AGV path optimization method and system based on digital twin

Abstract

The present application relates to the field of digital twin technology, and discloses an AGV path optimization method and system based on digital twin, the method comprising: performing data standardization processing on AGV real-time coordinate data, environmental obstacle distribution data and cargo storage location data, and constructing a storage environment state matrix; calculating AGV task time parameters, and obtaining a task evaluation matrix including task timeliness index and task priority coefficient; performing multi-objective parameter calculation on AGV motion energy consumption coefficient and task weight coefficient, and establishing an AGV path optimization objective function; performing path planning analysis using path exchange operator, path insertion operator and path reversal operator, and generating an AGV initial path set; performing variable parameter particle swarm annealing optimization and continuous convex approximation dual iterative calculation, and obtaining an AGV optimal path planning set, the present application improves the flexibility and optimization effect of path planning, and significantly improves the convergence speed and solution quality of path optimization.

Description

Technical Field

[0001] The present application relates to the field of digital twin technology, and in particular to an AGV path optimization method and system based on digital twin. Background Art

[0002] With the rapid development of intelligent warehousing, the application of multi-AGV systems in dynamic warehousing environments is becoming more and more widespread. Multi-AGV systems have the advantages of high efficiency and low cost, but in actual operation, they face challenges such as complex path planning and difficult task coordination. The introduction of digital twin technology provides a new idea for solving these problems. By creating a digital representation of the physical warehouse system, the system status can be reflected in real time, and the monitoring and control of the multi-AGV system can be realized.

[0003] However, in a dynamic warehousing environment, there are still many difficulties in the path optimization problem of multi-AGV systems. Communication and coordination problems between AGVs may lead to task delays and increased energy consumption; secondly, traditional path planning methods are difficult to take into account both task timeliness and energy efficiency; the obstacle distribution and task requirements in a dynamic environment continue to change, which puts higher requirements on the real-time and robustness of path planning. Summary of the invention

[0004] The present application provides an AGV path optimization method and system based on digital twins, which improves the flexibility and optimization effect of path planning, and significantly improves the convergence speed and solution quality of path optimization.

[0005] The first aspect of the present application provides an AGV path optimization method based on digital twins, and the AGV path optimization method based on digital twins includes:

[0006] Standardize the AGV real-time coordinate data, environmental obstacle distribution data, and cargo storage location data to build a warehouse environment status matrix;

[0007] Based on the warehouse environment state matrix, the AGV task time parameters are calculated to obtain a task evaluation matrix including task timeliness indicators and task priority coefficients;

[0008] According to the task evaluation matrix, multi-objective parameter calculation is performed on the AGV motion energy consumption coefficient and the task weight coefficient, and an AGV path optimization objective function is established;

[0009] Based on the AGV path optimization objective function, a path exchange operator, a path insertion operator and a path reversal operator are used to perform path planning analysis to generate an AGV initial path set;

[0010] The AGV initial path set is subjected to variable parameter particle swarm annealing optimization and continuous convex approximation double iterative calculation to obtain the AGV optimal path planning set.

[0011] The second aspect of the present application provides an AGV path optimization system based on digital twins, and the AGV path optimization system based on digital twins includes:

[0012] The standardization module is used to standardize the AGV real-time coordinate data, environmental obstacle distribution data, and cargo storage location data to build a storage environment status matrix;

[0013] A calculation module, used to calculate the AGV task time parameters based on the warehouse environment state matrix to obtain a task evaluation matrix including task timeliness indicators and task priority coefficients;

[0014] A processing module, used to perform multi-objective parameter calculation on the AGV motion energy consumption coefficient and the task weight coefficient according to the task evaluation matrix, and establish an AGV path optimization objective function;

[0015] A planning module, used to perform path planning analysis based on the AGV path optimization objective function using a path exchange operator, a path insertion operator, and a path reversal operator to generate an AGV initial path set;

[0016] The iterative module is used to perform variable parameter particle swarm annealing optimization and continuous convex approximation double iterative calculation on the AGV initial path set to obtain the AGV optimal path planning set.

[0017] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned digital twin-based AGV path optimization method.

[0018] The fourth aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned digital twin-based AGV path optimization method.

[0019] Compared with the prior art, the present application has the following beneficial effects: through multi-level data standardization processing and state information merging, the efficient integration of AGV real-time coordinates, environmental obstacles and cargo storage location data is realized, and the data quality and availability are improved. The dual evaluation mechanism of task timeliness index and priority coefficient is adopted, combined with the comprehensive calculation of task complexity, load status and urgency, so that the task allocation is more reasonable. Through the precise modeling of motion energy consumption and waiting energy consumption, combined with multi-objective parameter optimization, the precise control and optimization of energy consumption are achieved. The three operators of path exchange, insertion and inversion are used, and the dynamic update mechanism of operator selection weight is combined to improve the flexibility and optimization effect of path planning. Combined with the dual iterative calculation of variable parameter particle swarm annealing optimization and continuous convex approximation, the convergence speed and solution quality of path optimization are significantly improved. Through real-time deviation analysis and conflict detection, combined with the dynamic path correction mechanism within the time window, the stable operation of the system in a dynamic environment is guaranteed. Through the design of conflict handling priority vector and conflict avoidance strategy, the coordination problem of multi-AGV system is effectively solved. A hierarchical optimization strategy is adopted to decompose the complex path planning problem into multiple sub-problems, which reduces the computational complexity and improves the algorithm efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0022] Figure 1 It is a flowchart of an AGV path optimization method based on digital twins provided in an embodiment of the present invention;

[0023] Figure 2 It is a schematic block diagram of the structure of an AGV path optimization system based on digital twins provided in an embodiment of the present invention;

[0024] Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0027] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0028] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 , an embodiment of the AGV path optimization method based on digital twin in the embodiment of the present application includes:

[0029] Step 100: Perform data standardization processing on the AGV real-time coordinate data, environmental obstacle distribution data and cargo storage location data to construct a storage environment status matrix;

[0030] It is understandable that the execution subject of the present application can be an AGV path optimization system based on digital twins, or a terminal or a server, which is not limited here. The embodiment of the present application is described by taking the server as the execution subject as an example.

[0031] Specifically, for the real-time coordinate data of AGV collected by multiple sensors, the data is processed by the zero mean standardization method to remove the offset and scale effects of the data and ensure the centralization and unitization of the AGV coordinate data. The AGV standardized position matrix is ​​generated by calculating the difference between each coordinate point and the overall mean and dividing it by the standard deviation. The outliers in the AGV standardized position matrix are median filtered, and the isolated noise points or outliers are effectively removed by replacing the value of each data point with the median of all values ​​in its neighborhood window to obtain the AGV outlier filter matrix. At the same time, for the environmental obstacle distribution data, binary processing is performed, and each grid point in the storage area is classified according to whether there is an obstacle, and 1 is used to represent the obstacle position and 0 is used to represent the idle position to generate an obstacle distribution binary matrix. The cargo storage location data is coded and mapped, and different cargo types are mapped to corresponding digital numbers according to the type, priority or other relevant characteristics of the cargo to generate a cargo location coding matrix. For example, some high-priority cargo is mapped to a lower number, while ordinary cargo is mapped to a higher number. This mapping method can intuitively reflect the cargo distribution and its priority information in the environmental matrix. The AGV outlier filter matrix, obstacle distribution binary matrix, and cargo location encoding matrix are concatenated to obtain a multi-dimensional initial environment state tensor. During the tensor concatenation process, the corresponding positions of each matrix are aligned and the data is expanded in the new dimension to ensure that the information in each matrix remains independent and interrelated. The initial environment state tensor is reduced in dimension. The dimension reduction process uses methods such as principal component analysis or singular value decomposition to extract the characteristic dimensions with the most concentrated information in the tensor, map the original high-dimensional data to a low-dimensional space, and generate an environment state reduction matrix. The environment state reduction matrix is ​​normalized to ensure that the data values ​​of all grid points fall within the same range (such as [0,1]) to avoid calculation errors caused by data scale differences. After normalization, the data of each grid point represents a unified expression of its relative state in the storage environment. The state information of each grid point in the normalized matrix is ​​merged. By performing weighted summation or logical combination of various information such as AGV position, obstacle status, and cargo coding at a single grid point, a warehouse environment status matrix is ​​generated to reflect the status of each location in the warehouse area, including the current distribution of AGVs, the specific location of obstacles, and the distribution and type information of cargo.

[0032] Step 200: Calculate the AGV task time parameters based on the warehouse environment state matrix to obtain a task evaluation matrix including task timeliness index and task priority coefficient;

[0033] Specifically, based on the warehouse environment state matrix, the time parameters of the i-th task performed by the k-th AGV are calculated. The environmental information related to the task is extracted from the warehouse environment state matrix, such as the starting point, target point, and obstacle distribution on the path. Using this information, combined with the current position state of the AGV, the start time and completion time of the task are calculated. The start time of the task depends on the path time between the actual position of the AGV after completing the previous task and the starting point of the current task, and the completion time is based on the path time of the AGV from the starting point to the target point, as well as the loading and unloading operation time of the goods. The timeliness index of the task is calculated by performing a difference operation on the start time and the completion time of the task. The task timeliness index is a key parameter to measure the urgency of the task completion time. It is expressed in the form of a relative value, reflecting the gap between the actual completion time of the task and the expected completion time. The task complexity parameter is extracted from the warehouse environment state matrix. The task complexity is decomposed into three aspects: path length, number of turns, and number of obstacle avoidances. The path length is directly calculated from the shortest path from the starting point of the task to the target point of the AGV. The path search algorithm (such as A* or Dijkstra) is combined with the obstacle distribution data in the warehouse environment state matrix for calculation. The number of turns refers to the number of times the AGV changes direction on the path. This indicator is obtained by analyzing the angle change between the path nodes. The more turns, the higher the task complexity. The number of obstacle avoidances is the number of trajectory adjustments made by the AGV in path planning to avoid obstacles, which is recorded by the path search algorithm during the obstacle avoidance optimization process. Combining the above three indicators into a task complexity vector can fully reflect the complexity of the task. At the same time, the current load state of the kth AGV is evaluated to calculate its execution ability for the i-th task. The load state of the AGV is calculated by the ratio of the current weight of the goods it carries to its maximum carrying capacity to form the AGV load state vector. In the warehouse environment, the urgency of the task varies depending on the priority of the goods or changes in external demand, so the urgency of the i-th task of the k-th AGV is calculated separately. The task urgency coefficient is a parameter that comprehensively reflects the importance and time sensitivity of the task. It is estimated by combining the priority weight of the cargo type, the deadline of the task, and the relative ratio between the task start time and the completion time. According to the task complexity vector, the AGV load state vector and the task urgency coefficient, the priority weight coefficient is weighted and calculated to obtain the task priority coefficient. The setting of each weight is adjusted according to the priority requirements in the actual application scenario. The task timeliness index and the task priority coefficient are combined in a matrix to generate the initial task evaluation matrix. The initial task evaluation matrix is ​​a two-dimensional matrix, in which the rows correspond to each AGV and the columns correspond to the tasks that the AGV needs to perform. Each element in the matrix is ​​composed of the task timeliness index and the task priority coefficient, which can comprehensively reflect the completion time and importance of the task.After the initial task evaluation matrix is ​​generated, in order to ensure that high-priority tasks are ranked first in the scheduling, the matrix is ​​dynamically prioritized. The sorting process is based on the comprehensive evaluation results of each task. By dynamically adjusting the sorting order of tasks, tasks with higher priorities are assigned to the AGV task queue in advance to generate the final task evaluation matrix.

[0034] Step 300: Perform multi-objective parameter calculation on the AGV motion energy consumption coefficient and the task weight coefficient according to the task evaluation matrix, and establish the AGV path optimization objective function;

[0035] It should be noted that the motion state data of AGV is extracted based on the task evaluation matrix. A motion state vector containing AGV speed, acceleration and load weight is obtained, which reflects the dynamic behavior of AGV when performing specific tasks. In the motion state vector, speed represents the motion level of AGV on the path, acceleration reflects the dynamic changes of AGV between different path nodes, and load weight directly affects the basic value of motion energy consumption. These parameters are updated in real time through digital twin technology to ensure that they are consistent with the actual operating state. The energy consumption coefficient is multiplied by the AGV speed, acceleration and load weight in the motion state vector to obtain the basic energy consumption coefficient matrix of AGV. Based on the execution time of the kth AGV and the i-th task in the basic energy consumption coefficient matrix, the integral calculation of motion energy consumption is performed. The basic energy consumption value at each moment is accumulated to obtain the total motion energy consumption matrix of AGV during the task completion process, which reflects the energy consumption accumulation of each AGV in all task execution stages, and the main source of energy consumption is identified by analyzing the matrix. The AGV task execution sequence in the task evaluation matrix is ​​analyzed to extract the waiting time series of AGV. Waiting time refers to the period of time that the AGV stays because it needs to coordinate other tasks or avoid path conflicts. Based on the waiting time series, the product of the unit time energy consumption coefficient is performed to generate the AGV waiting energy consumption matrix. The waiting energy consumption matrix reflects the energy consumption of each AGV due to non-motion reasons. The total motion energy consumption matrix and the waiting energy consumption matrix are accumulated to obtain the total energy consumption of the AGV system. On the basis of obtaining the total energy consumption of the system, the weight coefficient in the optimization objective function is configured and calculated in combination with the task timeliness index in the task evaluation matrix. The calculation of the weight coefficient vector needs to balance the importance of task timeliness and system energy consumption. For example, for tasks with high timeliness requirements, the weight coefficient should be inclined to the task timeliness index, while for energy-sensitive scenarios, the weight coefficient should consider the impact of total energy consumption more. After completing the calculation of the weight coefficient vector, the task timeliness index and the total energy consumption of the AGV system are weighted and combined to construct the path optimization objective function. The objective function is expressed as a linear combination of the weight coefficient, the task timeliness index and the total energy consumption, and the specific form is: ,in is the task timeliness indicator, is the total energy consumption of the AGV system, and is the weight coefficient. By adjusting these two coefficients, the preference of the objective function is optimized according to actual needs, for example, focusing on timeliness during peak hours and focusing more on energy consumption optimization during non-peak hours. Generate the path optimization objective function.

[0036] Step 400: Based on the AGV path optimization objective function, a path exchange operator, a path insertion operator and a path reversal operator are used to perform path planning analysis to generate an AGV initial path set;

[0037] Specifically, based on the AGV path optimization objective function, the path point sequence of the kth AGV is extracted. The path point sequence consists of a series of key nodes that the AGV needs to pass through from the starting position to the target position. These nodes contain the geometric position of the path and the dynamic constraints determined by the environmental characteristics. After the path point sequence is extracted, the turning angle and obstacle avoidance distance between each pair of adjacent path points are calculated. The turning angle is used to measure the degree of change in the direction of the AGV on the path, and the obstacle avoidance distance reflects the distance margin of the AGV when passing through obstacles. This information is integrated into a path feature vector to form a key data basis for describing the path performance. The path exchange operator is operated on the turning angle in the path feature vector. The path exchange operator reduces energy consumption and improves path smoothness by optimizing the turning angle in the path. For path segments with turning angle values ​​greater than the preset threshold in the path feature vector, local optimization is achieved by exchanging the positions of these segments. This operation can effectively reduce sharp turns in the path, thereby improving the smoothness and coherence of the path. After optimization by the path exchange operator, the generated angle-optimized path is more geometrically smooth and more suitable for AGV operation in actual environments. Based on the angle-optimized path, the path insertion operator operation of the obstacle avoidance distance is performed. The obstacle avoidance distance is one of the parameters that need to be focused on in path planning. When the obstacle avoidance distance is less than a certain threshold, the AGV faces the risk of collision, so it is necessary to optimize it by inserting new path nodes. At the location where the obstacle avoidance distance is greater than the preset threshold, additional path nodes are inserted to bypass obstacles and generate an obstacle avoidance optimized path. The obstacle avoidance optimized path is operated by the path reversal operator, and the path direction is locally adjusted by the path reversal operator to optimize the path length. The path reversal operator reduces unnecessary detours by reversing the direction of local segments in the path, thereby reducing the total length of the path. For parts of the path with large redundancy or loops, the path reversal operator can rearrange the node order to generate a more compact length-optimized path. The length-optimized path is not only more reasonable in space, but also can significantly reduce the total driving distance and time of the AGV in actual operation. After completing the above optimization steps, the path quality score is calculated based on the length-optimized path. The path quality score is a quantitative evaluation of the comprehensive performance of the path, including indicators such as path smoothness, obstacle avoidance efficiency, and length optimization degree. At the same time, in order to improve the efficiency of operator operations, the operation frequency is calculated in combination with historical path data to obtain the selection weights of path exchange, path insertion and path reversal operators. These weights reflect the actual effect of each operator in different path optimization scenarios. In order to continuously improve the effect of operator combination, the operator selection weights are iteratively updated. In each iteration, the performance improvement of the operator combination is judged by comparing the difference in the objective function values ​​after two adjacent optimizations. The operator weights are adjusted according to the optimization effect, and the optimal operator combination sequence is dynamically updated. This process can effectively avoid falling into local optimality and improve the global performance of path planning.The AGV path is iterated multiple times based on the optimal operator combination sequence. After each iteration, the generated paths are sorted according to the path quality score, and the N paths with the highest scores are selected as the initial path set of the AGV.

[0038] Step 500: Perform variable parameter particle swarm annealing optimization and continuous convex approximation dual iterative calculation on the AGV initial path set to obtain the AGV optimal path planning set.

[0039] Specifically, each path in the AGV initial path set is encoded with rectangular coordinates, and the path point sequence is converted into a position vector expressed in rectangular coordinates. This conversion extracts the spatial coordinates of the path points and arranges them in a matrix form, namely the path encoding matrix. Each row in the path encoding matrix represents the complete coordinate sequence of a path. Based on the path encoding matrix, the particle swarm is initialized for the path of the kth AGV, and the position of each particle is represented as the coordinate sequence of the AGV at each time, generating a particle swarm position matrix. Each row of the particle swarm position matrix represents the initial position of a particle (i.e., the path solution), and the column vector represents the spatial coordinates of the path point. After the particle swarm is initialized, the dynamic behavior of the particle swarm is configured, and the movement mode of the particles in the search space is defined by setting parameters such as inertia weight, cognitive factor, and social factor. The inertia weight controls the tendency of the particle to maintain the current direction of movement, the cognitive factor reflects the ability of the particle to approach its own historical optimal solution, and the social factor describes the interaction between the particle and the global optimal solution of the group. Combining these parameters, a speed update parameter matrix is ​​constructed by generating random numbers to dynamically adjust the speed change of the particles in the search space. The velocity update parameter matrix provides randomness and flexibility for the moving direction and step size of each particle, thereby ensuring that the particle swarm can conduct extensive global search in the early stage and gradually converge to the optimal solution in the later stage. The particle swarm velocity is iteratively updated according to the velocity update parameter matrix, and the temperature parameter is introduced for annealing. The annealing process dynamically reduces the temperature parameter to gradually reduce the degree of disturbance of the solution, so that the particle swarm transitions from the global search stage to the local convergence stage. This variable parameter particle swarm annealing optimization (VPPSA) combines the global search capability of the particle swarm algorithm with the local optimization capability of the simulated annealing algorithm, so that it can effectively avoid falling into the local optimum when optimizing the path, and the generated VPPSA optimized path sequence has more global superiority. The VPPSA optimized path sequence is input into the continuous convex approximation (SCA) algorithm to solve the non-convex constraint problem in the path. Since the path planning problem contains complex nonlinear constraints (such as obstacle avoidance constraints, path smoothness constraints, etc.), it is difficult to solve directly. By performing a first-order Taylor expansion on the non-convex path constraints, it is locally linearized to generate a local linearized constraint matrix. Based on this matrix, the resource allocation of the AGV path is solved by convex optimization, and the original path planning problem is decomposed into a series of convex subproblems that are easy to solve. In the path optimization subproblem set, each subproblem represents a local optimization path. By iteratively solving these subproblems, the global optimal solution is gradually approached. In each iteration, the linearization point is updated to adapt to the current state of the path planning, and the convergence condition is checked. If the convergence condition is met, the iteration is stopped and the local optimal path sequence is output; otherwise, the constraint matrix is ​​continued to be updated and the next round of iteration is performed. Through multiple iterations, the overall performance of the path is gradually optimized, while ensuring that it meets the solution accuracy requirements of the continuous convex approximation algorithm.The local optimal path sequence is verified for AGV motion constraints, obstacle avoidance constraints, and path smoothness constraints to ensure that the generated path is feasible in practical applications. For example, AGV motion constraints need to ensure that the acceleration and speed in the path meet the equipment performance; obstacle avoidance constraints require that path points do not collide with obstacles; and path smoothness constraints optimize the continuity and smoothness of the path. Only paths that meet all constraints can be selected into the final path set. After double iterative calculations and multiple constraint verifications, the path set that meets all conditions is selected as the AGV optimal path planning set.

[0040] The Euclidean distance between the optimal path planning set of AGV and the real-time path of AGV in actual operation is calculated. By comparing the deviation of each position point with the preset threshold, the deviation degree of the actual position of AGV and the expected path is quantified, and a path deviation marking matrix is ​​generated, in which each matrix element represents the deviation state of the corresponding position point. If the deviation of a point exceeds the preset threshold, it is marked as a position point that needs to be adjusted. Based on the path deviation marking matrix, the path segment from the current moment to the time window is extracted, and the path outside the time window remains unchanged, ensuring that the optimization process only focuses on the path in the near period without affecting the stability of the path in the long period, thereby reducing unnecessary computational burden. The extracted path segment is defined as a sequence of path segments to be optimized, and each path segment contains key points that need to be adjusted. In order to analyze the dynamic characteristics of the path, the sequence of path segments to be optimized is mapped in the time dimension, and the coordinate point sequence of AGV at different time points is generated to form a spatiotemporal trajectory matrix. The spatiotemporal trajectory matrix describes the spatiotemporal characteristics of the path segment by combining the spatial position and the time series. The minimum distance between any two paths is calculated based on the spatiotemporal trajectory matrix to detect the conflict between the paths. By analyzing the minimum distance between two paths in the spatial dimension, when the distance is less than the safety threshold, it is marked as a conflict point and recorded in the conflict detection result set. These conflict points reflect potential AGV path conflict problems and need to be further processed to ensure operational safety. For all conflict points in the conflict detection result set, the relevant AGVs are sorted based on the task priority coefficients contained in the task evaluation matrix to assign a priority level to each conflicting AGV. The priority level is assigned according to comprehensive factors such as the urgency, importance and time sensitivity of the task to form a conflict handling priority vector. According to the conflict handling priority vector, the low-priority AGV path is adjusted. The adjustment process needs to consider two strategies: one is to calculate the waiting time of the low-priority AGV to ensure that the high-priority AGV can pass first; the other is to plan a detour path for the low-priority AGV to avoid the conflict area. These strategies are integrated into a conflict avoidance strategy set, which contains all possible path adjustment schemes. Based on the conflict avoidance strategy set, local replanning is performed to generate an optimized correction path set. The local replanning process ensures that the path adjustment meets the safety requirements without significantly increasing the path length or task delay by re-optimizing the affected path segments. After the generation of the optimized and corrected path set is completed, it is replaced with the path segment within the corresponding time window in the AGV optimal path planning set, while the path segment outside the time window remains unchanged. After multi-level processing of path deviation detection, conflict handling, path adjustment and local replanning, the dynamically updated AGV optimal path planning set achieves efficient path optimization and real-time adjustment. The new path set achieves a balance in energy consumption, time and safety, and can adapt to complex and changing storage environments.

[0041] In the embodiment of the present application, through multi-level data standardization processing and state information merging, the efficient integration of AGV real-time coordinates, environmental obstacles and cargo storage location data is achieved, and the data quality and availability are improved. The dual evaluation mechanism of task timeliness index and priority coefficient is adopted, combined with the comprehensive calculation of task complexity, load status and urgency, so that the task allocation is more reasonable. Through the accurate modeling of motion energy consumption and waiting energy consumption, combined with multi-objective parameter optimization, the precise control and optimization of energy consumption are achieved. The three operators of path exchange, insertion and inversion are used, and the dynamic update mechanism of operator selection weight is combined to improve the flexibility and optimization effect of path planning. Combined with the dual iterative calculation of variable parameter particle swarm annealing optimization and continuous convex approximation, the convergence speed and solution quality of path optimization are significantly improved. Through real-time deviation analysis and conflict detection, combined with the dynamic path correction mechanism within the time window, the stable operation of the system in a dynamic environment is guaranteed. Through the design of conflict handling priority vector and conflict avoidance strategy, the coordination problem of multi-AGV system is effectively solved. The hierarchical optimization strategy is adopted to decompose the complex path planning problem into multiple sub-problems, which reduces the computational complexity and improves the algorithm efficiency.

[0042] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0043] The real-time coordinate data of the AGV collected by multiple sensors is normalized to zero mean to obtain the AGV standardized position matrix, and the outliers in the AGV standardized position matrix are median filtered to obtain the AGV outlier filter matrix;

[0044] Binarize the environmental obstacle distribution data, mark the obstacle position as 1, and mark the idle position as 0, to obtain the obstacle distribution binary matrix, and encode and map the cargo storage location data, map different cargo types to corresponding numbers, to obtain the cargo location coding matrix;

[0045] The AGV outlier filter matrix, the obstacle distribution binary matrix, and the cargo location encoding matrix are concatenated to obtain the initial environment state tensor, and the initial environment state tensor is subjected to dimensionality reduction processing to obtain the environment state dimensionality reduction matrix;

[0046] The environmental state dimension reduction matrix is ​​normalized to obtain the environmental state normalized matrix, and the state information of each grid point in the environmental state normalized matrix is ​​merged to obtain the storage environment state matrix.

[0047] Specifically, for the real-time coordinate data of AGV, assume that The data collected by sensors include The AGV position coordinates at a certain moment are recorded as , where each element Indicates Moment To eliminate the measurement deviation between different sensors, Perform zero mean standardization. The zero mean standardization formula is:

[0048] ;

[0049] in, is the standardized value, It is is its standard deviation. The matrix obtained after standardization is This is the AGV standardized position matrix. Perform median filtering. Median filtering is a neighborhood-based nonlinear smoothing method that replaces each data point with the median of its neighborhood data. Assume that the filter window is ,for , filtering results It is expressed as:

[0050] ;

[0051] in . The matrix after median filtering This is the AGV outlier filter matrix, which effectively removes isolated outliers while retaining the overall trend of the data. When processing environmental obstacle distribution data, it is assumed that the obstacle distribution matrix Indicates that and are the number of rows and columns of the storage environment, and each element Indicates the occupancy status of the grid. Perform binarization, mark the obstacle position as 1, and the idle position as 0. The formula is:

[0052] ;

[0053] in is the threshold for obstacle detection, and the result matrix is the obstacle distribution binary matrix. For the cargo storage location data, assume that the cargo distribution matrix Indicates that Representation Grid Cargo type. Cargo storage locations are coded and mapped, and different cargo types are mapped to unique numbers to obtain a cargo location coding matrix. The mapping rule uses a function type Z means, for example:

[0054] ;

[0055] in is the original cargo type, is the corresponding number value. AGV outlier filter matrix , obstacle distribution binary matrix and cargo location coding matrix Perform tensor concatenation. Assume Map to The spatial scale of the initial environment state tensor is obtained after splicing , where each dimension corresponds to the AGV position, obstacle state, and cargo distribution. In order to reduce computational complexity and storage pressure, the initial environment state tensor Perform dimensionality reduction. The dimensionality reduction method uses principal component analysis, which calculates the covariance matrix of the tensor and extracts the principal component with the largest eigenvalue to map the tensor from three dimensions to two dimensions. The specific dimensionality reduction formula is:

[0056] ;

[0057] in is the dimension reduction matrix, is the environmental state matrix after dimension reduction. The matrix after dimension reduction Normalization is performed to ensure that data of different feature dimensions are in the same numerical range. The normalization formula is:

[0058] ;

[0059] The normalized matrix is the normalized matrix of the environmental state. The state information of each grid point in is merged, and the features such as AGV position, obstacle status and cargo distribution are weighted and integrated. For example, the weights are , the combined formula is:

[0060] ;

[0061] Results Matrix That is the final storage environment status matrix, which comprehensively reflects the environmental status information of each grid point.

[0062] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0063] Based on the warehouse environment state matrix, the time parameter of the i-th task of the k-th AGV is calculated to obtain the task start time and task completion time, and the difference operation is performed based on the task start time and task completion time to obtain the task timeliness index;

[0064] Extract the task complexity parameters of the kth AGV in the warehouse environment state matrix to obtain a task complexity vector including the task path length, number of turns, and number of obstacle avoidances;

[0065] Based on the warehouse environment state matrix, the current load state of the k-th AGV is calculated to obtain the AGV load state vector, and the urgency of the i-th task of the k-th AGV is calculated to obtain the task urgency coefficient;

[0066] According to the task complexity vector, the AGV load state vector and the task urgency coefficient, a weighted calculation of the priority weight coefficient is performed to obtain the task priority coefficient;

[0067] The task timeliness index and the task priority coefficient are combined into a matrix to obtain the initial task evaluation matrix, and the initial task evaluation matrix is ​​dynamically prioritized to ensure that high-priority tasks are prioritized first to obtain the task evaluation matrix.

[0068] Specifically, based on the warehouse environment status matrix , for The first AGV The time parameters of each task are calculated. Assume that the starting coordinates of the task are , the target point coordinates are , the average speed of AGV is . Calculate from the path planning algorithm (such as Dijkstra or A*) arrive The shortest path length , the path length formula is:

[0069] ;

[0070] in is the number of waypoints, Indicates the coordinates of the path point. Task start time Defined as the timestamp after the AGV completes the current task, task completion time is calculated as:

[0071] ;

[0072] in It is the time to load and unload cargo, which depends on the type and weight of cargo. The task completion time and the maximum task time allowed by the system The ratio of is calculated as:

[0073] ;

[0074] if A value close to 1 indicates that the task time urgency is low, and vice versa. Extract task complexity parameters and analyze the The AGV is executing The path length, number of turns, and number of obstacles to avoid during each mission. Already obtained in the above time parameter calculation. Number of turns is the number of direction changes in the path, calculated as the angular change of the path point sequence:

[0075] ;

[0076] in Is the first The direction angle of the segment, is the indicator function, It is the angle threshold for judging turning. Each point on the path is calculated by marking the obstacle distribution in the warehouse environment state matrix. If satisfied , it is regarded as an obstacle avoidance operation, and the calculation formula is:

[0077] ;

[0078] Task Complexity Vector It is expressed as:

[0079] ;

[0080] Based on the warehouse environment status matrix, evaluate the The current load state of each AGV is used to obtain the load state vector . Assume that the current load of the AGV is , the maximum load is , load ratio is calculated as:

[0081] ;

[0082] At the same time, the AGV's task urgency coefficient Combine the task timeliness index and cargo priority. Assume that the cargo priority is , the urgency coefficient is:

[0083] ;

[0084] in and is a weight coefficient used to balance the impact of time and cargo priority. Combined with the task complexity vector , load state vector and the urgency factor , perform weighted calculation of the priority weight coefficient to obtain the task priority coefficient :

[0085] ;

[0086] in It is to adjust the weight, and the specific value is set according to the system requirements. and the task priority coefficient Combined into the initial task evaluation matrix , where each row corresponds to a task and the columns represent the timeliness index and priority coefficient respectively:

[0087] ;

[0088] Initial Task Evaluation Matrix Perform dynamic priority sorting, rearrange the tasks from high to low according to their priority coefficients, ensure that high-priority tasks are ranked first, and generate the final task evaluation matrix .

[0089] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0090] The data of AGV motion state is extracted according to the task evaluation matrix to obtain the motion state vector including AGV speed, acceleration and load weight;

[0091] The energy consumption coefficient of the AGV speed, acceleration and load weight in the motion state vector is multiplied to obtain the AGV basic energy consumption coefficient matrix;

[0092] Based on the i-th task execution time of the k-th AGV in the AGV basic energy consumption coefficient matrix, the motion energy consumption is integrated and calculated to obtain the AGV motion total energy consumption matrix;

[0093] The AGV waiting time is extracted according to the AGV task execution order in the task evaluation matrix to obtain the AGV waiting time series, and the AGV waiting time series is multiplied by the unit time energy consumption coefficient to obtain the AGV waiting energy consumption matrix;

[0094] The total energy consumption matrix of AGV movement and the total energy consumption matrix of AGV waiting are accumulated to obtain the total energy consumption of the AGV system, and the weight coefficient configuration calculation of the objective function is performed based on the task timeliness index in the task evaluation matrix and the total energy consumption of the AGV system to obtain the weight coefficient vector;

[0095] Based on the weight coefficient vector, the task timeliness index and the total energy consumption of the AGV system are weighted combined to obtain the AGV path optimization objective function.

[0096] Specifically, from the task evaluation matrix Extract the data information related to the AGV motion status. Assume No. The row corresponds to The task information of each AGV includes the task timeliness index and priority coefficient. The first AGV A task, whose motion state is determined by the speed , acceleration and current load weight Description, forming a motion state vector:

[0097] ;

[0098] in, It is The AGV is executing The average speed of the task, is the mean acceleration of the main path during task execution, is the current load weight of the AGV. The above parameters are extracted through sensor data collection or simulation data in the digital twin environment. The energy consumption coefficient is multiplied by the speed, acceleration and load weight in the motion state vector to calculate the basic energy consumption coefficient matrix of the AGV. Assume that the energy consumption coefficients are , , , respectively represent the weights of speed, acceleration and load on energy consumption. Then the basic energy consumption coefficient matrix The calculation formula for each element of is:

[0099] ;

[0100] This matrix can reflect the energy consumption level per unit time of each AGV when performing a specific task. Based on the basic energy consumption coefficient matrix , combined with the execution time of the task , calculate the total energy consumption matrix of AGV movement Assume that the execution time interval of the task is , the exercise energy consumption is calculated by integrating the basic energy consumption over time, and the formula is:

[0101] ;

[0102] In actual calculations, the integral is discretized into a cumulative form, for example:

[0103] ;

[0104] in is the time step, is the total number of discrete time points. At the same time, according to the task execution order in the task evaluation matrix, the waiting time sequence of AGV is extracted. Assume that The waiting time of an AGV between task executions is , then these waiting times constitute a waiting time sequence For each element in the waiting time series, combined with the unit time energy consumption coefficient , calculate the waiting energy consumption matrix of AGV :

[0105] ;

[0106] Total energy consumption matrix of exercise and waiting energy consumption matrix The total energy consumption of the AGV system is obtained by adding

[0107] ;

[0108] The total energy consumption matrix reflects the total energy consumption required by the AGV in completing all tasks. Based on the total energy consumption, combined with the timeliness index in the task evaluation matrix , calculate the weight coefficient of the objective function. Assume that the weight of the timeliness index is , the weight of total energy consumption is , then the weight coefficient vector w is:

[0109] ;

[0110] The configuration of the weight coefficient is adjusted according to the specific scenario, for example, in the scenario where efficiency is prioritized, value, and in energy-saving priority scenarios, increase By comparing the task timeliness index and total energy consumption The weighted combination operation of is used to construct the AGV path optimization objective function. The objective function is expressed as:

[0111] ;

[0112] in is the total number of AGVs, The number of tasks for each AGV.

[0113] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0114] Based on the AGV path optimization objective function, the path point sequence of the kth AGV is extracted, and the turning angle and obstacle avoidance distance between two adjacent points in the path point sequence are calculated to obtain the path feature vector;

[0115] A path exchange operator is performed on the turning angle in the path feature vector, and local optimization is performed by exchanging two path segments whose angle values ​​are greater than a preset threshold to obtain an angle-optimized path.

[0116] Based on the angle-optimized path, a path insertion operator is performed on the obstacle avoidance distance, and a new path node is inserted at a position where the obstacle avoidance distance is greater than a preset threshold to optimize the obstacle avoidance and obtain an optimized obstacle avoidance path.

[0117] Perform path reversal operator operation on the obstacle avoidance optimization path, reduce the path length by reversing the local path direction, and obtain the length optimized path;

[0118] The path quality score is calculated based on the length-optimized path, the operation frequency is calculated based on the historical path data, the operator selection weight is obtained, and the operator selection weight is iteratively updated. The parameter optimization is performed by calculating the difference between the objective function values ​​of two adjacent iterations to obtain the optimal operator combination sequence.

[0119] The AGV path is iteratively calculated based on the optimal operator combination sequence, and the N best-ranked paths are selected as the AGV initial path set.

[0120] Specifically, extract the first The path point sequence of an AGV . Assume that the path point sequence contains The coordinates of each point are , then the path point sequence is expressed as:

[0121] ;

[0122] Based on the path point sequence, calculate the turning angle and obstacle avoidance distance between two adjacent points. Turning Angle Indicates and The angle between the segment path directions is calculated as:

[0123] ;

[0124] in and They are and Euclidean distance of the path segment. Obstacle avoidance distance Indicates The shortest distance from a path point to an obstacle is calculated as:

[0125] ;

[0126] in is a set of obstacle coordinates. By calculating and , and get the path feature vector:

[0127] ;

[0128] The turning angle in the path feature vector Perform path exchange operator operation to optimize the smoothness of the path. The path exchange operator selects the path with a turning angle greater than the preset threshold. The path segments are exchanged, namely:

[0129] ;

[0130] By swapping the positions of these segments, the number of sharp turns is reduced, resulting in an angle-optimized path. For example, if and The path is optimized based on the angle, and the obstacle avoidance distance in the path feature vector is calculated. Perform path insertion operator operation to optimize the safety of the path. Insert a new path node at the position , whose coordinates are obtained by interpolation calculation:

[0131] ;

[0132] Generate obstacle avoidance optimized path after insertion , ensuring that the path can effectively avoid obstacles. Perform a path reversal operator to reduce the path length. The path reversal operator optimizes the overall length of the path by reversing the direction of the local path. There is a significant detour and the following conditions are met:

[0133] ;

[0134] in is the redundant length coefficient, yes The midpoint of the path is reversed to produce a length-optimized path. After completing the path optimization, calculate the path quality score , taking into account path length, smoothness and obstacle avoidance performance:

[0135] ;

[0136] in is the total path length, is the smoothness score (weighted sum of the inverse of the turning angles), is the obstacle avoidance score (the minimum obstacle avoidance distance). Indicates the relative importance of length, smoothness, and obstacle avoidance. Combined with historical path data, the frequency of use of path operators is counted to calculate the operator selection weight :

[0137] ;

[0138] in is the number of times the operator is used. The operator weight is iteratively updated according to the difference in the objective function value between two adjacent iterations. Make adjustments:

[0139] ;

[0140] in is the learning rate, and the updated weights ensure the optimality of the operator combination. The AGV path is iterated multiple times based on the optimal operator combination sequence, and the path with the highest quality score is selected. The paths are used as the initial path set of AGV.

[0141] The path exchange operator operation is performed on the turning angle in the path feature vector, and local optimization is performed by exchanging two path segments whose angle values ​​are greater than a preset threshold to obtain an angle optimized path, including: performing vector product calculation on three adjacent path points in the path feature vector to obtain a path turning angle sequence {θ1, θ2, ..., θn}; comparing each turning angle value with a preset threshold θmax based on the path turning angle sequence to obtain a large-angle turning point set P = {pi | θi>θmax}; performing spatial distance calculation on the turning points in the large-angle turning point set P, and sorting them in ascending order according to the distance value to obtain a set of point pairs to be exchanged S = {(pi, pj)}; verify the interchangeability of the path segments between each pair of points pi and pj in the set of exchanged point pairs S, and obtain the valid exchange point pair set V through collision detection and path connectivity calculation; perform exchange operation S(p1, p2) on the path segments based on the valid exchange point pair set V, and obtain the exchanged path set R by reconnecting the start and end points of the path segments; calculate the local optimization degree according to the turning angle sequence of each path in the exchanged path set R, and obtain the path optimization score matrix M; screen the optimal path for the path optimization score matrix M, and obtain the optimal exchange plan based on the weighted calculation of the turning angle reduction and the path length increment; reorganize the original path according to the optimal exchange plan, and obtain the angle optimized path by updating the path point sequence and the turning angle sequence.

[0142] Among them, a path insertion operator operation is performed on the obstacle avoidance distance based on the angle-optimized path, and obstacle avoidance optimization is performed by inserting a new path node at a position where the obstacle avoidance distance is greater than a preset threshold, so as to obtain an obstacle avoidance optimization path, including: performing Euclidean distance calculation on the obstacle avoidance distance between adjacent path points based on the angle-optimized path to obtain an obstacle avoidance distance sequence matrix; performing threshold comparison processing on the values ​​in the obstacle avoidance distance sequence matrix, selecting the path segment greater than the preset threshold, and obtaining a set of positions of points to be inserted; performing obstacle position mapping on the set of positions of points to be inserted based on the warehouse environment state matrix to obtain a set of obstacle distribution points; and performing a comparison on the obstacle distribution point set. Perform the shortest distance analysis, and obtain the obstacle avoidance safety distance matrix by calculating the shortest distance from the path point to the obstacle; divide the safety area around the obstacle into grids based on the obstacle avoidance safety distance matrix to obtain a set of candidate insertion points; score and calculate the obstacle avoidance effect of each insertion point according to the set of candidate insertion points, and obtain the insertion point scoring vector by comprehensively considering the path length increment and the obstacle avoidance effect; select the optimal value of the insertion point scoring vector, and select the point with the highest score as the path insertion point to obtain the path node insertion plan; use the path node insertion plan to perform node update processing on the angle optimized path to obtain the obstacle avoidance optimized path.

[0143] The obstacle avoidance optimization path is subjected to a path reversal operator operation, and the path length is reduced by reversing the local path direction to obtain a length optimized path, including: performing a cumulative summation operation on the distances between adjacent path points in the obstacle avoidance optimization path to obtain total path length data; performing segmentation on the total path length data, and dividing it in units of a fixed length L to obtain a plurality of local path sub-segments; performing path direction calculation on the local path sub-segments, and obtaining a sub-segment direction matrix by calculating the direction vector between the head and tail points of the sub-segments; calculating the direction angles of adjacent sub-segments based on the sub-segment direction matrix, and comparing them with a preset threshold to obtain a set of sub-segments to be reversed; performing path reversal on each sub-segment in the set of sub-segments to be reversed, and obtaining a local reversed path by adjusting the access order of the path points; recalculating the path length of the local reversed path, and obtaining a path improvement amount ΔL by comparing the path lengths before and after the reversal; making a choice and judgment on the reversal operation based on the path improvement amount ΔL, retaining the reversal results with a positive improvement amount, and obtaining a path optimization result set; splicing and reorganizing all local optimized paths in the path optimization result set, and reconstructing them according to the connection order of the original paths to obtain a length optimized path.

[0144] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0145] Cartesian coordinate encoding is performed on each path in the AGV initial path set, and the path point sequence is converted into a position vector to obtain a path encoding matrix;

[0146] Based on the path encoding matrix, the particle swarm is initialized for the path of the kth AGV, and the position of each particle is represented as the coordinate sequence of the AGV at each moment to obtain the particle swarm position matrix. The particle swarm position matrix is ​​configured with parameters of inertia weight, cognitive factor and social factor, and random numbers are generated to obtain the speed update parameter matrix.

[0147] According to the speed update parameter matrix, the particle swarm speed is iteratively updated, and annealing is performed according to the temperature parameter to obtain the VPPSA optimization path sequence;

[0148] The VPPSA optimized path sequence is input into the SCA algorithm. By performing a first-order Taylor expansion on the non-convex path constraints, a local linearized constraint matrix is ​​obtained. Based on the local linearized constraint matrix, a convex optimization solution is performed on the AGV path resource allocation. The path planning problem is converted into a sequence of convex subproblems to obtain a set of path optimization subproblems.

[0149] Iteratively solve the path optimization sub-problem set, update the linearization point and check the convergence condition in each iteration to obtain the local optimal path sequence;

[0150] The AGV motion constraints, obstacle avoidance constraints and path smoothness constraints are verified for the local optimal path sequence, and the path set that meets all constraints is selected as the AGV optimal path planning set.

[0151] Specifically, each path in the AGV initial path set is encoded with rectangular coordinates. Assume that the initial path set contains paths, each consisting of discrete path points, and the two-dimensional coordinates of each point are After encoding all paths, the path encoding matrix is ​​generated , where each row represents the vector of the flattened coordinates of a path point sequence. Path point sequence The formula for conversion to path vector is:

[0152] ;

[0153] All paths After stacking, it becomes the path encoding matrix . Based on the path encoding matrix , for The position of each particle is represented as a coordinate sequence of path points. The size of the particle group is . At initialization, the position matrix of each particle is Random perturbation generation, the first The position of a particle is expressed as:

[0154] ;

[0155] in is a random perturbation vector used to initialize the diversity of particles. In order to control the motion behavior of the particle swarm, the particle swarm position matrix Configuring Inertia Weight , cognitive factors and social factors At the same time, a random number matrix is ​​generated for the particles , used to adjust the random contribution of cognitive and social factors. Particle velocity matrix The update formula is:

[0156] ;

[0157] in It is a particle The best historical position, is the global optimal position. After the velocity is updated, the particle position matrix is ​​updated as:

[0158] ;

[0159] In order to improve the convergence performance of particle swarm optimization, annealing is added. Annealing dynamically adjusts the temperature parameters , gradually reducing the degree of disturbance of the solution. The temperature update formula is:

[0160] ;

[0161] in is the annealing coefficient. After each speed update, a poor solution is randomly accepted according to the current temperature to enhance the global exploration capability and generate a VPPSA optimized path sequence. The VPPSA optimized path sequence is input into the continuous convex approximation (SCA) algorithm to process the non-convex constraints in the path. Non-convex constraints such as path smoothness or obstacle avoidance performance are locally linearized through the first-order Taylor expansion to generate a local linearized constraint matrix . Assume that the non-convex objective function of path planning is , its first-order Taylor expansion is:

[0162] ;

[0163] Through the constraint matrix The non-convex problem is transformed into a convex optimization problem. Combined with the requirements of path resource allocation, the objective function is set as ,in is the reference path. The optimization goal is converted to:

[0164] ;

[0165] The path optimization sub-problem set is generated by solving the convex optimization algorithm. The sub-problem set is solved iteratively, and the linearization point is updated each iteration. And check the convergence conditions. If it satisfies , then the iteration stops and the local optimal path sequence is output. The local optimal path sequence is verified by motion constraints, obstacle avoidance constraints and path smoothness constraints. Motion constraints ensure that the speed and acceleration in the path meet the physical limitations of the AGV; obstacle avoidance constraints ensure that the distance between the path point and the obstacle is not less than the safety threshold; path smoothness constraints optimize the continuity and smoothness of the path. The paths that meet all constraints constitute the final AGV optimal path planning set.

[0166] The VPPSA optimized path sequence is input into the SCA algorithm, and the local linearized constraint matrix is ​​obtained by performing first-order Taylor expansion on the non-convex path constraints. Based on the local linearized constraint matrix, the AGV path resource allocation is solved by convex optimization, and the path planning problem is converted into a convex sub-problem sequence to obtain a path optimization sub-problem set, including: performing first-order Taylor expansion on the non-convex constraints in the VPPSA optimized path sequence, converting the AGV kinematic constraints and collision avoidance constraints into linear constraint forms, and obtaining an initial linearized constraint set; based on the initial linearized constraint set, the gradient of the AGV path points is calculated ∇f(x0), and the linearized objective function f(x) + ∇f(x0)^T(x-x0) of the path optimization is constructed to obtain a local convex optimization objective function; the local convex optimization objective function is replaced by variables, and the non-convex constraints of the AGV path points are converted into linear inequality constraints ax + b ≥ 0, and obtain the linearized constraint matrix; based on the linearized constraint matrix, divide the AGV path resources, allocate computing resources, communication resources and storage resources to different path segments, and obtain the resource allocation vector; substitute the resource allocation vector into the path planning problem, construct an optimization problem with the goal of minimizing the task timeliness index AoI and the total energy consumption EC of the system, and obtain the convex subproblem model; perform KKT condition analysis on the convex subproblem model, calculate the Lagrange multiplier, construct the dual problem, and obtain the convex subproblem solution sequence; based on the convex subproblem solution sequence, solve each subproblem by the interior point method, and obtain the path optimization candidate solution set by iteratively updating the obstacle parameters; perform constraint satisfaction verification on the path optimization candidate solution set, select the solution set that satisfies all linearized constraints and has the smallest objective function value, and obtain the path optimization subproblem set.

[0167] In a specific embodiment, the above-mentioned AGV path optimization method based on digital twins also includes the following steps:

[0168] The Euclidean distance between the AGV optimal path planning set and the AGV real-time path is calculated, and the deviation of each position point is compared with the preset threshold to obtain the path deviation marking matrix;

[0169] Based on the path deviation marking matrix, the path segments from the current moment to the time window are extracted, and the paths outside the time window are kept unchanged to obtain the sequence of path segments to be optimized. Each path in the sequence of path segments to be optimized is mapped to the time dimension to generate the coordinate point sequence of AGV at different moments to obtain the space-time trajectory matrix.

[0170] The minimum distance between any two paths is calculated based on the space-time trajectory matrix. When the distance is less than the safety threshold, it is marked as a conflict point to obtain a conflict detection result set. The AGVs in the conflict detection result set are sorted according to the task priority coefficient, and a priority level is assigned to each conflicting AGV to obtain a conflict handling priority vector.

[0171] According to the conflict handling priority vector, the path of the low-priority AGV is adjusted, the waiting time or the detour path length is calculated, and a conflict avoidance strategy set is obtained;

[0172] The conflict avoidance strategy set is locally replanned to obtain an optimized corrected path set, and the path segments in the optimized corrected path set replace the corresponding path segments in the time window of the AGV optimal path planning set, while the remaining path segments remain unchanged to obtain a dynamically updated AGV optimal path planning set.

[0173] Specifically, based on the AGV optimal path planning set and real-time path of AGV , calculate the Euclidean distance deviation of each path point. Assume that The planned path point sequence of an AGV is , the real-time path point sequence is The deviation matrix of the two Defined as the Euclidean distance of each point:

[0174] ;

[0175] For each deviation value in the deviation matrix With preset threshold For comparison, if , then it is marked as a path deviation point, generating a path deviation marking matrix ,in:

[0176] ;

[0177] Path Deviation Labeling Matrix , extract the current time into the time window The path segments within the time window are kept unchanged. Assume that the current time is , the time range of the path segment is , the corresponding path segment point sequence It is expressed as:

[0178] ;

[0179] For each path segment to be optimized Perform time dimension mapping and expand the path points into a space-time trajectory matrix according to time distribution , where each row represents the two-dimensional coordinates of a path point at a certain moment:

[0180] ;

[0181] Based on the space-time trajectory matrix , calculate the minimum distance between any two paths. and A set of path points for AGVs and , minimum distance Defined as:

[0182] ;

[0183] in represents the Euclidean distance. , then the path conflict is marked as a conflict point and recorded in the conflict detection result set. For each conflict point in the conflict detection result set, according to the task priority coefficient in the task evaluation matrix , sort the related AGVs and assign a priority level to each conflicting AGV. Priority level vector The calculation formula is:

[0184] ;

[0185] in is the number of conflicting AGVs, the higher Corresponding to a higher priority. According to the priority vector , adjust the path of low-priority AGVs and generate a set of conflict avoidance strategies. For low-priority AGVs, the adjustment strategy includes calculating the waiting time Or plan the detour path length The calculation formula for waiting time is:

[0186] ;

[0187] in is the average speed of the AGV. The detour path is optimized by reinserting the path nodes, and the detour path length is calculated by a path search algorithm (such as the A* algorithm). After generating a set of conflict avoidance strategies, the path is locally replanned based on these strategies to optimize the corrected path. Assume that the corrected path segment is , replace it with the original path set The corresponding time window path segment in the dynamic update is the optimal path planning set of AGV. It is expressed as:

[0188] .

[0189] The above describes the AGV path optimization method based on digital twin in the embodiment of the present application. The following describes the AGV path optimization system 10 based on digital twin in the embodiment of the present application. Figure 2In the embodiment of the present application, an embodiment of the AGV path optimization system 10 based on digital twins includes:

[0190] The standardization module 11 is used to perform data standardization processing on the AGV real-time coordinate data, the environmental obstacle distribution data and the cargo storage location data, and construct a storage environment state matrix;

[0191] A calculation module 12 is used to calculate the AGV task time parameters based on the warehouse environment state matrix to obtain a task evaluation matrix including task timeliness indicators and task priority coefficients;

[0192] The processing module 13 is used to perform multi-objective parameter calculation on the AGV motion energy consumption coefficient and the task weight coefficient according to the task evaluation matrix, and establish the AGV path optimization objective function;

[0193] The planning module 14 is used to perform path planning analysis based on the AGV path optimization objective function using a path exchange operator, a path insertion operator, and a path reversal operator to generate an AGV initial path set;

[0194] The iteration module 15 is used to perform variable parameter particle swarm annealing optimization and continuous convex approximation dual iterative calculation on the AGV initial path set to obtain the AGV optimal path planning set.

[0195] Through the cooperation of the above components, multi-level data standardization processing and state information merging, the efficient integration of AGV real-time coordinates, environmental obstacles and cargo storage location data is achieved, and the data quality and availability are improved. The dual evaluation mechanism of task timeliness index and priority coefficient is adopted, combined with the comprehensive calculation of task complexity, load status and urgency, so that the task allocation is more reasonable. Through the accurate modeling of motion energy consumption and waiting energy consumption, combined with multi-objective parameter optimization, the precise control and optimization of energy consumption are achieved. The three operators of path exchange, insertion and inversion are used, and the dynamic update mechanism of operator selection weight is combined to improve the flexibility and optimization effect of path planning. The dual iterative calculation of variable parameter particle swarm annealing optimization and continuous convex approximation is combined to significantly improve the convergence speed and solution quality of path optimization. Through real-time deviation analysis and conflict detection, combined with the dynamic path correction mechanism within the time window, the stable operation of the system in a dynamic environment is guaranteed. Through the design of conflict handling priority vector and conflict avoidance strategy, the coordination problem of multi-AGV system is effectively solved. The hierarchical optimization strategy is adopted to decompose the complex path planning problem into multiple sub-problems, which reduces the computational complexity and improves the efficiency of the algorithm.

[0196] See also Figure 3 , Figure 3This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a system bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0197] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned AGV path optimization methods based on digital twins.

[0198] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0199] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned AGV path optimization methods based on digital twins.

[0200] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the present application scheme, and does not constitute a limitation on the electronic device 300 involved in the present application scheme. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0201] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0202] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned AGV path optimization method based on digital twins, and will not be repeated here.

[0203] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the AGV path optimization method based on digital twins as provided in an embodiment of the present application.

[0204] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the electronic device 300.

[0205] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0206] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0207] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An AGV path optimization method based on digital twins, characterized in that: The method comprises: Standardize the AGV real-time coordinate data, environmental obstacle distribution data, and cargo storage location data to build a warehouse environment status matrix; Based on the warehouse environment state matrix, the AGV task time parameters are calculated to obtain a task evaluation matrix including task timeliness indicators and task priority coefficients; According to the task evaluation matrix, multi-objective parameter calculation is performed on the AGV motion energy consumption coefficient and the task weight coefficient, and an AGV path optimization objective function is established; Based on the AGV path optimization objective function, a path exchange operator, a path insertion operator and a path reversal operator are used to perform path planning analysis to generate an AGV initial path set; specifically, the following steps are performed: based on the AGV path optimization objective function, a path point sequence of the kth AGV is extracted, and the turning angle and obstacle avoidance distance between two adjacent points in the path point sequence are calculated to obtain a path feature vector; a path exchange operator operation is performed on the turning angle in the path feature vector, and local optimization is performed by exchanging two path segments whose angle values ​​are greater than a preset threshold to obtain an angle-optimized path; a path insertion operator operation is performed on the obstacle avoidance distance based on the angle-optimized path, and an angle-optimized path is obtained by performing local optimization on the obstacle avoidance distance in the obstacle avoidance vector. A new path node is inserted at a position where the obstacle distance is greater than a preset threshold to perform obstacle avoidance optimization, and an obstacle avoidance optimized path is obtained; a path reversal operator operation is performed on the obstacle avoidance optimized path, and the path length is reduced by reversing the local path direction to obtain a length optimized path; a path quality score is calculated according to the length optimized path, and an operation frequency is calculated based on historical path data to obtain an operator selection weight, and the operator selection weight is iteratively updated, and parameter optimization is performed by calculating the difference between the objective function values ​​of two adjacent iterations to obtain an optimal operator combination sequence; the AGV path is iteratively calculated based on the optimal operator combination sequence, and the N paths with the best sorting are selected as the AGV initial path set; The AGV initial path set is subjected to variable parameter particle swarm annealing optimization and continuous convex approximation double iterative calculation to obtain the AGV optimal path planning set.

2. The AGV path optimization method based on digital twin according to claim 1 is characterized in that: The data standardization process of the AGV real-time coordinate data, the environmental obstacle distribution data and the cargo storage location data is performed to construct a storage environment state matrix, including: Performing zero-mean normalization on the AGV real-time coordinate data collected by multiple sensors to obtain an AGV standardized position matrix, and performing median filtering on the outliers in the AGV standardized position matrix to obtain an AGV outlier filter matrix; Binarize the environmental obstacle distribution data, mark the obstacle position as 1, and mark the idle position as 0, to obtain the obstacle distribution binary matrix, and encode and map the cargo storage location data, map different cargo types to corresponding numbers, and obtain the cargo location coding matrix; Performing tensor splicing on the AGV outlier filter matrix, the obstacle distribution binary matrix, and the cargo position encoding matrix to obtain an initial environment state tensor, and performing dimensionality reduction processing on the initial environment state tensor to obtain an environment state dimensionality reduction matrix; The environmental state dimension reduction matrix is ​​normalized to obtain an environmental state normalized matrix, and the state information of each grid point in the environmental state normalized matrix is ​​merged to obtain a storage environment state matrix.

3. The AGV path optimization method based on digital twin according to claim 2 is characterized in that: Based on the warehouse environment state matrix, the AGV task time parameters are calculated to obtain a task evaluation matrix including task timeliness index and task priority coefficient, including: Based on the warehouse environment state matrix, a time parameter is calculated for the i-th task of the k-th AGV to obtain the task start time and the task completion time, and a difference operation is performed according to the task start time and the task completion time to obtain a task timeliness index; Extracting the task complexity parameters of the kth AGV in the warehouse environment state matrix to obtain a task complexity vector including the task path length, the number of turns, and the number of obstacle avoidances; Calculate the current load state of the kth AGV based on the storage environment state matrix to obtain the AGV load state vector, and calculate the urgency of the i-th task of the k-th AGV to obtain the task urgency coefficient; Performing a weighted calculation of a priority weight coefficient according to the task complexity vector, the AGV load state vector and the task urgency coefficient to obtain a task priority coefficient; The task timeliness index and the task priority coefficient are combined in a matrix to obtain an initial task evaluation matrix, and the initial task evaluation matrix is ​​dynamically prioritized to ensure that high-priority tasks are prioritized first, thereby obtaining a task evaluation matrix.

4. The AGV path optimization method based on digital twin according to claim 3 is characterized in that: According to the task evaluation matrix, multi-objective parameter calculation is performed on the AGV motion energy consumption coefficient and the task weight coefficient to establish the AGV path optimization objective function, including: Extract data of the AGV motion state according to the task evaluation matrix to obtain a motion state vector including the AGV speed, acceleration and load weight; Performing a product operation of energy consumption coefficient on the AGV speed, acceleration and load weight in the motion state vector to obtain an AGV basic energy consumption coefficient matrix; Integrate and calculate the motion energy consumption based on the i-th task execution time of the k-th AGV in the AGV basic energy consumption coefficient matrix to obtain the AGV motion total energy consumption matrix; Extracting the AGV waiting time according to the AGV task execution order in the task evaluation matrix to obtain an AGV waiting time sequence, and performing a product operation of the unit time energy consumption coefficient on the AGV waiting time sequence to obtain an AGV waiting energy consumption matrix; The AGV movement total energy consumption matrix and the AGV waiting energy consumption matrix are cumulatively calculated to obtain the total energy consumption of the AGV system, and the objective function weight coefficient configuration calculation is performed based on the task timeliness index in the task evaluation matrix and the total energy consumption of the AGV system to obtain a weight coefficient vector; Based on the weight coefficient vector, a weighted combination operation is performed on the task timeliness index and the total energy consumption of the AGV system to obtain an AGV path optimization objective function.

5. The AGV path optimization method based on digital twin according to claim 1 is characterized in that: The AGV initial path set is subjected to variable parameter particle swarm annealing optimization and continuous convex approximation double iterative calculation to obtain the AGV optimal path planning set, including: Cartesian coordinate encoding is performed on each path in the AGV initial path set, and the path point sequence is converted into a position vector to obtain a path encoding matrix; Based on the path encoding matrix, the particle swarm is initialized for the path of the kth AGV, the position of each particle is represented as the coordinate sequence of the AGV at each moment, and the particle swarm position matrix is ​​obtained. The particle swarm position matrix is ​​configured with parameters of inertia weight, cognitive factor and social factor, and random numbers are generated to obtain a speed update parameter matrix. Iteratively updating the particle swarm velocity according to the velocity update parameter matrix, and performing annealing according to the temperature parameter to obtain a VPPSA optimization path sequence; The VPPSA optimized path sequence is input into the SCA algorithm, and a local linearized constraint matrix is ​​obtained by performing a first-order Taylor expansion on the non-convex path constraint. Based on the local linearized constraint matrix, a convex optimization solution is performed on the AGV path resource allocation, and the path planning problem is converted into a convex sub-problem sequence to obtain a path optimization sub-problem set; Iteratively solving the path optimization sub-problem set, updating the linearization point and checking the convergence condition in each iteration, and obtaining a local optimal path sequence; The local optimal path sequence is verified for AGV motion constraints, obstacle avoidance constraints and path smoothness constraints, and a path set that meets all constraints is selected as the AGV optimal path planning set.

6. The AGV path optimization method based on digital twin according to claim 5 is characterized in that: The AGV path optimization method based on digital twins also includes: The Euclidean distance between the optimal path planning set of the AGV and the real-time path of the AGV is calculated, and the deviation of each position point is compared with a preset threshold to obtain a path deviation marking matrix; Based on the path deviation marking matrix, the path segments from the current moment to the time window are extracted, and the paths outside the time window are kept unchanged to obtain a sequence of path segments to be optimized, and each path in the sequence of path segments to be optimized is mapped in the time dimension to generate a sequence of coordinate points of the AGV at different moments to obtain a spatiotemporal trajectory matrix; The minimum distance between any two paths is calculated based on the space-time trajectory matrix, and when the distance is less than the safety threshold, it is marked as a conflict point to obtain a conflict detection result set, and the AGVs in the conflict detection result set are sorted according to the task priority coefficient, and a priority level is assigned to each conflicting AGV to obtain a conflict handling priority vector; Adjust the path of the low-priority AGV according to the conflict handling priority vector, calculate the waiting time or the detour path length, and obtain a conflict avoidance strategy set; The conflict avoidance strategy set is locally replanned to obtain an optimized corrected path set, and the path segments in the optimized corrected path set replace the corresponding path segments in the time window in the AGV optimal path planning set, while the remaining path segments remain unchanged to obtain a dynamically updated AGV optimal path planning set.

7. An AGV path optimization system based on digital twins, characterized in that: Used to execute the AGV path optimization method based on digital twins as described in any one of claims 1 to 6, the AGV path optimization system based on digital twins includes: The standardization module is used to standardize the AGV real-time coordinate data, environmental obstacle distribution data, and cargo storage location data to build a storage environment status matrix; A calculation module, used to calculate the AGV task time parameters based on the warehouse environment state matrix to obtain a task evaluation matrix including task timeliness indicators and task priority coefficients; A processing module, used to perform multi-objective parameter calculation on the AGV motion energy consumption coefficient and the task weight coefficient according to the task evaluation matrix, and establish an AGV path optimization objective function; A planning module, used to perform path planning analysis based on the AGV path optimization objective function using a path exchange operator, a path insertion operator, and a path reversal operator to generate an AGV initial path set; The iterative module is used to perform variable parameter particle swarm annealing optimization and continuous convex approximation double iterative calculation on the AGV initial path set to obtain the AGV optimal path planning set.

8. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes the digital twin-based AGV path optimization method as described in any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the AGV path optimization method based on digital twins as described in any one of claims 1 to 6 is implemented.

Images