AGV task dynamic assignment system and method based on multi-objective collaborative optimization

Abstract

This invention discloses a dynamic AGV task assignment system and method based on multi-objective collaborative optimization, relating to the field of AGV task assignment technology. The method includes: initiating the task assignment process using a dynamic event triggering mechanism; constructing and screening a candidate AGV set through constraints, generating multi-dimensional cost bidding vectors independently and in parallel from each AGV, and dynamically adjusting the evaluation weights of each bidding vector dimension; comprehensively evaluating the bidding vectors to determine the optimal AGV and complete the task assignment; and continuously updating the system state based on the assignment results within a rolling time-domain framework, and progressively optimizing and adjusting unexecuted tasks. This invention solves the problems of single multi-objective operation, insufficient dynamic response capability, and low efficiency of multi-AGV collaborative scheduling.

Description

Technical Field

[0001] This invention relates to the field of AGV task assignment technology, and more specifically, to an AGV task dynamic assignment system and method based on multi-objective collaborative optimization. Background Technology

[0002] With the development of intelligent manufacturing and smart logistics, Automated Guided Vehicle (AGV) systems are widely used in warehousing, transportation, production distribution, and flexible manufacturing. AGVs work in conjunction with task management systems, path planning systems, and scheduling control systems to achieve automated material handling and efficient flow. In practical applications, AGV task scheduling typically involves issues such as multi-task concurrency, resource sharing, and dynamic changes in complex environments. Therefore, achieving efficient, stable, and real-time responsive task assignment has become one of the key technologies in AGV systems.

[0003] However, existing technologies still have the following shortcomings in practical applications: Multi-objective singularity fails to adapt to complex needs: Most dispatching systems only use "shortest path" or "shortest completion time" as a single optimization objective, ignoring the ever-changing actual needs of the work site, such as the coordination and trade-off between multiple objectives, including task urgency, AGV energy consumption balance, charging scheduling, and traffic congestion avoidance.

[0004] Static or quasi-static assignment with poor dynamic response: Traditional solutions often assign tasks offline or in batches within fixed periods. When dynamic events such as the sudden insertion of new tasks, sudden AGV failures, or temporary path congestion occur, the system cannot quickly and globally re-optimize the allocation, leading to scheduling rigidity and a decrease in overall efficiency.

[0005] Insufficient coordination can easily lead to local congestion or resource idleness: The lack of global optimization for collaborative operations among multiple AGVs can easily cause multiple AGVs to compete for the same path node or resource, resulting in "deadlock" or congestion; at the same time, it may also cause some AGVs to be overloaded while others are idle, resulting in low overall system energy efficiency. Summary of the Invention

[0006] To overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an AGV task dynamic assignment system and method based on multi-objective collaborative optimization, which solves the problems of single multi-objective, insufficient dynamic response capability and low efficiency of multi-AGV collaborative scheduling in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: Firstly, this application provides a method for dynamic AGV task assignment based on multi-objective collaborative optimization. This method includes: initiating the task assignment process using a dynamic event triggering mechanism; constructing and filtering a candidate AGV set through constraints, generating multi-dimensional cost bidding vectors independently and in parallel from each AGV, and dynamically calculating the evaluation weights of each bidding vector dimension based on task urgency, remaining AGV power, average system queue length, and path occupancy rate; comprehensively evaluating the bidding vectors, calculating the relative proximity of each candidate AGV, and determining the AGV with the highest relative proximity as the optimal AGV and completing the task assignment; based on the assignment results, continuously updating the system state within a rolling time-domain framework, and progressively optimizing and adjusting unexecuted tasks; the progressive optimization and adjustment refers to adjusting the allocation structure changes based on the current and historical allocation states within a rolling time-domain framework, constraining changes in the allocation structure through optimal transmission distance, adjusting the allocation of unexecuted tasks only when the overall benefit is better than the historical solution, while keeping currently executed tasks unchanged.

[0008] In one embodiment, a dynamic event triggering mechanism is used to initiate the task assignment process, including: Calculate the deviation of the system state vector from the baseline state model; if the deviation exceeds a preset deviation threshold, it is determined that a dynamic event has occurred; determine the contribution values ​​of the task state sub-vector, AGV state sub-vector, and environment state sub-vector in the system state vector to the deviation, determine the sub-vector type corresponding to the maximum contribution value as the dynamic event type, and generate the corresponding assignment trigger signal.

[0009] In one embodiment, a candidate AGV set is constructed and filtered based on constraints, including constructing a task pool and an AGV state set based on the current system state: responding to a task assignment trigger signal to obtain a snapshot of the current system state; extracting unfinished tasks from the system state snapshot, performing multi-dimensional weighted fusion sorting based on task priority, spatial reachability, and time constraints to construct a task pool; and extracting AGV state information from the system state snapshot to form an AGV state set.

[0010] In one embodiment, constructing and filtering the candidate AGV set through constraints further includes: Select tasks to be assigned from the task pool and extract task location and time constraints; extract multi-dimensional status parameters of each AGV from the AGV status set, including current position, power level, load and operating status; set constraints on the multi-dimensional status parameters and sequentially perform multi-dimensional constraint screening on the AGVs; form a candidate AGV set by selecting AGVs that pass all constraint screenings.

[0011] In one embodiment, generating a multidimensional cost bidding vector includes: broadcasting the task information to be assigned to a set of candidate AGVs; each candidate AGV independently and in parallel performs local path planning based on its own local state and environmental information: performing local planning on the path from the current position to the task position, and calculating the multidimensional local cost of task execution and the path congestion index respectively, thereby constructing a multidimensional cost bidding vector and returning it to the central scheduling unit.

[0012] In one embodiment, returning to the central scheduling unit includes: interpolating the multidimensional cost bidding vector sequence with time as the horizontal axis for each dimension to obtain a bidding function, and calculating its first derivative as a derivative feature; each AGV constructing a derivative invariant based on the derivative feature; each AGV concatenating the derivative invariant with the original bidding vector to generate a derivative invariant bidding vector, and sending it to the central scheduling unit; the central scheduling unit utilizing the characteristic that the derivative invariant is insensitive to time delay to uniformly sort and compare bidding data arriving at different times.

[0013] In one embodiment, determining the optimal AGV and assigning the task includes: standardizing the received set of bidding vectors; dynamically adjusting the evaluation weights of each dimension in the bidding vectors based on the current system state and task urgency, wherein the dimensions include time cost, energy cost, task queue delay increment, and path congestion index; constructing positive and negative ideal solution vectors, and calculating the weighted Euclidean distance between each candidate AGV bidding vector and the positive and negative ideal solutions; calculating the relative proximity of each candidate AGV based on the weighted Euclidean distance, sorting them by relative proximity, and selecting the optimal AGV as the task assignment object.

[0014] In one embodiment, incremental optimization and adjustment are performed on unexecuted tasks, including: constructing a task-AGV allocation matrix and a historical allocation matrix at the current moment based on the task assignment results, forming an allocation state pair; filtering unexecuted tasks based on the current and historical allocation states, extracting the corresponding allocation relationship in the current allocation matrix to form an allocation sub-matrix to be optimized, and extracting the corresponding part in the historical allocation matrix to form a reference allocation sub-matrix; normalizing the allocation relationship in the allocation sub-matrix to be optimized and the reference allocation sub-matrix into a probability distribution form, and constructing a task-to-AGV allocation distribution model.

[0015] In one embodiment, the incremental optimization adjustment of unexecuted tasks further includes: constructing a task allocation cost matrix at the current moment based on an allocation distribution model; introducing the distribution structure corresponding to the historical allocation state, constructing a transmission cost function, and measuring the structural offset of the current allocation distribution relative to the historical allocation distribution through the optimal transmission distance; constructing a joint optimization objective function based on the allocation cost matrix and the transmission cost function; solving the allocation submatrix to be optimized, and adjusting the allocation only when the overall benefit is better than the historical solution; merging and updating the optimized allocation result with the current execution state, applying optimized allocation to unexecuted tasks, and maintaining the original allocation for currently executing tasks, forming an updated system task allocation state; using the updated system task allocation state as the historical allocation state for the next rolling cycle, and repeating the iteration to achieve continuous incremental optimization.

[0016] Secondly, this application provides a dynamic AGV task assignment system based on multi-objective collaborative optimization, including: a data acquisition module for initiating the task assignment process using a dynamic event triggering mechanism; a screening and weight adjustment module for constructing and screening a set of candidate AGVs through constraints, generating multi-dimensional cost bidding vectors independently and in parallel by the AGVs, and dynamically calculating the evaluation weights of each bidding vector dimension based on task urgency, AGV remaining power, system average queue length, and path occupancy rate; a multi-objective decision-making module for comprehensively evaluating the bidding vectors, calculating the relative proximity of each candidate AGV, and determining the AGV with the highest relative proximity as the optimal AGV and completing the task assignment; and an update module for continuously updating the system status within a rolling time-domain framework based on the assignment results, and progressively optimizing and adjusting unexecuted tasks.

[0017] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: Real-time status awareness and anomaly detection: By fusing and standardizing multi-source data, a system status vector is constructed; combined with historical benchmark models and dynamic thresholds, events such as task arrival, AGV status changes and environmental disturbances are accurately identified, reducing false triggering rate and improving robustness.

[0018] Efficient task assignment: Candidate AGVs are screened step by step based on constraints such as space, power, and load to narrow the search space; a multi-dimensional cost bidding model (time, energy consumption, queue, congestion) is introduced, and AGVs are computed in parallel and participate in global competition; to address the asynchronous communication problem, the time derivative invariant is used to eliminate the impact of delay and achieve distributed self-optimization and global coordination.

[0019] Adaptive multi-index evaluation: The bidding vector is normalized using the improved TOPSIS method, and the weights are dynamically adjusted in combination with task urgency, remaining power, queue load and congestion status to achieve coordinated optimization of time, energy consumption and traffic risk; through relative proximity ranking and secondary constraints, misassignment caused by single index extreme values ​​is avoided and decision robustness is improved.

[0020] Inertial asymptotic optimization in the rolling time domain: Based on the optimal transmission theory, a distance constraint for the allocation structure is constructed, and discrete assignment is extended to a probabilistic evolution model; by minimizing the joint objective of current cost and historical structural change cost, "informational optimization under inertial constraint" is achieved, avoiding the oscillation and waste caused by frequent reallocation; only unexecuted tasks are adjusted, while executing tasks remain unchanged, ensuring continuity and stability in dynamic environments. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of a dynamic AGV task assignment method based on multi-objective collaborative optimization provided in an embodiment of this application.

[0023] Figure 2 This is a schematic diagram of an AGV task dynamic assignment system based on multi-objective collaborative optimization, provided in an embodiment of this application.

[0024] Figure 3 The power consumption curves for different AGVs performing tasks are provided in the embodiments of this application.

[0025] Figure 4 A heat map of path congestion index provided for embodiments of this application. Detailed Implementation

[0026] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] It should be noted that, in this document, terms such as “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the article or device that includes the aforementioned element.

[0028] Reference Figure 1 As shown in the diagram, this invention provides a flowchart of a dynamic AGV task assignment method based on multi-objective collaborative optimization, which includes the following steps: S1, monitor the system's operating status in real time, and trigger the task assignment process when a preset dynamic event is detected. The dynamic events include the arrival of new tasks, changes in AGV status, and changes in the operating environment. Based on the current system status, a task pool and an AGV status set are constructed.

[0029] In this embodiment, the system's operating status is monitored in real time, and a task assignment process is triggered when a preset dynamic event is detected, including: S11. The system operation data is obtained in real time from the AGV controller, task management system and environmental monitoring equipment through the communication network, and marked by timestamp to form the original data sequence. The system operation data includes task data: task priority, task quantity, task distribution density and task location; AGV operation data: AGV position, speed, power consumption and task load rate; environmental data: path occupancy, node congestion degree and regional load information. The AGV operation data is uploaded periodically through the on-board controller, with a sampling period of 0.1 to 1 second; task data is obtained in real time through the task management interface; and environmental data is obtained through path node sensors.

[0030] S12, standardize the original data sequence. The standardization process includes time alignment, data cleaning and format unification to obtain a standardized state data set. S13. Based on the standardized state data set, extract key state feature parameters and combine them according to a preset structure to construct the system state vector at the current moment, which is used to characterize the overall operating state of the system. The preset structure refers to the state vector organization method predefined according to the system modeling requirements. That is, the task state features, AGV state features and environmental state features obtained after standardizing the original data sequence are spliced ​​and encoded in a fixed order and dimension rules to form a unified system state vector structure.

[0031] The system state vector is specifically as follows:

[0032] In the formula, Let be the system state vector. This is a sub-vector of the task state. This is the AGV state sub-vector. This is the environment state sub-vector.

[0033] S14. Based on historical operating data, the system state vector is statistically analyzed within a preset time window to obtain the mean vector and standard deviation vector of each dimension and construct a baseline state model. The specific calculation formula for the baseline state model is as follows:

[0034] In the formula, This is the baseline state model, representing the statistical characteristics of the system under normal operating conditions, used as a reference benchmark for comparing the current system state; this model reflects the typical distribution characteristics of each state variable of the system during stable operation. This is a mean vector, representing the average value of each state dimension within a preset time window. The standard deviation vector represents the degree of fluctuation in each state dimension.

[0035] The time window adopts a sliding window mechanism, and the window length is a preset time interval (such as 30 seconds to 10 minutes). When the system is running, the window is updated in real time as time progresses to ensure the dynamic adaptability of the baseline state model.

[0036] S15, based on the system state vector and the baseline state model, calculate the standardized deviation of each dimension and calculate the overall state deviation. The overall state deviation is calculated using the following formula:

[0037] In the formula, The state deviation is the overall degree of deviation of the current system state from the baseline state model. It is used to quantify whether the system is in an abnormal state. The larger the value, the more serious the deviation from the normal state. n is the total number of state dimensions, representing the number of dimensions of the system state vector, i.e., the number of state features involved in the deviation calculation. Let be the current state component, representing the value of the system state vector in the i-th dimension at the current moment. Let be the baseline mean, representing the average value of the i-th state dimension under normal operating conditions. Let be the baseline standard deviation, representing the fluctuation range of the i-th state dimension under normal conditions. Let be the weight coefficient of the i-th state dimension in the overall deviation calculation, which satisfies: .

[0038] S16. Based on the historical deviation distribution, the deviation threshold is determined by the statistical quantile method, wherein the deviation threshold is the sum of the historical deviation mean and a preset k times the standard deviation; S17, compare the state deviation with the deviation threshold. If the state deviation is greater than the deviation threshold, it is determined that a dynamic event has occurred. Then, determine the contribution values ​​of the task state sub-vector, AGV state sub-vector, and environment state sub-vector in the system state vector to the state deviation. The sub-vector type corresponding to the largest contribution value is determined as the dynamic event type. That is, by calculating the standardized deviation of each dimension of each sub-vector and performing a weighted sum, the sub-vector with the largest contribution value is the main factor causing the state deviation. If the task state subvector deviates to the maximum, it is determined as a new task arrival event; If the AGV state subvector deviates to the maximum, it is determined to be an AGV state change event; If the environmental state subvector deviates to the maximum, it is determined to be an event of change in the operating environment; S18, Based on the judgment result, generate a task assignment trigger signal.

[0039] It should be noted that, compared to traditional methods based on fixed rules or periodic scheduling, this approach can construct a unified state vector based on multi-dimensional system state data and combine it with a baseline state model formed by historical statistics to standardize and quantify the deviation of the current operating state, thereby achieving continuous and dynamic perception of the degree of system anomalies. On this basis, by comparing the state deviation with an adaptive threshold, it can not only accurately identify various types of dynamic events such as the arrival of new tasks, changes in AGV state, and environmental changes, but also avoid false triggering caused by fluctuations in a single indicator or noisy data, improving the accuracy and robustness of event identification. At the same time, by decomposing and analyzing the sources of deviation, it can accurately locate the event type, enabling the task assignment process to be triggered at the most needed moment, reducing ineffective scheduling and system oscillations, thereby significantly improving the real-time response capability, scheduling stability, and overall operating efficiency of the AGV system in complex dynamic environments.

[0040] Furthermore, based on the current system state, a task pool and AGV state set are constructed, including: S19, Based on the generation of the task assignment trigger signal, record the current system state vector and related raw data to form a system state snapshot; S110: Extract all unfinished tasks from the system status snapshot, sort them according to task priority, location and time constraints, and build a task pool; The process involves several steps. First, the priority of each task is converted into a corresponding numerical weight according to a preset mapping rule (e.g., through a level lookup table or linear mapping). The higher the priority, the greater its basic ranking weight. Second, based on the task location and the current distribution of each AGV, a spatial accessibility index for the task is calculated, such as the distance from the task point to the nearest available AGV. Third, a time constraint parameter is introduced, converting the task's deadline into a time urgency index. Finally, the priority weight, spatial accessibility weight, and time urgency weight are normalized to remove inconsistencies in units, and then weighted and merged according to a preset ratio to form a comprehensive ranking score for each task. Tasks are then ranked from highest to lowest based on this score, ensuring that urgent tasks are prioritized while also considering execution efficiency and overall system coordination. A specific example of task ranking scores is shown in Table 1.

[0041] Table 1 Example of Task Ranking and Scoring S111, based on the system status snapshot, extracts the current status information of all AGVs to form an AGV status set, including: current position, remaining power, current task queue and availability indicator.

[0042] S112 encapsulates the task pool and AGV status set into a unified data structure.

[0043] S2, based on the task pool and AGV status set, filters out a set of candidate AGVs that meet the constraints according to the task location, AGV current position, battery status, and current task execution status, in order to narrow down the scope of subsequent calculations, including: S21, Based on the task pool, select tasks to be assigned from it, and extract the target information of the tasks to be assigned. The target information includes task location, task priority, number of tasks, task distribution density and task time constraint parameters. S22, Based on the AGV status set, extract the current operating status information of each AGV. The operating status information includes the AGV's current position, speed, power level, current task queue length and task execution progress, load rate and operating status identifier. S23. Based on the task location and the current position of the AGV, calculate the Euclidean distance from the AGV to the task location, and remove AGVs whose Euclidean distance exceeds a preset threshold to obtain a subset of AGVs. S24, based on the AGV subset, combining the current power of each AGV and the estimated energy consumption for completing the task, calculate the power required to complete the task and return to the charging point, and eliminate AGVs with insufficient remaining power to form a feasible power set; such as Figure 3As shown, the simulation curves of power consumption for different AGVs during task execution and return to the charging point are displayed, where the horizontal axis represents task execution time and the vertical axis represents the remaining power percentage. It can be seen that some AGVs with low initial power levels are close to their power threshold before completing the task, thus being eliminated by the screening mechanism to avoid the risk of execution interruption. The specific formula for calculating the amount of electricity required to complete the task and return to the charging point is as follows:

[0044]

[0045]

[0046] In the formula, Execute task j for AGVi and return the required amount of power to the charging point. The power required for AGVi to perform task j. This refers to the amount of electricity AGVi needs to return to the charging point from its mission completion location. This represents the power consumption of AGVi per unit time. The estimated time for AGVi to complete task j. The estimated time for AGVi to return from the mission point to the charging point.

[0047] S25. Based on the feasible set of power consumption, calculate the current task load and the expected load after adding new tasks for each AGV, and remove AGVs that exceed the preset load threshold (such as task queue length not exceeding the maximum queue capacity or cumulative load rate not exceeding 80%) to obtain the load balancing candidate set. The specific calculation formula for the expected load is as follows:

[0048] In the formula, The estimated total load after adding new tasks to AGVi. The load of the newly added task j. This is the task priority weighting coefficient. This represents the current task load for AGVi.

[0049] S26. Based on the load balancing candidate set, the availability is determined according to the AGV status identifier. AGVs in a faulty, maintenance, communication interruption or unschedulable state are eliminated, and a candidate AGV set is formed based on the remaining available AGVs.

[0050] It should be noted that by performing a step-by-step screening of the task pool and AGV state set under multi-dimensional constraints, all AGVs originally participating in the scheduling calculation are gradually reduced to a candidate AGV set that meets the requirements of spatial reachability, power feasibility, load balancing, and operational availability. This significantly reduces the scale and complexity of subsequent task allocation and optimization calculations. At the same time, by introducing an energy consumption model for task completion and return to the charging point, as well as a new task load evaluation mechanism, AGVs with insufficient power or excessive load can be eliminated in advance. This avoids interruptions, congestion, or overload problems during actual execution, improves the executability and stability of the scheduling results, and ultimately achieves efficient, reliable, and real-time AGV task assignment in complex dynamic environments.

[0051] S3. Broadcast the task information to be assigned to the candidate AGV set. Each AGV independently and in parallel constructs a multi-dimensional cost bidding vector for the corresponding task based on its own local state and environmental information, and returns the bidding vector to the central scheduling unit. The multi-dimensional cost bidding vector includes the estimated completion time cost of the task, the energy consumption cost of executing the task, the delay increment caused to its own task queue, and the path congestion index.

[0052] In this embodiment, the task information to be assigned is broadcast to the candidate AGV set. Each AGV independently and in parallel constructs a multi-dimensional cost bidding vector for the corresponding task based on its own local state and environmental information, including: S31, Based on the candidate AGV set and the task information to be assigned, the central scheduling unit sends task broadcast information to each AGV in the candidate AGV set through the communication network. The task broadcast information includes task location, task priority, task time constraint and task execution requirements. S32, based on the received task broadcast information, each candidate AGV calls the local control unit to obtain its current local state and environmental information; The local state includes the current location, current task queue, remaining battery power, and operating status; the environmental information includes the local path topology, distribution of nearby AGVs, and path occupancy. S33. Based on the acquired local state and environmental information, each AGV performs local planning on the path from its current position to the task position. It uses a heuristic search algorithm (such as the A* algorithm) or a shortest path algorithm (such as the Dijkstra algorithm) to search for the path, obtain the optimal path, and calculate the estimated completion time of the task based on the path length and the current running speed to form a time estimation result. The specific formula for calculating the time estimation result is as follows:

[0053]

[0054] In the formula, This represents the estimated total time for AGVi to complete the task under the current path. The time taken to move from the current position to the task position. For task processing time, For path length, This represents the average operating speed of AGVi.

[0055] S34, Based on the time estimation results, calculate the estimated completion time cost of each AGV task; The specific formula for calculating the time cost is as follows:

[0056] In the formula, For time cost, This is the timeout penalty coefficient. This is the deadline for the task.

[0057] S35, based on the path planning results and power information, each AGV calculates the energy consumption cost of performing the task according to the path length, operating power and load conditions. The energy consumption cost includes the energy consumption during the task execution process and the necessary path travel energy consumption. The specific formula for calculating the energy consumption cost is as follows:

[0058] In the formula, For energy consumption costs, This is the load impact factor. This represents the current load rate.

[0059] S36. Based on the current task queue information, each AGV inserts the new task into the current task queue, calculates the impact of the new task on the completion time of the original task, and obtains the task queue delay increment. The delay increment is the increase in the completion time of subsequent tasks caused by the new task. The specific formula for calculating the task queue latency increment is as follows:

[0060] In the formula, For task queue delay increment, This is the current task queue for the AGV. This is the original task completion time. The completion time after inserting a new task.

[0061] S37. Based on the path planning results and environmental information, and according to the occupancy status of each node on the path analyzed by each AGV and the density of adjacent AGVs, the path congestion index is calculated. This path congestion index is used to characterize the path's resistance to passage and the degree of potential conflict. Figure 4 As shown, a heat map of congestion index for different paths is presented in a multi-AGV collaborative operation scenario. The congestion level is represented by a color mapping method from green to yellow to red: green represents low congestion area, yellow represents medium congestion area, and red represents high congestion area. The path congestion index is calculated using the following formula:

[0062] In the formula, This is the path congestion index. , These are the weighting coefficients. For node occupancy rate, The density of AGVs near the node. For path nodes, This is the set of AGV path nodes.

[0063] S38. Based on the multi-dimensional local cost including time cost, energy consumption cost, task queue delay increment, and path congestion index, each AGV constructs a multi-dimensional cost bidding vector in a preset order to ensure the consistency of the bidding vectors submitted by each AGV in terms of structure and dimension.

[0064] It should be noted that by constructing a multi-dimensional cost bidding vector for each task, the task execution cost is expanded from a single indicator to a comprehensive quantitative expression encompassing multiple dimensions such as time, energy consumption, queue impact, and path congestion. This allows each AGV to comprehensively evaluate its task execution performance from its own local optimum. Specifically, time cost ensures task response efficiency while considering deadline constraints; energy consumption cost helps optimize and balance the overall energy consumption of the system; task queue delay increments prevent excessive interference from local optimum decisions on existing tasks, thereby improving scheduling stability; and the path congestion index reflects potential conflicts and traffic pressure during multi-AGV collaboration, effectively reducing congestion and deadlock risks. Through the synergistic effect of these multi-dimensional indicators, not only is the rationality of individual AGV decisions improved, but a structured and comparable decision-making basis is also provided to the central scheduling unit. This achieves multi-objective optimization at the global level, improving the overall system efficiency, robustness, and real-time response capabilities.

[0065] Furthermore, since each AGV uploads its bidding vector via the network, the calculation completion time and network transmission delay of different AGVs vary, which may cause the bidding data received by the central scheduling unit to be inconsistent in time, resulting in a "data asynchrony" phenomenon. In highly dynamic scenarios, this delay may cause some bidding information to no longer reflect the current real state, thereby affecting the accuracy of decision-making.

[0066] Returning the bidding vector to the central scheduling unit includes: S39, each AGV caches and records the multi-dimensional cost bidding vectors generated at multiple consecutive time points locally, and combines the corresponding timestamp information to construct a discrete trajectory sequence of the bidding vectors changing over time, which serves as the basic data for subsequent derivative feature extraction; S310, based on the constructed discrete trajectory sequence, each AGV performs interpolation fitting on the bidding vector in the time dimension to obtain the bidding function in the continuous time domain, thus providing a continuous function basis for derivative calculation; The specific calculation formula for the bidding function is as follows: ,

[0067] In the formula, Let be the bidding value of the i-th AGV in continuous time t. For the i-th AGV at discrete time... The sampled bidding vector, For the i-th AGV at the next sampling time The bidding vector.

[0068] S311, According to the bidding function, each AGV calculates the first or second derivative of the bidding vector with respect to time to obtain derivative features. The derivative features include the bidding change rate and change acceleration features, which are used to characterize the dynamic evolution trend of the bidding state. It should be noted that the bidding vector sequence refers to a multi-dimensional bidding vector calculated and locally cached by the same AGV at multiple historical trigger times (each trigger is driven by a dynamic event mechanism), with each vector accompanied by a corresponding timestamp. Using time as the x-axis, piecewise low-order interpolation (such as linear interpolation or cubic spline interpolation) is performed on each dimension to obtain a continuous function of cost change over time for each dimension. Then, its first derivative (rate of change in bidding) and second derivative (acceleration of change in bidding) are calculated. The first derivative characterizes the dynamic trend of cost change in that dimension, while the second derivative characterizes the inflection point information of that trend; together, they constitute the derivative features.

[0069] S312, Based on the derivative characteristics, each AGV further constructs derivative invariant parameters that are insensitive to time delay. The derivative invariant is the normalized ratio between the bidding change rate and the bidding value, which is used to eliminate the influence of time offset on the bidding value itself, thereby obtaining a stable feature representation that reflects the bidding change structure. S313, each AGV, according to the preset dimensional structure, fuses or replaces the derivative invariant parameters with the original bidding vector to construct the corresponding derivative invariant bidding vector, which is used to characterize the changing trend of the bidding state and is insensitive to time delay or time translation, so as to form a bidding expression form that is robust to time delay. S314, each AGV encapsulates the derivative invariant bidding vector along with the corresponding time stamp information to form an enhanced bidding data packet, and sends it to the central scheduling unit through the communication network; The central scheduling unit receives the enhanced bidding data packets uploaded by each AGV, performs consistency comparison and sorting based on the difference between the bidding vectors of derivative invariants, and sorts each AGV according to the difference from smallest to largest. By utilizing the characteristic that derivative invariants are insensitive to time delay, the comparability of bidding data arriving at different times under a unified changing structure is achieved, thereby avoiding decision-making bias caused by time asynchrony.

[0070] The central scheduling unit receives the enhanced bidding vector uploaded by each AGV, which includes the original bidding value and the derivative invariants for each dimension. The original values ​​and derivative invariants are then normalized to eliminate dimensional differences.

[0071] The interpolation and derivative calculations described above all employ local linear approximation or low-order difference methods. The computational complexity is linearly related to the dimension of the bidding vector, and the time taken for a single calculation is typically in the microsecond range. The AGV onboard controller possesses sufficient real-time computing capabilities and will not affect the real-time performance of critical tasks such as chassis control and communication due to the execution of this method.

[0072] It should be noted that by transforming the original bidding vector into a changing structural expression based on derivative invariants, the bidding information is transformed from a numerical comparison sensitive to absolute time to a structural comparison with a stable trend. This maintains the comparability and consistency between bidding data from different AGVs even in the presence of computational delays and asynchronous communication, effectively reducing the impact of time asynchrony on decision results and improving the accuracy, robustness, real-time performance, and stability of task assignment.

[0073] S4, the central scheduling unit receives the bidding vectors returned by each AGV, dynamically adjusts the weights of each bidding vector based on the TOPSIS method and the task urgency and system status, comprehensively evaluates the bidding vectors, determines the optimal AGV, and completes task assignment, including: S41, standardize the bidding vector set, normalize the indicators of each dimension to a unified dimension range, and the normalization adopts the minimum-maximum normalization method to obtain the candidate AGV bidding vector set; S42 dynamically adjusts the weights of each bidding dimension based on the current system status and task urgency, including: S43, for the time cost dimension, adjust the time cost weight based on the remaining executable time of the task; The specific formula for calculating the time cost weight is as follows:

[0074] In the formula, Weighted by time cost, As a basis for time cost weighting, The remaining executable time for the task. It is a local minimum constant.

[0075] S44, for the energy consumption cost dimension, dynamically adjust the weight based on the proportion of remaining power of AGV to obtain the energy consumption cost weight; The specific calculation formula for the energy consumption cost weight is as follows:

[0076] In the formula, As a weighted factor for energy consumption costs, As a basic weight for energy consumption costs, This represents the current remaining battery power of the AGV. This represents the maximum capacity of the AGV battery.

[0077] S45, for the task queue delay increment dimension, the weight is dynamically adjusted based on the system average queue length and the current AGV queue length to obtain the task queue delay increment weight; The specific calculation formula for the task queue delay increment weight is as follows:

[0078] In the formula, Incremental weight for task queue delay. As the base weight for queue delay, This represents the current task queue length of the AGV. This represents the average task queue length of the system.

[0079] S46, for the path congestion index dimension, the weights are dynamically adjusted based on the local path occupancy rate and the density of nearby AGVs to obtain the path congestion weights. The specific formula for calculating the path congestion weight is as follows:

[0080] In the formula, For path congestion weights, The basic weights for the congestion dimension.

[0081] S47, construct positive ideal solution vector and negative ideal solution vector, where the positive ideal solution vector is a vector composed of the optimal values ​​of each dimension, and the negative ideal solution vector is a vector composed of the worst values ​​of each dimension; Specifically, the standardized indicators of each candidate AGV in each evaluation dimension can be selected according to the principle of "maximum value for benefit-type indicators and minimum value for cost-type indicators" to form a positive ideal solution vector. At the same time, the worst values ​​can be selected according to the opposite principle (minimum value for benefit-type indicators and maximum value for cost-type indicators) to form a negative ideal solution vector, thereby forming a reference benchmark for TOPSIS evaluation.

[0082] S48, calculate the weighted Euclidean distance between each candidate AGV bidding vector and the positive and negative ideal solutions; The specific formula for calculating the weighted Euclidean distance is as follows:

[0083]

[0084] In the formula, Let be the weighted Euclidean distance from the i-th AGV bidding vector to the positive ideal solution vector. Let be the weighted Euclidean distance from the i-th AGV bidding vector to the negative ideal solution vector. Let i be the standardized metric for the i-th AGV in each evaluation dimension. For the j-th index, the positive ideal value is... The negative ideal value of the j-th index. Let be the dynamic weight of the j-th indicator.

[0085] S49, calculate the relative proximity of each candidate AGV based on weighted Euclidean distance; The relative closeness is calculated using the following formula:

[0086] In the formula, Let be the relative proximity of the i-th AGV, representing the degree to which the AGV approaches the ideal solution.

[0087] S410: Sort the candidate AGV set according to their relative proximity. The higher the relative proximity, the better. Select the optimal AGV as the task assignment object. For candidate AGVs with the same proximity, further sort them based on real-time power ratio, queue length ratio and path congestion index. By considering these real-time constraints, we can avoid selecting AGVs with insufficient power, excessive task load or excessive path congestion. This ensures that the finally selected AGV can complete the task efficiently and ensure the reliability and safety of the execution, thus achieving multi-objective optimization. S411: The optimal AGV selected and the task to be assigned are combined to form a task assignment result, and the result is sent to the AGV execution unit.

[0088] It should be noted that by standardizing the AGV bidding vector and dynamically adjusting the weights of each dimension in conjunction with real-time status factors such as task urgency, remaining AGV battery power, task queue length, and path congestion, an improved TOPSIS method is used for comprehensive evaluation and multi-level ranking to achieve precise selection of the optimal AGV. This scheme can balance task response efficiency, energy consumption, load balancing, and path accessibility, effectively preventing AGVs with insufficient power, overload, or congestion from being assigned tasks. This significantly improves the reliability of task assignment, execution efficiency, and overall system scheduling stability, achieving multi-objective optimization and enhanced dynamic real-time response capabilities.

[0089] S5, based on the assignment results, continuously update the system state within a rolling time-domain framework, and progressively optimize and adjust unexecuted tasks. This progressive optimization and adjustment refers to adjusting the allocation of unexecuted tasks only when the overall benefit is better than the historical solution, while keeping currently executing tasks unchanged, based on the current and historical allocation states within the rolling time-domain framework and constraining changes in the allocation structure by the optimal transmission distance. S51. Based on the task assignment result, construct the task and AGV allocation matrix at the current moment. The allocation matrix is ​​used to describe the matching relationship between each task and AGV and serves as the current allocation state. At the same time, record the allocation matrix at the previous moment as the historical allocation state, thereby forming an allocation state pair between adjacent moments. S52, based on the current allocation state and the historical allocation state, the unexecuted tasks are filtered, all the task sets that have not yet started execution are extracted, and the corresponding allocation relationship is retained in the current allocation matrix to form an allocation submatrix to be optimized. At the same time, the corresponding part of the historical allocation matrix is ​​extracted as a reference allocation submatrix. S53, based on the allocation submatrix to be optimized and the reference allocation submatrix, the task allocation relationship is normalized into a probability distribution form, wherein the relationship of each task being allocated to each AGV is represented as a quality distribution, the task set is mapped to the source distribution, the AGV set is mapped to the target distribution, and a task-to-AGV allocation distribution model is constructed. Specifically, the AGV allocation relationships corresponding to each task in the allocation submatrix to be optimized and the reference allocation submatrix are used as non-negative weight matrix elements. The AGV allocation weights corresponding to each task are normalized according to the task dimension, so that the sum of all AGV allocation weights corresponding to the same task is equal to 1. This transforms the original discrete allocation relationship into a discrete probability distribution representation of tasks to AGVs that satisfies the probability normalization constraint.

[0090] S54. Based on the allocation distribution model, construct the task allocation cost matrix at the current moment. The cost matrix is ​​used to characterize the immediate execution cost of each task allocated to each AGV. The execution cost includes the weighted sum of time cost, energy consumption cost, queue delay and path congestion. S55, based on the allocation cost matrix, introduce the distribution structure corresponding to the historical allocation state, construct the transmission cost function of allocation change, which is used to measure the degree of structural offset of the current allocation distribution relative to the historical allocation distribution, wherein the structural offset is characterized by the optimal transmission distance between the allocation distributions; The specific calculation formula for the transmission cost function is as follows:

[0091] In the formula, For the current allocation distribution and historical allocation distribution The optimal transmission distance (cost of allocation structure change) is used to measure the minimum cost of transforming the allocation structure from the previous time step to the current structure. Assign probability distribution matrices to the task and AGV at the current moment. Assign probability distribution matrices to the task and AGV from the previous time step. A set of feasible transmission plans, Let i be the amount of data transferred from the i-th task unit in distribution P to the j-th AGV unit in distribution Q. The unit transmission cost assigned to task i by AGVj.

[0092] S56. Based on the current allocation cost matrix and the transmission cost function, a joint optimization objective function with allocation inertia constraints is constructed, wherein the objective function simultaneously minimizes the current task allocation cost and the cost of allocation structure change, thereby optimizing the current efficiency while maintaining the continuity of the allocation structure. The joint optimization objective function is calculated using the following formula:

[0093] In the formula, The overall cost of the current task allocation scheme. Assigning results to the tasks actually performed. To assign inertia weight coefficients.

[0094] S57. Based on the joint optimization objective function, the allocation submatrix to be optimized is solved to obtain new task and AGV allocation results. The solution process weighs the current allocation cost against the historical allocation structure, so that the allocation adjustment only occurs when the new allocation scheme is better than the historical scheme in terms of overall benefit, thereby achieving adaptive suppression of allocation changes. The overall benefit is the difference between the overall cost of the current allocation scheme and the historical allocation scheme, which is used to measure the relative improvement brought about by optimization.

[0095] S58, merge and update the new allocation result with the current execution state, where the optimized allocation result is used for tasks that have not been executed, and the original allocation remains unchanged for tasks that are being executed, thus forming the updated system task allocation state; S59, the updated system task allocation status is used as the historical allocation status for the next rolling cycle, and the current allocation status is reconstructed in combination with real-time system status data. Steps S52 to S59 are repeated to achieve continuous and progressive optimization based on optimal transmission allocation inertia, thereby realizing smooth evolution and efficient collaboration of task allocation in a dynamic environment.

[0096] It should be noted that, within the rolling time-domain framework, combined with the optimal transmission allocation inertia mechanism, structural continuity constraints are introduced to the task allocation changes. This ensures the optimization of current scheduling costs while suppressing frequent fluctuations in allocation results, enabling smooth adjustment of unexecuted tasks. At the same time, real-time state updates are used to continuously correct allocation results, enhancing scheduling stability and robustness while improving the system's dynamic response capabilities, thereby improving overall task execution efficiency and resource utilization.

[0097] Reference Figure 2 As shown in the diagram, this invention provides a schematic of an AGV task dynamic assignment system based on multi-objective collaborative optimization, including a data acquisition module, a filtering and weight adjustment module, a multi-objective decision-making module, and an update module. The modules are interconnected. The data acquisition module is used to initiate the task assignment process using a dynamic event triggering mechanism; The screening and weight adjustment module is used to construct and screen a set of candidate AGVs based on constraints. The AGVs independently and in parallel generate multi-dimensional cost bidding vectors, and dynamically calculate the evaluation weights of each dimension of the bidding vector based on task urgency, AGV remaining power, system average queue length and path occupancy rate. The multi-objective decision-making module is used to comprehensively evaluate the bidding vectors, calculate the relative proximity of each candidate AGV, and determine the AGV with the highest relative proximity as the optimal AGV and complete the task assignment. The update module is used to continuously update the system status based on the assignment results within a rolling time-domain framework, and to progressively optimize and adjust unexecuted tasks.

[0098] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0099] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0100] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0101] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0102] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic task assignment method for AGVs based on multi-objective collaborative optimization, characterized in that, include: The task assignment process is initiated using a dynamic event triggering mechanism; A candidate AGV set is constructed and screened by constraints. The AGVs independently and in parallel generate multi-dimensional cost bidding vectors. The evaluation weights of each bidding vector dimension are dynamically calculated based on task urgency, AGV remaining power, system average queue length and path occupancy rate. The bidding vectors are comprehensively evaluated, the relative proximity of each candidate AGV is calculated, and the AGV with the highest relative proximity is determined as the optimal AGV and the task is assigned. Based on the assignment results, the system status is continuously updated within a rolling time-domain framework, and incremental optimizations and adjustments are made to unexecuted tasks. The incremental optimization adjustment refers to adjusting the allocation structure based on the current and historical allocation states within a rolling time-domain framework, constraining changes in the allocation structure by the optimal transmission distance, adjusting the allocation of unexecuted tasks only when the overall benefit is better than the historical scheme, while keeping the currently executed tasks unchanged.

2. The AGV task dynamic assignment method based on multi-objective collaborative optimization according to claim 1, characterized in that, The process of initiating task assignment using a dynamic event triggering mechanism includes: Calculate the deviation of the system state vector from the baseline state model; If the deviation of the state exceeds a preset deviation threshold, it is determined that a dynamic event has occurred; The contribution values ​​of the task state sub-vector, AGV state sub-vector, and environment state sub-vector in the system state vector to the state deviation are determined respectively. The sub-vector type corresponding to the maximum contribution value is determined as the dynamic event type, and the corresponding assignment trigger signal is generated.

3. The AGV task dynamic assignment method based on multi-objective collaborative optimization according to claim 1, characterized in that, The process of constructing and filtering the candidate AGV set through constraints includes constructing a task pool and an AGV state set based on the current system state: Respond to task assignment trigger signals and obtain a snapshot of the current system state; Unfinished tasks are extracted from system state snapshots and sorted by multi-dimensional weighted fusion based on task priority, spatial reachability, and time constraints to construct a task pool. AGV status information is extracted from the system status snapshot to form an AGV status set.

4. The AGV task dynamic assignment method based on multi-objective collaborative optimization according to claim 3, characterized in that, The process of constructing and filtering the candidate AGV set through constraints also includes: Select tasks to be assigned from the task pool and extract the task location and time constraints; Extract multi-dimensional status parameters for each AGV from the AGV status set, including current position, power level, load, and operating status; Set constraints for multi-dimensional state parameters, and then sequentially perform multi-dimensional constraint filtering on the AGV; The AGVs that pass all the constraints will form a candidate AGV set.

5. The AGV task dynamic assignment method based on multi-objective collaborative optimization according to claim 1, characterized in that, The generation of the multidimensional cost bidding vector includes: Broadcast the task information to be assigned to the candidate AGV set; Each candidate AGV independently and in parallel performs local path planning based on its own local state and environmental information: Local planning is performed on the path from the current location to the task location, and the multidimensional local cost and path congestion index of task execution are calculated respectively. Based on this, a multidimensional cost bidding vector is constructed and returned to the central scheduling unit.

6. The AGV task dynamic assignment method based on multi-objective collaborative optimization according to claim 5, characterized in that, The return to the central scheduling unit includes: For a multidimensional cost bidding vector sequence, interpolate each dimension with time as the horizontal axis to obtain the bidding function, and calculate its first derivative as the derivative feature; Each AGV constructs derivative invariants based on derivative characteristics; Each AGV concatenates the derivative invariant with the original bidding vector to generate a derivative invariant bidding vector, and sends it to the central scheduling unit; The central scheduling unit utilizes the characteristic that derivative invariants are insensitive to time delays to uniformly sort and compare bidding data arriving at different times.

7. The AGV task dynamic assignment method based on multi-objective collaborative optimization according to claim 1, characterized in that, The process of determining the optimal AGV and completing task assignment includes: The received set of bidding vectors is standardized. Based on the current system status and task urgency, the evaluation weights of each dimension in the bidding vector are dynamically adjusted. These dimensions include time cost, energy cost, task queue delay increment, and path congestion index. Construct positive and negative ideal solution vectors, and calculate the weighted Euclidean distance between each candidate AGV bidding vector and the positive and negative ideal solutions; The relative proximity of each candidate AGV is calculated based on the weighted Euclidean distance, and the AGVs are sorted according to their relative proximity to select the optimal AGV as the task assignment object.

8. The AGV task dynamic assignment method based on multi-objective collaborative optimization according to claim 1, characterized in that, The progressive optimization and adjustment of unexecuted tasks includes: Based on the task assignment results, construct the task and AGV allocation matrix for the current moment and the historical allocation matrix to form allocation state pairs; Based on the current and historical allocation status, unexecuted tasks are filtered out, and the corresponding allocation relationships in the current allocation matrix are extracted to form an allocation submatrix to be optimized. The corresponding parts in the historical allocation matrix are also extracted to form a reference allocation submatrix. The allocation relationship between the submatrix to be optimized and the reference allocation submatrix is ​​normalized into a probability distribution form to construct the task-to-AGV allocation distribution model.

9. The AGV task dynamic assignment method based on multi-objective collaborative optimization according to claim 8, characterized in that, The progressive optimization and adjustment of unexecuted tasks also includes: Based on the allocation distribution model, construct the task allocation cost matrix at the current moment; By introducing the distribution structure corresponding to the historical allocation state, a transmission cost function is constructed, and the structural offset of the current allocation distribution relative to the historical allocation distribution is measured by the optimal transmission distance. Based on the allocation cost matrix and transmission cost function, a joint optimization objective function is constructed; For solving the optimal allocation submatrix, the allocation is adjusted only when the overall benefit is better than the historical solution. The optimized allocation results are merged and updated with the current execution status. Optimized allocation is applied to tasks that have not yet been executed, while the original allocation is maintained for tasks that are currently being executed, resulting in an updated system task allocation status. The updated system task allocation status is used as the historical allocation status for the next rolling cycle, and this process is repeated iteratively to achieve continuous and incremental optimization.

10. A system using the AGV task dynamic assignment method based on multi-objective cooperative optimization as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to initiate the task assignment process using a dynamic event triggering mechanism; The screening and weight adjustment module is used to construct and screen a set of candidate AGVs based on constraints. The AGVs independently and in parallel generate multi-dimensional cost bidding vectors, and dynamically calculate the evaluation weights of each bidding vector dimension based on task urgency, AGV remaining power, system average queue length and path occupancy rate. The multi-objective decision-making module is used to comprehensively evaluate the bidding vectors, calculate the relative proximity of each candidate AGV, and determine the AGV with the highest relative proximity as the optimal AGV and complete the task assignment. The update module is used to continuously update the system status based on the assignment results within a rolling time-domain framework, and to perform incremental optimization and adjustments on unexecuted tasks.

Images