Multi-robot collaborative semiconductor material scheduling method and system

Abstract

The invention discloses a multi-robot collaborative semiconductor material scheduling method and system, and relates to the technical field of robot collaboration. The method comprises the following steps: constructing a scheduling prediction model, and generating an initial allocation and path planning scheme by using a regression network; scheduling efficiency coefficients are calculated in combination with transportation demand parameters, resource states are matched in a partition mode by dynamically adjusting the coefficients, load reduction is carried out in a resource surplus area, maintenance is carried out in a matching area, emergency capacity expansion is carried out in a tension area, and a standby robot is triggered; comparing the path conflict, the delay time, the utilization rate and the difference of a standard value in real time, generating a comparison result CR by adopting a dynamic weight algorithm, and adaptively enhancing the weight along with the conflict times and the delay proportion; and adjusting the path and the rhythm through a correction model based on a comparison result, if the CR does not reach a threshold value, performing iterative optimization, and starting exception processing when the CR is not improved. According to the method, through data-driven prediction, dynamic resource adaptation, real-time feedback correction and a fault-tolerant mechanism, the semiconductor material scheduling efficiency and reliability are improved, and the high-precision manufacturing requirement is met.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of robot cooperation, and particularly relates to a semiconductor material scheduling method and system for multi-robot cooperation. BACKGROUND

[0002] In the field of semiconductor manufacturing, the high precision and low delay requirements of material scheduling pose severe challenges to multi-robot cooperation systems. The traditional scheduling method has the following core problems: Lack of dynamic adaptability: static rules or simple heuristic algorithms are difficult to cope with real-time fluctuations in material urgency, quantity and robot state in semiconductor production, which easily leads to unbalanced resource allocation (such as some robots being idle while others being overloaded).

[0003] Low resource utilization: lack of dynamic adjustment mechanism based on efficiency coefficient, unable to actively reduce load when resources are excessive, unable to quickly expand capacity when resources are tight, resulting in high idle rate or insufficient emergency capacity.

[0004] Path conflict and delay accumulation: lack of real-time feedback correction after initial path planning, path conflict and transportation delay are easily exacerbated by production fluctuations, affecting overall scheduling efficiency.

[0005] Absence of exception handling: no iteration termination condition and fault tolerance mechanism are set, which easily falls into a dead loop when scheduling anomalies persist, lacks backup resource allocation and manual intervention process, and has poor reliability.

[0006] To solve the above problems, the application provides a semiconductor material scheduling method and system for multi-robot cooperation. SUMMARY

[0007] In order to overcome the shortcomings and deficiencies of the prior art, the first object of the application is to provide a semiconductor material scheduling method for multi-robot cooperation; the second object of the application is to provide a semiconductor material scheduling system for multi-robot cooperation.

[0008] The first object of the application adopts the following technical solutions: The semiconductor material scheduling method for multi-robot cooperation has the following process: Step one, obtain the material characteristic parameters of the semiconductor material and the robot resource parameters, input the material characteristic parameters and the robot resource parameters into the pre-constructed scheduling scheme prediction model, obtain the initial robot allocation scheme and the initial path planning parameters; wherein the scheduling scheme prediction model includes a robot allocation sub-model and a path planning sub-model; Step two, obtain the transportation demand parameters of the semiconductor material, calculate the scheduling efficiency coefficient based on the material characteristic parameters, the robot resource parameters and the transportation demand parameters, and determine the initial scheduling efficiency parameter according to the scheduling efficiency coefficient and the preset scheduling efficiency parameter; Step 3: Obtain real-time scheduling status data and compare it with standard scheduling status data to generate comparison results; Step 4: Input the comparison results, initial path planning parameters, and initial scheduling efficiency parameters into the scheduling correction model to obtain adjustment data; adjust the robot path and scheduling rhythm according to the adjustment data, and determine whether to return to step 3 or deliver materials to the next process based on the comparison results.

[0009] Preferably, the method for constructing the scheduling scheme prediction model includes: Obtain the sample scheduling dataset and divide it into a sample training set and a sample test set; A regression network is constructed, with historical material feature parameters and historical robot resource parameters in the sample training set as input data, and historical robot allocation schemes and historical path planning parameters as output data, to train the initial regression network. The initial regression network is tested using a sample test set, and the output regression network that meets the preset error requirements is used as the scheduling scheme prediction model.

[0010] Preferably, the method for calculating the scheduling efficiency coefficient includes: The calculation formula is constructed based on material urgency, material quantity, robot load rate, robot idle time, and transportation volume per unit time. Among them, the material quantity is processed using a logarithmic function to prevent numerical overflow, the robot load rate is mapped to the nonlinear range through an arcsine function, the material urgency is amplified by an exponential function, and the robot idle time is processed by adding 1 to the reciprocal to reflect the impact of busyness on efficiency.

[0011] Preferably, the method for determining the initial scheduling efficiency parameters includes: The preset efficiency coefficient range and threshold are defined. When the scheduling efficiency coefficient is less than or equal to the resource surplus threshold, the preset scheduling efficiency parameter is multiplied by the surplus resource adjustment coefficient. When the scheduling efficiency coefficient is between the resource surplus threshold and the resource shortage threshold, the preset scheduling efficiency parameter is multiplied by the benchmark adjustment coefficient. When the scheduling efficiency coefficient is greater than the resource shortage threshold, the preset scheduling efficiency parameter is multiplied by the emergency adjustment coefficient; Among them, the emergency adjustment coefficient is greater than the benchmark adjustment coefficient, and the benchmark adjustment coefficient is greater than the surplus resource adjustment coefficient.

[0012] Preferably, the method for generating the comparison results includes: Extract the real-time feature set of real-time scheduling status data and the standard feature set of standard scheduling status data. The real-time feature set includes the number of path conflicts, transportation delay time and robot utilization. The computational relationship is constructed based on the difference between the real-time feature set and the standard feature set. The weight of the path conflict term is dynamically adjusted with the number of conflicts, the weight of the delay time term is positively correlated with the delay ratio, and the utilization rate term adopts a fixed weight and a reward and punishment mechanism is set.

[0013] Preferably, the method for constructing the scheduling correction model includes: Obtain the sample calibration dataset and divide it into a sample training set and a sample test set; A regression network is constructed, with historical comparison results, historical initial path planning parameters, and historical initial efficiency parameters in the sample training set as input data, and historical corrected path planning parameters and historical corrected efficiency parameters as output data, to train the initial regression network. The initial regression network is tested using a sample test set, and the regression network that meets the preset error requirements is used as the scheduling correction model.

[0014] Preferably, the method for determining whether to return to step three or deliver materials to the next process based on the comparison results includes: If the comparison result is greater than or equal to the preset comparison threshold, the initial path planning parameters and initial scheduling efficiency parameters will be replaced with the corrected parameters, and the process will return to step three. If the comparison result is less than the preset comparison threshold, the material scheduling task will be delivered to the next process. The preset comparison threshold is determined based on the average deviation value of the historical best scheduling batches.

[0015] Preferably, a maximum iteration count limit mechanism is also included: If the comparison results do not improve after a preset number of adjustments, the current path planning is paused and a backup robot is allocated first, and the scheduling efficiency coefficient is recalculated. At the same time, an alarm signal is sent to the dispatch center to prompt manual intervention, and abnormal data is recorded in the log for subsequent model optimization.

[0016] The second objective of this invention is achieved through the following technical solution: A multi-robot collaborative semiconductor material scheduling system, used to implement a method for multi-robot collaborative semiconductor material scheduling, the system includes... Data acquisition module: Acquires material characteristic parameters and robot resource parameters of semiconductor materials; material characteristic parameters include material urgency and material quantity; robot resource parameters include robot load rate and robot idle time. Scheduling prediction module: Based on a pre-built scheduling scheme prediction model, it takes historical material characteristic parameters and historical robot resource parameters as input and outputs initial robot allocation scheme and initial path planning parameters; the scheduling scheme prediction model is obtained through regression network training; Efficiency Calculation Module: Calculates scheduling efficiency coefficients based on material characteristic parameters, robot resource parameters, and transportation demand parameters. Combines preset scheduling efficiency parameters with dynamically adjusted excess resource adjustment coefficient A, benchmark adjustment coefficient B, and emergency adjustment coefficient C to determine the initial scheduling efficiency parameters. Status Comparison Module: Acquires real-time scheduling status data and generates comparison results with standard scheduling status data; real-time scheduling status data includes path conflict count, transportation delay time, and robot utilization rate. Scheduling correction module: Based on the comparison results, initial path planning parameters, and initial scheduling efficiency parameters, the scheduling correction model outputs corrected path planning parameters and corrected scheduling efficiency parameters; the scheduling correction model is trained through a regression network. Iterative execution module: Adjusts the robot path and scheduling rhythm according to the correction parameters. If the comparison result does not reach the preset threshold, it will perform cyclic correction. If there is no improvement after a set number of consecutive times, it will trigger the abnormal handling process.

[0017] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention constructs a scheduling scheme prediction model based on a regression network. Using historical material characteristic parameters and robot resource parameters as input, it accurately predicts the initial robot allocation scheme and path planning parameters, solving the problem of initial scheme deviation caused by traditional scheduling relying on experience. Simultaneously, a dynamic adjustment mechanism for the scheduling efficiency coefficient is designed. The adjustment coefficient is adaptively selected based on a coefficient range, and the coefficient value is optimized in real time through a feedback calibration process. This achieves intelligent matching between resource status and scheduling rhythm, improving scheduling adaptability under different resource load scenarios.

[0018] 2. This invention introduces a mechanism for comparing real-time scheduling status with standard status. By quantifying the differences in features such as path conflicts, transportation delays, and robot utilization, and combining this with a scheduling correction model to dynamically output correction parameters, a closed-loop optimization process of prediction-execution-monitoring-correction is formed. A weight adjustment strategy is adopted to dynamically assign weights to key indicators such as conflicts and delays, ensuring correction accuracy in deviation-sensitive scenarios. Simultaneously, the correction model is optimized through sample training to continuously improve the accuracy of parameter adjustments and reduce the risk of fluctuations during the scheduling process.

[0019] 3. This invention incorporates a maximum iteration limit and an exception handling mechanism. When continuous adjustments fail to improve scheduling performance, it automatically triggers backup robot allocation, manual intervention alarms, and exception log recording, resolving the failure and stagnation issues of traditional scheduling in continuously abnormal scenarios. By linking the scheduling efficiency coefficient with a dynamic robot expansion mechanism, an emergency mode is quickly activated when resources are scarce. Combined with full-process data recording to support model iterative optimization, this enhances the robustness of the scheduling system in complex scenarios such as high urgency and resource fluctuations, meeting the stringent requirements of semiconductor manufacturing for precise scheduling. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart of the multi-robot collaborative semiconductor material scheduling method of the present invention is shown; Figure 2 A block diagram of the multi-robot collaborative semiconductor material scheduling system of the present invention is shown; Figure 3 A flowchart of step three of the present invention is shown. Detailed Implementation

[0022] 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.

[0023] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of exemplary embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0024] Example 1: See Figure 1 As shown, the process of the multi-robot collaborative semiconductor material scheduling method in this embodiment is as follows: Step 1: Obtain the material characteristic parameters and robot resource parameters of the semiconductor materials, and input the material characteristic parameters and robot resource parameters into the pre-built scheduling scheme prediction model to obtain the initial robot allocation scheme and initial path planning parameters; the scheduling scheme prediction model includes a robot allocation sub-model and a path planning sub-model.

[0025] The method for constructing a scheduling scheme prediction model includes: acquiring a sample scheduling dataset, dividing the sample scheduling dataset into a sample training set and a sample test set, constructing a regression network, using historical material characteristic parameters and historical robot resource parameters in the sample training set as input data for the regression network, using historical robot allocation schemes and historical path planning parameters in the sample training set as output data for the regression network, training the regression network to obtain an initial regression network for predicting real-time initial robot allocation schemes and initial path planning parameters, testing the initial regression network using the sample test set, and outputting an initial regression network that satisfies a preset error value as the scheduling scheme prediction model. Material characteristic parameters include material urgency and quantity; robot resource parameters include robot load rate and robot idle time; and transportation demand parameters are the transportation volume per unit time. Methods for calculating the scheduling efficiency coefficient based on material characteristic parameters, robot resource parameters, and transportation demand parameters include: ; This is the scheduling efficiency coefficient; The amount of transportation per unit time is directly positively correlated with efficiency. Robot load rate; Robot idle time; Let be the quantity of materials, and log10(・) be the logarithmic function to base 10. To ensure the positive effect of material quantity on efficiency, a logarithmic function is used to prevent excessive quantity from causing overflow; sin⁻¹(・) is the arcsine function, and e is the natural constant. Mapping the robot load rate to the nonlinear range (0, π / 2) demonstrates the positive impact of a high load rate on efficiency; The urgency of materials is amplified exponentially to prioritize urgent tasks. The lower the idle rate (the busier the worker), the larger the denominator and the lower the efficiency.

[0026] Step 2: Obtain the transportation demand parameters of semiconductor materials; calculate the initial scheduling efficiency parameters based on material characteristic parameters, robot resource parameters, transportation demand parameters, and preset scheduling efficiency parameters; calculate the scheduling efficiency coefficient based on material characteristic parameters, robot resource parameters, and transportation demand parameters; and determine the initial scheduling efficiency parameters based on the scheduling efficiency coefficient and preset scheduling efficiency parameters.

[0027] Methods for determining the initial scheduling efficiency parameters based on the scheduling efficiency coefficient and preset scheduling efficiency parameters include: Construct a coefficient prediction sub-model, taking historical scheduling efficiency coefficients, material urgency, robot type, etc. as inputs, and output recommended values ​​for excess resource adjustment coefficient A, baseline adjustment coefficient B, and emergency adjustment coefficient C in real time (e.g., for high-urgency materials, C is automatically increased to 1.4~1.5; for low-urgency materials, A is reduced to 0.6). Feedback calibration process: After each batch of scheduling is completed, A, B, and C are optimized in reverse using the actual scheduling effect (such as latency rate and resource utilization rate). For example, if C=1.3 still results in frequent delays in resource-scarce areas, C will be automatically increased to 1.4.

[0028] When the scheduling efficiency coefficient is less than or equal to the preset resource surplus threshold coefficient, the preset scheduling efficiency parameter is multiplied by the surplus resource adjustment coefficient A to obtain the initial scheduling efficiency parameter. When the scheduling efficiency coefficient is greater than the preset resource surplus threshold coefficient and less than the preset resource shortage threshold coefficient, the preset scheduling efficiency parameter is multiplied by the benchmark adjustment coefficient B to obtain the initial scheduling efficiency parameter. When the scheduling efficiency coefficient is greater than the preset resource shortage threshold coefficient, the preset scheduling efficiency parameter is multiplied by the emergency adjustment coefficient C to obtain the initial scheduling efficiency parameter, where C > B > A > 0.

[0029] Initial scheduling efficiency parameter = preset scheduling efficiency parameter ; Preset efficiency coefficient range: [Thr1, Thr2] Preset efficiency thresholds: Thr1=0.6, Thr2=0.8 (needs to be calibrated based on historical data).

[0030] When the scheduling efficiency coefficient is less than or equal to Thr1, it is marked as a resource surplus zone: prioritize reducing the robot idle rate, and the initial scheduling rhythm parameter is multiplied by A (A=0.7). When Thr1 < scheduling efficiency coefficient ≤ Thr2, it is marked as a resource matching zone: maintain the current rhythm, parameter × B (B = 1.0). When the scheduling efficiency coefficient > Thr2, it is marked as a resource-scarce area: the emergency scheduling mode is activated, parameter × C (C=1.3), and the robot dynamic expansion mechanism is triggered.

[0031] Step 3: Obtain real-time scheduling status data, and generate a comparison result based on the real-time scheduling status data and standard scheduling status data. Specifically: See Figure 3 As shown, the steps are as follows: The real-time feature set of real-time scheduling status data and the standard feature set of standard scheduling status data are obtained respectively. The comparison results are calculated based on the real-time feature set and the standard feature set. The real-time feature set includes the number of path conflicts, transportation delay time and robot utilization rate, while the standard feature set includes the number of standard conflicts, standard delay time and standard utilization rate.

[0032] Methods for calculating comparison results based on real-time feature sets and standard feature sets include: A database is constructed based on a set of standard features: the number of standard conflicts is divided into multiple standard conflict segments, and the standard latency and standard utilization value corresponding to each standard conflict segment are obtained to form a standard conflict-latency-utilization mapping table; the standard latency time is divided into multiple standard latency segments, and the standard conflict and standard utilization value corresponding to each standard latency segment are obtained to form a standard latency-conflict-utilization mapping table; the standard utilization rate is divided into multiple standard utilization segments, and the standard conflict and standard latency value corresponding to each standard utilization segment are obtained to form a standard utilization-conflict-latency mapping table.

[0033] The real-time features are segmented as follows: the real-time conflict count is divided into intervals based on the standard conflict segment to obtain the real-time conflict segment and the corresponding real-time delay and real-time utilization value; the real-time delay time is divided into intervals based on the standard delay segment to obtain the real-time delay segment and the corresponding real-time conflict and real-time utilization value; the real-time utilization rate is divided into intervals based on the standard utilization segment to obtain the real-time utilization segment and the corresponding real-time conflict and real-time delay value.

[0034] Calculate the similarity between real-time collision segments and standard collision segments. : ;in, This represents the delay amount corresponding to the real-time conflict segment; This represents the delay amount corresponding to the standard conflict segment; This represents the utilization value corresponding to the real-time conflict segment; This is the utilization value corresponding to the standard conflict segment. When the similarity exceeds the preset threshold (e.g., 0.8), it is recorded as a valid match, and the number of conflict matches is counted.

[0035] Calculate the similarity between the real-time delay segment and the standard delay segment. : ;in, This represents the conflict quantity corresponding to the real-time delay segment. This represents the collision amount corresponding to the standard delay segment; This represents the utilization value corresponding to the real-time delay segment. This represents the utilization rate value corresponding to the standard delay segment. When the similarity exceeds the preset threshold, it is recorded as a valid match, and the number of delayed matches is counted. Calculate the similarity between the real-time utilization segment and the standard utilization segment. : ;in, To utilize the corresponding conflict amount of segments in real time; The standard utilization segment corresponds to the conflict quantity; To utilize the corresponding delay value of the segment in real time; The standard segment uses the corresponding delay value. When the similarity exceeds the preset threshold, it is recorded as a valid match, and the number of matches used is counted.

[0036] Calculation based on comprehensive comparison results: ; where CR represents the alignment result; Weight the number of conflict matches (0.4); Weight for delayed matching quantity (0.4); To utilize the matching quantity weight (0.2). , , These represent the number of valid matches for the conflict, delay, and exploitation dimensions, respectively. , , These represent the maximum possible number of matches for each dimension.

[0037] The comparison results (CR), initial path planning parameters, and initial scheduling efficiency parameters are input into the scheduling correction model to obtain adjustment data. The adjustment data includes the corrected path planning parameters and the corrected scheduling efficiency parameters. Real-time scheduling status data is generated by executing material scheduling based on the initial robot allocation scheme and initial path planning parameters, and by monitoring based on the initial scheduling efficiency parameters. The construction method of the scheduling correction model includes: obtaining a sample correction dataset, dividing the sample correction dataset into a sample training set and a sample test set, constructing a regression network, using historical comparison results, historical initial path planning parameters, and historical initial efficiency parameters from the sample training set as input data for the regression network, and using historical corrected path planning parameters and historical corrected efficiency parameters from the sample training set as output data for the regression network, training the regression network to obtain an initial regression network used to predict real-time corrected path parameters and corrected efficiency parameters, testing the initial regression network using the sample test set, and outputting an initial regression network that satisfies a preset error value as the scheduling correction model.

[0038] Step 4: Adjust the robot path according to the calibration path planning parameters and optimize the scheduling rhythm according to the calibration scheduling efficiency parameters. Based on the comparison results, return to Step 3 or deliver the material scheduling task to the next process.

[0039] The method for determining whether to return to step three based on the comparison result, or to deliver the material scheduling task to the next process, includes: determining whether the comparison result is less than a preset comparison threshold; if not, replacing the value corresponding to the initial path planning parameter with the value corresponding to the corrected path planning parameter, and replacing the value corresponding to the initial scheduling efficiency parameter with the value corresponding to the corrected scheduling efficiency parameter, and returning to step three; if yes, delivering the material scheduling task to the next process. The preset comparison threshold is determined based on the average deviation value of the historical best scheduling batch, meeting the precision scheduling requirements of semiconductor production.

[0040] Maximum number of iterations limit: If CR does not improve after 3 consecutive adjustments, force exit and trigger the following actions: a. Pause the current path planning, prioritize the allocation of backup robots, and recalculate the scheduling efficiency coefficient.

[0041] b. Send an alarm signal to the dispatch center to prompt manual intervention.

[0042] c. Record abnormal data to the log for subsequent model optimization.

[0043] The beneficial effects of this embodiment are as follows: Efficient collaborative transportation of semiconductor materials is achieved through dynamic resource allocation and real-time correction mechanisms. A regression model-based prediction mechanism quickly generates initial plans, improving response speed; resource utilization is optimized by dynamically adjusting resource parameters through efficiency coefficient partitioning; real-time comparison and correction models combined with dynamic weights accurately adapt to production fluctuations; emergency expansion and manual intervention ensure scheduling reliability, forming a data-driven closed-loop optimization system that meets the high-precision, low-latency requirements of semiconductor manufacturing.

[0044] Example 2 See Figure 2 As shown, the multi-robot collaborative semiconductor material scheduling system of this embodiment includes: Data acquisition module: Acquires material characteristic parameters (material urgency, material quantity) of semiconductor materials and robot resource parameters (robot load rate, robot idle time); Scheduling prediction module: Based on a pre-built scheduling scheme prediction model, it takes historical material characteristic parameters and historical robot resource parameters as input and outputs initial robot allocation scheme and initial path planning parameters; the scheduling scheme prediction model is obtained through regression network training; Efficiency Calculation Module: Calculates scheduling efficiency coefficients based on material characteristic parameters, robot resource parameters, and transportation demand parameters. Combines preset scheduling efficiency parameters and dynamic adjustment constants (excess resource adjustment coefficient A, baseline adjustment coefficient B, and emergency adjustment coefficient C) to determine initial scheduling efficiency parameters. Status comparison module: Obtains real-time scheduling status data (number of path conflicts, transportation delay time, robot utilization) and generates comparison results with standard scheduling status data; The scheduling correction module, based on the comparison results, initial path planning parameters, and initial scheduling efficiency parameters, outputs corrected path planning parameters and corrected scheduling efficiency parameters through a scheduling correction model; the scheduling correction model is trained through a regression network. Iterative execution module: Adjusts the robot path and scheduling rhythm according to the correction parameters. If the comparison result does not reach the preset threshold, it will perform cyclic correction. If there is no improvement after 3 consecutive times, it will trigger the abnormal handling process.

[0045] The beneficial effects of this embodiment are as follows: the combination of data acquisition and prediction models can quickly generate initial plans and reduce manual intervention; dynamic adjustment of coefficients can accurately match resource status and improve load balancing capabilities; real-time status comparison and correction models form a feedback loop to continuously optimize paths and rhythms; and the anomaly handling mechanism ensures system stability, meets the stringent requirements of semiconductor production for high precision and low latency, and improves overall resource utilization and scheduling reliability.

[0046] The weights in this invention are used to measure the degree of influence of different factors or variables on a particular outcome or decision. A weight is defined as a numerical value assigned to each factor when comparing and evaluating multiple factors to reflect their importance or priority. These weights can be determined based on specific circumstances and needs, and are typically developed and confirmed jointly by professionals or relevant stakeholders. By setting appropriate weights, programs or systems can make more accurate decisions or predictions.

[0047] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0048] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A semiconductor material scheduling method for multi-robot collaboration, characterized in that, The method flow is as follows: Step 1: Obtain the material characteristic parameters and robot resource parameters of the semiconductor material, and input the material characteristic parameters and robot resource parameters into the pre-built scheduling scheme prediction model to obtain the initial robot allocation scheme and initial path planning parameters; wherein, the scheduling scheme prediction model includes a robot allocation sub-model and a path planning sub-model; Step 2: Obtain the transportation demand parameters of semiconductor materials, calculate the scheduling efficiency coefficient based on the material characteristic parameters, robot resource parameters, and transportation demand parameters, and determine the initial scheduling efficiency parameters according to the scheduling efficiency coefficient and the preset scheduling efficiency parameters. Step 3: Obtain real-time scheduling status data, and compare the real-time scheduling status data with standard scheduling status data to generate comparison results; Step 4: Input the comparison results, initial path planning parameters, and initial scheduling efficiency parameters into the scheduling correction model to obtain adjustment data; adjust the robot path and scheduling rhythm according to the adjustment data, and determine whether to return to step 3 or deliver materials to the next process based on the comparison results.

2. The semiconductor material scheduling method for multi-robot collaboration according to claim 1, characterized in that, The method for constructing the scheduling scheme prediction model includes: Obtain the sample scheduling dataset and divide it into a sample training set and a sample test set; A regression network is constructed, with historical material feature parameters and historical robot resource parameters in the sample training set as input data, and historical robot allocation schemes and historical path planning parameters as output data, to train the initial regression network. The initial regression network is tested using a sample test set, and the output regression network that meets the preset error requirements is used as the scheduling scheme prediction model.

3. The semiconductor material scheduling method for multi-robot collaboration according to claim 1, characterized in that, The method for calculating the scheduling efficiency coefficient includes: The calculation formula is constructed based on material urgency, material quantity, robot load rate, robot idle time, and transportation volume per unit time. Among them, the material quantity is processed using a logarithmic function to prevent numerical overflow, the robot load rate is mapped to the nonlinear range through an arcsine function, the material urgency is amplified by an exponential function, and the robot idle time is processed by adding 1 to the reciprocal to reflect the impact of busyness on efficiency.

4. The semiconductor material scheduling method for multi-robot collaboration according to claim 1, characterized in that, The method for determining the initial scheduling efficiency parameters includes: The preset efficiency coefficient range and threshold are defined. When the scheduling efficiency coefficient is less than or equal to the resource surplus threshold, the preset scheduling efficiency parameter is multiplied by the surplus resource adjustment coefficient. When the scheduling efficiency coefficient is between the resource surplus threshold and the resource shortage threshold, the preset scheduling efficiency parameter is multiplied by the benchmark adjustment coefficient. When the scheduling efficiency coefficient is greater than the resource shortage threshold, the preset scheduling efficiency parameter is multiplied by the emergency adjustment coefficient; Among them, the emergency adjustment coefficient is greater than the benchmark adjustment coefficient, and the benchmark adjustment coefficient is greater than the surplus resource adjustment coefficient.

5. The semiconductor material scheduling method for multi-robot collaboration according to claim 1, characterized in that, The method for generating the comparison results includes: Extract the real-time feature set of real-time scheduling status data and the standard feature set of standard scheduling status data. The real-time feature set includes the number of path conflicts, transportation delay time, and robot utilization. The computational relationship is constructed based on the difference between the real-time feature set and the standard feature set. The weight of the path conflict term is dynamically adjusted with the number of conflicts, the weight of the delay time term is positively correlated with the delay ratio, and the utilization rate term adopts a fixed weight and a reward and punishment mechanism is set.

6. The semiconductor material scheduling method for multi-robot collaboration according to claim 1, characterized in that, The method for constructing the scheduling correction model includes: Obtain the sample calibration dataset and divide it into a sample training set and a sample test set; A regression network is constructed, with historical comparison results, historical initial path planning parameters, and historical initial efficiency parameters in the sample training set as input data, and historical corrected path planning parameters and historical corrected efficiency parameters as output data, to train the initial regression network. The initial regression network is tested using a sample test set, and the regression network that meets the preset error requirements is used as the scheduling correction model.

7. The semiconductor material scheduling method for multi-robot collaboration according to claim 1, characterized in that, The method for determining whether to return to step three or deliver materials to the next process based on the comparison results includes: If the comparison result is greater than or equal to the preset comparison threshold, the initial path planning parameters and initial scheduling efficiency parameters will be replaced with the corrected parameters, and the process will return to step three. If the comparison result is less than the preset comparison threshold, the material scheduling task will be delivered to the next process. The preset comparison threshold is determined based on the average deviation value of the historical best scheduling batches.

8. The semiconductor material scheduling method for multi-robot collaboration according to claim 1, characterized in that, It also includes a maximum iteration limit mechanism: If the comparison results do not improve after a preset number of adjustments, the current path planning is paused and a backup robot is allocated first, and the scheduling efficiency coefficient is recalculated. At the same time, an alarm signal is sent to the dispatch center to prompt manual intervention, and abnormal data is recorded in the log for subsequent model optimization.

9. A multi-robot collaborative semiconductor material scheduling system, used to implement the multi-robot collaborative semiconductor material scheduling method as described in claim 1, characterized in that, The system includes Data acquisition module: Acquires material characteristic parameters and robot resource parameters of semiconductor materials; material characteristic parameters include material urgency and material quantity; robot resource parameters include robot load rate and robot idle time. Scheduling prediction module: Based on a pre-built scheduling scheme prediction model, it takes historical material characteristic parameters and historical robot resource parameters as input and outputs initial robot allocation scheme and initial path planning parameters; the scheduling scheme prediction model is obtained through regression network training; Efficiency Calculation Module: Calculates scheduling efficiency coefficients based on material characteristic parameters, robot resource parameters, and transportation demand parameters. Combines preset scheduling efficiency parameters with dynamically adjusted excess resource adjustment coefficient A, benchmark adjustment coefficient B, and emergency adjustment coefficient C to determine the initial scheduling efficiency parameters. Status Comparison Module: Acquires real-time scheduling status data and generates comparison results with standard scheduling status data; real-time scheduling status data includes path conflict count, transportation delay time, and robot utilization rate. The scheduling correction module, based on the comparison results, initial path planning parameters, and initial scheduling efficiency parameters, outputs corrected path planning parameters and corrected scheduling efficiency parameters through the scheduling correction model; the scheduling correction model is obtained through regression network training. Iterative execution module: Adjusts the robot path and scheduling rhythm according to the correction parameters. If the comparison result does not reach the preset threshold, it will perform cyclic correction. If there is no improvement after a set number of consecutive times, it will trigger the abnormal handling process.

Images