Multi-task cooperative control robot scheduling method and system based on quantum optimization

Abstract

The application discloses a multi-task cooperative control robot scheduling method and system based on quantum optimization, and relates to the technical field of computers.The technical scheme is as follows: structured data flow is generated according to equipment operation parameters decision; process constraint conditions are generated through a digital twin engine; the process constraint conditions are coded into a feature vector based on federated learning and mapped to a QUBO model; real-time production line topology structure hard constraints are injected, and the optimal scheduling scheme is solved through a quantum annealing algorithm; multi-physical field simulation verification is performed; and the model weight parameters of the federated learning and the update gradient of the process constraint conditions generated by the digital twin engine are dynamically updated according to the deviation degree.The application realizes safe, efficient and flexible scheduling in a complex manufacturing scene through a cooperative technology system of quantum optimization global solution, federated learning heterogeneous compatibility, digital twin virtual and real verification, and dynamic feedback continuous iteration.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer technology, more specifically, it relates to a multi-task cooperative control robot scheduling method and system based on quantum optimization. BACKGROUND

[0002] Multi-task cooperative control robots refer to intelligent systems capable of simultaneously executing multiple tasks and achieving efficient and flexible operation by coordinating resource allocation, priority, and interaction between different tasks. Such robots often require the integration of perception, decision-making, execution, and communication technologies to address challenges in complex dynamic environments.

[0003] In existing technologies, multi-task cooperative control generally adopts a hierarchical control architecture composed of high-level planning, middle-level scheduling, and bottom-level execution. The high-level planning mainly performs task decomposition and priority sorting to generate a global task sequence, for example, an industrial robot simultaneously performing tasks such as grabbing, assembling, and detecting. The middle-level scheduling mainly allocates resources (such as mechanical arms, sensors, and computing units) to different tasks to avoid resource conflicts. The bottom-level execution mainly controls the execution mechanism (such as motors and end effectors) to complete specific actions in real time while ensuring precision and safety. However, the traditional hierarchical control architecture (planning-scheduling-execution) has response delays and is difficult to handle high-frequency sudden tasks (such as emergency order insertion and equipment failure). For example, scheduling systems based on heuristic rules need to recalculate the global solution when resource competition is intense, leading to production line downtime. Although some existing technologies use reinforcement learning for dynamic scheduling, the state space dimension explosion problem limits its real-time performance, making it unable to meet the millisecond-level response requirement. In addition, existing optimization methods (such as genetic algorithms and simulated annealing) are prone to local optimization when dealing with high-dimensional process constraints (electromagnetic compatibility, thermodynamics, and kinematics). Although some existing technologies propose mixed integer programming models, the quadratic coupling constraints lead to exponential growth of solving time with the number of devices, making it difficult to support real-time scheduling of large-scale production lines.

[0004] Therefore, how to design a multi-task cooperative control robot scheduling method and system based on quantum optimization that can overcome the above-mentioned defects is a problem we need to solve urgently. SUMMARY

[0005] To solve the problems in the prior art, the application aims to provide a multi-task cooperative control robot scheduling method and system based on quantum optimization, which combines quantum annealing algorithm with federated learning, encodes process constraint conditions into a quadratic unconstrained binary optimization problem through a QUBO model, and uses the parallel search characteristics of quantum annealing to significantly shorten the solving time compared with traditional methods in a 1000-variable scale; in addition, the domain adaptive network is used to align the feature distribution of multiple devices, cross-line model sharing is realized under the premise of protecting data privacy, and the scheduling accuracy can be improved in a multi-type heterogeneous robot scene.

[0006] The above technical purpose of the application is achieved by the following technical scheme:

[0007] In a first aspect, a multi-task cooperative control robot scheduling method based on quantum optimization is provided, comprising the following steps:

[0008] Real-time acquisition of device operation parameters of the target device, and decision-making of generating a structured data stream containing resource request signals, task queues and sensor abnormal values according to the device operation parameters;

[0009] Generating process constraint conditions including task priority mapping rules and physical space collision probability thresholds through a digital twin engine to analyze the structured data stream;

[0010] Encoding the process constraint conditions into a feature vector based on federated learning, and mapping the feature vector to a QUBO model;

[0011] Injecting real-time line topology structure hard constraints into the QUBO model, and solving the optimal scheduling scheme through a quantum annealing algorithm to obtain a binary solution set;

[0012] Carrying out multi-physics field simulation verification on the binary solution set through a digital twin engine, and issuing the control instruction set generated after verification to the target device;

[0013] Determining the deviation degree between the device execution result of the target device and the digital twin prediction value, and dynamically updating the model weight parameters of the federated learning and the update gradient of the process constraint conditions generated by the digital twin engine according to the deviation degree.

[0014] Further, the task priority mapping rules are generated based on resource competition intensity or generated through a Markov decision process.

[0015] Further, the physical space collision probability threshold is determined through a space-time Transformer model.

[0016] Further, the federated learning encodes the process constraints into a feature vector using a domain adaptive network, comprising:

[0017] Applying a maximum mean difference constraint to the device historical data;

[0018] The robust feature vector is generated based on a maximum mean difference constraint.

[0019] Further, the process of mapping the feature vector to the QUBO model comprises:

[0020] generating a linear term coefficient determined by a single constraint weight in the feature vector;

[0021] and generating a quadratic term coefficient determined by a coupling relationship between different constraints in the feature vector.

[0022] Further, the multi-physical field simulation verification comprises verification of any one or more of:

[0023] electromagnetic compatibility verification: electromagnetic interference strength is less than a first threshold value;

[0024] thermodynamic verification: temperature rise of the robot joint is less than a second threshold value;

[0025] kinematic verification: path planning error is less than a third threshold value.

[0026] Further, the deviation degree is quantitatively determined by any one or more of operation trajectory deviation, task completion time deviation, and energy consumption deviation.

[0027] Further, the process of dynamically updating the model weight parameters of the federated learning according to the deviation degree comprises:

[0028] controlling the update rate of the federated learning weight through the deviation degree.

[0029] Further, the process of dynamically updating the update gradient of the process constraint condition generated by the digital twin engine according to the deviation degree comprises:

[0030] controlling the process constraint condition and the loss function of feature alignment in the federated learning according to the deviation degree, respectively obtaining the gradient contribution of the process constraint condition to the federated learning and the total loss value of the feature alignment;

[0031] determining the gradient update direction of the digital twin engine according to the gradient contribution of the process constraint condition to the federated learning and the total loss value of the feature alignment.

[0032] In a second aspect, a multi-task cooperative control robot scheduling system based on quantum optimization is provided, which is used to implement the multi-task cooperative control robot scheduling method based on quantum optimization as described in any one of the first aspect, comprising:

[0033] a data processing module, configured to collect device running parameters of a target device in real time, and generate a structured data stream comprising a resource request signal, a task queue, and a sensor abnormal value according to the device running parameters;

[0034] a constraint generation module configured to generate process constraint conditions including task priority mapping rules and physical space collision probability thresholds after analyzing the structured data stream by the digital twin engine;

[0035] a feature encoding module configured to encode the process constraint conditions into a feature vector based on federated learning and map the feature vector to a QUBO model;

[0036] a quantum optimization module configured to inject real-time production line topology hard constraints into the QUBO model and solve an optimal scheduling scheme by a quantum annealing algorithm to obtain a binary solution set;

[0037] a simulation verification module configured to perform multi-physics field simulation verification on the binary solution set by the digital twin engine and issue a control instruction set generated after verification to a target device;

[0038] a dynamic updating module configured to determine a deviation degree between a device execution result of the target device and a digital twin prediction value and dynamically update model weight parameters of the federated learning and an update gradient of the process constraint conditions generated by the digital twin engine according to the deviation degree.

[0039] Compared with the prior art, the present application has the following beneficial effects:

[0040] 1. The multi-task collaborative control robot scheduling method based on quantum optimization provided by the present application combines quantum annealing algorithm and federated learning, encodes the process constraint conditions into a quadratic unconstrained binary optimization problem through a QUBO model, and uses the parallel search characteristics of quantum annealing to significantly shorten the solving time under high variable scale compared with traditional methods. In addition, the domain adaptive network aligns the feature distribution of multiple devices, realizes cross-production line model sharing under the premise of protecting data privacy, and can improve the scheduling accuracy in multiple heterogeneous robot scenarios.

[0041] 2. The present application constructs a space-time Transformer and multi-physics field simulation verification system. The space-time Transformer captures the space-time correlation of device motion through a self-attention mechanism, effectively reduces the collision probability prediction error, and simultaneously verifies electromagnetic compatibility (<3V / m), thermodynamics (temperature rise <25℃), and kinematics (error <0.5mm), thereby reducing the failure probability of control instructions.

[0042] 3. The present application dynamically adjusts the update rate of federated learning through deviation degree, which can improve the model convergence speed when there is a sudden load fluctuation. In addition, a gradient contribution model of process constraint conditions and feature alignment loss is established, so that the prediction accuracy of the digital twin is exponentially improved with the number of iterations. BRIEF DESCRIPTION OF DRAWINGS

[0043] The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:

[0044] Figure 1 is a flowchart in embodiment 1 of the present application;

[0045] Figure 2 is a system block diagram in embodiment 2 of the present application. DETAILED DESCRIPTION

[0046] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application in combination with embodiments and drawings, the illustrative embodiments of the present application and the description thereof are only used to explain the present application, and do not limit the present application.

[0047] Embodiment 1: A multi-task collaborative control robot scheduling method based on quantum optimization, as shown in Figure 1 includes the following steps:

[0048] S1: Real-time acquisition of device running parameters of the target device, and decision-making generation of structured data stream containing resource request signal, task queue and sensor abnormal value according to the device running parameters;

[0049] S2: Generating process constraint conditions including task priority mapping rules and physical space collision probability threshold through digital twin engine after analyzing the structured data stream;

[0050] S3: Encoding the process constraint conditions into a feature vector based on federated learning, and mapping the feature vector to a QUBO model;

[0051] S4: Injecting real-time production line topology structure hard constraints into the QUBO model, and solving the optimal scheduling scheme through quantum annealing algorithm to obtain a binary solution set;

[0052] S5: Multi-physical field simulation verification of the binary solution set through the digital twin engine, and issuing the control instruction set generated after the verification to the target device;

[0053] S6: Determining the deviation degree between the device execution result of the target device and the digital twin prediction value, and dynamically updating the model weight parameters of the federated learning and the update gradient of the process constraint conditions generated by the digital twin engine according to the deviation degree.

[0054] The present application is applicable to intelligent production lines in automobile manufacturing, 3C electronics and other industries, such as robots cooperating to complete assembly, welding, detection and other tasks. The above steps will be described in detail taking the automobile welding production line as an example.

[0055] In step S1, the target devices in the automobile welding production line generally include welding robots, handling robots, glue applying robots, clamp systems, conveyor belt systems, quality detection devices, and other key devices, which are responsible for different processes and cooperatively complete the body welding task.

[0056] The device operation parameters are real-time state data of each target device. For example, the device operation parameters of the welding robot include but are not limited to servo motor current, welding gun pressure, welding temperature, electrode wear degree, etc. The device operation parameters of the handling robot include but are not limited to motor torque, end effector position, battery power, etc. The device operation parameters of the glue applying robot include but are not limited to glue amount sensor data, glue gun temperature. The device operation parameters of the clamp system include but are not limited to clamping force, positioning accuracy. The device operation parameters of the conveyor belt system include but are not limited to speed, load state; the device operation parameters of the quality detection device include but are not limited to image processing result, defect type statistics, etc.

[0057] The resource request signal, the task queue, and the sensor abnormal value are obtained based on the analysis and processing of the device operation parameters.

[0058] In some examples, the resource request signal is generated by triggering a condition to judge an abnormal situation. For example, if the electrode wear degree of the welding robot is greater than 50,000 times, the electrode resource is requested to be replaced; if the battery power of the handling robot is less than 20%, the charging pile is requested to be dispatched; if the clamping force of the clamp is less than 2000N, intervention is requested; if the servo motor current is abnormal, the maintenance resource is triggered to generate the resource request signal.

[0059] In some examples, the task queue refers to a job sequence assigned to each robot or device. In some examples, the task queue contains task type, execution order, time constraint, and resource allocation information, which mainly realizes the functions of coordinating multi-device cooperation, dynamically adjusting priority, and optimizing resource allocation. In some examples, the coordination of multi-device cooperation mainly ensures the seamless connection of welding, handling, glue applying, and other processes, the dynamic priority adjustment mainly responds to the insertion of emergency orders, device failure, and other sudden situations, and the resource optimization allocation mainly allocates tasks according to the real-time state of the device (such as power, load) to maximize the efficiency of the production line.

[0060] In some examples, the sensor abnormal value detection is realized by threshold or model judgment, such as marking an abnormality if the welding temperature exceeds the range.

[0061] It should be noted that the structured data stream needs a standardized format, generally using JSON or ProtocolBuffers format, and the format generally contains fields such as timestamp, device ID, parameter type, value, and state label.

[0062] In step S2, the task priority mapping rule is mainly to determine the execution order and resource allocation weight of different tasks (such as welding, carrying, and gluing).

[0063] In some examples, if the task is an emergency task, the resource competition intensity can be generated. In addition, whether the task is an emergency task or not, the task priority mapping rule can be generated through a Markov decision process.

[0064] The implementation logic based on the resource competition intensity is as follows: real-time acquisition of equipment operation parameters such as load, number of emergency tasks, etc.; then calculate the resource competition intensity, the larger the value, the higher the task priority; then arrange the task queue in descending order according to the resource competition intensity, and dynamically adjust the execution order.

[0065] Specifically, the calculation formula of the resource competition intensity is as follows:

[0066] ;

[0067] Among them, represents the resource competition intensity; represents the current load rate of the equipment; represents the rated maximum power of the equipment; is an exponential term, which represents the competition effect in the nonlinear amplification high load scenario; represents the number of emergency tasks; represents an exponential decay adjustment factor of the number of emergency tasks on resource competition; represents a bias weight coefficient; represents a position bias degree.

[0068] The present application can quickly respond to sudden changes through resource competition intensity when handling emergency orders, equipment failures and other sudden situations.

[0069] The implementation logic generated through the Markov decision process is as follows: pre-train the strategy network through digital twin simulation; input the real-time state (such as equipment power, load rate, task queue length, etc.) into the strategy network, and output the optimal action (such as task allocation, equipment start-stop, etc.). In addition, the strategy can also be adjusted according to the bias degree. The task priority mapping rule generated through the Markov decision process can avoid local optimization, improve global efficiency, and also balance equipment life and total energy consumption through multi-objective function, which is suitable for high-cost equipment scheduling and sustainable manufacturing scenarios.

[0070] The physical space collision probability threshold contains two pieces of information, collision probability and judgment threshold, and is the highest allowed probability value for determining the risk of collision between the device and other objects on the motion path. The physical space collision probability threshold has functions of safety control and scheduling optimization. In some examples, when the collision probability exceeds the judgment threshold, an obstacle avoidance or shutdown instruction is triggered. In some examples, the judgment threshold is used as a hard constraint condition in the quantum optimization model to avoid high-risk task allocation.

[0071] In some examples, the physical space collision probability threshold is determined by a space-time Transformer model. The specific process is as follows: a three-dimensional model of the workshop is constructed by a digital twin engine, and real-time data such as device motion trajectory, speed, and acceleration are synchronized; the space-time Transformer model is used to predict the collision probability at a future time; and the collision probability is compared with the judgment threshold to adjust the task scheduling strategy, for example, when the collision probability is greater than the judgment threshold, an obstacle avoidance instruction can be triggered.

[0072] In some examples, the expression of the collision probability is as follows:

[0073] ;

[0074] wherein, represents the device and the device collision probability in a future time interval; represents the motion trajectory sequence of the device such as position, speed, and direction; represents the motion trajectory sequence of the device such as position, speed, and direction; represents the Sigmoid function, which smoothes the probability transition; represents the real-time distance between the device and the device ; represents the safety distance.

[0075] In some examples, if multiple tasks are rejected due to collision risk, the judgment threshold can be adaptively adjusted, and the expression is as follows:

[0076] ;

[0077] wherein, represents the adaptively adjusted judgment threshold; represents the judgment threshold before adaptive adjustment; represents the decay coefficient, for example, the value is 0.1; represents the number of tasks rejected due to collision risk.

[0078] In some examples, the judgment threshold can also be adjusted according to the deviation degree, and if the actual deviation increases, the threshold is lowered to improve safety.

[0079] In some examples, the expression of feedback adjustment is as follows:

[0080] ;

[0081] wherein, represents the judgment threshold after feedback adjustment; represents the basic threshold before feedback adjustment; represents the deviation degree; represents an adjustment coefficient, used to control the sensitivity of the threshold to the deviation.

[0082] In step S3, since the process constraint conditions generally include task priority mapping rules, physical space collision probability threshold values, device dynamic parameters (such as joint load, battery power, electromagnetic interference intensity) and the like, these constraints involve multi-dimensional and multi-type data, and direct processing can easily lead to an explosion of model complexity. Therefore, the application converts heterogeneous constraints (numerical values, rules, probabilities) into mathematical vectors of a unified dimension, and reduces redundant information by extracting key constraint features.

[0083] In addition, the application considers that multiple production lines or different types of robots are involved in automobile manufacturing, and different types of robots work in the same workshop, their sensor data distributions are different, but they need to cooperate to complete the welding task, and direct aggregation of the model can lead to performance degradation, and all the application is based on federated learning for encoding processing. Federated learning allows each device to locally train a model and then aggregate model parameters.

[0084] In some examples, the training process of the local model includes data preparation and feature extraction, each robot locally stores historical data including but not limited to process parameters and constraint indicators, and then each device extracts a feature vector through a local neural network, each task or device state corresponding to a feature vector.

[0085] In addition, the application also adjusts the feature distribution of the data of each device through a domain adaptation network, so that they are closer in the feature space, so that the model can better generalize.

[0086] In some examples, the feature distribution can be aligned by minimizing the MMD (Maximum Mean Discrepancy) between the source domain (robot A) and the target domain (robot B).

[0087] In some examples, an adversarial discriminator can also be introduced in the domain adaptation network to further ensure consistency of the feature distribution. For example, an application discriminator attempts to distinguish whether the features are from robot A or robot B; a feature extractor is trained to deceive the discriminator, determine a first function loss; and a joint optimization is performed in combination with the first function loss, a function loss of the maximum mean difference, and a loss of the welding quality prediction task.

[0088] In some examples, the process of mapping the feature vector to the QUBO model includes: generating linear term coefficients determined by single constraint weights in the feature vector, such as the priority of task economy; and generating quadratic term coefficients determined by coupling relationships between different constraints in the feature vector, such as the joint influence of device load and collision probability.

[0089] In step S4, the real-time production line topology hard constraint is a constraint that must be met, which can be achieved by maximizing the penalty term. In addition, some soft constraints that can be met as much as possible can also be injected, and the priority is achieved by weight adjustment, such as task delay penalty.

[0090] In step S5, the multi-physical field simulation verification includes verification of any one or more of the following: electromagnetic compatibility verification, thermodynamic verification, and kinematic verification.

[0091] For example, the kinematic verification requires that the electromagnetic interference strength be less than 3V / m; the thermodynamic verification requires that the robot joint temperature rise be less than 25℃; and the path planning error be less than ±0.5mm.

[0092] In step S6, the deviation degree is quantitatively determined by any one or more of the operation trajectory deviation, the task completion time deviation, and the energy consumption deviation.

[0093] In some examples, the process of dynamically updating the model weight parameters of the federated learning according to the deviation degree is specifically: controlling the update rate of the federated learning weight through the deviation degree.

[0094] For example, if the deviation degree is greater than 1, the weight reduction rate of the federated learning is 0.1 times the deviation degree.

[0095] In some examples, the process of dynamically updating the update gradient of the process constraint condition generated by the digital twin engine according to the deviation degree is specifically: adjusting the process constraint condition and the loss function of the feature alignment in the federated learning to obtain the gradient contribution of the process constraint condition to the federated learning and the total loss value of the feature alignment, respectively; and determining the gradient update direction of the digital twin engine according to the gradient contribution of the process constraint condition to the federated learning and the total loss value of the feature alignment.

[0096] The expression of the update gradient is as follows:

[0097] ;

[0098] wherein, denotes the gradient update direction of the digital twin engine, expressed as a vector; denotes the total loss value of feature alignment; denotes the set of trainable parameters of federated learning; denotes the mixing coefficient of process constraint gradient, such as 0.05; denotes the gradient contribution of process constraints to federated learning.

[0099] Experimental verification

[0100] I. The parameters and values of the experimental setup are shown in Table 1.

[0101] Table 1 Parameters and values

[0102] Parameter Value Production line scale 8 robots (welding x 3, carrying x 2, gluing x 2, quality inspection x 1) Task complexity Emergency task proportion 30%, resource competition intensity threshold 0.65 Environmental dynamics Equipment failure rate 5% / h, new task insertion frequency 15 times / min Comparison method 1. Traditional dynamic programming (DP) 2. Heuristic genetic algorithm (GA) 3. Deep reinforcement learning (DRL) 4. Invention scheme (Q-FedTwin)

[0103] II. The comparison results of key performance indicators are shown in Table 2.

[0104] Index DP GA DRL Q-FedTwin (invention) Improvement range Scheduling time (ms) 1520±210 680±95 320±45 85±12 73.4%↑ vs DRL Collision probability (%) 8.7 5.2 3.8 0.9 76.3%↓ vs DRL Energy consumption (kW·h) 42.5 38.1 35.7 28.3 20.7%↓ vs DRL Task completion rate (%) 82.3 88.5 91.2 98.6 8.1%↑ vs DRL Model update delay (s) - - 12.5 3.2 74.4%↓

[0105] Compared with the above-mentioned prior art, the scheduling time, collision probability, energy consumption, task completion rate and model update delay, etc. have obvious advantages.

[0106] III. The comparison results of quantum optimization efficiency are shown in Table 3.

[0107] Table 3 Comparison results of quantum optimization efficiency

[0108] Task size Traditional simulated annealing (SA) Quantum annealing (QA) Speedup ratio 50 variable problem 18.7s 0.9s 20.8× 200 variable problem Timeout (>300s) 5.2s >57.7×

[0109] The present application achieves a real-time response of 5.2s on a 200-variable scheduling problem (traditional method exceeds 300s), and the solution space search efficiency is improved by 2 orders of magnitude (QA vs SA).

[0110] Example 2: A multi-task collaborative control robot scheduling system based on quantum optimization, which is used to implement the multi-task collaborative control robot scheduling method based on quantum optimization as described in Example 1, including a data processing module, a constraint generation module, a feature encoding module, a quantum optimization module, a simulation verification module and a dynamic updating module.

[0111] The data processing module is configured to collect device operation parameters of the target device in real time, and generate a structured data stream including a resource request signal, a task queue and a sensor abnormal value according to the device operation parameters; the constraint generation module is configured to generate process constraint conditions including a task priority mapping rule and a physical space collision probability threshold by analyzing the structured data stream through a digital twin engine; the feature coding module is configured to encode the process constraint conditions into a feature vector based on federated learning, and map the feature vector to a QUBO model; the quantum optimization module is configured to inject real-time production line topology structure hard constraints into the QUBO model, and solve an optimal scheduling scheme through a quantum annealing algorithm to obtain a binary solution set; the simulation verification module is configured to perform multi-physics field simulation verification on the binary solution set through the digital twin engine, and distribute a control instruction set generated after verification to the target device; and the dynamic updating module is configured to determine a deviation degree between a device execution result of the target device and a digital twin prediction value, and dynamically update model weight parameters of the federated learning and an update gradient of the process constraint conditions generated by the digital twin engine according to the deviation degree.

[0112] Working principle: The collaborative technology system of quantum optimization global solution, federated learning heterogeneous compatibility, digital twin virtual-real verification, and dynamic feedback continuous iteration solves the bottlenecks of traditional methods in real-time, multi-objective optimization, cross-device collaboration and adaptive learning, and realizes safe, efficient and flexible scheduling in a complex manufacturing scene.

[0113] Those skilled in the art will understand that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0114] The present application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks

[0115] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods can be implemented on a single device or distributed across several devices. Figure 1

[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods can be implemented on a single device or distributed across several devices. Figure 1

[0117] The above detailed description merely describes the specific implementation of the application, and the purpose, technical solutions and advantages of the application are further described in detail. It should be understood that the above description is only a specific implementation of the application and is not intended to limit the protection scope of the application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.​​

Claims

1. A multi-task cooperative control robot scheduling method based on quantum optimization, characterized in that, The method comprises the following steps: real-time acquisition of device operation parameters of a target device, and generation of a structured data stream comprising resource request signals, task queues, and sensor abnormal values based on the device operation parameters; generation of process constraint conditions comprising task priority mapping rules and physical space collision probability thresholds by a digital twin engine after analyzing the structured data stream; encoding of the process constraint conditions as feature vectors based on federated learning, and mapping of the feature vectors to a QUBO model; injection of real-time production line topology hard constraints into the QUBO model, and solving of an optimal scheduling scheme by a quantum annealing algorithm to obtain a binary solution set; verification of the binary solution set by a digital twin engine, and generation of a control instruction set after verification, and delivery of the control instruction set to the target device; determination of a deviation degree between a device execution result of the target device and a digital twin prediction value, and dynamic updating of model weight parameters of the federated learning and an update gradient of the process constraint conditions generated by the digital twin engine based on the deviation degree.

2. The quantum optimization based multi-task cooperative control robotic scheduling method of claim 1, wherein, The task priority mapping rules are generated based on resource competition intensity or by a Markov decision process.

3. The quantum optimization based multi-task cooperative control robotic scheduling method of claim 1, wherein, The physical space collision probability threshold is determined by a space-time Transformer model.

4. The quantum optimization based multi-task cooperative control robotic scheduling method of claim 1, wherein, The federated learning encodes the process constraints as feature vectors by using a domain adaptation network, comprising: applying a maximum mean difference constraint to device historical data; generating a robust feature vector based on the maximum mean difference constraint.

5. The quantum optimization based multi-task cooperative control robotic scheduling method of claim 1, wherein, The process of mapping the feature vectors to the QUBO model comprises: generating linear term coefficients determined by single constraint weights in the feature vectors; and generating quadratic term coefficients determined by coupling relationships between different constraints in the feature vectors.

6. The quantum optimization based multi-task cooperative control robotic scheduling method of claim 1, wherein, The multi-physics field simulation verification comprises verification of any one or more of the following: electromagnetic compatibility verification: electromagnetic interference intensity is less than a first threshold value; thermodynamic verification: robot joint temperature rise is less than a second threshold value; kinematics verification: path planning error is less than a third threshold value.

7. The quantum optimization based multi-task cooperative control robotic scheduling method of claim 1, wherein, The deviation degree is quantitatively determined by any one or more of operation trajectory deviation, task completion time deviation, and energy consumption deviation.

8. The quantum optimization based multi-task cooperative control robotic scheduling method of claim 1, wherein, The process of dynamically updating the model weight parameters of the federated learning based on the deviation degree comprises: controlling the update rate of the federated learning weight by the deviation degree.

9. The quantum optimization based multi-task cooperative control robotic scheduling method of claim 1, wherein, The process of dynamically updating the update gradient of the process constraint conditions generated by the digital twin engine based on the deviation degree comprises: controlling the process constraint conditions and the loss function of feature alignment in the federated learning based on the deviation degree, to obtain the gradient contribution of the process constraint conditions to the federated learning and the total loss value of feature alignment, respectively; determining the gradient update direction of the digital twin engine based on the gradient contribution of the process constraint conditions to the federated learning and the total loss value of feature alignment.

10. A multi-task collaborative control robot scheduling system based on quantum optimization, characterized by, The system is used to implement the quantum optimization-based multi-task collaborative control robot scheduling method according to any one of claims 1-9, comprising: a data processing module for real-time acquisition of device operation parameters of a target device, and generation of a structured data stream comprising resource request signals, task queues, and sensor abnormal values based on the device operation parameters; The constraint generation module is configured to generate process constraint conditions including task priority mapping rules and physical space collision probability thresholds after the digital twin engine analyzes the structured data stream; The feature encoding module is configured to encode the process constraint conditions into a feature vector based on federated learning, and map the feature vector to a QUBO model; The quantum optimization module is configured to inject real-time production line topology hard constraints into the QUBO model, and solve an optimal scheduling scheme by using a quantum annealing algorithm to obtain a binary solution set; The simulation verification module is configured to perform multi-physics field simulation verification on the binary solution set by using the digital twin engine, and distribute a control instruction set generated after verification to target equipment; The dynamic updating module is configured to determine a deviation degree between a device execution result of the target equipment and a digital twin prediction value, and dynamically update model weight parameters of the federated learning and an update gradient of the process constraint conditions generated by the digital twin engine according to the deviation degree.

Images