Intelligent factory production optimization method and system based on AI scheduling

Abstract

The application provides an intelligent factory production optimization method and system based on AI scheduling, and belongs to the technical field of intelligent factory production management. First, a batch production order is disassembled to obtain an order task graph. A dynamic resource pool is constructed according to a real-time resource state of a factory. The order task graph is input into a pre-trained scheduling AI model. A plurality of initial scheduling paths are generated in combination with the dynamic resource pool. Conflict detection is performed on the initial scheduling paths. Conflict points are identified. Based on the conflict point types and influence ranges, an adaptive adjustment module is called to correct parameters. A final production optimization scheme is generated and imported into a factory execution system. The intelligentization and high efficiency of production scheduling are realized, and the production efficiency and benefits are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent factory production management technology, and more specifically, to an intelligent factory production optimization method and system based on AI scheduling. Background Technology

[0002] In the field of production management in traditional smart factories, production scheduling is a key link in ensuring efficient and orderly production. However, existing production scheduling methods have many limitations.

[0003] On the one hand, for processing batch production orders, traditional methods typically employ fixed process breakdowns, making it difficult to fully consider the complex pre-dependencies and parallel constraints between basic process units. This can easily lead to problems such as poor process integration and chaotic production flows during production, thereby affecting production efficiency and product quality.

[0004] On the other hand, in terms of resource management, traditional methods often construct static resource pools that cannot reflect the dynamic changes in resources such as production equipment clusters, material reserve sequences, and manpower allocation information in real time. The operating status of production equipment changes over time, material reserves decrease due to production consumption, and manpower allocation may be adjusted due to various unforeseen circumstances. Static resource pools cannot capture these changes in a timely manner, resulting in a lack of accuracy and real-time performance in production scheduling.

[0005] Furthermore, existing production scheduling methods lack effective conflict detection and adaptive adjustment mechanisms when faced with multiple initial scheduling paths. They cannot quickly and accurately identify and correct conflicts such as overlapping resource usage, violated process dependencies, and overload, resulting in production optimization schemes failing to meet actual production needs and reducing the overall production efficiency of the factory. Summary of the Invention

[0006] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide an AI-based intelligent factory production optimization method, the method comprising:

[0007] The batch production orders received by the smart factory are broken down into processes to obtain an order task map containing basic process units and their relationships. The relationships reflect the pre-dependencies and parallel constraints between basic process units.

[0008] A dynamic resource pool is constructed based on the real-time resource status of the factory. The dynamic resource pool includes production equipment clusters, material reserve sequences, and manpower configuration information. The status of the production equipment clusters is dynamically updated over the duration of operation.

[0009] The order task graph is input into a pre-trained scheduling AI model, and multiple initial scheduling paths are generated by combining the real-time status of the dynamic resource pool. Each initial scheduling path includes the correspondence between basic process units and resources and the execution sequence.

[0010] Conflict detection is performed on multiple initial scheduling paths to identify conflict points such as overlapping resource usage, violation of process dependencies, and overload. The conflict points include the basic process unit identifiers and associated resource information involved in the conflict.

[0011] Based on the type and scope of the conflict points, the adaptive adjustment module of the scheduling AI model is invoked to correct the parameters of the initial scheduling path, generating a final production optimization scheme that includes dynamic resource allocation rules and time series elastic intervals, and then importing the final production optimization scheme into the factory execution system.

[0012] In another aspect, embodiments of the present invention also provide an AI-based intelligent factory production optimization system, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0013] Based on the above, this embodiment of the invention constructs an order task graph containing basic process units and their relationships by accurately breaking down batch production orders into process steps. A dynamic resource pool is built based on the real-time resource status of the factory, reflecting the dynamic changes in production equipment, materials, and manpower in real time, thus improving the real-time performance and accuracy of scheduling. The order task graph is input into a pre-trained scheduling AI model, which, combined with the dynamic resource pool, generates multiple initial scheduling paths. This fully leverages the powerful computing and analytical capabilities of the AI ​​model, providing diverse scheduling schemes. Conflict detection is performed on the initial scheduling paths, promptly identifying issues such as overlapping resource usage, violation of process dependencies, and overload. Based on the conflict point type and its impact range, an adaptive adjustment module is invoked to correct parameters, generating a final production optimization scheme that includes dynamic resource allocation rules and time-series elastic intervals. This effectively solves the problem of insufficient conflict handling capabilities in traditional methods. Finally, the production optimization scheme is imported into the factory execution system, achieving production process optimization, improving production efficiency, and reducing production costs. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the execution flow of the AI-based intelligent factory production optimization method provided in an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of exemplary hardware and software components of the AI-based intelligent factory production optimization system provided in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an AI-based intelligent factory production optimization method according to an embodiment of the present invention. The following is a detailed description of the AI-based intelligent factory production optimization method.

[0017] Step S110: Decompose the batch production orders received by the smart factory into process steps to obtain an order task map containing basic process units and their relationships. The relationships reflect the prerequisite dependencies and parallel constraints between basic process units.

[0018] In the production of papermaking wire mesh, smart factories may receive batch production orders containing papermaking wire mesh of different specifications and materials. These orders need to be broken down into specific processes before subsequent production scheduling can be carried out. For example, for an order containing both polyester and nylon papermaking wire mesh, the specific processes required for the production of each type of wire mesh must first be clearly defined.

[0019] Step S111: Read the product type and quantity information in the batch production order, and call the corresponding standard process flow in the factory process database. The standard process flow includes the names and sequence of all processes required to complete the production of this product type.

[0020] From the received batch production orders, we extract the product types of polyester papermaking mesh and nylon papermaking mesh, along with their respective quantities. Then, by calling the process database through the factory's internal system, we obtain the standard process flow for polyester papermaking mesh. This standard process flow may include processes such as raw material screening, melt extrusion, weaving, shaping, and cutting, and these processes have a clear sequence. Similarly, we obtain the standard process flow for nylon papermaking mesh, although its processes may differ from those of polyester, for example, in aspects such as temperature control during melt extrusion.

[0021] Step S112: The standard process flow is broken down according to the number of products to obtain a single set of process sequences for each product. Each process in the single set of process sequences includes a process code, the type of processing equipment required, and a list of material consumption.

[0022] For polyester papermaking mesh, assuming a certain order quantity, the standard process flow is broken down into individual process sequences corresponding to the quantity. The raw material screening process within each individual process sequence might have the process code YL-001, requiring a raw material screening machine, and the material consumption list includes polyester raw materials and screening auxiliary materials. The individual process sequences for nylon papermaking mesh are similar, only the specific process codes, equipment types, and materials differ.

[0023] Step S113: Identify the process combinations that can be executed in parallel in a single process sequence, mark the process units that have the same equipment type requirements and no material dependence, and mark the process units as parallel candidate processes.

[0024] In a single process sequence of polyester papermaking mesh, each process is analyzed. For example, the raw material drying process after raw material screening and the additive preparation process cannot be paralleled if they require different types of equipment; however, if they are two different batches of raw material screening processes, both requiring raw material screening machines, and there is no material dependency between the two processes, i.e., the output of the previous screening process is not the input of the next screening process, then these two process units can be marked as parallel candidate processes.

[0025] Step S1131: Traverse each process unit in a single process sequence and extract the equipment type requirement field for each process unit. The equipment type requirement field includes the equipment function code and accuracy level requirements.

[0026] Iterate through each process unit in the single process sequence of polyester papermaking wire mesh, such as raw material screening, melt extrusion, and weaving. For the raw material screening process unit, extract its equipment type requirement field. The equipment function code may be SCX-01 (indicating screening function), and the accuracy level requirement is P1 (indicating a certain screening accuracy). The equipment function code for the melt extrusion process unit may be RCJ-02 (indicating melt extrusion function), and the accuracy level requirement is P2.

[0027] Step S1132: Group the process units according to the equipment type requirement field, and group process units with completely identical equipment function codes and accuracy level requirements into the same equipment requirement group.

[0028] Based on the extracted equipment type requirement field, the process units in a single process sequence for polyester papermaking wire mesh are grouped. For example, all raw material screening process units with equipment function code SCX-01 and precision level requirement P1 are grouped into one equipment requirement group; melt extrusion process units with equipment function code RCJ-02 and precision level requirement P2 are grouped into another equipment requirement group.

[0029] Step S1133: Within each equipment requirement group, analyze the material dependencies between process units and query the input bill of materials and output bill of materials for the process unit.

[0030] Within the equipment requirements group of the raw material screening process unit, examine the input and output bills of materials for each process unit. The input bill of materials for one raw material screening process unit might be unscreened polyester raw materials, while the output bill of materials might be screened polyester raw materials. Another raw material screening process unit in the same group might have an input of another batch of unscreened polyester raw materials and an output of the corresponding screened polyester raw materials. If there is no overlap between the inputs and outputs of these two process units, then they may have no material dependency.

[0031] Step S1134: When the input bills of materials of two process units have no overlap and the output bills of materials have no overlap, and there is no sequential constraint between the two in the standard process flow, it is determined that the two process units have no material dependency.

[0032] In the equipment requirement group of the above raw material screening process unit, if the input materials of the two process units come from different batches of unscreened polyester raw materials, and the output is the screened polyester raw materials of their respective units, and the input material list and output material list have no overlap, and the order of these two process units is not specified in the standard process flow, then it can be determined that these two process units have no material dependence.

[0033] Step S1135: Combine all process units within the equipment requirement group that are not material-dependent into process combinations that can be executed in parallel, and assign a unique parallel group identifier to each process combination.

[0034] The aforementioned raw material screening process units, which are independent of material requirements, are grouped into process combinations that can be executed in parallel, and each combination is assigned a unique parallel group identifier, such as BX-001. In this way, during production scheduling, the process units within this process combination can be performed simultaneously.

[0035] Step S1136: Further check the maximum simultaneous processing capacity of the equipment type within the process combination. Determine the upper limit of the number of process units that can be executed simultaneously within the process combination based on the total number of equipment of the corresponding equipment type in the smart factory. Mark the process units that meet the maximum simultaneous processing capacity as parallel candidate processes. Add the parallel group identifier and the maximum parallel quantity parameter to the attribute information of the process unit.

[0036] Check the total number of raw material screening machines in the smart factory. Assuming there are several machines, the maximum number of raw material screening process units that can be executed simultaneously within this process combination is determined by this total number. If the total number of machines is 3, then a maximum of 3 raw material screening process units can be executed simultaneously within this combination. Mark the process units that meet the above conditions as parallel candidate processes, and add the parallel group identifier BX-001 and the maximum parallel quantity parameter 3 to their attribute information.

[0037] Step S114: Analyze the material transfer relationship between process units in a single process sequence. When the output material of the current process is the input material of the subsequent process, establish a prerequisite dependency relationship between the two and record the dependency strength parameter.

[0038] In a single process sequence of polyester papermaking mesh, the output material of the melt extrusion process is molten polyester material, while the input material of the weaving process is the same molten polyester material. Therefore, there is a material transfer relationship between the melt extrusion process and the weaving process, and it is necessary to establish a prerequisite dependency relationship and record the dependency strength parameters.

[0039] Step S1141: Extract the output material code and output quantity of each process unit in a single process sequence to form an output material list.

[0040] For the melt extrusion process unit, the output material code may be R-001 (representing melted polyester material), and the output quantity is determined according to production needs, forming an output material list of [(R-001, output quantity)]. For the weaving process unit, the output material code is W-001 (representing the woven mesh semi-finished product), forming a corresponding output material list.

[0041] Step S1142: Extract the input material code and input quantity for each process unit to form an input material list.

[0042] The input material code for the weaving process unit is R-001, and the input quantity is related to the output quantity of the melt extrusion process, forming an input material list of [(R-001, input quantity)]. The input material code for the raw material screening process unit is Y-001 (indicating unscreened polyester raw material), forming its input material list.

[0043] Step S1143: Perform pairwise comparisons of the process units in a single process sequence to check whether there is any overlap between the output material list of the preceding process unit and the input material list of the subsequent process unit.

[0044] By comparing the melt extrusion process unit as the preceding process and the weaving process unit as the following process, it was found that both of their output material lists and input material lists contain R-001, indicating an overlap. However, when comparing the raw material screening process unit and the weaving process unit, their output and input material lists do not share any material codes, indicating no overlap.

[0045] Step S1144: When there is an intersection, calculate the quantity matching degree of the intersecting materials, that is, the ratio of the output quantity of the preceding process unit to the input quantity of the subsequent process unit.

[0046] In the example of melt extrusion and weaving processes, the intersecting material is R-001. The ratio of the output quantity of the melt extrusion process to the input quantity of the weaving process is calculated, and this ratio is the quantity matching degree.

[0047] Step S1145: Divide the dependency level according to the value range of the dependency strength parameter. When the ratio reaches a perfect match, the dependency level is strong dependency; when the ratio is in the partial match range, the dependency level is weak dependency.

[0048] Set the range of values ​​for the dependency strength parameter. If the ratio of the above quantity matching degree is 1 (complete match), the dependency level between the melt extrusion process and the weaving process is strong dependency; if the ratio is between 0 and 1 (partial match), it is weak dependency.

[0049] Step S1146: Establish a directed pre-dependency relationship between the preceding process unit and the subsequent process unit. The relationship attributes of the pre-dependency relationship include dependency level, intersection material code and quantity matching degree.

[0050] Establish a preceding dependency relationship in the direction from the melt extrusion process unit to the weaving process unit. The relationship attributes include dependency level (such as strong dependency), intersection material code R-001, and the calculated quantity matching degree.

[0051] Step S1147: For process units with multiple path dependencies, record the preceding dependencies and corresponding dependency strength parameters with each preceding process unit. The process unit with multiple path dependencies is used to indicate that the input material of a process unit comes from multiple preceding process units.

[0052] If the setting process of papermaking wire mesh receives input materials not only from the output of the weaving process but also from the output of another coating process, then the setting process is a process unit with multi-path dependencies. In this case, it is necessary to record the prerequisite dependencies between the setting process and the weaving process, and between the setting process and the coating process, as well as their respective dependency strength parameters.

[0053] Step S115: Summarize the process units of all single process sequences into a basic process unit set. Each basic process unit contains the original process code, the order identifier, and the product serial number.

[0054] The process units in all individual process sequences of polyester and nylon papermaking mesh are summarized to form a basic process unit set. Each basic process unit, such as the raw material screening process of a certain polyester mesh, includes information such as the original process code YL-001, the order identifier DD-001, and the product serial number CP-001-001.

[0055] Step S116: Using basic process units as nodes and prior dependencies and parallel constraints as edges, construct an order task graph. The attributes of the edges in the order task graph include dependency strength parameters and the upper limit of parallel execution. The upper limit of parallel execution is determined by the available number of equipment types.

[0056] Using the aggregated basic process units as nodes, and the previously established prerequisite dependencies and parallel constraints as edges, an order task graph is constructed. For example, the edge between the melt extrusion process node and the weaving process node has attributes including a dependency strength parameter indicating strong dependency and an upper limit for parallel execution determined by equipment limitations; while the edge between nodes composed of parallel candidate processes has attributes including the corresponding dependency strength parameter (which may be non-dependent) and an upper limit for parallel execution.

[0057] Step S120: Construct a dynamic resource pool based on the real-time resource status of the factory. The dynamic resource pool includes production equipment clusters, material reserve sequences, and manpower configuration information. The status of the production equipment clusters is dynamically updated over time.

[0058] In a papermaking fabrication plant, it is necessary to monitor the status of various resources in real time to build a dynamic resource pool. The operating status of production equipment such as raw material screening machines, melt extruders, and weaving machines, the inventory of materials such as polyester raw materials, nylon raw materials, and various additives, as well as the on-duty status and skill level of workers are all components of the dynamic resource pool, and the status of production equipment changes over time.

[0059] Step S121: Connect to the factory equipment management system and collect real-time operating data of all production equipment. The real-time operating data includes the current status of the equipment, the running time, the cumulative processing quantity, and the time of the most recent maintenance.

[0060] The system connects to the factory equipment management system via a system interface to collect real-time operating data of the raw material screening machine, including whether it is currently running, stopped, or in a faulty state, the running time since the last start, the cumulative amount of raw materials processed so far, and the time of the last maintenance. The system also collects real-time operating data of other production equipment such as melt extruders and weaving machines.

[0061] Step S122: Based on the real-time operating data, the production equipment is divided into available status, maintenance-pending status, and fault status. The production equipment in the available status is categorized by functional type to form a production equipment cluster. Each production equipment cluster contains the identifier of the same type of equipment and its current load rate.

[0062] Based on the collected real-time operating data, the raw material screening machine is classified as available if it is currently operating normally, without faults, and before its maintenance period. If the machine's operating time has reached its maintenance cycle and maintenance is required but it has not yet been shut down, it is classified as awaiting maintenance. If the machine malfunctions and cannot operate normally, it is classified as faulty. All raw material screening machines in available status are grouped into a production equipment cluster. This cluster includes the identifier of each raw material screening machine, such as SB-001, SB-002, etc., as well as the current load rate of each machine.

[0063] Step S1221: Analyze the real-time operating data of the production equipment, extract the equipment status code, and determine the equipment status as available when the status code indicates availability, as pending maintenance when the status code indicates maintenance, and as faulty when the status code indicates a fault.

[0064] Analyze the real-time operating data of the melt extruder to extract the equipment status code. A status code of "01" indicates the equipment is available; "02" indicates it requires maintenance; and "03" indicates a malfunction. Based on the extracted status code, determine the current status of the melt extruder.

[0065] Step S1222: For production equipment in an available state, read its function type code. The function type code consists of the equipment's processing method, processing material, and precision level. Equipment with the same code is classified into the same function type.

[0066] For a weaving machine in usable condition, read its function type code, assuming the code is "BZ-PB-P1", where "BZ" indicates the weaving process, "PB" indicates the papermaking mesh materials it can handle, and "P1" indicates the precision level. All weaving machines with the function type code "BZ-PB-P1" are classified as the same function type.

[0067] Step S1223: Group the production equipment in available status according to the functional type code, and each group forms a production equipment cluster. Assign a unique cluster identifier to each production equipment cluster.

[0068] According to the functional type coding, all raw material screening machines coded as "SC-PC-P1" (screening-polyester-precision 1) are grouped into one group to form a production equipment cluster, and the cluster is assigned the cluster identifier JQ-001; the melt extruders coded as "RC-PB-P2" (melting-polyester-precision 2) are grouped into another group to form a production equipment cluster, and the cluster is identified as JQ-002.

[0069] Step S1224: Calculate the current load rate of each production device by the ratio of the total duration of tasks assigned to the device to the maximum working duration of the device. The larger the ratio, the higher the current load rate.

[0070] For a raw material screening machine, calculate the total duration of its assigned tasks. Assuming the maximum working time of the equipment is the daily working time (excluding necessary rest and maintenance time), the current load rate is the total duration of assigned tasks divided by the maximum working time. The larger this ratio, the heavier the current load on the equipment.

[0071] Step S1225: Record the identifier, current load rate, and most recently processed product type of each device in the production equipment cluster. The product type is used to match the processing requirements of the process unit.

[0072] In the JQ-001 production equipment cluster (raw material screening machine cluster), the identifier of each piece of equipment, such as SB-001, its current load rate, and the type of product most recently processed, whether it is polyester raw material or nylon raw material, are recorded so that when allocating process units, they can be matched according to the type of raw material to be processed in the process.

[0073] Step S1226: Set a cluster load threshold for each production equipment cluster. When the average load rate of all equipment in the production equipment cluster exceeds the cluster load threshold, it is marked as a high-load cluster, and the allocation of new tasks is restricted during scheduling.

[0074] Set a cluster load threshold for the JQ-002 production equipment cluster (melt extruder cluster), assuming that this threshold is determined based on the equipment's performance and historical operating data. When the average load rate of all melt extruders in the cluster exceeds this threshold, it is marked as a high-load cluster. In subsequent scheduling processes, tasks requiring melt extrusion processes will be preferentially not assigned to this cluster.

[0075] Step S1227: Periodically check whether the equipment in the maintenance-waiting state and the fault state has been restored to availability. When the equipment status code changes to a status code indicating availability, add it to the corresponding production equipment cluster and update the load rate calculation result of the production equipment cluster.

[0076] Periodically check whether the knitting machines in the maintenance pending status have completed maintenance, and whether the knitting machines in the fault status have been repaired. When the status code of a knitting machine changes from "02" or "03" to "01", add it to the corresponding knitting machine production equipment cluster, recalculate the average load rate of all equipment in the cluster, and update the cluster's load rate information.

[0077] Step S123: Connect to the factory warehouse management system to obtain the current inventory quantity, storage location and entry time of various materials, and form a material reserve sequence. The material reserve sequence is sorted in descending order according to the material consumption frequency.

[0078] By connecting to the factory warehouse management system, the system obtains the current inventory quantity of polyester raw materials, their storage location (area A) in the warehouse, and their arrival time; it also obtains relevant information about nylon raw materials, stored in area B; and information about various additives. Based on the consumption frequency of these materials in the papermaking fabric production process, the material reserve sequence is sorted from high to low consumption frequency. For example, polyester raw materials, which have the highest consumption frequency, are listed first, while certain special additives, which have a low consumption frequency, are listed later.

[0079] Step S124: Extract the number of on-duty personnel, skill level, and current work tasks of each shift from the factory human resource management system to generate human resource allocation information. The skill level is associated with the type of operable equipment.

[0080] The system extracts the number of on-duty personnel in the knitting team, each person's skill level (e.g., beginner, intermediate, advanced), and their current work tasks, such as operating a specific knitting machine. Advanced skill level personnel can operate all models of knitting machines, intermediate skill level personnel can operate some models, and beginner skill level personnel can only operate basic models. After organizing this information, the system generates manpower allocation information for the knitting team, which specifies the types of equipment that personnel at different skill levels can operate.

[0081] Step S125: Set up a status update trigger mechanism for the dynamic resource pool. When the running time of the production equipment reaches a preset threshold, update its load rate parameter; when the material inventory quantity is lower than the safety line, mark an early warning icon in the material reserve sequence.

[0082] Preset thresholds for the operating time of production equipment are established; for example, the preset threshold for a melt extruder is a certain number of hours of continuous operation. When the operating time of the melt extruder reaches this threshold, the system automatically triggers a status update and recalculates its load rate parameters. Simultaneously, inventory safety lines are set for each material; for example, the safety line for polyester raw materials is a certain quantity. When the warehouse management system shows that the inventory quantity of polyester raw materials is lower than this safety line, a warning sign is marked on the polyester raw materials in the material reserve sequence, indicating that replenishment is needed.

[0083] Step S126: Establish an association index between production equipment clusters, material reserve sequences, and manpower configuration information. The association index is used to query the collaborative configuration relationship of equipment, materials, and manpower required to complete a specified process.

[0084] When establishing the association index, the raw material screening process is associated with the JQ-001 production equipment cluster (raw material screening machine cluster), polyester raw materials (in the material reserve sequence), and personnel with raw material screening machine operation skills (in the manpower configuration information). When it is necessary to query the resources required to complete the raw material screening process, this association index can be used to quickly find the corresponding equipment cluster, the required materials, and the appropriate manpower, and understand the collaborative configuration relationship between them.

[0085] Step S127: Store all information of the dynamic resource pool in a distributed database.

[0086] Detailed information about the production equipment cluster, including the identifier, status, and load rate of each device; inventory quantities and storage locations of material reserves; and personnel configuration information such as on-duty personnel and skill levels, are all stored in a distributed database. The distributed database provides efficient data access and storage capabilities, ensuring that the smart factory can quickly obtain the necessary resource information when scheduling production.

[0087] Step S130: Input the order task graph into the pre-trained scheduling AI model, and generate multiple initial scheduling paths in combination with the real-time status of the dynamic resource pool. Each initial scheduling path includes the correspondence between basic process units and resources and the execution sequence.

[0088] In the production of papermaking wire mesh, a pre-constructed order task graph, including the production processes and relationships of polyester and nylon wire mesh, is input into a pre-trained scheduling AI model. Simultaneously, the scheduling AI model acquires real-time resource status from a dynamic resource pool, such as the load status of each production equipment cluster, material inventory, and manpower allocation, thereby generating multiple initial scheduling paths. Each initial scheduling path specifies which basic process unit is operated by which equipment and which personnel, as well as the start and end times of that process.

[0089] Step S131: Convert the order task graph into structured data that can be processed by the scheduling AI model. The node information in the structured data includes the identifier of the basic process unit, the required equipment type and the bill of materials, and the edge information includes the dependency relationship type and constraint parameters.

[0090] The order task graph is transformed into data, with each basic process unit as a node. The node information includes the identifier of the process, such as YL-001 (raw material screening process), the required equipment type is raw material screening machine, and the bill of materials is polyester raw material. The edge information reflects the dependency relationship between nodes, such as the edge from melt extrusion process to weaving process. The dependency relationship type is strong dependency, and the constraint parameters include quantity matching degree, etc., making it structured data that the scheduling AI model can process.

[0091] Step S132: Extract the real-time status data of the dynamic resource pool and convert it into a resource feature vector. The resource feature vector includes the load distribution of the production equipment cluster, the inventory level of the material reserve sequence, and the skill matching degree of the human resource configuration information.

[0092] The system extracts the load distribution of production equipment clusters from the dynamic resource pool, such as the load rate distribution of each device in the JQ-001 cluster; the inventory level of material reserves, such as the proportion of polyester raw material inventory to the safety line; and the skill matching degree of human resource allocation information, such as the proportion of current on-duty personnel with knitting machine operation skills. This data is then converted into resource feature vectors and used as one of the inputs to the scheduling AI model.

[0093] Step S133: Input the structured data and resource feature vectors into the input layer of the scheduling AI model at the same time, and map the node information into a low-dimensional vector through the embedding layer, while preserving the correlation features between nodes.

[0094] The input layer of the scheduling AI model receives structured data and resource feature vectors from the order task graph. Next, the embedding layer processes the node information in the structured data, mapping each node to a low-dimensional vector. During this mapping process, the relationships between nodes can be preserved; for example, the strong dependency between melt extrusion process nodes and weaving process nodes will be represented as specific vector relationships in the low-dimensional vector.

[0095] Step S134: Call the path generation module of the scheduling AI model. The path generation module adopts a reinforcement learning framework to generate multiple sets of candidate scheduling paths with the goal of maximizing equipment utilization and minimizing process delay. Each set of candidate scheduling paths contains the correspondence between basic process units and production equipment.

[0096] The path generation module of the scheduling AI model is activated; this module is based on a reinforcement learning framework. In the scheduling of papermaking wire mesh production, the goal is to maximize the utilization of production equipment while minimizing the delay time of each process. The reinforcement learning framework, considering the order task graph and the dynamic resource pool state, attempts different matching methods between basic process units and production equipment, generating multiple sets of candidate scheduling paths. Each initial scheduling path clearly defines the correspondence between processes and equipment.

[0097] For example, step S1341: Initialize the state space of the reinforcement learning framework and generate initial state data. The initial state data includes the set of basic process units that have been completed, the current load status of the production equipment cluster, and the real-time inventory of the material reserve sequence. The initial state data serves as the initial input of the reinforcement learning agent.

[0098] When the reinforcement learning framework starts running, it first initializes the state space and generates initial state data. In the initial state, the set of completed basic process units is empty; the current load state of the production equipment cluster is the initial load rate of each device, such as 0% when it starts up; the real-time inventory of the material reserve sequence is the current actual inventory quantity. The above initial state data is input into the reinforcement learning agent as the starting point for the agent to begin learning and decision-making.

[0099] Step S1342: Define an action space based on the initial state data. The action space includes all possible combinations of production equipment, manpower and material resources to select suitable production equipment, manpower and material resources for unassigned basic process units. Each combination corresponds to a specific action. The parameter range of the action is determined by the current load status of the production equipment cluster and the real-time inventory of the material reserve sequence.

[0100] Based on the initial state data, define the action space. The action space encompasses all possible combinations of resources to select for all unassigned basic process units. For example, for the raw material screening process, select a raw material screening machine in the JQ-001 cluster, personnel with the corresponding skills, and a specified batch of polyester raw materials. Each combination is a specific action, and the parameter range of the action is restricted; for example, equipment with full load cannot be selected, and materials with insufficient inventory cannot be selected.

[0101] Step S1343: Set a reward function according to the action type in the action space. The input of the reward function is the state change data before and after the action is executed. The state change data includes the load change of the production equipment cluster, the delay time difference of the basic process unit, and the inventory change value of the material reserve sequence. The output of the reward function is the reward value corresponding to the state change data. When the action makes the load change tend to be balanced, the delay time difference decreases, and the inventory change value is within a reasonable range, a positive reward value is output; otherwise, a negative reward value is output.

[0102] A reward function is set up to evaluate the merits of an action. When an action is executed, if the load change of the production equipment cluster becomes more balanced (e.g., the load of equipment with high load rate decreases, while the load of equipment with low load rate increases); the delay time difference of basic process units decreases (i.e., the difference between the actual start time and the planned start time becomes smaller); and the inventory change value of the material reserve sequence is within a reasonable range (e.g., the amount of material consumed matches the production demand), the reward function will output a positive reward value. Conversely, if the action leads to a more unbalanced load, increased delay, or unreasonable inventory change, a negative reward value will be output.

[0103] Step S1344: The reinforcement learning agent receives the initial state data and action space, selects and executes the first action, generates the first state change data, inputs the first state change data into the reward function, and obtains the first reward value. The first state change data includes the load status update result of the production equipment cluster, the allocation status of the basic process unit, and the inventory update result of the material reserve sequence after executing the first action.

[0104] After acquiring initial state data and the action space, the reinforcement learning agent selects its first action, such as assigning a raw material screening process to the SB-001 device and a skilled worker in the JQ-001 cluster, and using a certain quantity of polyester raw material. After executing this action, first state change data is generated, including an increase in the load rate of the SB-001 device, the raw material screening process being marked as assigned, and a decrease in polyester raw material inventory. This data is then input into the reward function to obtain the first reward value.

[0105] Step S1345: Adjust the action selection strategy based on the first state change data and the first reward value. Adopt a greedy strategy that balances exploration and utilization. In the initial stage, randomly select the next action from the action space with a preset probability. Select the action with the highest historical reward value with the remaining probability. As the number of iterations increases, gradually increase the probability of selecting the action with the highest historical reward value.

[0106] Based on the first state change data and the first reward value, the agent adjusts its action selection strategy. In the initial stage, actions are randomly selected with a preset probability to explore different possibilities. The remaining probability is used to select the action with the highest historical reward value, utilizing existing experience. As the number of iterations increases, the probability of selecting the action with the highest historical reward value gradually increases, enabling the agent to select more actions that have proven to be better, thereby generating a better scheduling path.

[0107] Step S1346: Repeat the action selection and state update process. After each action is executed, new state change data is generated. The new state change data is input into the reward function to obtain a new reward value. The new reward value is added to the historical reward value to form a cumulative reward value until all basic process units are allocated and a complete scheduling path is generated. The complete scheduling path includes the execution order of all actions, the corresponding state change data, and the cumulative reward value.

[0108] The agent continuously repeats the process of selecting actions, executing actions, and updating its state. Each completed action generates new state change data, such as changes in equipment load or newly assigned tasks. This data is input into a reward function to obtain a new reward value, which is then added to the previous reward value. Once all basic task units have been allocated resources, a complete scheduling path is generated. This path records the execution order of all actions, the state changes resulting from each action, and the accumulated reward value.

[0109] Step S1347: Using the cumulative reward value of the complete scheduling path as an evaluation index, repeat the above steps to generate multiple sets of candidate scheduling paths with different cumulative reward values. The level of the cumulative reward value reflects the comprehensive performance score of the candidate scheduling path in terms of equipment utilization and process delay. The higher the cumulative reward value, the better the comprehensive performance score.

[0110] Using the cumulative reward value of the complete scheduling path as the evaluation criterion, steps S1341 to S1346 are repeated to generate multiple sets of candidate scheduling paths. Different paths have different cumulative reward values. Paths with higher cumulative reward values ​​indicate better overall performance in terms of equipment utilization and process delay, and are more likely to become high-quality initial scheduling paths.

[0111] Step S135: Assign execution timing information to each basic process unit. The execution timing information includes the planned start time and the planned end time. The planned end time of the preceding process is earlier than the planned start time of the subsequent process.

[0112] In the generated candidate scheduling paths, the execution sequence is determined for each basic process unit. For example, the planned start time of the raw material screening process (YL-001) is 8:00 AM on a certain day, and the planned end time is 10:00 AM; the planned start time of its subsequent melt extrusion process (RC-001) is 10:00 AM, and the planned end time is 12:00 PM, ensuring that the subsequent process only begins after the preceding process is completed.

[0113] Step S136: Perform preliminary screening on the generated multiple sets of candidate scheduling paths, eliminate candidate scheduling paths that violate resource load thresholds and parallel constraints, and retain a preset number of feasible candidate scheduling paths as initial scheduling paths.

[0114] Check whether any candidate scheduling paths have resource loads exceeding the cluster load threshold, such as the average load rate of the JQ-002 cluster in a certain path exceeding the set threshold; or whether they violate parallel constraints, such as the number of parallel candidate operations exceeding the maximum number of parallel operations. Paths with the above problems are removed, and a preset number of qualified paths are retained as the initial scheduling paths.

[0115] Step S140: Perform conflict detection on multiple initial scheduling paths to identify conflict points such as overlapping resource usage, violation of process dependencies, and overload. The conflict points include the identifiers of the basic process units involved in the conflict and the associated resource information.

[0116] In the initial scheduling path of papermaking wire mesh production, each initial scheduling path is checked to see if there is resource overlap where the same equipment is occupied by multiple processes in the same time period; if there are process dependency violations where the preceding process has not been completed but the subsequent process has started; and if the equipment cluster load exceeds the threshold. For the conflict points found, the basic process unit identifiers involved and related equipment, material and other resource information are recorded.

[0117] Step S141: Traverse each initial scheduling path, group the execution sequence of basic process units according to the production equipment identifier, check whether the same equipment is assigned multiple process units in the same time period, and mark it as resource occupation overlap conflict if it exists.

[0118] Traverse each initial scheduling path and group the basic process units under the same production equipment identifier according to their execution sequence. For example, check the allocation status of equipment SB-001 (raw material screening machine). If it is assigned to both process YL-001 and process YL-002 during the time period from 9:00 to 10:00, it is marked as a resource occupation overlap conflict.

[0119] Step S142: For resource occupation overlap conflicts, record the basic process unit identifier involved in the conflict, the corresponding production equipment identifier, and the overlapping time period information, wherein the time period information includes the conflict start time and conflict end time.

[0120] Regarding the resource overlap conflict of the aforementioned SB-001 equipment, the basic process unit involved is identified as YL-001 and YL-002, and the corresponding production equipment is identified as SB-001. The overlapping time period is from 9:00 to 10:00, with the conflict starting at 9:00 and ending at 10:00.

[0121] Step S143: Check whether the execution sequence of the basic process units in the initial scheduling path satisfies the preceding dependency relationship. If the planned end time of the preceding process is later than the planned start time of the subsequent process, mark it as a process dependency violation conflict.

[0122] Check the execution sequence of the melt extrusion process (RC-001) and the braiding process (BZ-001) in the initial scheduling path. If the planned end time of RC-001 is 12 o'clock and the planned start time of BZ-001 is 11 o'clock, the end time of the preceding process is later than the start time of the subsequent process, which does not satisfy the preceding dependency relationship. Then mark it as a process dependency violation conflict.

[0123] Step S144: For process dependency violation conflicts, record the identifiers of the preceding and following basic process units involved, their planned time information, and the dependency relationship type, whereby the dependency relationship type includes strong dependency or weak dependency.

[0124] For the process dependency violation conflict between RC-001 and BZ-001, the preceding basic process unit is recorded as RC-001 and the following one as BZ-001; the planned end time of RC-001 is 12:00 and the planned start time of BZ-001 is 11:00; the dependency relationship between the two is a strong dependency.

[0125] Step S145: Calculate the load rate change curve of the production equipment cluster in the initial scheduling path. When the load rate exceeds the cluster load threshold in any time period, it is marked as a load overload conflict.

[0126] Calculate the load rate change curve of the JQ-002 production equipment cluster (melt extruder cluster) in the initial scheduling path and observe the load rate in each time period. If the load rate of the cluster exceeds the set cluster load threshold between 2 PM and 4 PM, it is marked as a load overload conflict.

[0127] Step S146: For load overload conflicts, record the corresponding production equipment cluster identifier, overload time period, and overload magnitude, where the overload magnitude is the difference between the actual load rate and the cluster load threshold.

[0128] For the load overload conflict of the JQ-002 cluster, the production equipment cluster is recorded as JQ-002, the overload period is from 2 PM to 4 PM, and the overload range is the difference between the actual load rate during this period and the cluster load threshold.

[0129] Step S147: Classify and summarize all identified conflict points according to conflict type, and generate a conflict list for each group of initial scheduling paths. The conflict list contains detailed information about the conflict points and assessment information on their impact on the overall scheduling.

[0130] The identified resource overlap conflicts, process dependency violation conflicts, and load overload conflicts are categorized and summarized to generate a conflict list for each initial scheduling path. In addition to detailed information on each conflict point, the list can also assess the impact of each conflict on the overall scheduling. For example, a resource overlap conflict may only affect the execution of two processes, while a load overload conflict may affect multiple processes across the entire cluster.

[0131] Step S150: Based on the type and scope of the conflict points, the adaptive adjustment module of the scheduling AI model is invoked to correct the parameters of the initial scheduling path, generate a final production optimization scheme that includes dynamic resource allocation rules and time series elastic intervals, and import the final production optimization scheme into the factory execution system.

[0132] Based on the type and scope of impact of the conflict points in the conflict list, the adaptive adjustment module of the scheduling AI model will revise the initial scheduling path. For example, for resource overlap conflicts, equipment will be reallocated; for process dependency violation conflicts, the timing will be adjusted; for load overload conflicts, tasks will be diverted, etc. After revision, a dynamic resource allocation rule will be generated, such as alternative solutions for equipment failures, as well as the timing flexibility range for each process, such as the time range that can be advanced or delayed. The final production optimization plan will then be imported into the factory execution system to guide the production of papermaking wire mesh.

[0133] Step S151: Input the conflict list into the adaptive adjustment module of the scheduling AI model. The adaptive adjustment module calls the corresponding correction strategy according to the conflict type. For resource occupation overlap conflict, the resource reallocation strategy is adopted; for process dependency violation conflict, the timing adjustment strategy is adopted; and for load overload conflict, the task diversion strategy is adopted.

[0134] After the conflict list is input into the adaptive adjustment module, the module selects a correction strategy based on the conflict type. If there are resource overlap conflicts in the list, such as the conflict between SB-001 devices, the resource reallocation strategy is invoked; if there are process dependency violation conflicts, such as the conflict between RC-001 and BZ-001, the timing adjustment strategy is invoked; if there are load overload conflicts, such as the conflict between JQ-002 clusters, the task distribution strategy is invoked.

[0135] Step S152: For resource overlap conflicts, reselect equipment with a load rate lower than the set load rate within the same production equipment cluster, allocate the conflicting basic process unit to the new equipment, update the correspondence between the basic process unit and the resource, and adjust the execution sequence of related processes to adapt to the new equipment allocation.

[0136] In the papermaking wire mesh production scenario, if the SB-001 equipment (belonging to the JQ-001 raw material screening machine cluster) has resource overlap conflict, that is, the equipment is allocated two raw material screening basic process units in the same time period, namely YL-001-01 (corresponding to polyester raw material A batch screening) and YL-001-02 (corresponding to polyester raw material B batch screening).

[0137] The adaptive adjustment module queries the current load rate of each device in the same production equipment cluster JQ-001 and finds that the load rate of device SB-003 is significantly lower than that of other devices. At this point, the conflicting basic process unit YL-001-02 is reassigned to device SB-003, and the mapping relationship between basic process units and resources is updated, that is, the corresponding device for YL-001-02 changes from SB-001 to SB-003.

[0138] Due to a change in the equipment used in YL-001-02, the execution sequence of its subsequent associated melt extrusion unit RC-001-02 (which depends on the output material of YL-001-02) needs to be adjusted accordingly. Originally, YL-001-02 was scheduled to finish at T1 on SB-001. However, the processing rhythm of SB-003 differs from that of SB-001. Therefore, the planned finish time for YL-001-02 on SB-003 has been recalculated to T2. Accordingly, the planned start time for RC-001-02 has been adjusted from the original T1+Δt to T2+Δt (Δt being the preparation time for process connection) to ensure smooth process transitions.

[0139] Step S153: For process dependency violation conflicts, extend the planned end time of the preceding process or advance the planned start time of the subsequent process. When the problem cannot be solved by timing adjustment alone, split the preceding process into multiple sub-processes, allowing some sub-processes to be executed in parallel with the subsequent process. The split must meet the phased requirements of material output.

[0140] Taking the process dependency violation conflict between RC-001 (melt extrusion process) and BZ-001 (weaving process) as an example, the planned end time of RC-001 is T3, while the planned start time of BZ-001 is T4, and T3>T4, which violates the preceding dependency relationship.

[0141] First, try timing adjustments. If RC-001 still has room for extended processing, extend its planned end time to T5 (T5 < T4) so ​​that BZ-001 can start as originally planned. If RC-001 cannot be extended, check if BZ-001 can start earlier and advance its planned start time to T6 (T6 > T3).

[0142] When the aforementioned timing adjustment cannot be implemented, consider splitting RC-001 into two sub-processes, RC-001-1 and RC-001-2. RC-001-1 is responsible for producing the molten material initially required by BZ-001, while RC-001-2 processes the remaining material. After splitting, the planned end time of RC-001-1 is set to T7 (T7 < T4) to ensure that BZ-001 can start normally at T4. Simultaneously, RC-001-2 executes in parallel with BZ-001, and the output material of RC-001-2 can meet the material requirements of subsequent processing in BZ-001.

[0143] Step S154: For load overload conflicts, transfer the basic process units that exceed the load threshold to other production equipment clusters of the same type. If all clusters of the same type are under high load, activate the backup equipment cluster. The performance parameters of the backup equipment cluster must meet the processing requirements of the process unit.

[0144] The load overload conflict of the JQ-002 melt extruder cluster is as follows: the current average load rate of the cluster has exceeded the set threshold, and the cluster contains basic process units such as RC-002-01 to RC-002-05.

[0145] The query revealed that the current load rate of the JQ-003 melt extruder cluster of the same type is low. Therefore, the two basic process units RC-002-04 and RC-002-05 with higher load rates in JQ-002 were transferred to the JQ-003 cluster and assigned to the RC equipment in the cluster.

[0146] If cluster JQ-003 is also under high load, then activate the backup equipment cluster JQ-004 (backup melt extruder cluster). Before the transfer, check the performance parameters of the equipment in cluster JQ-004, such as melt temperature range and extrusion speed, and confirm that they meet the processing accuracy and efficiency requirements of process units RC-002-04 and RC-002-05. Then, transfer these two process units to cluster JQ-004.

[0147] Step S155: During the correction process, a timing flexibility range is set for each basic process unit. The timing flexibility range is used to represent the maximum extent to which the planned start time can be advanced and the maximum extent to which the planned end time can be delayed.

[0148] For the YL-001-01 basic process unit (raw material screening), based on the characteristics of the processed materials, the idle status of the equipment, and the buffer time of subsequent processes, the maximum advance of the planned start time is set to ΔT1, meaning that it can start as early as ΔT1 before the original planned start time; the maximum delay of the planned end time is set to ΔT2, meaning that it can be completed as late as ΔT2 after the original planned end time.

[0149] For the BZ-001 weaving process unit, considering the high requirement for continuity in the weaving process, the maximum advance of the planned start time is set to ΔT3 (ΔT3 < ΔT1), and the maximum delay of the planned end time is set to ΔT4 (ΔT4 < ΔT2), in order to avoid excessive impact on subsequent shaping processes.

[0150] The timing flexibility range information for each basic process unit will be recorded in the attributes of the scheduling path, serving as the basis for subsequent dynamic adjustments.

[0151] Step S156: Construct dynamic resource allocation rules, which include the selection order of alternative equipment when equipment fails, the priority allocation mechanism when materials are in short supply, and the emergency response process for manpower allocation.

[0152] The order of selecting alternative equipment in case of equipment failure: When a production equipment fails, such as the SB-002 raw material screening machine, the order of selecting alternative equipment is as follows: first, the equipment with the lowest load rate in the same production equipment cluster (such as SB-003); second, the equipment in other production equipment clusters of the same type (such as the JQ-005 raw material screening machine cluster); and finally, the equipment in the standby equipment cluster (such as JQ-006).

[0153] Material Shortage Priority Allocation Mechanism: When polyester raw materials are in short supply, the basic process units requiring polyester raw materials are prioritized based on the urgency of orders, the product's profit contribution rate, and production schedule. Priority is given to ensuring the material supply for process units such as YL-001-03 and YL-001-04 corresponding to urgent orders, followed by process units corresponding to orders with higher profit contribution rates.

[0154] Emergency response procedure for manpower allocation: When a weaving team experiences a temporary absence of staff, resulting in a manpower shortage, staff should first be allocated from other teams of the same skill level (such as auxiliary weaving teams). If this is not possible, cross-skill level allocation should be initiated, with senior skilled personnel temporarily taking on the work of junior skilled personnel. At the same time, the human resources management system should be notified to make corresponding adjustments to staff attendance and salaries.

[0155] Step S157: Integrate the modified scheduling path with the dynamic resource allocation rules and the time-series elastic interval to form the final production optimization plan. The final production optimization plan includes the final resource allocation results, execution sequence, and adjustment plans for various anomalies of all basic process units.

[0156] The revised scheduling path clarifies the final resource allocation for each basic process unit, such as YL-001-01 being allocated to SB-001, RC-001-02 being allocated to RC-003, and the planned start and end times for each process.

[0157] By integrating the dynamic resource allocation rules and time-series flexibility intervals with the aforementioned scheduling paths, a final production optimization plan is formed. This final production optimization plan includes corresponding adjustment contingency plans for various abnormal situations such as equipment failure, material shortages, and manpower shortages.

[0158] For example, when equipment SB-001 experiences a sudden malfunction, the contingency plan clearly stipulates that YL-001-01 will be transferred to SB-003 according to the order of alternative equipment selection, and its execution sequence will be adjusted according to the time-series flexibility range; when polyester raw materials are in short supply, the material supply for key processes will be guaranteed according to the priority allocation mechanism. After the final production optimization plan is generated, it will be imported into the factory execution system as the basis for the execution of papermaking wire mesh production.

[0159] Figure 2 The illustration shows exemplary hardware and software components of an AI-based scheduling smart factory production optimization system 100 that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the AI-based scheduling smart factory production optimization system 100 and to perform the functions in this application.

[0160] The AI-based intelligent factory production optimization system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the AI-based intelligent factory production optimization method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0161] For example, the AI-based intelligent factory production optimization system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the AI-based intelligent factory production optimization system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The AI-based intelligent factory production optimization system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0162] For ease of explanation, only one processor is described in the AI-based scheduling smart factory production optimization system 100. However, it should be noted that the AI-based scheduling smart factory production optimization system 100 of this application may also include multiple processors. Therefore, the steps executed by one processor as described in this application may also be executed jointly or individually by multiple processors. For example, if the processor of the AI-based scheduling smart factory production optimization system 100 executes steps A and B, it should be understood that steps A and B may also be executed jointly by two different processors or individually by one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0163] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned AI-based intelligent factory production optimization method is implemented.

[0164] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. An AI scheduling-based intelligent factory production optimization method, characterized in that, The method comprises: process decomposition of batch production orders received by the intelligent factory to obtain an order task graph comprising basic process units and associated relationships, the associated relationships embody pre-depending and parallel constraints between the basic process units; constructing a dynamic resource pool according to the real-time resource state of the factory, the dynamic resource pool comprising a production equipment cluster, a material reserve sequence and human resource configuration information, the state of the production equipment cluster being dynamically updated with running time; inputting the order task graph into a pre-trained scheduling AI model to generate multiple sets of initial scheduling paths in combination with the real-time state of the dynamic resource pool, each set of initial scheduling paths comprising the corresponding relationship between the basic process units and the resources and the execution time sequence; detecting conflicts in the multiple sets of initial scheduling paths to identify conflict points of resource occupation overlap, process dependency violation and load overrun, the conflict points comprising the basic process unit identifiers and associated resource information involved in the conflicts; based on the types and influence ranges of the conflict points, calling an adaptive adjustment module of the scheduling AI model to correct the parameters of the initial scheduling paths to generate a final production optimization scheme comprising resource dynamic allocation rules and time sequence elastic intervals, and importing the final production optimization scheme into a factory execution system. 2.The AI scheduling-based smart factory production optimization method of claim 1, wherein, The process decomposition of batch production orders received by the intelligent factory to obtain an order task graph comprising basic process units and associated relationships comprises: reading product type and quantity information in the batch production order, and calling a corresponding standard process flow in the factory process database, the standard process flow comprising all process names and sequences required to complete production of the product type; splitting the standard process flow according to the product quantity to obtain a single set of process sequences for each product, each process in the single set of process sequences comprising a process code, a required processing equipment type and a material consumption list; identifying combinations of processes that can be executed in parallel in the single set of process sequences, marking process units that have the same equipment type requirement and no material dependency, and marking the process units as parallel candidate processes; analyzing the material transfer relationship between the process units in the single set of process sequences, establishing a pre-depending relationship between the output material of the current process and the input material of the subsequent process when the output material of the current process is the input material of the subsequent process, and recording the dependency strength parameter; summarizing the process units of all single sets of process sequences into a basic process unit set, each basic process unit comprising an original process code, an order identifier and a product serial number; constructing an order task graph with the basic process units as nodes and the pre-depending relationship and parallel constraints as edges, the attributes of the edges in the order task graph comprising the dependency strength parameter and the parallel execution upper limit number, which is determined by the available number of equipment types. 3.The AI scheduling-based smart factory production optimization method of claim 2, wherein, The process of identifying combinations of processes that can be executed in parallel in the single set of process sequences, marking process units that have the same equipment type requirement and no material dependency, and marking the process units as parallel candidate processes comprises: traversing each process unit in the single set of process sequences, extracting the equipment type requirement field of each process unit, the equipment type requirement field comprising a device function code and an accuracy level requirement; Grouping the process units according to the equipment type requirement field, the process units with the same equipment function code and precision level requirement are grouped into the same equipment requirement group; In each equipment requirement group, analyzing the material dependency relationship between the process units, querying the input material list and output material list of the process units; When the input material list of two process units has no intersection and the output material list has no intersection, and there is no sequence constraint between the two in the standard process flow, it is determined that the two process units have no material dependency; All the process units without material dependency in the equipment requirement group are combined into a parallel process combination, and each process combination is assigned a unique parallel group identifier; Further checking the maximum simultaneous processing capacity of the equipment type in the process combination, determining the upper limit of the number of process units that can be simultaneously executed in the process combination according to the total number of corresponding equipment types in the intelligent factory, and marking the process units that meet the maximum simultaneous processing capacity as parallel candidate processes. The parallel group identifier and the maximum parallel number parameter are added to the attribute information of the process unit. 4.The AI scheduling-based smart factory production optimization method of claim 2, wherein, The analysis of the material transfer relationship between the process units in a single set of process sequence, when the output material of the current process is the input material of the subsequent process, a precedence dependency relationship is established between them, and the dependency strength parameter is recorded, including: Extracting the output material code and output quantity of each process unit in a single set of process sequence to form an output material list; Extracting the input material code and input quantity of each process unit to form an input material list; Comparing the output material list of the previous process unit with the input material list of the subsequent process unit in a single set of process sequence to check whether there is an intersection; When there is an intersection, calculate the quantity matching degree of the intersection material, that is, the ratio of the output quantity of the previous process unit to the input quantity of the subsequent process unit; According to the value range of the dependency strength parameter, the dependency level is divided, when the ratio reaches complete matching, the dependency level is strong dependency, when the ratio is in the partial matching interval, the dependency level is weak dependency; A directed precedence dependency relationship is established between the previous process unit and the subsequent process unit, and the dependency level, intersection material code and quantity matching degree are included in the relationship attribute of the precedence dependency relationship; For process units with multiple path dependencies, the precedence dependency relationship with each previous process unit and the corresponding dependency strength parameter are recorded respectively, wherein the process unit with multiple path dependencies means that the input material of a process unit comes from multiple previous process units. 5.The AI scheduling-based smart factory production optimization method of claim 1, wherein, The dynamic resource pool is constructed according to the real-time resource state of the factory, including: Connecting the factory equipment management system to collect real-time running data of all production equipment, the real-time running data including the current state of the equipment, the running time, the cumulative processing quantity and the last maintenance time; According to the real-time running data, the production equipment is divided into available state, maintenance state and fault state, and the production equipment in available state is classified according to function type to form a production equipment cluster, each production equipment cluster containing the identifier and current load rate of the same type of equipment; A connection factory warehouse management system is connected to obtain current inventory quantity, storage location and storage time of various materials, and a material storage sequence is formed, which is sorted in descending order according to material consumption frequency; The number of on-site personnel, skill level and current work task of each team are extracted from a factory human resource management system to generate human resource allocation information, and the skill level is associated with the type of operable equipment; A state update trigger mechanism is set for the dynamic resource pool, and when the running time of the production equipment reaches a preset threshold, the load rate parameter is updated; when the material inventory quantity is lower than the safety line, a warning mark is marked in the material storage sequence; An association index is established between the production equipment cluster, the material storage sequence and the human resource allocation information, and the association index is used to query the cooperative configuration relationship of the equipment, materials and human resources required to complete the specified process; All information of the dynamic resource pool is stored in a distributed database. 6.The AI scheduling-based smart factory production optimization method of claim 5, wherein, The production equipment is divided into available state, maintenance state and fault state according to the real-time running data, and the production equipment in available state is classified according to function type to form a production equipment cluster, including: The real-time running data of the production equipment is analyzed, and the device state code is extracted; when the device state code indicates availability, it is determined as available state; when the state code indicates maintenance, it is determined as maintenance state; and when the state code indicates fault, it is determined as fault state; For the production equipment in available state, the function type code of the equipment is read, which is composed of processing method, processing material and precision level; and the equipment with the same code is classified into the same function type; The production equipment in available state is grouped according to the function type code, and each group forms a production equipment cluster, and each production equipment cluster is assigned a unique cluster identifier; The current load rate of each production equipment is calculated, which is obtained by the ratio of the total duration of the assigned task to the maximum working duration of the equipment; the larger the ratio, the greater the current load rate; The identifier, current load rate and last processed product type of each device are recorded in the production equipment cluster, and the product type is used to match the processing demand of the process unit; A cluster load threshold is set for each production equipment cluster, and when the average load rate of all devices in the production equipment cluster exceeds the cluster load threshold, it is marked as a high load cluster, and the allocation of new tasks is preferentially limited during scheduling; Whether the devices in maintenance state and fault state are restored to available state is checked regularly, and when the device state code changes to an available state code, it is added to the corresponding production equipment cluster, and the load rate calculation result of the production equipment cluster is updated. 7.The AI scheduling-based smart factory production optimization method of claim 1, wherein, The order task graph is input into a pre-trained scheduling AI model to generate multiple initial scheduling paths in combination with the real-time state of the dynamic resource pool, including: The order task graph is converted into structured data that can be processed by the scheduling AI model, and the node information in the structured data includes the identifier of the basic process unit, the required equipment type and the material list, and the edge information includes the dependency type and the constraint parameter; Extracting real-time state data of the dynamic resource pool, converting it into a resource feature vector, which includes the load distribution of the production equipment cluster, the inventory level of the material reserve sequence, and the skill matching degree of the human resource allocation information; Inputting the structured data and the resource feature vector into the input layer of the scheduling AI model, mapping the node information into a low-dimensional vector through an embedding layer, and retaining the correlation characteristics between nodes; Calling the path generation module of the scheduling AI model, which uses a reinforcement learning framework to maximize equipment utilization and minimize process delays, to generate multiple groups of candidate scheduling paths; each group of candidate scheduling paths includes the correspondence between the basic process units and the production equipment; Assigning execution timing information to each basic process unit, which includes the planned start time and the planned end time, and ensuring that the planned end time of a previous process is earlier than the planned start time of a subsequent process; Performing preliminary screening on the generated multiple groups of candidate scheduling paths, eliminating candidate scheduling paths that violate resource load thresholds and parallel constraints, and retaining a preset number of feasible candidate scheduling paths as initial scheduling paths. 8.The AI scheduling based smart factory production optimization method of claim 1, wherein, The conflict detection on multiple initial scheduling paths identifies conflict points of resource occupation overlap, process dependency violation, and load overrun, including: Traversing each initial scheduling path, grouping the execution timing of the basic process units according to the production equipment identifier, and checking whether the same equipment is assigned multiple process units within the same time period. If so, it is marked as a resource occupation overlap conflict; For resource occupation overlap conflicts, record the basic process unit identifier involved, the corresponding production equipment identifier, and the overlapping time period information, which includes the conflict start time and the conflict end time; Check if the execution timing of the basic process units in the initial scheduling path satisfies the precedence dependency relationship. If the planned end time of a previous process is later than the planned start time of a subsequent process, it is marked as a process dependency violation conflict; For process dependency violation conflicts, record the identifiers of the involved previous and subsequent basic process units, their planned time information, and the dependency relationship type, which includes strong dependency or weak dependency; Calculate the load rate change curve of the production equipment cluster in the initial scheduling path. When the load rate in any time period exceeds the cluster load threshold, it is marked as a load overrun conflict; For load overrun conflicts, record the corresponding production equipment cluster identifier, the overrun time period, and the overrun amplitude, which is the difference between the actual load rate and the cluster load threshold; Classify and summarize all identified conflict points by conflict type, generate a conflict list for each initial scheduling path, and include detailed information of the conflict points and impact range evaluation information in the conflict list. 9.The AI scheduling based smart factory production optimization method of claim 1, wherein, Based on the type and impact range of the conflict points, call the adaptive adjustment module of the scheduling AI model to modify the parameters of the initial scheduling path, and generate a final production optimization scheme that includes resource dynamic allocation rules and timing flexibility intervals, including: The conflict list is input into an adaptive adjustment module of the scheduling AI model, which calls corresponding correction strategies according to the conflict type. The resource occupation overlap conflict adopts a resource reallocation strategy, the process dependency violation conflict adopts a timing adjustment strategy, and the load overrun conflict adopts a task shunting strategy. For the resource occupation overlap conflict, a device with a load rate lower than a set load rate is reselected within the same production device cluster, the basic process unit of the conflict is allocated to the new device, the correspondence between the basic process unit and the resource is updated, and the execution timing of the associated process is adjusted to adapt to the new device allocation. For the process dependency violation conflict, the planned end time of the preceding process is extended or the planned start time of the subsequent process is advanced. When timing adjustment alone cannot solve the problem, the preceding process is split into multiple sub-processes, allowing some sub-processes to execute in parallel with the subsequent process. Splitting must meet the phased requirements of material output. For the load overrun conflict, part of the basic process unit that exceeds the load threshold is transferred to other production device clusters of the same type. If all clusters of the same type are in a high-load state, a backup device cluster is enabled. The performance parameters of the backup device cluster must meet the processing requirements of the process unit. During the correction process, a timing flexibility interval is set for each basic process unit, which represents the maximum advance of the planned start time and the maximum delay of the planned end time. A resource dynamic allocation rule is constructed, which includes the replacement device selection order in the event of device failure, the priority allocation mechanism in the event of material shortage, and the emergency response process of human resource allocation. The corrected scheduling path is integrated with the resource dynamic allocation rule and the timing flexibility interval to form a final production optimization scheme, which includes the final resource allocation result, execution timing, and adjustment plan for each type of exception for all basic process units.

10. An AI scheduling based intelligent factory production optimization system, characterized in that, The device comprises a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to realize the AI scheduling-based intelligent factory production optimization method in any one of claims 1-9.

Images