A cloud computing-based power iron accessory production scheduling intelligent generation method and system

Abstract

This invention belongs to the field of intelligent production planning and management technology, specifically relating to a cloud-based intelligent generation method and system for power railway accessory production scheduling. The method includes: constructing an initial disjunction graph of power railway accessory production processes and their sequential constraints; setting virtual batch aggregation nodes and batch saturation parameters for batch processing processes; calculating the conjunction arc weights of adjacent processes; in iterative optimization, calculating the bottleneck index of all undirected disjunction arc pairs; prioritizing the disjunction arc with the largest bottleneck index; and generating an initial production scheduling plan for power railway accessories after full arc orienting; upon receiving a workshop production event, identifying the set of affected processes and defining the minimum disturbance subgraph containing processing sequence and resource relationships of adjacent processes; generating a feasible schedule with the goal of minimizing the timing impact of the unaffected parts of the original plan; and outputting the updated production scheduling plan. This invention can coordinate batch processing and regular processes to achieve rapid and stable disturbance response.

Description

Technical Field

[0001] This invention relates to the field of production planning and management technology. More specifically, this invention relates to a cloud computing-based intelligent generation method and system for power railway accessory production scheduling. Background Technology

[0002] The production of power railway accessories is a discrete manufacturing process, encompassing numerous processes such as material preparation, stamping, bending, welding, and hot-dip galvanizing. It is characterized by a high degree of customization, complex specifications, and tight delivery schedules. In actual production, it includes both routine assembly line processing for single pieces or small batches and batch processing processes such as hot-dip galvanizing, heat treatment, and shot blasting. This mixed production model makes the timing constraints between processes and the competition for equipment resources highly complex.

[0003] Currently, production tasks are typically abstracted into disjunctive graph networks to express the sequential constraints between processes and the competition for equipment resources. Optimization algorithms are then used to directionally solve the undirected disjunctive arcs in the network, thereby determining the processing sequence of each process and generating a production schedule.

[0004] However, existing disjunctive graph models cannot effectively model batch-processing operations, fail to evaluate uniformly named batch saturation parameters, and lack bottleneck process identification and priority targeting strategies during iterative solution, resulting in poor quality of the generated initial production schedule. At the same time, the workshop environment is full of uncertainties and often faces disturbances such as emergency order insertions and equipment failures. When responding to these disturbances, existing methods usually adopt a global rescheduling strategy to resolve the entire disjunctive graph. This global repair is not only computationally time-consuming, but also triggers a butterfly effect, causing timing impacts on processes that were not affected in the original plan, disrupting the stability of production execution, and failing to meet the needs of industrial sites for rapid response to abnormal events and smooth transition of production schedules. Summary of the Invention

[0005] To address the technical problems of existing methods being unable to coordinate batch processing procedures and bottleneck identification, and the time-consuming global rescheduling process that impacts the stability of the original plan, this invention proposes a cloud computing-based intelligent generation method and system for power railway accessory production scheduling. This method can coordinate batch processing and routine processes to achieve rapid and stable disturbance response.

[0006] In a first aspect, the present invention provides a cloud computing-based intelligent generation method for power railway accessory production scheduling, comprising: constructing an initial disjunction graph representing the processing steps and sequential constraints in the production of power railway accessories; wherein, for batch processing steps, virtual batch aggregation nodes are set to uniformly schedule multiple identical steps belonging to the same production batch, and a batch saturation parameter is set for each virtual batch aggregation node; calculating the conjunction arc weights connecting two adjacent steps in the initial disjunction graph; and, during the iterative optimization process of the initial disjunction graph, calculating the bottleneck index of all undirected disjunction arc pairs, and preferentially selecting the disjunction arc with the largest bottleneck index for orienting, based on... The orientation results determine the processing sequence of each processing step. After all dissociative arcs are oriented, an initial production schedule for power railway accessories is generated and output. When the scheduling system receives a workshop production event, it identifies the set of processes affected by the event and defines a minimum disturbance subgraph containing the processes and the processes adjacent to them in terms of processing timing and processing resources. Within the minimum disturbance subgraph, with the goal of minimizing the timing impact on the non-influenced parts of the original plan, a new feasible schedule is generated for the event by executing repair operators, including right shifting of processes, resource reallocation, sequence exchange within the allowed time window, and orientation of newly added unoriented dissociative arcs. The updated production schedule for power railway accessories is then output.

[0007] By adopting the above technical solutions, a disjunction graph model containing virtual batch aggregation nodes and batch saturation parameters was constructed, solving the problem of collaborative modeling of conventional processing and batch processing procedures. The bottleneck index was used to guide the priority orientation of disjunction arcs, improving the quality of initial scheduling. When disturbances occur, the minimum disturbance subgraph is defined and various repair operators are used to perform local rescheduling, which significantly reduces the computation time and maintains the execution stability of the original plan.

[0008] Preferably, for batch processing operations, setting up virtual batch aggregation nodes includes: acquiring operations in the pending processing plan of electric iron accessories that are surface treated on the same processing equipment and belong to the same manufacturing order batch, and aggregating them into the corresponding batch processing operation set, wherein batch processing operations include hot-dip galvanizing, heat treatment, shot blasting, or other operations that need to be completed centrally within the same batch processing equipment; creating virtual batch aggregation nodes containing proprietary identification information in the scheduling system, binding each operation task in the batch processing operation set to the corresponding virtual batch aggregation node, and establishing a batch dependency mapping relationship between the virtual batch aggregation node and each operation task; the batch dependency mapping relationship is used to represent the same batch processing relationship and does not represent a new processing sequence constraint separately.

[0009] By adopting the above technical solution, surface treatment processes of the same batch and equipment are aggregated and virtual batch aggregation nodes and subordinate mapping relationships are created, realizing unified modeling and centralized scheduling of batch processing processes. This avoids batch processing tasks being broken down into scattered scheduling objects, enabling the production scheduling model to accurately reflect the actual constraints of batch processing operations.

[0010] Preferably, setting a batch saturation parameter for each virtual batch aggregation node includes: reading the minimum three-dimensional bounding box volume data of the physical materials corresponding to each process in the batch processing process set, and performing cumulative processing in combination with a preset spatial loading porosity compensation coefficient to obtain the merged total physical volume.

[0011] By adopting the above technical solution, the minimum three-dimensional bounding box volume data of the physical material is read and accumulated in combination with the space loading porosity compensation coefficient to obtain the combined physical total volume. This provides a more accurate basis for calculating the batch saturation parameter and avoids the deviation caused by estimating the equipment loading rate only based on the number of processes.

[0012] Preferably, after obtaining the merged total physical volume, the method further includes: retrieving the maximum loading space volume of the corresponding processing equipment; dividing the merged total physical volume by the maximum loading space volume to obtain a volume percentage value; assigning the volume percentage value as a batch saturation parameter to the corresponding virtual batch aggregation node; and marking the corresponding batch as an overloaded batch when the volume percentage value is greater than 1, so as to facilitate subsequent batch splitting or re-merging.

[0013] By adopting the above technical solution, the total physical volume is compared with the maximum loading space of the equipment, and the volume ratio is calculated as a batch saturation parameter, thus realizing the assessment of the loading degree of the batch processing equipment. At the same time, by automatically marking the overloaded batches, it is convenient to split or re-merge the batches in the future, ensuring the feasibility of the production schedule.

[0014] Preferably, the calculation of the conjunctive arc weight connecting two adjacent processes in the initial disjunctive graph includes: locating the preceding and subsequent processes connected by the conjunctive arc in the initial disjunctive graph; determining whether the processing equipment for the preceding process has been determined; if the processing equipment has been determined, reading the processing time constant value of the preceding process on that processing equipment; if the unique processing equipment has not yet been determined, reading the processing time corresponding to each candidate processing equipment in the available machine set of the preceding process, and determining the baseline processing time constant value according to the preset equipment selection rules; reading the one-way transportation time constant value required for the component to move to the location of the subsequent process after the preceding process is completed; adding the processing time constant value or the baseline processing time constant value to the one-way transportation time constant value to obtain the total time constant value; and assigning the total time constant value as the conjunctive arc weight to the corresponding conjunctive arc.

[0015] By adopting the above technical solution, the processing time of the preceding process and the single-trip transportation time between processes are integrated into the conjunctive arc weight, which solves the problem of repeated or omitted time calculations that are easy to occur in subsequent time propagation calculations; for the preceding process whose processing equipment has not yet been determined, a reference time determination method based on preset rules is provided to ensure the consistency of weight assignment in the disjunctive graph model in the early stage of solution.

[0016] Preferably, the calculation of the bottleneck index for all undirected disjunction arc pairs includes: traversing all undirected disjunction arc pairs in the initial disjunction graph that have resource usage conflicts and whose processing order has not yet been determined; for two processes associated with an undirected disjunction arc pair, performing forward calculation from the starting node of the initial disjunction graph to obtain the earliest start time of each process; performing reverse calculation from the ending node of the initial disjunction graph to obtain the latest start time of each process; calculating the difference between the latest start time and the earliest start time of each process as the relaxation time of each process; calculating the sum of the relaxation times of the two processes; calculating a value with the natural constant as the base and the negative of the product of the sum of relaxation times and the preset time normalization coefficient as the exponent as the relaxation index type bottleneck index, and assigning it to the corresponding undirected disjunction arc pair.

[0017] Preferably, defining the minimum perturbation subgraph, which includes processes and processes adjacent to each other in terms of processing timing and processing resources, includes: real-time monitoring of the workshop production event message queue; when an emergency insertion command is received, extracting the set of processes affected by the emergency insertion command; in the initial disjunction graph, finding the preceding and succeeding processes of each process in the process set based on timing constraints; in the initial disjunction graph, finding the preceding and succeeding processing processes adjacent to each process in the process set on the same processing equipment based on resource constraints; merging the affected set of processes, preceding processes, succeeding processes, and adjacent preceding and succeeding processing processes to extract the minimum perturbation subgraph, wherein the minimum perturbation subgraph is used to limit the local rescheduling range, and its number of nodes is adjusted according to a preset neighborhood depth, maximum number of nodes, or event impact range.

[0018] Preferably, the step of generating a new feasible schedule for the event includes: extracting each process and time constraint parameter contained in the minimum disturbance subgraph; executing a process right shift operator on the affected processes to postpone the start time of the processes to avoid resource conflicts, and verifying the delivery date constraints, subsequent process timing constraints, and equipment availability constraints after the right shift; finding replaceable idle processing equipment in the workshop, executing a resource reallocation operator on the affected processes, and scheduling the processes to idle processing equipment, wherein the idle processing equipment must meet the processing capacity, tooling fixtures, process parameters, and quality requirements of the corresponding processes; executing a sequence exchange operator within the allowable range of the time constraint parameters to adjust the processing order between processes on the same processing equipment; executing a direction operator on newly added undirected disjunctive arcs to determine the processing order of the processes, and performing directed cycle detection after directionation to avoid forming circular dependencies.

[0019] Preferably, the step of generating new feasible schedules for events and outputting updated power railway accessory production plans includes: calculating the time deviation between each new scheduling scheme generated after executing the repair operator and the non-affected part of the original plan; selecting the feasible new scheduling scheme with the smallest time deviation as the feasible schedule to update the initial disjunction graph.

[0020] Secondly, the present invention provides a cloud computing-based intelligent generation system for power railway accessory production scheduling, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned cloud computing-based intelligent generation method for power railway accessory production scheduling is implemented.

[0021] By adopting the above technical solution, a computer program is generated from the above-mentioned intelligent generation method for power railway accessory production scheduling based on cloud computing, and stored in a memory for loading and execution by a processor. This allows for the creation of terminal devices based on the memory and processor, making them convenient to use.

[0022] The beneficial effects of this invention are as follows: This invention achieves centralized and unified scheduling of batch processing operations by constructing a disjunctive graph model that incorporates virtual batch aggregation nodes and batch saturation parameters, and effectively improves the quality of the initial production scheduling plan by leveraging a bottleneck index priority orientation strategy.

[0023] Furthermore, this invention defines the minimum disturbance subgraph of the affected process and comprehensively applies repair operators such as resource reallocation, process right shift, and sequence exchange within this local area. With the goal of minimizing the timing impact on the non-affected parts of the original plan, the invention maintains the stability of the original plan while generating an updated scheduling scheme, thus avoiding the computational power consumption and production site disorder caused by global reordering. Attached Figure Description

[0024] Figure 1A flowchart of a cloud-based intelligent generation method for power railway accessory production scheduling; Figure 2 A schematic diagram comparing the volume saturation parameters of aggregation nodes in different virtual batches; Figure 3 This is a schematic diagram illustrating the trend of the bottleneck index of the undirected dissociative arc as a function of total relaxation time. Figure 4 This diagram illustrates the impact of different rescheduling and repair schemes on the original timeline deviation. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0026] This invention discloses a cloud computing-based intelligent generation method for power railway accessory production scheduling, referring to... Figure 1 This includes steps S1-S3: S1. Construct an initial disjunction graph representing the processing steps and sequential constraints in the production of electric railway accessories. For batch processing steps, set up virtual batch aggregation nodes to uniformly schedule multiple identical steps belonging to the same production batch, and set a batch saturation parameter for each virtual batch aggregation node; calculate the conjunction arc weight connecting two adjacent steps in the initial disjunction graph.

[0027] The cloud computing platform includes a cloud data storage module, a distributed computing scheduling module, and an elastic resource allocation module, used to achieve centralized management of manufacturing data and parallel solution of production scheduling models. Through a cloud interface, it reads power ferrule work order data and process route data from the Manufacturing Execution System (MES) in real-time or near real-time, obtaining the workpiece's process set, the standard processing time for each process, and the set of available machines. An initial disjunction graph object is created, which stores directed conjunction arcs representing the process sequence. Simultaneously, a set of undirected disjunction arc pairs is established to record resource conflicts on the same processing resource where the sequence has not yet been determined.

[0028] Traverse the set of processes, creating a regular node for each process and setting its processing time attribute. Based on the sequence in the process route, establish directed conjunctive arcs between adjacent process nodes on the same workpiece. For any two processes whose available machine sets intersect and cannot occupy the same processing resource simultaneously, register resource conflict relationships in the set of undirected disjunctive arc pairs. The processing order of these resource conflict relationships is initially uncertain and gradually becomes oriented during subsequent iterative solutions.

[0029] For batch processing operations that need to be completed centrally within the same batch processing equipment, tank, or furnace, a virtual batch aggregation node is established. This virtual batch aggregation node represents the batch processing unit of that batch, and a batch dependency mapping relationship is established between each regular process node belonging to that batch and the virtual batch aggregation node. This batch dependency mapping relationship is used for unified scheduling and capacity verification, and is not used as a constraint edge for ordinary process sequence in the critical path duration accumulation, to avoid duplicate calculation of batch processing operations. The virtual batch aggregation node itself has a batch processing time attribute, which is determined based on the standard batch processing time, loading and unloading time, and necessary waiting time of the corresponding batch processing equipment. In time propagation calculations, this virtual batch aggregation node is used as the actual batch processing node participating in production scheduling, and its preceding and succeeding processes are connected by conjunctive arcs to determine the planned start time and planned completion time of the batch processing operation.

[0030] The total volume of all processed parts allocated to the virtual batch aggregation node is calculated, or the total volume after loading correction is calculated, and then divided by the maximum loading capacity of the batch processing equipment. The resulting ratio is stored as a batch saturation parameter in the attributes of the virtual batch aggregation node. When the batch saturation parameter is greater than 1, it indicates that the batch exceeds the allowable loading capacity of the corresponding batch processing equipment, and the scheduling system marks the batch as needing to be split or re-merged.

[0031] For all conjuncts in the diagram, the weight of the conjunct is calculated based on the equipment status of the preceding process. When the processing equipment for the preceding process has been determined, the standard processing time of the source node on that processing equipment and the logistics transportation time between processes are extracted, and the two are summed to obtain the weight of the conjunct. This weight is then updated in the weight attribute of the corresponding conjunct. The conjunct weight represents the minimum time interval between the start time of the preceding process and the allowed start time of the subsequent process. During subsequent forward time propagation based on this conjunct weight, the processing time attribute of the preceding process node is not repeatedly accumulated. When the preceding process has not yet determined a unique processing equipment, the baseline processing time is determined according to a preset equipment selection rule, or the candidate conjunct weights corresponding to the candidate equipment are stored separately and updated to the actual conjunct weights after the subsequent equipment allocation is determined.

[0032] As one possible implementation, for batch processing operations, virtual batch aggregation nodes are set up to uniformly schedule multiple identical operations belonging to the same production batch, and a batch saturation parameter is set for each virtual batch aggregation node, including: The process of surface treatment on the same processing equipment and belonging to the same manufacturing order batch in the pending processing plan for electric railway accessories is collected and grouped into the corresponding batch processing process set. The batch processing processes include hot-dip galvanizing, heat treatment, shot blasting, or other processes that need to be completed centrally within the same batch processing equipment. A virtual batch aggregation node containing proprietary identification information is created in the production scheduling system. Each process task in the batch processing process set is bound to the corresponding virtual batch aggregation node, and a batch dependency mapping relationship is established between the virtual batch aggregation node and each process task. This batch dependency mapping relationship is used to represent the same batch processing relationship, not... A new processing sequence constraint is represented separately; the minimum three-dimensional bounding box volume data of the physical materials corresponding to each process in the batch processing process set are read and accumulated in combination with the preset space loading porosity compensation coefficient to obtain the merged physical total volume; the maximum loading space volume quantitative value of the corresponding processing equipment is retrieved; the merged physical total volume is divided by the maximum loading space volume quantitative value to obtain the volume ratio value; the volume ratio value is used as the batch saturation parameter and assigned to the corresponding virtual batch aggregation node, and when the volume ratio value is greater than 1, the corresponding batch is marked as an overloaded batch for subsequent batch splitting or re-batch processing.

[0033] In typical batch-processing operations such as hot-dip galvanizing or shot blasting surface treatment of power railway accessories, the scheduling system scans the pending plans in the MES system and uses SQL queries to filter out orders from the same processing equipment, such as hot-dip galvanizing tank number 2, and order batch numbers. Identical processes are grouped into a batch-type process set. A unique identifier is created for this set using a UUID generator, such as "". "As a virtual batch aggregation node, this virtual node and each sub-process within the set are established in the graph structure table of the database, such as..." , The hierarchical mapping relationship enables unified scheduling management, allowing batch processing operations to be treated as the same batch processing unit for capacity verification and production scheduling calculations, avoiding their fragmentation into multiple independent and scattered scheduling objects. Simultaneously, the scheduling system reads the standard immersion treatment time, loading time, and unloading time of the No. 2 hot-dip galvanizing tank from the equipment parameter library, using the sum of these three as the virtual node " The batch processing time is calculated, and the virtual node is connected to the timing chain of the corresponding process route so that the batch processing process can obtain a clear planned start time and planned completion time.

[0034] When determining batch saturation parameters, the 3D modeling data is called to read the minimum 3D bounding box volume of each entity material within the process set. For example, the bounding box of component A has dimensions of [length, width, and height]. The volume is The system utilizes a preset spatial loading porosity compensation coefficient. This coefficient compensates for gap loss caused by irregular shapes, and its preferred range is generally between 1.2 and 1.5, specifically 1.35. Assuming the batch contains 50 parts A, the cumulative total volume is calculated as follows: .

[0035] Retrieve the maximum loading space volume value corresponding to the No. 2 hot-dip galvanizing tank from the equipment parameter database, for example, set it to... The calculated combined physical volume Divide by the maximum loading space volume. The resulting volume percentage was 0.45. This value of 0.45 was then assigned as the batch saturation parameter to the virtual node. When the batch saturation parameter is below a threshold, such as 0.3, an early warning can be triggered, indicating that the batch needs to be merged to improve the actual utilization rate of the equipment. The volume saturation distribution of different virtual batch nodes is as follows: Figure 2 As shown, the equipment loading and utilization levels of each batch processing step are presented.

[0036] As one possible implementation, calculating the weight of the conjunctive arc connecting two adjacent processes in the initial disjunctive graph includes: locating the preceding and following processes connected by the conjunctive arc in the initial disjunctive graph; determining whether the processing equipment for the preceding process has been determined; if the processing equipment has been determined, reading the processing time constant value of the preceding process on that processing equipment; if a unique processing equipment has not yet been determined, reading the processing time corresponding to each candidate processing equipment in the available machine set of the preceding process, and determining the baseline processing time constant value according to a preset equipment selection rule.

[0037] After the preceding process is completed, the constant value of the single-trip transportation time required for the component to move to the location of the following process is read; the constant value of the processing time or the reference processing time constant value is added to the constant value of the single-trip transportation time to obtain the constant value of the total time consumption; the constant value of the total time consumption is used as the weight of the conjunct arc and assigned to the corresponding conjunct arc, wherein the weight of the conjunct arc is used to represent the minimum time interval between the start of the preceding process and the completion of the starting conditions of the following process, and the processing time constant value of the preceding process is not repeatedly added in subsequent time propagation calculations.

[0038] When constructing the initial disjunctive graph at the graph theory modeling level, the conjunctive arc represents the forced sequential constraint in the processing of the same workpiece. The scheduling system locates adjacent preceding and succeeding processes using a graph traversal algorithm. For example, if the preceding process is "CNC punching" of angle steel, the node ID is: The subsequent process is "CNC cutting", node ID: Connecting to the standard process time reference parameter library via the ODBC interface, the system reads the parameters for angle steel of this specification in a specific punching device, such as... The processing time constant value is a combination of machine tool feed rate and material thickness, usually accurate to the second, with an example value of 450s.

[0039] Calculate the workpiece's logistics transfer time within the workshop by extracting the position coordinates of the preceding punching equipment. , ) and the position coordinates of the subsequent cutting equipment ( , By combining the average travel speed of the workshop AGV or overhead crane, preferably within the range of 1 m / s to 2.5 m / s, and taking an average speed of 1.5 m / s as an example, the constant value of the one-way transportation time required for the component to move to the location of the subsequent process is calculated. Taking into account the fixed motion losses of loading and unloading, this constant value of transportation time can be evaluated as 150 s in this example.

[0040] The underlying calculation module performs an addition operation, adding the processing time constant value of 450s to the transportation time constant value of 150s, resulting in a total time constant value of 600s. This value of 600s is then assigned to the connection. and The conjunctive arcs are fixed as weight parameters of the graph. During subsequent forward topology propagation using these conjunctive arc weights, The earliest allowed start time can be determined by The result is obtained by adding 600 seconds to the start time, so it will not be added again. The processing time is 450 seconds, thus avoiding repeated timing. If the subsequent directional solution is determined... If the hole is actually assigned to another candidate punching device, the processing time and transfer time corresponding to that candidate device are reread, and the conjunct arc weight is updated.

[0041] S2, in the iterative solution optimization process of the initial disjunction diagram, the bottleneck index of all unoriented disjunction arc pairs is calculated, and the disjunction arc with the largest bottleneck index is selected for orientation first. Based on the orientation results, the processing sequence of each processing step is determined. After all disjunction arcs are oriented, the initial production schedule of electric railway accessories is generated and output.

[0042] In one feasible main implementation, a path-incremental bottleneck index is used to calculate the bottleneck degree of undirected disjunctive arc pairs; in another alternative implementation, a relaxation-index bottleneck index is used to calculate the bottleneck degree of undirected disjunctive arc pairs. Either method is chosen for the same iterative solution process and is not cross-compared in the same sorting process to ensure consistency in the calculation of the bottleneck index.

[0043] The critical path algorithm is used to calculate the earliest start time from the virtual source node to each process node in the initial disjunctive graph, and the latest start time of each process node is calculated backwards from the virtual sink node. The latest start time uses the order delivery date as the priority time boundary; when the corresponding order does not have a specified delivery date, it is obtained by backward recursion using the critical path length of the currently oriented graph as the time boundary of the virtual sink node; in the same iteration, the same type of time boundary is used for the same order or the same set of processes. During time propagation, the edge weights determined in S1 are used for calculation. When the conjunctive arc weight already includes the processing time of the preceding process, the processing time of that preceding process is not accumulated again.

[0044] For any two unordered process nodes competing on the same machine, forming an undirected disjunction arc pair, calculate the path length increment under two possible orientation methods. During temporary orientation, if process A is oriented before process B, an orientation disjunction arc indicating A precedes B is set; the weight of this orientation disjunction arc is determined based on the processing time of process A and the necessary changeover, waiting, or transfer time, representing the minimum start interval of process B relative to process A. If an orientation method would result in a directed cycle in the graph, this orientation method is marked as infeasible and not used as the final orientation direction. When using a path incremental bottleneck index, the sum of the path length increments of the two feasible directions is divided by a reference time scale to obtain the path incremental bottleneck index for the disjunction arc pair; the reference time scale can be the critical path length of the currently oriented graph, order delivery date, or a preset production scheduling time unit, used to ensure a unified comparison caliber for the path incremental bottleneck index. If there is only one feasible direction, the orientation of the disjunctive arc pair is uniquely determined. The scheduling system directly adopts the feasible direction for orientation, or divides the path length increment of the feasible direction by the reference time scale as the path increment bottleneck index for sorting, and no longer introduces undefined preset penalty values.

[0045] Based on the bottleneck index comparison criteria, the undirected disjunctive arc pair with the largest bottleneck index is searched from all undirected disjunctive arc pairs. The directional evaluation values ​​of this disjunctive arc pair with the largest bottleneck index are evaluated for both forward and reverse orientation schemes. The directional evaluation value is defined as the critical path length increment caused by the corresponding orientation direction; the smaller the directional evaluation value, the smaller the impact of that direction on the additional time of the current production schedule. The direction that minimizes the increase in the critical path length of the entire disjunctive graph and does not create directed cycles is selected as the orientation direction, and this directed edge is added. Simultaneously, the corresponding resource conflict relationship is removed from the set of undirected disjunctive arc pairs. If both directions are not feasible, the process either backtracks to the previous branch or reselects the next undirected disjunctive arc pair with the next bottleneck index for processing.

[0046] Repeat the above earliest start time calculation, bottleneck index calculation, and orientation operations, using a depth-first search or backtracking search framework, until all unoriented disjunctive arcs on all machines in the graph are successfully oriented. After each orientation, directed cycle detection is performed to ensure that the current graph remains a directed acyclic graph (DAG). At this point, the disjunctive graph is transformed into a DAG. Generate the equipment processing sequence, planned start time, and planned completion time for each processing step. Use the Matplotlib library to plot and output a Gantt chart with a specific time axis as the initial production schedule for power railway accessories. By prioritizing resource conflicts with large path increments or short relaxation times, critical resource conflicts can be identified early in the iteration, reducing the passive lengthening of the critical path by subsequent orientations and providing a calculation basis for reducing order delays and improving the stability of the initial production schedule.

[0047] As one possible implementation, the bottleneck index of all undirected disjunctive arc pairs is calculated during the iterative optimization process of the initial disjunctive graph. This includes: traversing all undirected disjunctive arc pairs in the initial disjunctive graph that have resource usage conflicts and whose processing order has not yet been determined; for two processes associated with an undirected disjunctive arc pair, forward calculation is performed from the starting node of the initial disjunctive graph to obtain the earliest start time of each process. The forward calculation is propagated according to the edge weights of the currently directed conjunctive arc and the directed disjunctive arc to avoid repeatedly accumulating the processing time of the same process. The latest start time of each process is obtained by reverse calculation from the termination node of the initial disjunction graph; the difference between the latest start time and the earliest start time of each two processes is calculated as the relaxation time of each process; the sum of the relaxation times of the two processes is calculated; a value with the natural constant as the base and the negative of the product of the sum of the relaxation times and the preset time normalization coefficient as the exponent is calculated as the relaxation exponential bottleneck index and assigned to the corresponding undirected disjunction arc pair, wherein the preset time normalization coefficient is used to eliminate the time dimension, making the exponent term a dimensionless value.

[0048] The goal of iterative optimization of the initial disjunction graph for power railway accessories is to resolve resource conflicts arising from multiple processes competing for the same processing equipment. A graph search algorithm is used to traverse all undirected disjunction arc pairs. For processes A and B associated with a pair of disjunction arcs, bidirectional network deduction is employed. From the virtual starting node Source, a forward topology propagation is performed along the currently directed conjunction and disjunction arcs, accumulating edge weights to obtain the earliest start time of each process. When the edge weight already includes the processing time of the preceding process, the processing time of the preceding node is not additionally added, thus obtaining the earliest start time of process A. For example... And the earliest start time of process B, for example s.

[0049] Starting from the virtual termination node Sink, the reverse recursion is initiated, prioritizing the order delivery date as the time boundary for the reverse recursion. When the corresponding order does not have a specified delivery date, the length of the critical path in the current oriented graph is used as the time boundary for the reverse recursion. The time consumed by subsequent nodes is deducted level by level to obtain the latest start time of process A. For example... The latest start time of process s and process B, for example After obtaining the timing boundaries, substitute the values ​​into the formula: Relaxation time = Latest start time - Earliest start time. Therefore, the relaxation time for process A is 500s, and the relaxation time for process B is 100s.

[0050] The relaxation index type bottleneck index is calculated using the formula. Where k is a preset time normalization coefficient used to eliminate the time dimension. A normalization adjustment coefficient k, ranging from 0.01 to 0.05, is used. When the time unit is seconds, k can be determined according to the production scheduling time scale. For example, k = 0.01, and the calculation formula is as follows: The result of 0.00247 is assigned to the unoriented dissociative arc pair. A higher index value indicates that the process pair is more likely to become a bottleneck in the entire production chain, and the scheduler will assign priority in the dissociative calculation accordingly. The relationship between the total relaxation time of the dissociative arc and the bottleneck index is as follows: Figure 3 As shown, this verifies the rule that the shorter the total relaxation time, the higher the bottleneck index.

[0051] After selecting the disjunctive arc pairs to be oriented based on the bottleneck index, the scheduling system attempts both forward and reverse orientation directions, performing directed cycle detection after each temporary addition of a oriented edge. If an orientation direction leads to a directed cycle, it is abandoned. If both directions are feasible, the direction with the smaller evaluation value is selected, i.e., the direction that minimizes the increase in the critical path length of the currently oriented graph, is chosen as the final orientation direction. After orientation is complete, the oriented graph and the set of unoriented disjunctive arc pairs are updated, and the earliest start time, latest start time, and bottleneck index required for subsequent iterations are recalculated until all unoriented disjunctive arc pairs have been oriented. When evaluating the initial scheduling effect, critical path length, average order delay time, number of resource conflict backtracking attempts, or scheduling calculation time can be used as evaluation indicators to illustrate the improvement effect of the bottleneck index-priority orientation rule compared to random orientation or first-come, first-served orientation rules.

[0052] S3, when the production scheduling system receives a workshop production event, it identifies the set of processes affected by the event and defines a minimum disturbance subgraph that includes the processes and the processes adjacent to each other in terms of processing timing and processing resources. Within the minimum disturbance subgraph, with the goal of minimizing the timing impact on the parts that are not affected by the original plan, it generates a new feasible schedule for the event by executing repair operators, including process right shift, resource reallocation, sequence exchange within the allowed time window, and orientation of newly added unoriented disjunction arcs, and outputs the updated power railway accessory production schedule.

[0053] In one primary implementation, the scheduling system can listen to workshop production events based on the workshop production event message queue and define the minimum disturbance subgraph using a graph-structured neighborhood retrieval method. In another alternative detailed implementation, the minimum disturbance subgraph can also be extracted based on emergency insertion instructions, equipment failure instructions, or production status change instructions in the workshop production event message queue, combined with an equipment timeline retrieval method. The workshop production event message queue refers to a message queue used to receive production event data sent by the manufacturing execution system, the equipment IoT system, or a manual scheduling terminal, and does not refer to a physical process queue in the workshop.

[0054] The system monitors in real-time machine failure or emergency order insertion events sent by the manufacturing execution system via a Kafka message queue. Upon receiving an event, it parses the event data format and matches it with the corresponding process node in the initial disjunction graph. Processes that cannot be processed on time due to machine failure or newly added processes due to emergency orders are added to the affected process set. The system queries the predecessor and successor nodes of each node in the affected process set in terms of process constraints, and simultaneously traverses the current machine processing sequence to query adjacent nodes in terms of resource constraints. The affected process nodes and their immediate neighbors are extracted together to generate a minimum perturbation subgraph. The node size of the minimum perturbation subgraph is typically smaller than the full disjunction graph; the specific number of nodes can be determined based on the number of affected processes, the neighborhood expansion depth, and the preset maximum number of nodes.

[0055] Within the minimum disturbance subgraph, the fitness function is set as the sum of the absolute values ​​of the start time deviations of the newly generated schedule and the initial production plan for indirectly affected processes. Feasibility is verified by combining the delivery date constraints, resource availability constraints, and process sequence constraints of the affected processes to ensure that the selected repair plan reduces the timing impact on indirect parts while meeting the actual execution conditions in the workshop. In this embodiment, the start time deviation metric is used uniformly for scheme comparison. If the completion time deviation or total order delay needs to be used as an evaluation indicator, it should be defined separately as another alternative implementation method and not mixed with the start time deviation metric for comparison.

[0056] The optimization module of the SciPy library is called in conjunction with a heuristic rescheduling algorithm to try four single repair operators in turn, and corresponding candidate scheduling schemes are generated for each: triggering the process right shift operator, keeping the original processing order unchanged, and shifting and postponing the start time of the affected process and the subsequent adjacent processes to the right according to the machine failure recovery time; triggering the resource reallocation operator, searching for idle alternative machines with the same processing capacity in the workshop, re-determining the processing time according to the process capacity parameters of the alternative equipment, and using the shortest path algorithm to evaluate the logistics transportation time caused by the change of equipment location, and modifying the machine mapping attributes of the process node; triggering the sequence exchange operator, generating candidate exchange sequences or local permutations and combinations within the allowed time window and evaluating their legality; for the newly added undirected disjunction arcs generated by emergency orders, the bottleneck index orientation rule consistent with S2 is used for orientation; when the bottleneck index or direction evaluation value of multiple candidate directions is the same, the shortest processing time (SPT) rule or the earliest delivery date (EDD) rule can be used as the parallel decision rule, and the branch and bound algorithm not described in the expanded description is no longer called separately.

[0057] After the newly added undirected disjunction arc is oriented, the scheduling system performs directed cycle detection. If a directed cycle is detected, the current orientation direction is canceled and another feasible direction is tried. If both the forward and reverse directions will lead to a directed cycle, the current orientation scheme is abandoned and the system rolls back to the previous candidate repair operator or re-executes the right shift of the process and resource reallocation to release the feasible time window, until a cycle-free candidate scheduling scheme is obtained or the scheme is marked as infeasible.

[0058] After the resource reallocation operator schedules a process to an alternative processing device, the scheduling system recalculates the processing time of that process on the alternative device, the logistics transportation time from the preceding process to the alternative device, and the logistics transportation time from the alternative device to the location of the successor process. It also updates the edge weights of the relevant conjunctive arcs or directed disjunctive arcs. Subsequently, starting from the reallocated process, forward time propagation is performed on the successor processes within the minimum perturbation subgraph to update the earliest start time, planned start time, and planned completion time of each successor process. After the update is completed, order delivery dates, equipment occupancy conflicts, and process sequence constraints are verified.

[0059] Comparing the local scheduling schemes generated by the four repair operators, the feasible repair scheme with the smallest fitness function value is selected. The minimum disturbance subgraph after optimal repair is re-integrated with the unaffected global graph, and the topology sorting algorithm is run to calculate the updated time parameters of each global process. The results are exported as JSON format and the updated power railway accessory production schedule is output.

[0060] As one possible implementation, when the production scheduling system receives a workshop production event, it identifies the set of processes affected by the event and defines a minimum disturbance subgraph containing the processes and adjacent processes in terms of processing timing and processing resource relationships. This includes: real-time monitoring of the workshop production event message queue; when an emergency insertion command is received, extracting the set of processes affected by the emergency insertion command; in the initial disjunction graph, finding the preceding and succeeding processes of each process in the set of processes based on timing constraints; in the initial disjunction graph, finding the preceding and succeeding processing processes adjacent to each process in the set of processes on the same processing equipment based on resource constraints; merging the affected set of processes, preceding processes, succeeding processes, and adjacent preceding and succeeding processing processes to extract a minimum disturbance subgraph, wherein the minimum disturbance subgraph is used to limit the local rescheduling range, and its number of nodes is adjusted according to a preset neighborhood depth, maximum number of nodes, or event impact range.

[0061] To address the unpredictable factors such as expedited orders or sudden equipment downtime commonly encountered in the power railway accessories workshop, a daemon process resides in the backend of the production scheduling system, constantly monitoring the workshop's production message middleware, such as Kafka or RabbitMQ message queues. When a high-priority (e.g., Priority=1) urgent insertion command for UHV power transmission tower accessories is detected, the kernel parser immediately extracts the set of affected processes from the command's payload data, such as the typing process containing expedited orders. and drilling process The graph analysis algorithm will locate the immediate predecessor and successor processes of the existing full disjunctive graph network by performing a one-hop bidirectional search along the directed conjunctive arc, with two nodes as the root nodes.

[0062] Based on this, the process route transcends the workpiece level and delves into the machine resource dimension for horizontal correlation. Based on the resource constraints determined by the machine's arc extraction, the target processing unit that will undertake the aforementioned urgent task is identified, such as CNC typewriter No. 1 and 3D drilling machine No. 3. Then, the pre-processing and post-processing operations originally planned to be executed on the two machines and whose time windows are closest to the urgent part are extracted from the existing timeline sequence.

[0063] The union operation operator of the graph computation library is invoked to merge three main categories of nodes: the affected expedited processes themselves, the immediate / immediate preceding / following processes on the vertical process route, and the preceding / following processes on the horizontal equipment queue, and deduplication is filtered out. A local miniaturized network structure composed of the nodes and their internally connected logical edges is successfully extracted as a minimum perturbation subgraph. The node size of this subgraph is usually significantly smaller than that of the full disjunctive graph, and can preferably be controlled within the range of 5 to 15 nodes. When there are many affected processes, truncation, layering, or batch processing can be performed according to the preset expansion depth or maximum number of nodes to ensure that the system can complete the local rescheduling calculation in a short time, avoiding the increased computational overhead and decreased plan stability caused by a single event triggering a global plan overhaul and rearrangement.

[0064] As one possible implementation, within the minimum disturbance subgraph, with the goal of minimizing the timing impact on the originally unaffected parts, a new feasible schedule is generated for the event by executing repair operators, including right shifting of processes, resource reallocation, sequence exchange within allowed time windows, and orienting newly added unoriented disjunction arcs. This includes: extracting the processes and time constraint parameters contained within the minimum disturbance subgraph; executing right shifting operators on affected processes to postpone the start time of the processes to avoid resource conflicts, and verifying the delivery constraints, subsequent process timing constraints, and equipment availability constraints after the right shift; and finding replaceable idle processing equipment in the workshop to address the impact. The process execution resource reallocation operator is used to schedule the process to idle processing equipment, wherein the idle processing equipment must meet the processing capacity, tooling, process parameters and quality requirements of the corresponding process; within the time limit parameters, the sequence exchange operator is executed to adjust the processing order between processes on the same processing equipment; the orientation operator is executed on newly added unoriented disjunction arcs to determine the processing order of the processes, and directed cycle detection is performed after orientation to avoid the formation of circular dependencies; the time deviation between each new scheduling scheme generated after the execution of the repair operator and the original plan non-affected part is calculated; the feasible new scheduling scheme with the smallest time deviation is selected as the feasible scheduling and updated to the initial disjunction graph.

[0065] After defining the minimum disturbance subgraph, the internal scheduling engine extracts the time constraint parameters of each affected node from this local graph, including the original planned start and stop times of each process and the maximum tolerable delay grace, such as preset grace variables. Minutes. Faced with time overlaps and resource conflicts caused by expedited orders, the scheduling engine sequentially executes four types of single repair operators: right shift of processes, resource reallocation, sequence exchange, and new disjunction arc orientation, and forms candidate scheduling schemes for each, without adopting a combined mutation method with an unlimited search range.

[0066] The process right shift operator shifts the originally scheduled task to the right on the timeline. For example, the angle steel punching task originally scheduled to start at 14:00 can be postponed to 14:45, while meeting the constraints of subsequent processes and delivery dates, to make way for urgent orders. Simultaneously, the resource reallocation operator calls the workshop equipment IoT real-time status pool via API. If the system detects that the second standby punch press of the same model is in standby idle state, it immediately binds the displaced process task to this idle processing equipment. After binding, the system recalculates the processing time of the process on the second standby punch press, the logistics transportation time from the preceding position to the second standby punch press, and the logistics transportation time from the second standby punch press to the immediately following process position. The system then propagates the changed time parameters forward to the immediately following processes, re-verifying the planned start time, planned completion time, and order delivery date of each immediately following process.

[0067] Furthermore, while ensuring that the 30-minute grace period does not exceed the limit, the sequence exchange operator is used to partially swap the order of existing processes with sufficient relaxation time in the same equipment queue. For newly added undirected disjunctive arcs obtained after expedited orders are initiated, the orientation operator uses the bottleneck index orientation rule consistent with S2 to determine the preceding and following order logic. When the evaluation values ​​of candidate directions are the same, the SPT rule or EDD rule is used as the parallel decision rule, and directed cycle detection is performed after orientation. If a circular dependency is detected, the current direction is canceled and another direction is tried; if the other direction still forms a circular dependency, the candidate solution is marked as infeasible, and the process is reverted to right shift, resource reallocation, or the sequence exchange operator to regenerate candidate solutions, ensuring that the updated graph structure can perform topological sorting and time propagation calculations.

[0068] After processing by the operators, the scheduling algorithm library generates multiple candidate new scheduling schemes. Using a constructed evaluation function, the sum of the absolute values ​​of the start time deviations of the indirectly affected processes relative to the original global baseline plan is calculated for each new scheme, and verified in conjunction with delivery constraints and resource feasibility. For example, in scheme A, the cumulative start time deviation of the indirectly affected processes is 120 minutes, in scheme B it is 40 minutes, and in scheme C it is 10 minutes. Based on this, a greedy criterion is used to select scheme C, which has the smallest time deviation and meets the constraints, as the target feasible scheduling scheme, and it is updated and integrated into the initial disjunctive graph, completing the smooth reconstruction of the production plan. A comparison of the candidate scheduling schemes generated by different repair operators with the cumulative start time deviations of the indirectly affected parts of the original plan is shown below. Figure 4 As shown, the time-series impact of each scheme on the non-influenced parts of the original plan is represented, and the target repair strategy is selected.

[0069] Experimental conditions: The test set was 30 days of production data from a large power equipment manufacturing enterprise's UHV tower accessory production workshop. The test environment was deployed on an industrial edge server based on an 8-core processor and 64GB of memory. The workshop's lower level contained 50 CNC machining machines and 5 large hot-dip galvanizing tanks. The daily average production order processing was set at 200 batches, with five high-priority emergency order insertions and two sudden equipment shutdown failures randomly injected daily. The comparative schemes were divided into a control group based on a global reordering production model using traditional heuristic algorithms, an ablation group removing the minimum perturbation subgraph local repair module, and the complete group of this application, which includes virtual batch aggregation and a minimum perturbation subgraph multi-operator repair mechanism.

[0070] Specific experimental data and results: In the control group, the average time for a single production scheduling recalculation when dealing with disturbances was 152.3 seconds, the overall equipment utilization rate was 65.4%, and the average order delivery delay caused by the sudden event was 310 minutes. In the ablation group, after adopting global rescheduling but retaining the virtual batch aggregation calculation module, the time for a single production scheduling calculation reached 168.5 seconds, the overall equipment utilization rate increased to 82.1%, and the average order delivery delay decreased to 240 minutes. In the complete group of this application, after the event is triggered, the local disturbance network is extracted, the time for a single local rescheduling calculation is reduced to 0.35 seconds, the overall equipment utilization rate is stabilized at 84.3%, and the average order delivery delay is compressed to 42 minutes.

[0071] In this invention, the virtual batch aggregation mechanism utilizes the minimum three-dimensional bounding box volume data and the spatial loading porosity compensation coefficient to achieve unified aggregation and scheduling of similar processes in multiple batches. This reduces the situation where batch processing processes are broken down into multiple independent scheduling objects, thereby improving the actual loading utilization rate of batch processing equipment. The minimum disturbance subgraph algorithm, combined with process right shift and resource reallocation repair operators, isolates scheduling repair actions within the dimensions of the affected nodes and adjacent preceding and following resources, preventing the chain reaction of a single sudden event triggering a global plan overhaul and rescheduling. This mechanism enables the scheduling engine to have sub-second or short-term local reconstruction capabilities, reducing the timing impact on non-disturbance area plans and reducing delivery delay time. This invention also discloses a cloud-based intelligent generation system for power railway accessory scheduling, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a cloud-based intelligent generation method for power railway accessory scheduling according to this invention.

[0072] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0073] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise expressly and specifically defined.

Claims

1. A cloud computing-based intelligent generation method for power railway accessory production scheduling, characterized in that, include: An initial disjunctive graph is constructed to represent the processing steps and their sequential constraints in the production of electric railway accessories. For batch-processing steps, virtual batch aggregation nodes are set to uniformly schedule multiple identical steps belonging to the same production batch, and a batch saturation parameter is assigned to each virtual batch aggregation node. The weight of the conjunctive arc connecting two adjacent steps in the initial disjunctive graph is calculated. During the iterative optimization of the initial disjunctive graph, the bottleneck index of all undirected disjunctive arc pairs is calculated, and the disjunctive arc with the largest bottleneck index is preferentially selected for orienting. Based on the orienting results, the processing sequence of each processing step is determined. After the dissociated arcs are oriented, an initial production schedule for power railway accessories is generated and output. When the scheduling system receives a workshop production event, it identifies the set of processes affected by the event and defines a minimum disturbance subgraph containing the processes and the processes adjacent to them in terms of processing sequence and processing resources. Within the minimum disturbance subgraph, with the goal of minimizing the timing impact on the non-influenced parts of the original plan, a new feasible schedule is generated for the event by executing repair operators, including right shifting of processes, resource reallocation, sequence exchange within the allowed time window, and orienting newly added unoriented dissociated arcs, and an updated production schedule for power railway accessories is output.

2. The method according to claim 1, characterized in that, For batch processing operations, setting up virtual batch aggregation nodes includes: For the pending processing plan of electric railway accessories, the processes that are surface treated on the same processing equipment and belong to the same manufacturing order batch are grouped into the corresponding batch processing process set. Among them, batch processing processes include hot-dip galvanizing, heat treatment, shot blasting or other processes that need to be completed in the same batch processing equipment. In the production scheduling system, a virtual batch aggregation node containing proprietary identification information is created. Each process task in the batch processing process set is bound to the corresponding virtual batch aggregation node, and a batch dependency mapping relationship is established between the virtual batch aggregation node and each process task. The batch dependency mapping relationship is used to represent the same batch processing relationship and does not represent a new processing sequence constraint on its own.

3. The method according to claim 2, characterized in that, Setting a batch saturation parameter for each virtual batch aggregation node includes: reading the minimum three-dimensional bounding box volume data of the physical materials corresponding to each process in the batch processing process set, and performing cumulative processing in combination with a preset spatial loading porosity compensation coefficient to obtain the merged total physical volume.

4. The method according to claim 3, characterized in that, After obtaining the combined physical total volume, the process also includes: Retrieve the maximum loading space volume of the corresponding processing equipment; Divide the total combined physical volume by the maximum loading space volume to obtain the volume percentage value; The volume percentage value is used as a batch saturation parameter and assigned to the corresponding virtual batch aggregation node. When the volume percentage value is greater than 1, the corresponding batch is marked as an overloaded batch for subsequent batch splitting or re-merging.

5. The method according to claim 1, characterized in that, The calculation of the conjunction arc weight connecting two adjacent processes in the initial disjunction diagram includes: locating the preceding and following processes connected by the conjunction arc in the initial disjunction diagram; determining whether the processing equipment for the preceding process has been determined; if the processing equipment has been determined, reading the processing time constant value of the preceding process on that processing equipment; if the unique processing equipment has not yet been determined, reading the processing time corresponding to each candidate processing equipment in the available machine set of the preceding process, and determining the baseline processing time constant value according to the preset equipment selection rules; reading the one-way transportation time constant value required for the component to move to the location of the following process after the preceding process is completed; adding the processing time constant value or the baseline processing time constant value to the one-way transportation time constant value to obtain the total time constant value; and assigning the total time constant value as the conjunction arc weight to the corresponding conjunction arc.

6. The method according to claim 1, characterized in that, The calculation of the bottleneck index for all undirected disjunction arc pairs includes: traversing all undirected disjunction arc pairs in the initial disjunction graph that have resource usage conflicts and whose processing order has not yet been determined; for two processes associated with an undirected disjunction arc pair, performing forward calculation from the starting node of the initial disjunction graph to obtain the earliest start time of each process; performing reverse calculation from the ending node of the initial disjunction graph to obtain the latest start time of each process; calculating the difference between the latest start time and the earliest start time of each process as the relaxation time of each process; calculating the sum of the relaxation times of the two processes; calculating a value with the natural constant as the base and the negative of the product of the sum of relaxation times and the preset time normalization coefficient as the exponent as the relaxation index type bottleneck index, and assigning it to the corresponding undirected disjunction arc pair.

7. The method according to claim 1, characterized in that, The definition of the minimum perturbation subgraph, which includes processes and processes adjacent to each other in terms of processing timing and processing resources, includes: real-time monitoring of the workshop production event message queue; when an emergency insertion command is received, extracting the set of processes affected by the emergency insertion command; in the initial disjunction graph, finding the preceding and succeeding processes of each process in the process set based on timing constraints; in the initial disjunction graph, finding the preceding and succeeding processing processes adjacent to each process in the process set on the same processing equipment based on resource constraints; merging the affected set of processes, preceding processes, succeeding processes, and adjacent preceding and succeeding processing processes to extract the minimum perturbation subgraph, wherein the minimum perturbation subgraph is used to limit the local rescheduling range, and its number of nodes is adjusted according to the preset neighborhood depth, maximum number of nodes, or event impact range.

8. The method according to claim 1, characterized in that, The process of generating a new feasible schedule for the event includes: extracting each process and time constraint parameter contained in the minimum disturbance subgraph; executing a process right shift operator on the affected processes to postpone the start time of the processes to avoid resource conflicts, and verifying the delivery date constraints, subsequent process timing constraints, and equipment availability constraints after the right shift; finding replaceable idle processing equipment in the workshop, executing a resource reallocation operator on the affected processes, and scheduling the processes to idle processing equipment, wherein the idle processing equipment must meet the processing capacity, tooling, process parameters, and quality requirements of the corresponding process; executing a sequence exchange operator within the time constraint parameter's allowable range to adjust the processing order between processes on the same processing equipment; executing a direction operator on newly added undirected disjunctive arcs to determine the processing order of the processes, and performing directed cycle detection after directionation to avoid forming circular dependencies.

9. The method according to claim 8, characterized in that, The process of generating new feasible schedules for events and outputting updated power railway accessory production plans includes: Calculate the time deviation between each new scheduling scheme generated after executing the repair operator and the unaffected part of the original plan; The feasible new scheduling scheme with the smallest time deviation is selected as the feasible scheduling and updated to the initial disjunction graph.

10. A cloud computing-based intelligent generation system for power railway accessory production scheduling, characterized in that, It includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, it implements the intelligent generation method for power railway accessory scheduling based on cloud computing as described in any one of claims 1 to 9.

Images