, planned work period, and application parameters per unit area; the plot slope is read from the plot archive database; and robot resource status parameters are obtained from the robot status interface. The work order scheduling module performs duplicate request verification based on the work request identifier; if a work order with the same work request identifier has already been generated, the existing work order is read, and a new work order is not generated.

[0071] The job order scheduling module first reads the planned job time period and compares its start time with the effective and expiration times of the rule version. If the start time of the planned job time period is after the effective time of the rule version but before its expiration time, the rule version is set as a candidate rule version for this job request. If multiple candidate rule versions exist, the rule version with the later effective time is selected. If no candidate rule version exists, the basic scenario verification rule record is read, and a "no matching rule" status and the corresponding rule source are written into the job order. After a candidate rule version is matched, the rule version identifier is written into the job order. This rule version identifier remains unchanged after the job order is generated, and subsequent new rule versions do not change the rule version identifier already bound to the job order.

[0072] The work order includes a work order identifier, a work request identifier, a plot identifier, plot boundaries, declared work area, crop type, slope grade, planned work period, application parameters per unit area, rule version identifier, and work order status. When a work order is generated, its status is set to "pending splitting." The work order is associated with and saved with the original fields of the work request. The plot slope in the plot archive database is mapped to a slope grade by the grade mapping record described in S100 and then written into the work order. The wind speed grade is not fixed as a final verification scenario field in this step. S300 generates the actual wind speed grade based on the wind speed data during the actual work period and matches the corresponding scenario verification rule record within the rule version already bound to the work order.

[0073] The work order scheduling module splits the work orders based on the land parcel boundaries and work zone records. The work zone records include a zone identifier, zone boundary, target work area, allowed work time period, and adjacent zone relationships. Work zones originate from the zone boundaries configured in the work request, or are split from the boundary records in the land parcel file according to the robot's single-operation area configuration. After splitting, each work zone corresponds to a subtask; the subtask includes a subtask identifier, work order identifier, zone identifier, target work area, theoretical reagent consumption, planned work time period, and subtask status.

[0074] The work order scheduling module calculates the theoretical pesticide consumption based on the product of the target operating area of ​​the subtask and the application parameters per unit area. The application parameters per unit area are read from the application parameter record based on crop type and pest / disease level. When multiple application parameter versions exist for the same crop type and pest / disease level, the version in effect is selected according to the planned operating period and written to the subtask. The theoretical pesticide consumption is written to the corresponding subtask and used as the calculation benchmark for the second deviation rate in S300. If the sum of the target operating areas of multiple subtasks is inconsistent with the declared operating area in the work order, the work order is set to a zone data inconsistency state, all subtask records are retained, and task assignment is stopped until the plot boundary or operating zone record is revised.

[0075] After a subtask is generated, the job order scheduling module obtains resource status parameters for multiple robots from the robot status interface. These parameters include robot identifier, robot operating status, remaining battery power, remaining reagent quantity, robot position, and data acquisition time. The robot operating status includes executable, in-process, maintenance, and communication interruption states; only robots in the executable state are added to the candidate robot set. The job order scheduling module compares the difference between the current system time and the data acquisition time with the data validity period; if the difference is not greater than the data validity period, the robot resource status parameters meet the data validity condition; if the difference is greater than the data validity period, the corresponding robot is set to have timed out and is not added to the candidate robot set.

[0076] For robots with valid status data, the work order scheduling module filters based on power constraints and reagent reserve constraints. The power constraints consist of the estimated power consumption of the sub-task and the safe return power. In one implementation, the estimated power consumption of the sub-task is calculated based on the target working area, the robot's power consumption per unit area, and the distance the robot travels from its position to the working area. The power constraint is considered successful if the robot's remaining power is not less than the sum of the estimated power consumption of the sub-task and the safe return power. The reagent reserve constraint consists of the theoretical reagent consumption of the sub-task and the safe reagent reserve. The reagent reserve constraint is considered successful if the robot's remaining reagent reserve is not less than the sum of the theoretical reagent consumption and the safe reagent reserve.

[0077] Robots that pass the constraints of power consumption, remaining reagent supply, and data validity are added to the candidate robot set. The candidate robot set includes a subtask identifier, multiple robot identifiers, corresponding remaining power consumption, remaining reagent supply, robot position, and resource status acquisition time. When multiple candidate robots exist, the job order scheduling module sorts them according to the distance between the robot's position and the job area; if the distances are the same, they are sorted from highest to lowest remaining power consumption. The robot ranked first is bound to a subtask, generating a task assignment record. The task assignment record includes a task assignment record identifier, job order identifier, subtask identifier, robot identifier, assignment time, resource status parameters, and assignment status.

[0078] The work order scheduling module generates subtask scheduling execution instructions based on the task assignment record. These instructions include the work order identifier, subtask identifier, robot identifier, work zone boundary, planned work period, application parameters per unit area, and execution round. After the subtask scheduling execution instruction is sent to the corresponding robot controller, the assignment status of the task assignment record is set to "issued." Upon receiving a receipt from the robot controller, the subtask status is set to "pending execution." If no receipt is received within the specified waiting time, the task assignment record is set to "receipt timeout," the binding relationship between the robot and the subtask is released, and a new task assignment record is generated based on the next-ranked robot in the candidate robot set.

[0079] In one implementation, a plot of land with a declared work area of ​​60 hm² is divided into three work zones, each generating a sub-task. The work order scheduling module reads the resource status parameters of four robots; one robot is in maintenance mode, one robot's remaining power does not meet the power constraint of the third sub-task, and the remaining two robots meet the power constraint and reagent remaining constraint of their respective sub-tasks. The work order scheduling module sorts candidate robots based on their distance from the work zone and generates three task assignment records; when the same robot executes two sub-tasks consecutively, it generates execution rounds and sub-task scheduling execution instructions for each sub-task.

[0080] This step outputs the work order, subtasks, task assignment records, and subtask scheduling execution instructions. The rule version identifier, crop type, slope grade, and declared work area in the work order serve as inputs for S300's scenario matching and rule verification records and for calculating the first deviation rate; the theoretical pesticide consumption in the subtask serves as input for S300's calculation of the second deviation rate; the work order identifier, subtask identifier, robot identifier, and execution round number serve as associated fields for S300's aggregated execution feedback data. The work order scheduling module writes assignments, receipts, reassignments, and subtask status changes to the execution log.

[0081] S300: Based on the work order, receive execution feedback data, calculate the operation coverage area using the motion trajectory data of the spraying in the on state, convert the actual pesticide consumption according to the pesticide metering data, calculate the first deviation rate and the second deviation rate, match the first verification threshold and the second verification threshold from the scene verification rule record bound to the work order according to crop type, slope level and actual wind speed level, generate verification result record and update the verification status.

[0082] This step is performed by the job feedback verification module. The job feedback verification module receives the job order, subtask, task assignment record, and subtask scheduling execution instruction generated by S200, and receives execution feedback data from the robot controller. The execution feedback data includes the job order identifier, subtask identifier, robot identifier, execution round, motion trajectory data, spraying status data, pesticide metering data, job start time, job end time, subtask completion status, and feedback sequence number. The motion trajectory data includes multiple trajectory points, each including position coordinates, acquisition time, and positioning accuracy; the spraying status data includes spray on status, spray off status, and corresponding status time; the pesticide metering data includes pesticide consumption pulse count, pulse equivalent, calibration version, and data acquisition time.

[0083] The job feedback verification module initiates execution feedback aggregation upon receiving a subtask completion status or reaching the feedback waiting time. Specifically, it queries the task assignment record using the job order identifier, subtask identifier, robot identifier, and execution round. If the execution feedback data matches the corresponding field in the task assignment record, the execution feedback data is written to the execution feedback record of that subtask. If the feedback sequence number already exists, this feedback is set to a duplicate feedback status, preventing the repeated accumulation of motion trajectory data and drug consumption pulse counts. If the robot identifier in the feedback does not match the task assignment record, the feedback is set to a source inconsistency status, while the original feedback fields are retained for subsequent verification.

[0084] After the feedback data is associated, the operation feedback verification module extracts the motion trajectory data corresponding to the spraying on-state within the operation period. The operation period is defined by the start and end times of the operation. When the motion trajectory data is collected within the operation period and the corresponding spraying state is "on," the corresponding trajectory point is written into the trajectory set to be verified. The trajectory points in the trajectory set to be verified are arranged according to the collection time. Trajectory points whose positioning accuracy does not meet the data validity conditions are set to invalid trajectory states. When the distance change between adjacent trajectory points exceeds the trajectory jump condition, the corresponding trajectory point is written into the exception reason record. Verified trajectory points constitute valid trajectory data.

[0085] The job feedback and verification module forms a coverage area based on valid trajectory data and robot spray width. The robot spray width is read from the robot equipment file or subtask scheduling execution instructions. In one implementation, a coverage band is formed on both sides of the line segment between adjacent valid trajectory points, extending along half of the robot spray width. Multiple coverage bands combine to form the coverage area for that execution round. If multiple coverage areas overlap, only one area is retained for the overlapping portion. When the same subtask is completed in multiple execution rounds, the coverage areas of multiple execution rounds are aggregated according to the subtask identifier, and overlapping portions are removed. When a work order includes multiple subtasks, the coverage areas of multiple subtasks are aggregated according to the work order identifier, generating a coverage area calculation record and the job coverage area.

[0086] The coverage area calculation record includes the work order identifier, subtask identifier, robot identifier, execution round, effective trajectory data range, robot spray width, coverage area, overlap removal area, work coverage area, and calculation time. If a subtask lacks motion trajectory data, or if the effective trajectory data cannot form a continuous coverage area, the work feedback and verification module sets the subtask to an incomplete trajectory data state. If a work order contains any incomplete trajectory data subtask that has not been supplemented, the work order's data integrity state is set to incomplete, the first deviation rate of the work order is not calculated, and a corresponding supplementary recording instruction is generated.

[0087] The operation feedback and verification module calculates the actual pesticide consumption based on the pesticide metering data. First, it extracts the pesticide consumption pulse count corresponding to the spraying on state during the operation period; pesticide consumption pulse counts corresponding to the spraying off state, equipment cleaning state, or pesticide emptying state are not added to the valid pesticide pulse set. Second, it reads the calibration version from the pesticide metering data and queries the pulse equivalent corresponding to that calibration version; if the calibration version is active during the operation period, the product of the pesticide consumption pulse count and the pulse equivalent is written into the single pesticide consumption amount. If the calibration version does not exist or is not active, the pesticide metering status is set to an abnormal pesticide metering status.

[0088] After the single-use reagent consumption is generated, the operation feedback and verification module aggregates the consumption based on the operation order identifier, subtask identifier, robot identifier, and execution round. When multiple pulse count segments exist within the same execution round, the single-use reagent consumption for each segment is summarized. When a subtask has supplementary flight execution rounds, the reagent consumption for multiple execution rounds that are already in a valid state is added together. When a single operation order includes multiple subtasks, the reagent consumption for each subtask is added together, generating a reagent consumption conversion record and the actual reagent consumption. The reagent consumption conversion record includes the calibration version, pulse equivalent, valid reagent consumption pulse count, excluded pulse count, execution round, and actual reagent consumption.

[0089] After the work coverage area and actual reagent consumption are generated, the work feedback verification module reads the declared work area and the theoretical reagent consumption of multiple sub-tasks from the work order. The theoretical reagent consumption of multiple sub-tasks is added together to form the theoretical reagent consumption of the work order. The first deviation rate is the absolute value of the difference between the work coverage area and the declared work area divided by the declared work area; the second deviation rate is the absolute value of the difference between the actual reagent consumption and the theoretical reagent consumption divided by the theoretical reagent consumption. The first deviation rate and the second deviation rate are respectively written into the verification evidence set; the verification evidence set also includes the work order identifier, rule version identifier, scene group identifier, coverage area calculation record, reagent consumption conversion record, data integrity status, and reagent measurement status.

[0090] The operation feedback verification module reads wind speed data during the operation period and generates the actual wind speed level according to the level mapping record used in S100. The wind speed data is preferentially read from the sampling records of the robot's meteorological sensor during the operation period; if the robot's meteorological sensor records are missing, wind speed data from the site meteorological station during the same operation period is read; if both types of wind speed data are missing, the scene matching status is set to "no matching rule". The operation feedback verification module forms a scene group identifier based on the crop type and slope level in the work order and the actual wind speed level generated in this step, and matches the corresponding scene verification rule record in the rule version already bound to the work order. If no corresponding scene verification rule record is matched, the rules of the higher-level scene group are read according to the small sample rollback rule, and the rule source is written into the verification evidence set.

[0091] After matching the scenario verification rule records, the job feedback verification module reads the first verification threshold and the second verification threshold. When the first deviation rate is not greater than the first verification threshold and the second deviation rate is not greater than the second verification threshold, a verification pass status is generated; when the first deviation rate is greater than the first verification threshold and the second deviation rate is not greater than the second verification threshold, a first deviation anomaly is generated; when the first deviation rate is not greater than the first verification threshold and the second deviation rate is greater than the second verification threshold, a second deviation anomaly is generated; when both the first and second deviation rates are greater than their respective verification thresholds, a double deviation anomaly is generated. When the data integrity status is incomplete, the aforementioned double threshold comparison is not performed, and a supplementary entry waiting status and a supplementary entry instruction are generated; when the drug measurement status is an abnormal drug measurement status, an anomaly reason record corresponding to the second deviation anomaly is generated, and the drug measurement data supplementary entry field is written into the supplementary entry instruction.

[0092] The reconciliation result record includes the work order identifier, rule version identifier, scenario group identifier, first deviation rate, second deviation rate, first reconciliation threshold, second reconciliation threshold, data integrity status, drug dosage status, reconciliation status, anomaly reason record, and generation time. After the reconciliation pass status is written to the work order, the work order's reconciliation status changes from pending reconciliation to reconciliation pass. After the first deviation anomaly, second deviation anomaly, or double deviation anomaly is written to the work order, the work order's reconciliation status changes to an anomaly status. After the supplementary data is written to the work order, the work order remains in the incomplete reconciliation status. After the supplementary data arrives, the work feedback reconciliation module reads the rule version identifier originally bound to the work order, re-aggregates and executes the feedback, and generates a new reconciliation result record. The original reconciliation result record retains its record version.

[0093] In one implementation, a work order includes two sub-tasks; the first sub-task is executed by robot R1, and the second sub-task is executed by robot R2. The work feedback verification module extracts the effective trajectory data of both robots in the spraying state, forming two coverage areas and removing overlaps to obtain the work coverage area. Then, it calculates the actual pesticide consumption based on the pulse equivalent of the corresponding calibration versions of the two robots and aggregates the data according to the work order identifier. The crop type in the work order is crop type A, the slope level is slope level 2, and the actual wind speed level is wind speed level 1. The work feedback verification module reads the double verification threshold corresponding to the scenario group from the rule version bound to the work order and generates a verification result record.

[0094] This step outputs the reconciliation result record, the reconciliation evidence set, and the updated reconciliation status. The scenario grouping identifier, double deviation rate, reconciliation status, data integrity status, and drug dosage status in the reconciliation result record serve as inputs for S400 to generate candidate feedback samples and perform sample admission verification; the rule version identifier serves as the grouping basis for S400 to accumulate the number of newly added valid samples since the current rule version took effect; and the coverage area calculation record and drug consumption conversion record serve as the data source records for the candidate feedback samples.

[0095] S400. Generate valid feedback samples based on the reconciliation result record, data integrity status, and drug dosage status; write the valid feedback samples into the corresponding scenario group; recalculate the preset quantile when the number of newly added valid samples meets the update trigger condition, generate a new rule version, and record the version migration relationship.

[0096] This step is executed by the rule feedback update module. The rule feedback update module receives the verification result record and verification evidence set generated by S300, and reads the work order's review status, duplicate feedback verification results, and the scenario group sample statistics record corresponding to the current rule version. The rule feedback update module first forms candidate feedback samples based on the verification result record; the candidate feedback samples include candidate feedback sample identifier, work order identifier, scenario group identifier, crop type, slope level, actual wind speed level, declared operation area, operation coverage area, theoretical pesticide consumption, actual pesticide consumption, first deviation rate, second deviation rate, verification status, data integrity status, pesticide dosage status, duplicate feedback verification results, rule version identifier, and sample generation time.

[0097] After candidate feedback samples are generated, the rule feedback update module generates sample admission status according to the sample admission criteria. These criteria include: data integrity status (not incomplete); drug dosage status (not abnormal); repeated feedback verification result (verification passed); and verification status (verification passed or review completed). When a candidate feedback sample simultaneously meets all the aforementioned conditions, it is set to the sample admission status and becomes a valid feedback sample; if any condition is not met, it is set to the sample exclusion status, and the unmet condition is written into the exception reason record.

[0098] Candidate feedback samples with a verified status directly enter the sample admission verification; candidate feedback samples with a verification status of first deviation anomaly, second deviation anomaly, or double deviation anomaly do not directly participate in threshold updates. In one implementation, after abnormal work orders are supplemented or reviewed on-site, the review record includes a review completion status and a reviewed work quantity field; when the review completion status indicates that the work order data can be used for historical samples, the rule feedback update module regenerates candidate feedback samples based on the currently valid fields after review and writes the verification status as a review completion status. Abnormal work orders caused by equipment failure, missing trajectory, or abnormal drug dosage remain excluded even after the business is completed and closed.

[0099] The rule feedback update module writes valid feedback samples to the corresponding scenario group based on the scenario group identifier. During writing, the source rule version identifier of the valid feedback sample is recorded; if a sample admission status record already exists for the same work order, the number of newly added valid samples is not increased repeatedly. Scenario group sample statistics records include the scenario group identifier, current rule version identifier, current rule version effective time, number of newly added valid samples, and most recent update time. If the sample generation time of a valid feedback sample is after the current rule version effective time, and its source rule version identifier is consistent with the current rule version identifier, the number of newly added valid samples is increased by 1.

[0100] The rule feedback update module compares the number of newly added valid samples with the update trigger value. When the number of newly added valid samples is less than the update trigger value, the rule validity status of the current scenario cancellation rule record is maintained, and the scenario group sample statistics record is updated. When the number of newly added valid samples reaches the update trigger value, a threshold update task is generated. The threshold update task includes the scenario group identifier, the current rule version identifier, the valid feedback sample identifier participating in the update, the range of update source samples, and the task status. If a threshold update task is already in execution for the same scenario group, a new threshold update task will not be generated again; valid feedback samples for subsequent admissions are retained in this scenario group for use in the next update round.

[0101] After the threshold update task is generated, the rule feedback update module locks the range of valid feedback samples participating in this update. The range of valid feedback samples includes valid feedback samples admitted between the time the current rule version takes effect and the time the threshold update task is generated. In one embodiment, the valid feedback samples and the historical operation samples corresponding to the current scenario write-off rule record in S100 are combined to form an update sample set. In another embodiment, valid feedback samples within a preset time range are used to form the update sample set. Each valid feedback sample participating in the update has a first deviation rate and a second deviation rate, and the rule feedback update module forms new sets of first and second deviation rates respectively.

[0102] The rule feedback update module recalculates the quantiles according to the first and second preset quantiles in S100, generating candidate first and candidate second write-off thresholds. During recalculation, the quantile parameters are read from the quantile configuration record corresponding to the current rule version; if the quantile configuration record has not changed, the original first and second preset quantiles are used. The candidate first and candidate second write-off thresholds are compared with the current first and second write-off thresholds, respectively, to generate the first threshold change and the second threshold change.

[0103] The rule feedback update module reads the threshold change limiting conditions. These conditions include the maximum single change of the first write-off threshold and the maximum single change of the second write-off threshold. When the change of a candidate write-off threshold relative to the current write-off threshold is not greater than the corresponding maximum single change, the candidate write-off threshold is written as the new write-off threshold. When the change of a candidate write-off threshold relative to the current write-off threshold is greater than the corresponding maximum single change, a new write-off threshold is generated by increasing or decreasing the maximum single change in the direction of change of the candidate write-off threshold, starting from the current write-off threshold. The first and second candidate write-off thresholds are compared as described above, and the processing result of the other threshold is not changed because one threshold exceeds the maximum single change.

[0104] After the new reimbursement threshold is generated, the rule feedback update module generates a new rule version. The new rule version includes a new rule version identifier, a scenario group identifier, a new first reimbursement threshold, a new second reimbursement threshold, an updated source sample range, quantile parameters, an effective time, and a rule effectiveness status. Before the effective time arrives, the new rule version is set to a candidate version pending activation; after the effective time arrives, it is set to an active state, and the current rule version is set to an inactive state. Already generated work orders retain their original bound rule version identifier; the new rule version is only bound to work orders generated after its effective time.

[0105] The rule feedback update module records the version migration relationship between the new rule version and the current rule version. The version migration record includes the scenario group identifier, current rule version identifier, new rule version identifier, current double-check threshold, new double-check threshold, threshold change, update source sample range, generation time, and effective time. When the generation of a new rule version fails, the threshold update task is set to a rule update failure state, and the current rule version remains effective; the candidate threshold, the identifier of the valid feedback sample participating in the update, and the reason for failure are written to the exception reason record. If incomplete fields or rule record reading failures occur after the new rule version takes effect, the previous rule version is restored to an effective state based on the version migration record, and a version rollback record is generated.

[0106] In one implementation, the current rule version for a certain scenario group is V1. After V1 takes effect, the number of newly added valid feedback samples reaches the configured update trigger value. The rule feedback update module generates a threshold update task and recalculates the first preset quantile and the second preset quantile based on the range of update source samples. If the change in the candidate first verification threshold relative to the current first verification threshold is less than the first maximum single change, the candidate first verification threshold is directly written as the new first verification threshold. If the change in the candidate second verification threshold relative to the current second verification threshold is greater than the second maximum single change, a new second verification threshold is generated according to the second maximum single change. The new dual verification threshold, the update source sample range, and the effective time are written into rule version V2, and a version migration record from V1 to V2 is established.

[0107] This step generates valid feedback samples, a new rule version, and a version migration record. The valid feedback samples retain their association with the reconciliation result record, the reconciliation evidence set, and the source work order. The new rule version is returned to S200 after taking effect, for binding when generating work orders in subsequent work requests. The version migration record is used by S300 to read the original rule version when reconciling historical work orders, and allows the next threshold update task to trace the update source of the current rule version. Thus, S100 generates the initial scenario reconciliation rule record, S200 writes the rule version identifier into the work order, S300 generates the reconciliation result record based on the version bound to the work order, and S400 generates subsequent rule versions using the valid feedback samples for admission.

[0108] Example 2: Figure 2 A structural block diagram of a robot artificial intelligence scheduling system based on job feedback according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0109] The scenario rule generation module 01 is used to acquire historical operation samples, form scenario groups according to crop type, slope level, and wind speed level, calculate the first historical deviation rate and the second historical deviation rate, and generate scenario verification rule records containing the first verification threshold, the second verification threshold, and the rule version identifier according to preset quantiles. Specifically, the scenario rule generation module receives historical operation samples, which include crop type, slope level, wind speed level, declared operation area, operation coverage area, theoretical pesticide consumption, and actual pesticide consumption. The scenario rule generation module matches the crop type, slope level, and wind speed level in the historical operation samples, and writes historical operation samples with the same matching result into the same scenario group. It generates the first historical deviation rate based on the operation coverage area and the declared operation area, and generates the first historical deviation rate based on the actual pesticide consumption. The consumption and theoretical reagent consumption are used to generate a second historical deviation rate; multiple first historical deviation rates and multiple second historical deviation rates are aggregated according to scenario groups, and a preset quantile is calculated on the aggregation results to form a first reversal threshold and a second reversal threshold; the scenario group, the first reversal threshold, the second reversal threshold, and the rule version identifier are associated to form a scenario reversal rule record; when the input field in the historical job sample is missing, the historical job sample is retained and writing it into the corresponding scenario group is stopped; when the scenario group does not form a corresponding first reversal threshold or second reversal threshold, the scenario reversal rule record of the scenario group is kept in an inactive state; the scenario rule generation module provides the job work order scheduling module with the scenario reversal rule record and rule version identifier in an active state, and retains the association record between the scenario group and the historical job sample.

[0110] The work order scheduling module 02, connected to the scene rule generation module, is used to receive work requests based on the scene verification rule record, generate work orders bound to the rule version identifier, split the work orders, and generate task assignment records and subtask scheduling execution instructions according to robot resource status parameters. Specifically, the work order scheduling module receives the scene verification rule record and rule version identifier output by the scene rule generation module, and receives work requests containing crop type, slope level, declared work area, and planned work period. The work order scheduling module reads the scene verification rule record that is in effect according to the planned work period, writes the corresponding rule version identifier into the work order, and maintains the binding relationship between the work order and the rule version identifier. The work order is split into multiple subtasks according to the work area and declared work area in the work request, and the target work area and target work area are recorded for each subtask. Theoretical reagent consumption; the work order scheduling module receives robot resource status parameters, including robot working status, remaining power, reagent balance, robot position, and data acquisition time; it performs status matching on the robot working status, remaining power, reagent balance, and data acquisition time, and writes the robots that meet the sub-task execution conditions into the corresponding sub-task task assignment record; when robot resource status parameters are missing or the data acquisition time does not meet the data validity conditions, no task assignment record is generated for that robot; a sub-task scheduling execution instruction is formed based on the task assignment record, and the sub-task scheduling execution instruction is sent to the corresponding robot; the work order scheduling module provides the work order, rule version identifier, task assignment record, and sub-task scheduling execution instruction to the work feedback verification module, wherein the work order identifier, robot identifier, and sub-task identifier are used as matching fields for the execution feedback data.

[0111] The job feedback and verification module 03, connected to the job order scheduling module, is used to receive execution feedback data based on the job order and task assignment record, calculate the job coverage area using the motion trajectory data corresponding to the spraying start state, convert the actual pesticide consumption based on the pesticide metering data, calculate the first deviation rate and the second deviation rate, match the first verification threshold and the second verification threshold from the scene verification rule record bound to the job order according to crop type, slope level, and actual wind speed level, generate a verification result record, and update the verification status; specifically, the job feedback and verification module receives the job output by the job order scheduling module. The system records work orders, rule version identifiers, and task assignments, and receives execution feedback data according to the work order identifier, subtask identifier, and robot identifier. The execution feedback data includes motion trajectory data, spraying status data, pesticide metering data, and actual wind speed level. The work feedback verification module matches the motion trajectory data with the spraying status data according to the collection time, extracts the motion trajectory data corresponding to the spraying start state, and forms a coverage area based on the motion trajectory data. Multiple coverage areas belonging to the same work order are aggregated, and overlapping parts are deduplicated to generate the work coverage area. The pesticide metering data is then compared with the spraying status data. Status data is matched according to collection time, and the pesticide metering data corresponding to the spraying start status is extracted and aggregated according to the work order identifier and sub-task identifier to generate the actual pesticide consumption. A first deviation rate is generated based on the operation coverage area and the declared operation area in the work order, and a second deviation rate is generated based on the actual pesticide consumption and the theoretical pesticide consumption in the work order. The operation feedback verification module reads the crop type and slope level in the work order and combines them with the actual wind speed level in the execution feedback data to form corresponding scene groups. The scene verification rule corresponding to the rule version identifier bound to the work order is then used for verification. The record matches the first and second verification thresholds for the scenario group; the first and second deviation rates are compared with the corresponding verification thresholds to generate verification result records, and the comparison results are written to the verification status; when motion trajectory data is missing, the data integrity status is recorded as the data incomplete status; when drug dosage data is missing or its recording status is abnormal, the drug dosage status is recorded as the drug dosage abnormal status; the operation feedback verification module provides the rule feedback update module with verification result records, data integrity status, drug dosage status, scenario group, first deviation rate, second deviation rate, and rule version identifier.

[0112] The rule feedback update module 04, connected to the job feedback verification module, is used to generate valid feedback samples based on the verification result record, data integrity status, and drug dosage status. It writes the valid feedback samples into the corresponding scenario group. When the number of newly added valid samples meets the update trigger condition, it recalculates the preset quantile, generates a new rule version, and records the version migration relationship. Specifically, the rule feedback update module receives the verification result record, data integrity status, drug dosage status, scenario group, first deviation rate, second deviation rate, and rule version identifier output by the job feedback verification module. It matches the data integrity status, drug dosage status, and verification status. When the data integrity status is not the data incomplete status, the drug dosage status is not the drug dosage abnormal status, and the verification status meets the sample admission condition, it converts the corresponding verification result record into a valid feedback sample. The valid feedback sample records the job order identifier, scenario group, first deviation rate, second deviation rate, and source rule version identifier. The rule feedback update module writes valid feedback samples according to scenario group and accumulates the number of newly added valid samples in the corresponding scenario group after the current rule version takes effect. When a valid feedback sample has already been generated for the same job order... The number of newly added valid samples is not repeatedly accumulated; when the number of newly added valid samples does not meet the update trigger condition, the current rule version remains unchanged; when the number of newly added valid samples meets the update trigger condition, multiple first deviation rates and multiple second deviation rates of the corresponding scenario group are aggregated, and the preset quantiles are recalculated to form new first and second verification thresholds; the new first verification threshold, second verification threshold, scenario group, effective time, and new rule version identifier are associated to generate a new rule version; the current rule version identifier, new rule version identifier, and effective time are written into the version migration relationship; the rule feedback update module provides the new rule version in the effective state to the scenario rule generation module and the job work order scheduling module, so that subsequent job requests can form new job work orders, and job work orders that have been bound to the current rule version identifier continue to retain the original rule version identifier.

Claims

1. A robot artificial intelligence scheduling method based on task feedback, characterized in that, include: S100. Obtain historical operation samples and form scene groups according to crop type, slope level and wind speed level; calculate the first historical deviation rate and the second historical deviation rate, and generate scene cancellation rule records containing the first cancellation threshold, the second cancellation threshold and the rule version identifier according to the preset quantile; S200: Receive job requests based on the scenario verification rule record, and generate a job order bound to the rule version identifier; split the job order and generate task assignment records and subtask scheduling execution instructions based on robot resource status parameters; S300: Receive execution feedback data based on the work order, calculate the work coverage area using the motion trajectory data of the spraying in the on state, and convert the actual pesticide consumption based on the pesticide metering data. Calculate the first deviation rate and the second deviation rate, match the first and second verification thresholds from the scenario verification rule records bound to the work order according to crop type, slope level, and actual wind speed level, generate verification result records, and update the verification status. S400. Generate a valid feedback sample based on the verification result record, data integrity status, and drug dosage status. Valid feedback samples are written into the corresponding scenario group. When the number of newly added valid samples meets the update trigger condition, the preset quantile is recalculated, a new rule version is generated, and the version migration relationship is recorded.

2. The robot artificial intelligence scheduling method based on job feedback according to claim 1, characterized in that, The historical operation samples are derived from completed and verified work orders, and include crop type, plot slope, wind speed data, declared operation area, operation coverage area, theoretical pesticide consumption, actual pesticide consumption, verification status, data integrity status, and pesticide measurement status. Perform field integrity verification, duplicate feedback verification, and reagent dosage status verification on the historical operation samples, and set the historical operation samples that pass the verification to the sample valid status; The plot slope is mapped to a slope grade, and the wind speed data is mapped to a wind speed grade. Scene grouping identifiers are generated based on crop type, slope grade, and wind speed grade.

3. The robot artificial intelligence scheduling method based on job feedback according to claim 2, characterized in that, The first historical deviation rate and the first deviation rate are the absolute values ​​of the difference between the operating coverage area and the declared operating area, divided by the declared operating area. The second historical deviation rate and the second deviation rate are the absolute value of the difference between the actual drug consumption and the theoretical drug consumption, divided by the theoretical drug consumption; A first preset quantile is calculated on the first historical deviation rate set to obtain a first write-off threshold, and a second preset quantile is calculated on the second historical deviation rate set to obtain a second write-off threshold. When the number of samples in a scenario group is less than the minimum sample size, the scenario cancellation rule record of the higher-level scenario group is matched according to the small sample rollback rule.

4. The robot artificial intelligence scheduling method based on job feedback according to claim 1, characterized in that, Based on the planned work period, filter the effective status of the rules and write the version identifier of the rules that are in effect into the work order; The rule version identifier remains unchanged after the work order is generated; The new rule version is linked to the work orders generated after its effective date; When re-verifying the work order, the rule version identifier originally bound to the work order is read.

5. The robot artificial intelligence scheduling method based on job feedback according to claim 1, characterized in that, The work order is divided into sub-tasks according to the work area, and the theoretical pesticide consumption is calculated based on the target work area and unit area application parameters of the sub-tasks. The robot resource status parameters include robot identifier, robot working status, remaining battery power, remaining medicine, robot position, and data acquisition time. A candidate robot set is generated by screening robots whose working status is executable, whose remaining power meets the power constraint, whose remaining drug quantity meets the drug quantity constraint, and whose data acquisition time meets the data validity condition. A task assignment record is generated based on the set of candidate robots.

6. The robot artificial intelligence scheduling method based on job feedback according to claim 1, characterized in that, The execution feedback data includes work order identifier, subtask identifier, robot identifier, execution round, motion trajectory data, and spraying status data; Aggregate execution feedback data according to the work order identifier, subtask identifier, robot identifier, and execution round; Extract the motion trajectory data corresponding to the spraying activation status during the operation period to generate valid trajectory data; Based on the effective trajectory data and the robot's spray pattern, a coverage area is formed. The overlapping parts of multiple coverage areas are deduplicated to generate a coverage area calculation record and the operational coverage area.

7. The robot artificial intelligence scheduling method based on job feedback according to claim 1, characterized in that, The drug metering data includes drug consumption pulse count, pulse equivalent, calibration version, and spraying status data; Extract the pesticide consumption pulse count corresponding to the spraying activation state during the operation period, and generate the single pesticide consumption amount based on the product of the pesticide consumption pulse count and the pulse equivalent. The single-time drug consumption is aggregated according to the work order identifier, sub-task identifier, and execution round to generate a drug consumption conversion record and the actual drug consumption.

8. The robot artificial intelligence scheduling method based on job feedback according to claim 1, characterized in that, When the first deviation rate is not greater than the first write-off threshold and the second deviation rate is not greater than the second write-off threshold, a write-off pass status is generated. When the first deviation rate is greater than the first write-off threshold and the second deviation rate is not greater than the second write-off threshold, a first deviation anomaly is generated. When the first deviation rate is not greater than the first write-off threshold and the second deviation rate is greater than the second write-off threshold, a second deviation anomaly is generated. When both the first deviation rate and the second deviation rate are greater than the corresponding write-off threshold, a double deviation anomaly is generated. When the data integrity status changes to the data incomplete status, a supplementary entry waiting status and a supplementary entry instruction are generated. Write the verification status, first deviation anomaly, second deviation anomaly, double deviation anomaly, or supplementary entry waiting status into the verification result record.

9. The robot artificial intelligence scheduling method based on job feedback according to claim 1, characterized in that, The valid feedback samples meet the following sample admission criteria: The data is complete, not incomplete; the dosage status is not abnormal; the repeated feedback verification result is verified; and the cancellation status is either cancelled or completed. Accumulate the number of newly valid samples since the current rule version took effect, and generate a threshold update task when the number of newly valid samples reaches the update trigger value; Recalculate the first preset quantile and the second preset quantile to generate candidate first write-off threshold and candidate second write-off threshold; When the change in the candidate write-off threshold relative to the current write-off threshold is greater than the maximum single change, a new write-off threshold is generated based on the maximum single change. The new write-off threshold, updated source sample range, and effective time will be written into the new rule version.

10. A robot artificial intelligence scheduling system based on job feedback, characterized in that, include: Modules for generating scene rules, scheduling work orders, verifying work feedback, and updating rules; The system is used to implement the method described in any one of claims 1-9.

Images