Multi-objective Optimization Scheduling Method and System for Multi-AGV Flexible Job Shop

Abstract

The present invention relates to a multi-objective optimization scheduling method and system for a multi-AGV flexible job shop. The method includes: collecting job shop layout data, AGV operation parameters, and production task data, and constructing a basic model of the scheduling system; performing recursive neural network training and physical constraint model integration to obtain a hybrid scheduling prediction model; performing subtask decomposition and path search processing, and obtaining a bi-objective optimization model through constraint linearization transformation; performing particle swarm optimization to solve the bi-objective optimization model, and obtaining a non-dominated solution set through two-stage iterative calculation of an adaptive weight strategy and constraint transformation; performing multi-dimensional evaluation calculation on the non-dominated solution set to obtain an optimal scheduling execution plan; and performing model feedback update on the hybrid scheduling prediction model to obtain a target scheduling prediction model. The present invention realizes the multi-objective balance of task completion time and energy consumption, and improves the real-time response ability and operation stability of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of AGV scheduling, and particularly relates to a multi-objective optimization scheduling method and system for a multi-AGV flexible job shop. Background Art

[0002] With the rapid development of intelligent manufacturing, the multi-AGV scheduling optimization in a flexible job shop has become a key technology to improve production efficiency. Traditional AGV scheduling methods mainly rely on static planning and manual experience, and it is difficult to cope with the dynamic and complex production environment. Especially in the scenario of multi-AGV collaborative operation, the coupling relationship between task allocation and path planning makes the scheduling optimization problem more complex.

[0003] Existing AGV scheduling algorithms generally have problems of insufficient adaptability to the real-time dynamic environment. The scheduling method based on filters overly relies on prior information, and filter parameters need to be redesigned for job shops of different scales; the method based on heuristic algorithms is difficult to balance multiple optimization objectives such as task completion time and energy consumption, and is prone to falling into local optimal solutions. Summary of the Invention

[0004] The main object of the present invention is to provide a multi-objective optimization scheduling method and system for a multi-AGV flexible job shop. The present invention realizes the multi-objective balance of task completion time and energy consumption, and improves the real-time response ability and operation stability of the system.

[0005] To achieve the above object, the present invention provides a multi-objective optimization scheduling method for a multi-AGV flexible job shop, including the following steps:

[0006] Collect job shop layout data, AGV operation parameters and production task data, and construct a basic model of the scheduling system;

[0007] Based on the AGV historical operation data in the basic model of the scheduling system, perform recursive neural network training and physical constraint model integration to obtain a hybrid scheduling prediction model;

[0008] Obtain the AGV state feature vector and task demand features output by the hybrid scheduling prediction model, and perform subtask decomposition and path search processing, and obtain a bi-objective optimization model through constraint linearization transformation;

[0009] Perform particle swarm optimization on the bi-objective optimization model, and obtain a non-dominated solution set through two-stage iterative calculation of an adaptive weight strategy and constraint transformation;

[0010] Perform multi-dimensional evaluation calculation on the non-dominated solution set to obtain an optimal scheduling execution plan;

[0011] Input the optimal scheduling execution plan into the AGV control system to perform model feedback update on the hybrid scheduling prediction model, and obtain the target scheduling prediction model.

[0012] The present invention also provides a multi-objective optimal scheduling system for a multi-AGV flexible job shop, including:

[0013] An acquisition module, configured to acquire job shop layout data, AGV operation parameters, and production task data, and construct a basic model of the scheduling system;

[0014] A training module, configured to perform recurrent neural network training and physical constraint model integration based on the AGV historical operation data in the basic model of the scheduling system to obtain a hybrid scheduling prediction model;

[0015] A processing module, configured to obtain the AGV state feature vector and task requirement features output by the hybrid scheduling prediction model, and perform subtask decomposition and path search processing, and obtain a bi-objective optimization model through constraint linearization transformation;

[0016] A solving module, configured to perform particle swarm optimization on the bi-objective optimization model, and obtain a non-dominated solution set through two-stage iterative calculation of an adaptive weight strategy and constraint transformation;

[0017] A calculation module, configured to perform multi-dimensional evaluation calculation on the non-dominated solution set to obtain an optimal scheduling execution plan;

[0018] An update module, configured to input the optimal scheduling execution plan into the AGV control system to perform model feedback update on the hybrid scheduling prediction model, and obtain the target scheduling prediction model.

[0019] The present invention also provides a computer device, including a memory and a processor, where a computer program is stored in the memory, and when the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0020] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in any one of the above are implemented.

[0021] In summary, the technical solution provided by the present invention constructs a hybrid scheduling prediction model through the deep integration of a recurrent neural network and a physical constraint model, improving the adaptability of the scheduling system to dynamic environments and avoiding the over-reliance on prior parameters in traditional methods; adopts a task decomposition method based on similarity and a path planning strategy that improves the A* algorithm, reducing the coupling complexity of task allocation and path planning and solving the deadlock problem in multi-AGV collaborative scheduling; introduces a two-stage optimization strategy of an adaptive particle swarm algorithm and the ε-constraint method to achieve multi-objective balance of task completion time and energy consumption, overcoming the defect that traditional algorithms are prone to falling into local optima; designs a multi-dimensional evaluation system based on time dimension, efficiency dimension, and energy consumption dimension, and conducts comprehensive evaluation in combination with the analytic hierarchy process method to ensure the scientificity and reliability of the scheduling plan; through the online data collection and model dynamic update mechanism, realizes the continuous optimization and adaptive learning of the scheduling system, and significantly improves the real-time response ability and operation stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a schematic diagram of the steps of a multi-objective optimization scheduling method for a multi-AGV flexible job shop in an embodiment of the present invention;

[0023] Figure 2 is a block diagram of the structure of a multi-objective optimization scheduling system for a multi-AGV flexible job shop in an embodiment of the present invention;

[0024] Figure 3 is a schematic block diagram of the structure of a computer device in an embodiment of the present invention.

[0025] The realization, functional features, and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] In order to make the object, technical solution, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0027] Referring to Figure 1 , this embodiment provides a multi-objective optimization scheduling method for a multi-AGV flexible job shop, including the following steps:

[0028] S1, collect the layout data of the job shop, the operating parameters of the AGVs, and the production task data, and construct a basic model of the scheduling system;

[0029] Among them, the layout data of the collection workshop is collected, including the coordinate data of processing equipment and the location data of AGV charging stations. The topological structure of the AGV's driving path is modeled. By performing topological analysis on the layout of the paths in the workshop, the layout topological data of the workshop is obtained. The coordinate mapping of the AGV's motion trajectory is carried out according to the workshop layout topological data. By converting the coordinate information of each important location, equipment, and path in the workshop into a unified workshop layout matrix, a matrix model representing the spatial distribution of the workshop is obtained. The operating parameters of the AGV are collected. This includes the collection of data such as the driving speed, turning radius, load capacity, and battery capacity of the AGV equipment. These parameters are monitored in real time through on-vehicle sensors, and combined with data fitting techniques such as the least squares method, a dynamic parameter model of the AGV is established. This model can accurately describe the motion performance and constraints of the AGV under different working conditions. The production tasks are scheduled according to the process flow data in the workshop. The workpiece processing operation data, the transfer time data between operations, and the priority data of the tasks in the workshop are input into the task scheduler. The task scheduler calculates the optimal production task sequence through a linear programming model. This task sequence will determine the execution order and priority of different tasks to ensure the smoothness and efficiency of the production process. In this process, the linear programming model not only considers the time constraints between tasks but also combines the priority information of the tasks to maximize the overall benefit of the workshop production. Based on the workshop layout matrix, the dynamic parameter model of the AGV, and the production task sequence, a kinematic model of the AGV is established. In the kinematic model, the speed vector and position vector of the AGV are substituted as core variables into the Euler equations, and the motion constraint conditions of the AGV during operation are obtained by solving these equations. The motion constraint conditions refer to the physical limitations that the AGV must satisfy when performing tasks, such as the maximum speed, minimum turning radius, etc. These constraints ensure the safety and efficiency of the AGV when performing tasks. The motion constraint conditions of the AGV and the task time window are modeled for constraints to obtain a scheduling constraint model. The time window means that each task needs to be completed within a certain time range, and it must be ensured during the scheduling process that the AGV can complete the task within the specified time. At the same time, considering the energy consumption of the AGV, in the case of long-term operation, the battery capacity of the AGV will become a limiting factor. The integral operation of energy consumption is performed on the speed data and battery capacity data in the dynamic parameter model of the AGV to obtain the corresponding energy consumption function, quantifying the energy requirements of the AGV when performing different tasks. The motion constraint model of the AGV, the task time window constraint model, and the energy consumption function are integrated to obtain the basic model of the scheduling system.

[0030] S2. Based on the historical operation data of the AGV in the basic model of the scheduling system, recursive neural network training and physical constraint model integration are carried out to obtain a hybrid scheduling prediction model;

[0031] Specifically, the AGV operation trajectory data, task execution time data, and energy consumption data in the basic model of the scheduling system are partitioned into datasets, and the data is converted into a training sample set suitable for time series analysis through time series segmentation. The training samples are standardized. To improve the quality of the data and the efficiency of neural network training, the training sample set is input into a batch normalization layer for data standardization to ensure that the means and variances of the input features are relatively balanced, which helps to accelerate training and avoid problems such as gradient vanishing or explosion. To enhance the diversity of the training data, data augmentation techniques are used to randomly rotate and translate the trajectory data to generate an augmented training data matrix. In this way, the scale of the training set is effectively expanded, and at the same time, the model's adaptability and generalization ability to different scenarios are enhanced. Time series feature extraction is performed through long short-term memory network units (LSTM). LSTM can effectively capture the long-term and short-term dependencies in time series data and is suitable for processing the historical trajectory data of AGVs. By batch processing the augmented training data matrix, LSTM can extract the state feature vectors of AGVs, and these feature vectors will be used as inputs to construct a recurrent neural network model. The AGV state features include dynamic change features such as position, speed, and acceleration, which can reflect the running state of AGVs at different time periods. The AGV state feature vectors are input into a recurrent neural network with 3 hidden layers. This recurrent neural network performs deep learning and processing on the input data through multiple layers of neurons to extract more complex patterns. To optimize the training effect of the network, the Adam optimizer is used to optimize the parameters of the neural network. The Adam optimizer is an adaptive optimization algorithm that can effectively adjust the learning rate and accelerate convergence during training, avoiding the defects of traditional gradient descent methods. After training, the first neural network model obtained can initially predict the state of AGVs. Dropout processing is performed on the hidden layer of the first neural network model. Dropout is a regularization technique that randomly discards some neurons during training to avoid overfitting of the model to the training data. When the validation loss does not decrease for several consecutive times (e.g., N times), an early stopping mechanism is triggered to prevent the model from being over-adjusted during training, resulting in performance degradation. After optimization, the second neural network model obtained has better generalization ability and higher stability. Based on the kinematic equation, time window constraint, and energy consumption function of AGVs in the basic model of the scheduling system, these physical constraint conditions are constructed as Lagrangian constraint terms. The Lagrangian constraint terms effectively introduce the physical constraint conditions into the optimization problem, ensuring that the scheduling decision not only conforms to the actual physical limitations but also can balance the requirements of multiple objectives. For example, the kinematic equation of AGVs can ensure that they do not exceed the physical limitations during driving, the time window constraint can ensure that tasks are completed on time, and the energy consumption function limits the energy consumption of AGVs, thus ensuring that AGVs can complete tasks efficiently.The physical constraint loss function and the prediction loss of the neural network are weighted and combined to obtain a hybrid optimization objective function. Under the guidance of this objective function, the backpropagation algorithm is used for iterative calculation. By gradually adjusting the parameters of the neural network, the hybrid scheduling prediction model can maximize the prediction accuracy while satisfying the physical constraints. After multiple iterations, the finally obtained hybrid scheduling prediction model can accurately predict the operating state and scheduling scheme of the AGV according to the state characteristics and task requirements of the AGV.

[0032] S3. Obtain the AGV state feature vector and task requirement features output by the hybrid scheduling prediction model, and perform subtask decomposition and path search processing. Through constraint linearization transformation, a bi-objective optimization model is obtained;

[0033] It should be noted that the output of the hybrid scheduling prediction model is processed, and the similarity between the state feature vector of the AGV and the task demand feature is calculated. The similarity between tasks is measured by a weighted combination of the cosine distance and the Euclidean distance to obtain a task similarity matrix, which reflects the similarity degree between different tasks. According to the set similarity threshold, tasks are clustered to obtain the decomposition sequence of tasks. Each subtask in the task decomposition sequence is mapped and matched with the AGV device. To ensure a reasonable pairing of tasks and AGVs, a 0-1 integer programming method is used to construct an M×N-dimensional task assignment matrix, where M is the total number of tasks and N is the total number of AGVs. This matrix clearly represents the assignment relationship between each task and each AGV, and a preliminary task assignment scheme is obtained. Based on the preliminary assignment scheme, the execution task sequence of each AGV is encoded for optimization. Through an optimization method based on the genetic algorithm, single-point crossover and two-point mutation operations are used to generate a new task optimization sequence. The single-point crossover operation can explore more task arrangement methods by exchanging some tasks in the task sequence, while the two-point mutation operation increases the diversity of the solution space by randomly changing the order of tasks. These operations help to improve the diversity of task scheduling and thus the optimization effect of the scheduling scheme. Path planning is carried out. According to the workshop layout matrix, the path planning space of the A* algorithm is constructed. The A* algorithm is a graph search algorithm that can find the shortest path in a given grid environment. When constructing the path planning space, the layout space of the workshop is rasterized, and the distance between each node is set to the length of an AGV body to ensure the accuracy and practical feasibility of path calculation. Through the rasterization process of the grid method, a path search network is obtained. A heuristic function is set for the path search network to optimize the path search process. The design of the heuristic function is crucial for the performance of the A* algorithm. An evaluation function of the path is constructed by combining the weighted sum of the Manhattan distance and the dynamic obstacle avoidance factor. The Manhattan distance can provide an estimate of the straight-line distance, while the dynamic obstacle avoidance factor takes into account the influence of real-time obstacles. In this way, the collision risk of the AGV during the path planning process is effectively avoided. Using the path evaluation function, a cost matrix of the path is obtained, which reflects the costs of different paths and helps to optimize the path selection. At the same time, the motion constraint conditions of the AGV are linearized. The kinematic equation, turning constraint, and speed constraint of the AGV are usually non-linear. The first-order Taylor expansion is used to transform the non-linear constraints into a system of linear inequalities. So that the originally complex motion constraint conditions can be expressed in the form of linear constraints, which is convenient for subsequent optimization calculations. The task completion time function is calculated based on the task optimization sequence. This function can reflect the time requirements for task completion and provide a timeliness basis for scheduling optimization. At the same time, according to the dynamic parameter model of the AGV, the energy consumption function of each AGV is calculated, which reflects the energy consumption of the AGV during the execution of tasks.To ensure reasonable comparison of the objective function at different scales, the task completion time function and the energy consumption function are normalized to convert them into standardized objective functions. The standardized objective functions, linearized constraint equations, and path cost matrices are combined to obtain a bi-objective optimization model.

[0034] S4. Perform particle swarm optimization to solve the bi-objective optimization model, and through two-stage iterative calculations of the adaptive weight strategy and constraint conversion, obtain the non-dominated solution set;

[0035] Specifically, the first-stage initialization of the bi-objective optimization model is carried out. According to the total number of tasks M and the total number of AGVs N, an initial solution of the particle swarm with dimensions of M×N is generated. The initial solution space represents a preliminary matching between tasks and AGVs, covering all possible task allocation methods. In the framework of particle swarm optimization, each particle represents a task allocation scheme, and the solution space is the scope of search for the particle swarm. The initial solution space is input into the adaptive particle swarm algorithm for optimization. The adaptive particle swarm algorithm controls the exploration and exploitation of the search process by adjusting the inertia weight of the particle swarm. In this process, the linear decreasing method is used to calculate the inertia weight value. The adjustment of the inertia weight value helps to balance the local search and global search of particles in the search space, enabling particles to maintain a larger search scope in the early stage and being able to finely adjust the accuracy of the solution in the later stage, thus accelerating the convergence process. The positions and velocities of the particle swarm are updated through the adaptive particle swarm algorithm, effectively guiding the particles to move towards the direction of the optimal solution. In this process, the movement of particles is not only affected by their own experience but also optimized by the guidance of the global optimal solution. To improve the efficiency of the particle swarm algorithm, the tournament selection strategy is adopted to screen the global optimal solution. This strategy selects the current most excellent particle through a competition mechanism and assigns it a higher selection probability, improving the search quality of the particle swarm. After the search calculation in the first stage, a set of search solution sets in the first stage is obtained, and these solutions represent the possible solutions found in the preliminary optimization process. The ε-constraint method is applied to the search solution set in the first stage. The ε-constraint method is used to perform constraint transformation on the objective function, converting the energy consumption objective into an inequality constraint condition to obtain a constraint transformation model. The original multi-objective optimization problem is transformed into a single-objective optimization problem. The energy consumption objective is regarded as a constraint condition, while the task completion time is optimized as the main objective. By taking energy consumption as a constraint, it is ensured that while optimizing the task completion time, the energy consumption limit is not exceeded, achieving the rational use of resources. Based on the constraint transformation model, the population is evolved by using the binary crossover operator and the Gaussian mutation operator to obtain the initial population in the second stage. The binary crossover operator can exchange some genes of particles, thus generating new solutions, while the Gaussian mutation operator increases the diversity of the search space by slightly perturbing the positions of particles, thus avoiding the risk of falling into local optimal solutions. Local search optimization is carried out on the initial population in the second stage. The particle positions are finely adjusted through the variable neighborhood search strategy, which ensures that better solutions can be found in the local solution space by dynamically adjusting the search scope and neighborhood structure. The introduction of variable neighborhood search can effectively improve the local search ability of the particle swarm algorithm, enhancing the accuracy and optimization effect of the solution. After local search optimization, a set of local optimal solution sets is obtained, and these solution sets represent the optimal solutions in the current stage of the search process. To select the most representative solution set from the local optimal solution sets, non-dominated sorting is carried out.Non-dominated sorting is a sorting method based on the Pareto optimization idea, which determines which solutions are optimal by comparing the dominance relationships between solutions. In non-dominated sorting, if a solution is not inferior to another solution in all objectives and is superior to another solution in at least one objective, then this solution is considered to dominate the other solution. Through non-dominated sorting, the local optimal solution set is divided into multiple levels to form a priority sorting. The solution set is screened based on the elitist retention strategy. The elitist retention strategy ensures the continuous existence of the best solutions during the evolutionary process and their further optimization in subsequent iterations by retaining the current optimal solutions. Selection is based on the crowding distance of the solutions. The crowding distance is used to measure the density of solutions in the solution space. The smaller the crowding distance of a solution, the sparser it is in the objective space and the more potential it has, so it should be retained preferentially. After non-dominated sorting, calculation of the crowding distance, and screening by the elitist retention strategy, a candidate non-dominated solution set is obtained. By determining the Pareto dominance relationship among these candidate solutions, the comprehensive optimal solution for task completion time and energy consumption is selected. These final non-dominated solution sets represent the solutions that can achieve the best balance among multiple objectives in task scheduling and path planning, that is, the non-dominated solution sets.

[0036] S5. Perform multi-dimensional evaluation calculations on the non-dominated solution set to obtain the optimal scheduling execution plan;

[0037] Among them, for each solution in the non-dominated solution set, time-related indicators are extracted. The time indicators include the total time for task completion and the waiting time of the AGV. By calculating the total time for task completion, the scheduling efficiency of the entire production task is evaluated, while the waiting time of the AGV reflects the idle degree of the vehicle or the irrationality of task arrangement. Standardize all the time indicators to enable fair comparison between different solutions, and obtain the standardized time indicators. Calculate the efficiency indicators for the non-dominated solution set. The efficiency indicators include the utilization rate of AGV equipment and the completion rate of production tasks. The utilization rate of AGV equipment is obtained by measuring the ratio of the active time of each AGV during the production process to the total working time, reflecting the resource utilization efficiency of AGV in the workshop. The completion rate of production tasks is measured by comparing the ratio between the actual number of completed tasks and the planned number of tasks to measure the completion of workshop tasks. Input these two efficiency indicators into the linear programming model for standardization calculation to obtain the corresponding standardized efficiency indicators. For the AGV equipment operation energy consumption data in each non-dominated solution set, cumulative calculation is performed. By calculating the total energy consumption of each AGV and dividing it by the number of tasks completed by the AGV, the standardized energy consumption indicator of each AGV is obtained, reflecting the energy use efficiency of the AGV under specific task allocation and scheduling. Input the standardized time indicators, standardized efficiency indicators, and standardized energy consumption indicators into the evaluation system of the analytic hierarchy process. The analytic hierarchy process is a comprehensive evaluation method for multi-criteria decision-making, which can rank multiple alternative solutions according to the weights of each indicator. To establish the analytic hierarchy network, a three-layer evaluation structure is constructed. The three-layer structure includes the goal layer, criterion layer, and alternative layer. The goal layer is the evaluation goal of the final scheduling execution plan, the criterion layer is the evaluation indicators such as time, efficiency, and energy consumption, and the alternative layer is each solution in the non-dominated solution set. Through substituting the standardized indicators into this structure, comprehensive evaluation is carried out. Calculate the consistency of the judgment matrix in the three-layer evaluation structure. The construction of the judgment matrix is a key step in the analytic hierarchy process, which is used to express the relative importance between various evaluation indicators. Through consistency calculation, it is ensured that the weight distribution in the judgment matrix is reasonable and the consistency between different evaluation indicators is relatively high. The consistency calculation is completed by calculating the consistency ratio (CR). If the consistency ratio is lower than the set threshold, it indicates that the judgment matrix has good logical consistency and the weight distribution is reasonable. Based on the weight matrix obtained through consistency calculation, comprehensive calculation is performed on the standardized time indicators, efficiency indicators, and energy consumption indicators to obtain the evaluation score of each non-dominated solution set. This score reflects the comprehensive advantages and disadvantages of each solution under multi-dimensional objectives such as time, efficiency, and energy consumption. Based on the comprehensive evaluation score sequence, multi-AGV simulation is carried out to verify the execution effect of the selected solution in the actual scheduling process. The simulation can simulate the scheduling process of multiple AGVs in the actual job shop and evaluate the execution effects of different solutions.Verify the feasibility and effectiveness of the optimization plan through simulation to ensure that the selected optimal scheduling execution plan can provide the expected performance in the actual environment.

[0038] S6. Input the optimal scheduling execution plan into the AGV control system to perform model feedback update on the hybrid scheduling prediction model, and obtain the target scheduling prediction model.

[0039] Specifically, the optimal scheduling execution plan is decomposed into a time series. The task sequence is converted into speed control instructions that AGV can understand and execute. By inputting the task sequence into the kinematic equation of the AGV, according to the time nodes of each task and the intervals between tasks, the corresponding speed instructions are calculated to provide real-time control instructions for the AGV. The real-time control sequence of the AGV is monitored for status and data filtering. By collecting the position information and task completion status data generated during the real-time operation of the AGV, the operation status matrix of the AGV is obtained. To ensure the accuracy and reliability of the data, a third-order Kalman filter is used to filter these data. By setting the noise coefficient to 0.1, the Kalman filter can effectively reduce the influence of external interference and sensor errors, and calculate the true position and state changes of the AGV. The AGV operation status matrix is input into a recurrent neural network model for analysis and prediction. Using long short-term memory (LSTM) units, the model can capture the time dependence during the movement of the AGV and predict the future movement trajectory of the AGV. By detecting the collision risk for the state prediction sequence, the model can identify potential collision risks. Based on the workshop layout matrix, the minimum safe distance between AGVs is calculated, and the safety threshold is set to 2 times the vehicle body length to determine whether a collision occurs. If the predicted trajectory intersects with the movement trajectories of other AGVs and does not meet the requirement of the safe distance, the system will trigger an obstacle avoidance mechanism and adjust the path or speed to ensure that the safe distance between AGVs is maintained. To improve the driving safety of the AGV, the local path of the AGV is replanned according to the obstacle avoidance constraint conditions. During this process, the Bellman equation is used to calculate the path cost, and the path points are continuously iteratively updated to find the optimal path sequence. The Bellman equation helps the AGV calculate the path with the minimum cost according to the current state and environmental information, enabling the AGV to complete the task with the optimal path while avoiding collisions. After the path planning is completed, the AGV operation status matrix and the optimized path sequence are combined in time series to form a new real-time execution dataset. These data include important information such as the real-time position of the AGV, task completion status, and energy consumption. By statistically calculating these real-time data, the task completion rate and energy consumption rate of the AGV are obtained. The hybrid scheduling prediction model is updated online. This update process is based on real-time data. By means of batch processing, every 32 samples are used as a batch and input into the hybrid scheduling prediction model. Through this operation, the system can optimize and update the model parameters according to the latest operation data, so that the scheduling prediction model can always adapt to the dynamically changing production environment in the workshop. The updated model parameters and the physical constraint loss function are weighted and integrated to ensure that the model can not only accurately predict the movement trajectory and task completion of the AGV, but also fully consider physical constraint conditions such as energy consumption and task time windows during optimal scheduling, resulting in the target scheduling prediction model.

[0040] In one example, the layout data of the production workshop, the operating parameters of the AGV, and the production task data are collected, and a basic model of the scheduling system is constructed, including:

[0041] Collect the coordinate data of the processing equipment and the location data of the AGV charging stations in the production workshop, and perform topological structure modeling on the AGV driving channels to obtain the workshop layout topological data;

[0042] Perform coordinate mapping on the AGV movement trajectory according to the workshop layout topological data to obtain the workshop layout matrix;

[0043] Collect the driving speed, turning radius, load capacity, and battery capacity of the AGV equipment through sensors and perform least squares fitting to obtain the AGV dynamic parameter model. Then, input the workpiece processing procedure data, the procedure transfer time data, and the task priority data into the task scheduler for linear programming modeling to obtain the production task sequence;

[0044] Based on the workshop layout matrix, the AGV dynamic parameter model, and the production task sequence, establish the AGV kinematic model, and substitute the AGV speed vector and position vector into the Euler equation to obtain the AGV movement constraint conditions;

[0045] Perform constraint modeling on the AGV movement constraint conditions and the task time window to obtain the scheduling constraint model. Then, perform energy consumption integration operations on the speed data and battery capacity data in the AGV dynamic parameter model to obtain the energy consumption function, and integrate the energy consumption function with the scheduling constraint model to obtain the basic model of the scheduling system.

[0046] In this example, the coordinate data of the processing equipment in the workshop and the location data of the AGV charging stations are collected, and topological structure modeling is performed on the AGV driving channels. By collecting the physical layout information of the workshop, a complete workshop layout topological data model is constructed to help the system understand the spatial position relationship between various equipment and charging stations in the workshop. Measure the positions of all processing equipment, charging stations, and AGV driving channels in the workshop to form the coordinate information of each position point. For example, assume that the position coordinates of the processing equipment are , where is the equipment number. The position of the charging station is calibrated as , and the AGV driving channels form different topological structures according to the actual layout. Each channel is connected physically or logically to form a graphical structure. According to the workshop layout topological data, perform coordinate mapping on the AGV movement trajectory to obtain the workshop layout matrix. The workshop layout matrix is a discretized spatial representation. By dividing the area where the AGV may travel into grids and mapping each grid to a specific coordinate point, the layout of the entire workshop is represented. Assume that the workshop layout matrix is , where Indicates the row and the column grid status (such as whether it is an obstacle, a passage area, etc.). Through matrix modeling, the movement trajectory of the AGV is combined with the actual layout of the workshop to provide support for path planning. During the dynamic modeling of the AGV, the driving speed, turning radius, load capacity, and battery capacity of the AGV equipment are collected by sensors. At the same time, in order to fit these parameters, the least squares method is used for data analysis, and a set of optimal parameters is solved by minimizing the fitting error. Let the speed of the AGV be , the turning radius be , the load capacity be , and the battery capacity be , then these parameters are fitted through the following equations:

[0047] ;

[0048] where , , , are functions of the fitted speed, turning radius, load capacity, and battery capacity, is the corresponding sampling point, is the number of sampling points. In this way, an accurate AGV dynamic parameter model is obtained. The workpiece processing operation data, operation transfer time data, and task priority data are input into the task scheduler for linear programming modeling to determine the execution order of production tasks. The linear programming model is modeled based on multiple factors such as the operation transfer time of tasks, task priorities, and AGV operation limitations. Let the operation transfer time of the task be , and the task priority be , then the task scheduling is optimized through the following linear programming objective function:

[0049] ;

[0050] where is the priority of the th task, is the operation transfer time of the th task. The purpose of this objective function is to minimize the total time to complete tasks while giving priority to high-priority tasks. Based on the workshop layout matrix, the AGV dynamic parameter model, and the production task sequence, a kinematic model of the AGV is established. The AGV kinematic model predicts its future movement trajectory by calculating the speed and position vector of the AGV. Let the position vector of the AGV be , and the speed vector be , then the kinematic model of the AGV is described by the Euler equation:

[0051] ;

[0052] wherein, represents the change rate of the AGV position, represents the velocity vector of the AGV. By substituting the velocity vector and position vector of the AGV into the above equation, the motion constraint conditions of the AGV are obtained. These constraint conditions reflect the spatial and time limitations of the AGV when performing tasks, such as speed limit, acceleration limit, maximum turning radius, etc. When constructing the scheduling constraint model, the motion constraint conditions of the AGV and the time window constraint of the task are considered. The time window constraint of the task is represented by the following inequality:

[0053] ;

[0054] wherein, is the start time of task , is the completion time of task , is the end time of task . By modeling these constraints, it is ensured that the scheduling system not only meets the physical constraints of the AGV but also meets the timeliness requirements of the tasks. An energy consumption integration operation is performed on the speed data and battery capacity data in the dynamic parameter model of the AGV to obtain an energy consumption function. The energy consumption function is related to factors such as the speed, acceleration, driving distance, and load of the AGV. Let the energy consumption of the AGV be , then the energy consumption function is expressed as:

[0055] ;

[0056] wherein, , , γ are coefficients related to the specific model of the AGV, is the speed, is the acceleration, is the load, is the total time of task execution. By integrating the energy consumption, the total energy consumption of the AGV during task execution is obtained. The energy consumption function is integrated with the scheduling constraint model to obtain the basic model of the scheduling system.

[0057] In an example, based on the AGV historical operation data in the basic model of the scheduling system, recursive neural network training and physical constraint model integration are performed to obtain a hybrid scheduling prediction model, including:

[0058] Partition the AGV operation trajectory data, task execution time data, and energy consumption data in the basic model of the scheduling system, and perform time series segmentation to obtain a training sample set;

[0059] The training sample set is input into the batch normalization layer for data standardization processing, and data augmentation is performed on the trajectory data through random rotation and translation to obtain an augmented training data matrix;

[0060] The augmented training data matrix is batched, and temporal feature extraction is performed through long short-term memory network units to obtain the AGV state feature vector;

[0061] The AGV state feature vector is input into a recurrent neural network with 3 hidden layers, and parameter optimization is performed through the Adam optimizer to obtain the first neural network model;

[0062] Dropout processing is performed on the hidden layer of the first neural network model, and the early stopping mechanism is triggered when the validation loss does not decrease for N consecutive times to obtain the second neural network model;

[0063] Based on the basic model of the scheduling system, the AGV kinematic equation, time window constraint, and energy consumption function are constructed as Lagrangian constraint terms to obtain the physical constraint loss function;

[0064] The physical constraint loss function and the neural network prediction loss are weighted and combined to obtain a hybrid optimization objective function, and the hybrid optimization objective function is input into the second neural network model, and the hybrid scheduling prediction model is obtained through iterative calculation by the backpropagation algorithm.

[0065] In this example, the AGV running trajectory data, task execution time data, and energy consumption data in the basic model of the scheduling system are partitioned into a dataset and temporally sliced to obtain a training sample set. The AGV running trajectory data contains the coordinate information of each position passed by each AGV during task execution, the task execution time data contains the time required for each task from start to completion, and the energy consumption data includes the real-time energy consumption information of the AGV during task execution. Through temporal slicing, these continuous data are cut according to a fixed time step to obtain a set of training samples. Suppose the trajectory data of the AGV during task execution is , where is the position coordinate of the AGV at time , the task execution time data is , and the energy consumption data is . By slicing these data into multiple time windows, multiple training sample sets are obtained. The training sample set is input into the batch normalization layer for data standardization processing, so that the data ranges of different features tend to be consistent, avoiding some features having too much impact on model training due to large data magnitudes. The data is processed with zero mean and unit variance. Let the training sample set be , where is the th sample. After standardization, the standardized data of the sample obtained by the following formula:

[0066] ;

[0067] where, and are the mean and standard deviation of the training set respectively. To enhance the diversity and generalization ability of the data, data augmentation is performed by randomly rotating and translating the trajectory data. Suppose the trajectory data set is , through the rotation transformation, the trajectory data is rotated along a certain angle to obtain the augmented data , and its rotation angle is randomly generated. During the translation process, random translation operations are performed on the trajectory data to obtain a new trajectory data set . Through these data augmentation techniques, multiple variants of training samples are generated to increase the robustness of the model. The augmented training data matrix is input into a long short-term memory network (LSTM) unit for temporal feature extraction. LSTM is a variant of the recurrent neural network that can effectively capture long-term dependencies in time series data. In the LSTM network, the input data at each time step will be processed by a memory unit and output the hidden state of the next time step. The core of LSTM includes an input gate, a forget gate, and an output gate, which control the flow of information through these gates to effectively avoid the problems of gradient vanishing and explosion in traditional RNNs. Suppose the input data is , after being processed by LSTM, the state feature vector is obtained, and these feature vectors can effectively represent the state information of the AGV. The AGV state feature vectors are input into a recurrent neural network (RNN) with 3 hidden layers. RNN is a type of neural network used to process sequence data, and it can propagate information between multiple time steps by sharing weights to capture long-term dependencies in the sequence. In this model, the output of each hidden layer will be used as the input of the next layer, and finally the prediction result is obtained through an output layer. During the training process, the Adam optimizer is used to optimize the parameters of the network. The Adam optimizer dynamically adjusts the learning rate by calculating the mean and variance of the gradients, thereby accelerating the convergence process. The update formula of the Adam optimization algorithm is:

[0068] ;

[0069] ;

[0070] ;

[0071] ;

[0072] where, and are the first - order and second - order moment estimates respectively, and is the decay rate, is a small constant to prevent division - by - zero errors, is the learning rate. Through this process, the parameters of the neural network can be effectively optimized to obtain the first neural network model. After obtaining the first neural network model, in order to improve the generalization ability of the model, dropout processing is performed on the hidden layer of the network. Dropout is a technique to prevent overfitting. By randomly discarding the outputs of some neurons during the training process, the network uses different sub - networks for each update, improving the robustness of the model. During the training process, an early - stopping mechanism is set. When the loss function on the validation set does not decrease significantly in successive iterations, the training is stopped in advance to prevent overfitting. On this basis, based on the basic model of the scheduling system, the AGV kinematic equation, time - window constraint, and energy - consumption function are constructed as Lagrangian constraint terms to obtain the physical - constraint loss function. The Lagrangian method is an optimization method that combines the constraint conditions into the objective function and deals with the constraint problem by introducing Lagrangian multipliers. Suppose the objective function is and the constraint condition is , then the Lagrangian function is expressed as:

[0073] ;

[0074] where, is the Lagrangian multiplier, is the constraint condition. By minimizing the Lagrangian function, the optimal solution is obtained. The physical - constraint loss function and the neural - network prediction loss are weighted and combined to obtain the hybrid - optimization objective function. The hybrid - optimization objective function combines the physical constraints and the errors of neural - network predictions and can optimize the scheduling results on the basis of considering system constraints. The expression of the hybrid - optimization objective function is:

[0075] ;

[0076] where, is the neural - network prediction loss, is the physical - constraint loss, and are the weight coefficients, controlling the relative importance of the two. The hybrid - optimization objective function is input into the second neural - network model, and iterative calculations are performed through the back - propagation algorithm to obtain the hybrid - scheduling prediction model.

[0077] In one example, the AGV state feature vector and task requirement features output by the hybrid scheduling prediction model are obtained, and sub-task decomposition and path search processing are performed. Through constraint linearization transformation, a bi-objective optimization model is obtained, including:

[0078] Calculate the similarity between the AGV state feature vector and task requirement features output by the hybrid scheduling prediction model. A task similarity matrix is obtained through a weighted combination of the cosine distance and Euclidean distance. Task clustering is performed according to the similarity threshold to obtain a task decomposition sequence;

[0079] Map and match each sub-task in the task decomposition sequence with the AGV devices. A task assignment matrix of M×N dimensions is constructed through 0-1 integer programming, where M is the total number of tasks and N is the total number of AGVs, to obtain an initial assignment scheme;

[0080] Encode the execution task sequence of each AGV in the initial assignment scheme, and through single-point crossover and two-point mutation operations, obtain a task optimization sequence;

[0081] Construct an A* algorithm path planning space based on the workshop layout matrix. Set the distance between nodes to the length of one AGV body, and perform rasterization processing on the space through the grid method to obtain a path search network;

[0082] Set a heuristic function for the path search network, and use the weighted sum of the Manhattan distance and the dynamic obstacle avoidance factor as the path evaluation function to obtain a path cost matrix;

[0083] Perform a first-order Taylor expansion on the AGV motion constraint conditions, and transform the AGV kinematic equation, turning constraint, and speed constraint into a system of linear inequalities to obtain a linearized constraint equation;

[0084] Calculate the task completion time function based on the task optimization sequence, calculate the energy consumption function for the AGV dynamics parameter model, and obtain a standardized objective function through normalization processing. Combine the standardized objective function, the linearized constraint equation, and the path cost matrix to obtain a bi-objective optimization model.

[0085] In this example, calculate the similarity between the AGV state feature vector and task requirement features output by the hybrid scheduling prediction model. A task similarity matrix is obtained through a weighted combination of the cosine distance and Euclidean distance. The calculation of task similarity is based on the feature similarity between tasks, and these features include the execution time of tasks, the resource priority required for tasks, etc. Through the weighted combination of the cosine distance and Euclidean distance, the similarity between different tasks is measured. Let the task feature vectors be and , where is the dimension of the task features. The cosine distance and Euclidean distance are calculated through the following formulas respectively: Cosine distance:

[0086] ;

[0087] Among them, represents the inner product of the task and the task feature vectors, and are respectively the norms of the task and the task feature vectors. Euclidean distance:

[0088] ;

[0089] Among them, and are respectively the values of the task and the task on the th feature dimension. By weighted combination of the cosine distance and the Euclidean distance, the final task similarity is obtained:

[0090] ;

[0091] Among them, and are the weighting coefficients, . According to the set similarity threshold, the tasks are clustered, and the tasks with higher similarity are grouped into the same class to obtain the task decomposition sequence. The subtasks after decomposing each task are mapped and matched with the AGV devices, and a task assignment matrix is constructed through 0-1 integer programming. Suppose there are tasks and AGVs, and the matching of tasks and AGVs is represented as a matrix , where indicates that the task is assigned to the AGV , otherwise . This task assignment matrix is optimized through a 0-1 integer programming model, and the goal is to minimize the total task completion time or maximize the task execution efficiency. The specific objective function is expressed as:

[0092] ;

[0093] Among them, is the time for the task on the AGV The execution cost (such as task execution time) on it, and an initial task allocation plan is obtained by solving this optimization problem. Optimize the initial task allocation plan. Encode the task execution sequence of each AGV, and use the single-point crossover and two-point mutation operations in the genetic algorithm to search for the optimized task sequence. Assume that the task execution sequence of each AGV is represented as a permutation vector , where is the number of tasks that the AGV needs to execute. Through the single-point crossover operation, the task sequences of two parent individuals generate offspring individuals by randomly swapping some genes:

[0094] ;

[0095] ;

[0096] The two-point mutation operation generates new individuals by randomly swapping the positions of two tasks in the sequence. In the continuous iterative optimization process, the genetic algorithm can find the optimal combination of the task sequence through operations such as selection, crossover, and mutation. After the task allocation is optimized, path planning is carried out. According to the workshop layout matrix, construct the path planning space of the A* algorithm. The A* algorithm is a heuristic search algorithm that estimates the cost from the current node to the target node and selects the path with the minimum cost for search. Assume that the workshop layout is a two-dimensional grid, each grid cell represents a position in the workshop, and the path cost consists of two factors: the actual cost and the heuristic estimate. The actual cost is the path cost from the starting point to the current node, and the heuristic cost is the estimated cost from the current node to the target node.

[0097] ;

[0098] where is the actual cost from the starting point to the current node , is the current node to the heuristic estimated cost of the target node. The result of path planning is to find the shortest path from the starting point to the target node. In the path planning process, the workshop layout matrix is rasterized by the grid method, and the workshop space is divided into several small grids. Each grid represents a node, and the connection between nodes represents the feasible path of the AGV. To increase the feasibility of the path, a dynamic obstacle avoidance factor is set, and the influence of obstacles is incorporated into the path evaluation function. In the path evaluation process, the weighted sum of the Manhattan distance and the obstacle avoidance factor is used as the path cost:

[0099] ;

[0100] where is the Manhattan distance, obstacle is the influence of obstacles on the path, and is the weight coefficient. To meet the motion constraints of the AGV, linearization is performed on the AGV kinematic equation, turning constraint, and speed constraint. Assume the AGV kinematic equation is:

[0101] ;

[0102] where, is the speed of the AGV, is the distance traveled by the AGV, is the travel time. Through Taylor expansion, these non-linear constraints are transformed into linear inequality forms. For example, the turning radius constraint of the AGV is obtained through Taylor expansion as:

[0103] ;

[0104] where, is the turning angle, is the turning angular velocity, is the turning radius. Based on the task optimization sequence, the task completion time function is calculated, and the energy consumption function is calculated in combination with the AGV dynamics model. The energy consumption function is related to factors such as speed, travel distance, and load. Assume the energy consumption is proportional to the speed, expressed as:

[0105] ;

[0106] where, is the energy consumption, is the energy consumption coefficient, is the speed. After obtaining the standardized objective function through normalization, the task completion time function, linearized constraint equations, and path cost matrix are combined to finally obtain a bi-objective optimization model. This model aims to optimize the execution time and energy consumption of the task simultaneously.

[0107] In an example, the bi-objective optimization model is solved by particle swarm optimization. Through two-stage iterative calculations of the adaptive weight strategy and constraint conversion, a non-dominated solution set is obtained, including:

[0108] Perform the first-stage initialization of the bi-objective optimization model. Generate the initial solution of the particle swarm with M×N dimensions according to the total number of tasks M and the total number of AGVs N to obtain the initial solution space;

[0109] Input the initial solution space into the adaptive particle swarm algorithm, calculate the inertia weight value through the linear decreasing method, and obtain the adaptive weight parameter;

[0110] Update the position and velocity of the particle swarm based on the adaptive weight parameter, and use the tournament selection strategy to screen the global optimal solution to obtain the search solution set of the first stage;

[0111] Apply the ε-constraint method to the search solution set of the first stage, transform the energy consumption target into inequality constraint conditions, and obtain the constraint transformation model;

[0112] Based on the constraint transformation model, conduct population evolution through the binary crossover operator and the Gaussian mutation operator to obtain the initial population of the second stage;

[0113] Perform local search optimization on the initial population of the second stage, and use the variable neighborhood search strategy to fine-tune the particle positions to obtain the local optimal solution set;

[0114] Stratify the local optimal solution set according to the non-dominated sorting method, calculate the crowding distance and combine the elite retention strategy to screen the solution set, obtain the candidate non-dominated solution set, and determine the Pareto dominance relationship of the candidate non-dominated solution set, and select the solution with the comprehensive optimal task completion time and energy consumption to obtain the non-dominated solution set.

[0115] In this example, during the initialization process of the first stage, according to the total number of tasks and the total number of AGVs generate an initial solution space of a particle swarm with dimensions. These particles represent possible task allocation schemes, and each particle represents a task allocation matrix. Assume that the position of each particle is represented by where is a binary variable indicating whether task is assigned to AGV . That is:

[0116] ;

[0117] The position of each particle corresponds to a task allocation scheme, and the solution space in the entire particle swarm is all possible task allocation schemes. Input the initial solution space into the adaptive particle swarm algorithm, which is an optimization method that simulates the foraging behavior of bird flocks in nature. In this process, the particles adjust their positions based on their personal experience and global experience and approach the optimal solution. Calculate the inertia weight value by the linearly decreasing method to update the velocity and position of the particles. Assume that the velocity update formula of the particle swarm algorithm is:

[0118] ;

[0119] where is the velocity of particle in dimension , is the inertia weight, is the learning factor, is a randomly generated value between [0, 1], is the particle at the historical optimal position in dimension is the dimension of the global optimal solution and the value on it. By continuously updating the velocity and position of the particles, the particle swarm optimization algorithm can make the particles gradually approach the global optimal solution. In this process, the tournament selection strategy is adopted to screen out the global optimal solution in the particle swarm, and the search solution set of the first stage is obtained. The tournament selection strategy means randomly selecting several particles in the particle swarm and then choosing the optimal particle among them as the representative of the next generation. Through this strategy, the particle swarm can focus on exploring the optimal solution region and avoid falling into the local optimum. Apply the ε - constraint method to transform the energy consumption target into inequality constraint conditions. The ε - constraint method is an effective method for dealing with multi - objective optimization problems. It takes one objective function as the optimization target and the other objective functions as constraint conditions to ensure that other constraints are satisfied under the optimization target. For example, assume the original bi - objective optimization model is as follows:

[0120] ;

[0121] where, represents the task completion time, represents the energy consumption. By transforming into a constraint condition, a new optimization model is obtained:

[0122] ;

[0123] where, is the pre - set maximum energy consumption value. Through the constraint transformation method, it is ensured that the optimization process not only focuses on the task completion time but also takes into account the energy consumption. Based on the constraint transformation model, population evolution is carried out through the binary crossover operator and the Gaussian mutation operator. The binary crossover operator generates new solutions by exchanging part of the solutions of two particles, while the Gaussian mutation operator fine - tunes the solutions of the particles by adding Gaussian noise. The binary crossover operator is expressed as:

[0124] ;

[0125] The update method of the Gaussian mutation operator is:

[0126] ;

[0127] where, represents a normal distribution with a mean of 0 and a variance of Gaussian noise. Through these operations, the particle swarm can generate new solutions in the solution space. In the second stage, local search optimization is performed on the initial population. The local search optimization adopts a variable neighborhood search strategy, and improves the quality of the solution by searching for the optimal solution near the current solution. The variable neighborhood search gradually adjusts the neighborhood structure, enabling the algorithm to avoid local optima and find the global optimal solution. During each local search, the solution of the particle is slightly adjusted within a certain range. The specific constraints of task scheduling, such as the task completion time and energy consumption of each AGV, are taken into account during this process. Through local search optimization, the solution of the particle gradually approaches the optimal solution of the problem. After the local search optimization is completed, non-dominated sorting is performed on the particle swarm. The non-dominated sorting method is a method of stratifying the solution set, and determines the superiority and inferiority of the solutions by comparing the dominance relationships between the solutions. Let solution A dominate solution B, that is, if solution A is not inferior to solution B in all objectives and is superior to solution B in at least one objective, then solution A dominates solution B. The result of non-dominated sorting is to divide the solution set into multiple levels. The closer the solution is to the top of the ranking, the closer it is to the optimal solution. After the non-dominated sorting is completed, the crowding distance of each solution is calculated to measure the density of each solution in the solution set. Solutions with a smaller crowding distance tend to be in the marginal area of the solution space, representing the global optimal solution. By combining the crowding distance and the elite retention strategy, the most potential solutions are selected to obtain a candidate non-dominated solution set. The Pareto dominance relationship is determined for the candidate non-dominated solution set, and the solution with the best comprehensive balance of task completion time and energy consumption is selected as the final solution. These final solutions represent the best scheduling plan that can effectively balance the task completion time and energy consumption in a multi-AGV job shop. Through the particle swarm optimization algorithm process, a non-dominated solution set is obtained.

[0128] In one example, multi-dimensional evaluation calculations are performed on the non-dominated solution set to obtain an optimal scheduling execution plan, including:

[0129] Extract time indicators for each solution in the non-dominated solution set, including the total task completion time and the AGV waiting time, to obtain standardized time indicators;

[0130] Calculate efficiency indicators for the non-dominated solution set, substitute the AGV equipment utilization rate and the production task completion rate into the linear programming model to obtain standardized efficiency indicators;

[0131] Cumulatively calculate the operating energy consumption data of each AGV in the non-dominated solution set, and divide the total energy consumption by the number of tasks completed to obtain standardized energy consumption indicators;

[0132] Input the standardized time indicators, standardized efficiency indicators, and standardized energy consumption indicators into the analytic hierarchy network, establish a three-layer evaluation structure, and calculate the consistency of the judgment matrix in the three-layer evaluation structure to obtain an index weight matrix;

[0133] Based on the index weight matrix, the standardized time index, standardized efficiency index, and standardized energy consumption index are comprehensively calculated to obtain an evaluation score sequence, and multi-AGV simulation is performed based on the evaluation score sequence to select the optimal scheduling execution plan.

[0134] In this example, the time index is extracted for each solution in the non-dominated solution set, including the total task completion time and the AGV waiting time. The total task completion time is expressed as , where is the completion time of the th task. The total task completion time refers to the time from the start of the task to the completion of the last task. The AGV waiting time refers to the idle time spent by each AGV during the entire task execution process except when performing tasks. It is calculated by the following formula:

[0135] ;

[0136] where, refers to whether task is assigned to AGV , if assigned, it is 1, otherwise it is is the execution time of task , is the total time for all tasks to be completed. For each solution, the calculated total task completion time and AGV waiting time are standardized. The standardized time index is calculated by the following formula:

[0137] ;

[0138] where, is the time index of the current solution, and are the minimum and maximum time indices among all solutions respectively. Through standardization, the influence caused by scale differences between different solutions is eliminated, enabling the time index to be compared on a unified scale. The efficiency index of the non-dominated solution set is calculated. The efficiency index is measured by the utilization rate of AGV equipment and the production task completion rate. The utilization rate of AGV equipment refers to the ratio of the actual working time of the AGV during the entire operation process to the total available time, and is calculated by the following formula:

[0139] ;

[0140] where, is the actual working time of AGV during the task execution process, is the total time for tasks to be completed. The production task completion rate Refers to the ratio of the number of tasks completed within a specified time to the total number of tasks, expressed as:

[0141] ;

[0142] Among them, is an indicator variable indicating whether the task is completed. If completed, it is 1; otherwise, it is 0. Substitute the AGV equipment utilization rate and production task completion rate into the linear programming model to obtain the standardized efficiency index . The standardization of the efficiency index is achieved by comparing it with the minimum and maximum values among all solutions, similar to the standardization process of time indicators:

[0143] ;

[0144] Among them, is the efficiency index of the current solution, and are the minimum and maximum efficiency indices among all solutions, respectively. Cumulatively calculate the operating energy consumption data of each AGV in the non-dominated solution set, and divide the total energy consumption by the number of completed tasks to obtain the standardized energy consumption index. The energy consumption of an AGV is related to factors such as its working time, load, and movement trajectory. Assume that the energy consumption of each AGV is expressed as:

[0145] ;

[0146] Among them, is the energy consumption function of the AGV at time , is the speed of the AGV at time , is the load of the AGV at time , is the working time of the AGV . The total energy consumption is the sum of the energy consumption of all AGVs:

[0147] ;

[0148] Divide the total energy consumption by the number of completed tasks to obtain the standardized energy consumption index . Its calculation formula is:

[0149] ;

[0150] Similarly, through standardization, energy consumption indicators can be compared on the same scale. The standardized time indicator, standardized efficiency indicator, and standardized energy consumption indicator are input into the analytic hierarchy process network to establish a three-layer evaluation structure. The analytic hierarchy process network is a multi-criteria decision-making method, including an objective layer, a criterion layer, and a scheme layer. The objective layer is the optimal scheduling execution plan. The criterion layer includes three indicators: time, efficiency, and energy consumption. The scheme layer includes all possible scheduling plans. The key step of the analytic hierarchy process network is to establish a judgment matrix and calculate its consistency. Assuming that we want to compare the relative importance among time, efficiency, and energy consumption, a judgment matrix is constructed, where each element represents the importance of criterion relative to criterion . According to the principle of the analytic hierarchy process network, the judgment matrix should meet the consistency requirement. The consistency ratio is used to verify whether the consistency is satisfied. The formula is:

[0151] ;

[0152] where is the consistency index, and is the random consistency index. If is less than 0.1, it is considered that the judgment matrix has good consistency. Otherwise, the judgment matrix needs to be adjusted. Based on the judgment matrix, the weight matrix of each criterion is calculated, where are the weights of time, efficiency, and energy consumption respectively. The weight matrix is obtained by the eigenvalue method or directly solving the linear equations. Based on the index weight matrix, comprehensive calculations are performed on the standardized time indicator, standardized efficiency indicator, and standardized energy consumption indicator. The comprehensive calculation formula is:

[0153] ;

[0154] where is the comprehensive score of each scheduling plan, are the standardized time, efficiency, and energy consumption indicators respectively, and are their weights. According to the comprehensive score sequence , all scheduling plans are sorted, and the actual performance of different scheduling plans is verified through multi-AGV simulation. Simulate the operation of different scheduling plans in the actual workshop environment, and evaluate the task completion situation, AGV operation efficiency, and energy consumption performance. Through the simulation results, the scheduling plan with the highest comprehensive score is selected as the optimal scheduling execution plan.

[0155] In an example, the optimal scheduling execution plan is input into the AGV control system to perform model feedback update on the hybrid scheduling prediction model, and the target scheduling prediction model is obtained, including:

[0156] Perform time series decomposition on the optimal scheduling execution plan, convert the task sequence into speed control instructions through the AGV kinematic equation, and obtain the AGV real-time control sequence;

[0157] Sample the position coordinate data and task completion status data in the AGV real-time control sequence, perform data filtering through a third-order Kalman filter with a noise coefficient of 0.1, and obtain the AGV operating state matrix;

[0158] Input the AGV operating state matrix into the recurrent neural network model, calculate the AGV motion trajectory through the long short-term memory unit, obtain the state prediction sequence, perform collision risk detection on the state prediction sequence, calculate the minimum safety distance between AGVs based on the workshop layout matrix, set the safety threshold to 2 times the vehicle body length, and obtain the obstacle avoidance constraint conditions;

[0159] Re-plan the AGV local path according to the obstacle avoidance constraint conditions, calculate the optimal path cost through the Bellman equation, and iteratively update the path points to obtain the optimized path sequence;

[0160] Combine the AGV operating state matrix and the optimized path sequence according to the time sequence, and perform statistical calculations on the task completion rate and energy consumption rate to obtain the real-time execution data set;

[0161] Perform online update on the hybrid scheduling prediction model, input the real-time execution data set into the hybrid scheduling prediction model in batches of 32 samples for parameter update, obtain the updated model parameters, and perform weighted integration on the updated model parameters and the physical constraint loss function to obtain the target scheduling prediction model.

[0162] In this example, perform time series decomposition on the optimal scheduling execution plan and convert the task sequence into the speed control instructions of the AGV. Convert the start and end times of the task, task priority and other information into the speed and position instructions that the AGV should execute. Suppose there is a task sequence , each task includes the start time , end time , task location , etc. In order to convert the task sequence into speed control instructions, combine the AGV kinematic equation, that is:

[0163] ;

[0164] Among them, is the speed when the AGV executes task , and are the positions of tasks and respectively, and are the start and end times of the task. In this way, the task sequence is transformed into a series of speed control instructions to generate the real-time control sequence of the AGV. Data sampling is performed on the real-time control sequence of the AGV, and a third-order Kalman filter is used to filter the data. The purpose of the Kalman filter is to remove noise and improve the accuracy of the system state. The general form of the Kalman filter is expressed as:

[0165] ;

[0166] where is the filtered state estimate, is the current measurement value, is the Kalman gain, is the measurement matrix. Through this process, the operating state matrix of the AGV is obtained, which contains the position information, speed information, and other important control parameters of the AGV. The noise coefficient in the third-order Kalman filter is set to 0.1 to ensure good noise suppression effect during the filtering process. The filtered AGV operating state matrix is input into the recurrent neural network model, and the long short-term memory unit (LSTM) is used to calculate the motion trajectory of the AGV. The LSTM network is a model for processing time series data, which can remember long-term dependence information and is suitable for predicting the motion trajectory of the AGV. Assume that the state matrix of the AGV is where represents the AGV state data (such as position, speed, etc.) at time , and the input of the LSTM network is , and its output is the state prediction at the next time. In this way, the LSTM network can predict the future motion trajectory of the AGV. Collision risk detection is performed on the state prediction sequence. The minimum safety distance between AGVs is calculated based on the workshop layout matrix. Assume that the body length of each AGV is , and the minimum safety distance between each pair of AGVs in the workshop must satisfy:

[0167] ;

[0168] If the actual distance between AGVs is less than the minimum safety distance, there is a collision risk. Collision risk detection is achieved by calculating the distance between AGVs and comparing it with the set safety distance to ensure that AGVs can avoid collisions during operation. Under the obstacle avoidance constraint conditions, the path planning of the AGV needs to be readjusted to avoid collisions while ensuring the optimality of the path. The Bellman equation is used for path cost calculation. The basic form of the Bellman equation is:

[0169] ;

[0170] where is the optimal cost in the state under, is the control input, is the current state and the control input under the cost, is the discount factor, is the state transition function. On this basis, the AGV iteratively updates the path cost through the Bellman equation to obtain an optimized path sequence. After the path planning is completed, the running state matrix of the AGV is combined with the optimized path sequence to form a time series combination. By statistically calculating the task completion rate and the energy consumption rate, a real-time execution dataset is obtained. The task completion rate is expressed as:

[0171] ;

[0172] where is the task whether the completion indicator variable, is the total number of tasks. And the energy consumption rate is:

[0173] ;

[0174] where is the energy consumption of the AGV of, is the total time for task completion. Based on the real-time execution data, the hybrid scheduling prediction model is updated online. Assuming that 32 samples are used as a batch for training, the real-time execution dataset is input into the hybrid scheduling prediction model for parameter update. The updated model parameters are calculated through the backpropagation algorithm. In this process, the loss function of the model is obtained by weighted combination of the prediction error and the physical constraint loss function. Assuming that the physical constraint loss function includes physical limitations such as path constraints and speed constraints, while the prediction loss function is the prediction error of the model, and the weighted combination loss function of the two is:

[0175] ;

[0176] where and is the weight coefficient. By adjusting these two coefficients, the impacts of physical constraints and prediction errors are balanced. The parameters of the target scheduling prediction model are optimized through the backpropagation algorithm to obtain an updated hybrid scheduling prediction model. This model can dynamically adjust the scheduling strategy according to real-time data, ensuring that in a complex job environment, the AGV can execute tasks efficiently and stably.

[0177] Referring to Figure 2 , this embodiment provides a multi-objective optimization scheduling system for a multi-AGV flexible job shop, including:

[0178] An acquisition module 1, configured to acquire job shop layout data, AGV operation parameters, and production task data, and construct a basic model of the scheduling system;

[0179] A training module 2, configured to perform recursive neural network training and physical constraint model integration based on the historical operation data of AGVs in the basic model of the scheduling system to obtain a hybrid scheduling prediction model;

[0180] A processing module 3, configured to obtain the AGV status feature vector and task requirement features output by the hybrid scheduling prediction model, and perform subtask decomposition and path search processing, and obtain a bi-objective optimization model through constraint linearization transformation;

[0181] A solving module 4, configured to perform particle swarm optimization on the bi-objective optimization model, and obtain a non-dominated solution set through two-stage iterative calculations of an adaptive weight strategy and constraint transformation;

[0182] A calculation module 5, configured to perform multi-dimensional evaluation calculations on the non-dominated solution set to obtain an optimal scheduling execution plan;

[0183] An update module 6, configured to input the optimal scheduling execution plan into the AGV control system, and perform model feedback update on the hybrid scheduling prediction model to obtain a target scheduling prediction model.

[0184] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to that described in the above method embodiment, and details are not described herein again.

[0185] Referring to Figure 3 , this embodiment of the present invention also provides a computer device, which may be a server, and its internal structure may be as Figure 3As shown in the figure. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements the above method.

[0186] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0187] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0188] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to the memory, storage, database, or other media provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0189] It should be noted that, in this text, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, device, article or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, device, article or method including such element.

[0190] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. A multi-objective optimization scheduling method for a multi-AGV flexible job shop, characterized in that Including the following steps: Collect the layout data of the operation workshop, the running parameters of AGVs, and the production task data, and construct the basic model of the scheduling system; Based on the historical running data of AGVs in the basic model of the scheduling system, perform recursive neural network training and physical constraint model integration to obtain a hybrid scheduling prediction model; specifically including: divide the AGV running trajectory data, task execution time data, and energy consumption data in the basic model of the scheduling system into data sets, and perform time series segmentation to obtain a training sample set; Input the training sample set into a batch normalization layer for data standardization processing, and perform data augmentation on the trajectory data through random rotation and translation to obtain an augmented training data matrix; batch the augmented training data matrix, extract time series features through long short-term memory network units to obtain AGV state feature vectors; input the AGV state feature vectors into a recursive neural network with 3 hidden layers, and perform parameter optimization through the Adam optimizer to obtain a first neural network model; perform dropout processing on the hidden layers of the first neural network model, and trigger an early stopping mechanism when the validation loss does not decrease continuously for N times to obtain a second neural network model; based on the basic model of the scheduling system, construct the AGV kinematic equation, time window constraint, and energy consumption function as Lagrangian constraint terms to obtain a physical constraint loss function; perform weighted combination on the physical constraint loss function and the neural network prediction loss to obtain a hybrid optimization objective function, and input the hybrid optimization objective function into the second neural network model, and perform iterative calculation through the backpropagation algorithm to obtain a hybrid scheduling prediction model; Obtain the AGV state feature vectors and task demand features output by the hybrid scheduling prediction model, perform sub-task decomposition and path search processing, and obtain a bi-objective optimization model through constraint linearization transformation; Perform particle swarm optimization on the bi-objective optimization model to solve, and obtain a non-dominated solution set through two-stage iterative calculation of the adaptive weight strategy and constraint transformation; Perform multi-dimensional evaluation calculation on the non-dominated solution set to obtain an optimal scheduling execution plan; Input the optimal scheduling execution plan into the AGV control system, and perform model feedback update on the hybrid scheduling prediction model to obtain a target scheduling prediction model.

2. The multi-objective optimal scheduling method for a multi-AGV flexible job shop according to claim 1, wherein The collecting the layout data of the operation workshop, the running parameters of AGVs, and the production task data, and constructing the basic model of the scheduling system includes: Collect the coordinate data of the processing equipment and the location data of the AGV charging stations in the operation workshop, and perform topological structure modeling on the AGV driving channels to obtain the workshop layout topological data; Perform coordinate mapping on the AGV movement trajectories according to the workshop layout topological data to obtain a workshop layout matrix; Collect the driving speed, turning radius, load capacity, and battery capacity of AGV equipment through sensors and perform least squares fitting to obtain an AGV dynamics parameter model, and input the workpiece processing process data, process transfer time data, and task priority data into a task scheduler for linear programming modeling to obtain a production task sequence; An AGV kinematic model is established based on the workshop layout matrix, the AGV dynamic parameter model, and the production task sequence. The AGV velocity vector and position vector are substituted into the Euler equation to obtain the AGV motion constraint conditions; Constraint modeling is performed on the AGV motion constraint conditions and the task time window to obtain a scheduling constraint model. Energy consumption integration operations are performed on the velocity data and battery capacity data in the AGV dynamic parameter model to obtain an energy consumption function. The energy consumption function is integrated with the scheduling constraint model to obtain a basic model of the scheduling system.

3. The multi-objective optimization scheduling method for a multi-AGV flexible job shop according to claim 1, wherein, The AGV state feature vector and task demand features output by the hybrid scheduling prediction model are obtained, and sub-task decomposition and path search processing are performed. Through constraint linearization transformation, a bi-objective optimization model is obtained, including: The similarity between the AGV state feature vector and task demand features output by the hybrid scheduling prediction model is calculated. A task similarity matrix is obtained through a weighted combination of the cosine distance and the Euclidean distance. Task clustering is performed according to the similarity threshold to obtain a task decomposition sequence; Each sub-task in the task decomposition sequence is mapped and matched with the AGV device. A task assignment matrix of M×N dimensions is constructed through 0-1 integer programming, where M is the total number of tasks and N is the total number of AGVs, to obtain an initial assignment plan; The execution task sequence of each AGV in the initial assignment plan is encoded, and through single-point crossover and two-point mutation operations, a task optimization sequence is obtained; An A* algorithm path planning space is constructed according to the workshop layout matrix. The distance between nodes is set to the length of one AGV body, and the space is rasterized by the grid method to obtain a path search network; A heuristic function is set for the path search network, and the weighted sum of the Manhattan distance and the dynamic obstacle avoidance factor is used as the path evaluation function to obtain a path cost matrix; The first-order Taylor expansion of the AGV motion constraint conditions is performed, and the AGV kinematic equation, turning constraint, and speed constraint are transformed into a system of linear inequalities to obtain a linearized constraint equation; Based on the task optimization sequence, a task completion time function is calculated. The energy consumption function is calculated for the AGV dynamic parameter model, and a normalized objective function is obtained through normalization processing. The normalized objective function, the linearized constraint equation, and the path cost matrix are combined to obtain a bi-objective optimization model.

4. The multi-objective optimal scheduling method for a multi-AGV flexible job shop according to claim 3, wherein The bi-objective optimization model is solved by particle swarm optimization. Through two-stage iterative calculations of the adaptive weight strategy and constraint transformation, a non-dominated solution set is obtained, including: The first-stage initialization of the bi-objective optimization model is performed. An initial solution of a particle swarm of M×N dimensions is generated according to the total number of tasks M and the total number of AGVs N to obtain an initial solution space; The initial solution space is input into the adaptive particle swarm algorithm, and the inertia weight value is calculated by the linear decreasing method to obtain an adaptive weight parameter; Based on the adaptive weight parameter, the position and velocity of the particle swarm are updated, and the global optimal solution is selected by the tournament selection strategy to obtain a first-stage search solution set; Apply the ε-constraint method to the first-stage search solution set, transform the energy consumption target into inequality constraint conditions, and obtain a constraint transformation model; Based on the constraint transformation model, perform population evolution through a binary crossover operator and a Gaussian mutation operator to obtain an initial population for the second stage; Perform local search optimization on the initial population of the second stage, and use a variable neighborhood search strategy to fine-tune the particle positions to obtain a locally optimal solution set; Stratify the locally optimal solution set according to the non-dominated sorting method, calculate the crowding distance, and combine the elite retention strategy to screen the solution set to obtain a candidate non-dominated solution set. Then, determine the Pareto dominance relationship for the candidate non-dominated solution set, and select the solution with the best combination of task completion time and energy consumption to obtain a non-dominated solution set.

5. The multi-objective optimization scheduling method for a multi-AGV flexible job shop according to claim 4, wherein The multi-dimensional evaluation calculation for the non-dominated solution set to obtain an optimal scheduling execution plan includes: Extract time indicators for each solution in the non-dominated solution set, including the total task completion time and the AGV waiting time, to obtain standardized time indicators; Calculate efficiency indicators for the non-dominated solution set. Substitute the AGV equipment utilization rate and the production task completion rate into a linear programming model to obtain standardized efficiency indicators; Cumulatively calculate the operating energy consumption data of each AGV in the non-dominated solution set, and divide the total energy consumption by the number of tasks completed to obtain standardized energy consumption indicators; Input the standardized time indicators, the standardized efficiency indicators, and the standardized energy consumption indicators into an analytic hierarchy network, establish a three-layer evaluation structure, and calculate the consistency of the judgment matrix in the three-layer evaluation structure to obtain an index weight matrix; Based on the index weight matrix, comprehensively calculate the standardized time indicators, the standardized efficiency indicators, and the standardized energy consumption indicators to obtain an evaluation score sequence, and perform multi-AGV simulation based on the evaluation score sequence to select an optimal scheduling execution plan.

6. The multi-objective optimal scheduling method for a multi-AGV flexible job shop according to claim 5, characterized in that Input the optimal scheduling execution plan into the AGV control system, and perform model feedback update on the hybrid scheduling prediction model to obtain a target scheduling prediction model, including: Perform time series decomposition on the optimal scheduling execution plan, and convert the task sequence into a speed control instruction through the AGV kinematic equation to obtain an AGV real-time control sequence; Sample the position coordinate data and task completion status data in the AGV real-time control sequence, and perform data filtering through a third-order Kalman filter with a noise coefficient of 0.1 to obtain an AGV operating state matrix; Input the AGV operating state matrix into a recurrent neural network model, calculate the AGV motion trajectory through long short-term memory units to obtain a state prediction sequence, and perform collision risk detection on the state prediction sequence. Calculate the minimum safety distance between AGVs based on the workshop layout matrix, and set the safety threshold to 2 times the vehicle body length to obtain obstacle avoidance constraint conditions; Re-plan the local path of the AGV according to the obstacle avoidance constraint conditions, calculate the optimal path cost through the Bellman equation, and iteratively update the path points to obtain an optimized path sequence; Combine the AGV operation status matrix and the optimized path sequence according to the time series, and perform statistical calculations on the task completion rate and energy consumption rate to obtain a real-time execution data set; Perform online update on the hybrid scheduling prediction model. Input the real-time execution data set into the hybrid scheduling prediction model in batches of 32 samples for parameter update to obtain updated model parameters, and perform weighted integration on the updated model parameters and the physical constraint loss function to obtain the target scheduling prediction model.

7. A multi-objective optimal scheduling system for a multi-AGV flexible job shop, characterized in that, Steps for implementing the multi-objective optimization scheduling method of the multi-AGV flexible job shop according to any one of claims 1 to 6. The multi-AGV flexible job shop multi-objective optimization scheduling system includes: An acquisition module, configured to acquire job shop layout data, AGV operation parameters, and production task data, and construct a basic model of the scheduling system; A training module, configured to perform recursive neural network training and physical constraint model integration based on the AGV historical operation data in the basic model of the scheduling system to obtain a hybrid scheduling prediction model; A processing module, configured to obtain the AGV status feature vector and task requirement features output by the hybrid scheduling prediction model, and perform subtask decomposition and path search processing, and obtain a bi-objective optimization model through constraint linearization transformation; A solution module, configured to perform particle swarm optimization on the bi-objective optimization model, and obtain a non-dominated solution set through two-stage iterative calculation of an adaptive weight strategy and constraint transformation; A calculation module, configured to perform multi-dimensional evaluation calculation on the non-dominated solution set to obtain an optimal scheduling execution plan; An update module, configured to input the optimal scheduling execution plan into the AGV control system, and perform model feedback update on the hybrid scheduling prediction model to obtain the target scheduling prediction model.

8. A computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that, When the processor executes the computer program, it implements the steps of the multi-objective optimization scheduling method of the multi-AGV flexible job shop according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-objective optimization scheduling method of the multi-AGV flexible job shop according to any one of claims 1 to 6.

Images