AI-Driven Adaptive Production Scheduling Method for Cable Production Lines

Abstract

The present invention relates to an AI-driven adaptive production scheduling method for cable production lines, comprising the following steps: S1: Collect and preprocess production line data; S2: Based on the preprocessed production line data, establish a data-driven production process model; S3: Based on the constructed production process model, use the particle swarm optimization algorithm to generate an initial scheduling plan; S4: Based on the initial scheduling plan, assign tasks to specific workstations and equipment, and the production line executes the plan; S5: Monitor the execution process of tasks, collect process deviations; combined with deep reinforcement learning, adjust the scheduling according to the real-time production status. The present invention can generate an efficient scheduling plan in a complex multi-constrained environment, and through real-time monitoring of task execution and feedback of process deviations, realize dynamic adaptive scheduling optimization, so as to effectively adapt to emergencies and rapidly changing production requirements in cable production lines.

Description

Technical Field

[0001] The present invention relates to the field of intelligent scheduling, and particularly to an AI-driven adaptive production scheduling method for cable production lines. Background Art

[0002] In modern manufacturing, production scheduling is one of the core links to ensure the efficient operation of the production system. Especially in cable production lines, due to their complex technological processes, limited equipment resources, and diverse customer requirements (such as multiple varieties, small batches, and rapid delivery), higher requirements for scheduling optimization are put forward. In the actual production process, there are many challenges, such as fluctuations in equipment operating status, uncertainties in task execution time, dynamic changes in inventory usage, and the impact of environmental factors (such as temperature and humidity) on the production line performance. Therefore, traditional static scheduling methods often struggle to adapt to these dynamic changes, easily leading to low equipment utilization, production delays, or resource waste. Summary of the Invention

[0003] To solve the above problems, the purpose of the present invention is to provide an AI-driven adaptive production scheduling method for cable production lines, which can generate an efficient scheduling plan in a complex multi-constraint environment and achieve dynamic adaptive scheduling optimization by real-time monitoring of task execution and feedback of process deviations, thereby effectively adapting to emergencies and rapidly changing production requirements in cable production lines.

[0004] To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] An AI-driven adaptive production scheduling method for cable production lines, comprising the following steps:

[0006] S1: Collect production line data, including equipment operation data, production task data, inventory data, and environmental data, and preprocess it;

[0007] S2: Based on the preprocessed production line data, establish a data-driven production process model;

[0008] The production process model is a two-layer model architecture, including a task flow model and a resource status model, specifically as follows:

[0009] The task flow model dynamically describes the execution sequence, process requirements, and scheduling constraints of tasks, and is modeled using a directed acyclic graph. Each task is a node T of the graph a , and the dependency relationship between tasks is represented by an edge e ab ;

[0010] The resource status model describes the status of materials and equipment in the production process through resource consumption and allocation rules; regards resources as constraint conditions to limit the allocation and execution of tasks;

[0011] S3: Based on the constructed production process model, use the particle swarm optimization algorithm to generate an initial scheduling plan;

[0012] S4: Based on the initial scheduling plan, allocate tasks to specific workstations and equipment, and the production line executes the plan;

[0013] S5: Monitor the execution process of tasks, collect process deviations; combine deep reinforcement learning to adjust the scheduling according to the real-time production status.

[0014] Furthermore, collect production line data, including equipment operation data, production task data, inventory data, and environmental data, as follows:

[0015] Collect equipment operation data through the intelligent sensor integrated with the PLC interface, including equipment status, extrusion temperature, tension, line speed, equipment rotation speed, and power;

[0016] Integrate with the production management database through the MES system interface to obtain real-time production task data, including current order requirements, progress, and priority;

[0017] Integrate the intelligent warehousing system with the MES system to synchronize inventory data regularly, including raw material inventory and remaining consumables;

[0018] Collect environmental data in real time through environmental Internet of Things sensors, including workshop temperature, humidity, and air data.

[0019] Furthermore, perform preprocessing, as follows: Use moving average filtering to filter out instantaneous abnormal fluctuations, and introduce rule detection to find error values; use the mean filling interpolation method to repair missing values; align the periodic data collected by the PLC to avoid errors caused by different sampling rates, and use the NTP protocol to ensure time consistency to ensure that various types of data are synchronized based on a unified timestamp.

[0020] Furthermore, the task process model is as follows:

[0021] Model all production tasks as a directed acyclic graph G=(V,E), where:

[0022] is the task set, and each task represents a node;

[0023] is the task and the task the set of dependencies between them; represents the task is the task the prerequisite task of, ;

[0024] Acyclicity:

[0025]

[0026] That is, to avoid circular dependencies and ensure the legality of the topological order of the task flow;

[0027] Among them, task T a corresponds to a node in the DAG graph, and its attribute set includes:

[0028] , where are respectively the task end time, task start time, task duration, task resource set, and task priority of task ;

[0029] Task constraints include time constraints and resource constraints, specifically as follows:

[0030] , that is, the start time of the successor task must be later than or equal to the end time of its predecessor task;

[0031] , that is, the start time of task T a must be within its earliest start time and latest start time ;

[0032] , that is, at any time t, for the kth type of resource R k , the total resource usage of the running tasks cannot exceed the maximum resource capacity ; active(t) is the set of tasks being executed at time t; is the consumption of task for resource R k ;

[0033] If the resource supply decreases due to real-time changes and the remaining inventory R k (t) is lower than the demand, then adjust the start time of the affected tasks:

[0034] ;

[0035] where t replenish (k) represents the time point when resource R k is replenished.

[0036] Furthermore, S3 is specifically as follows:

[0037] Based on the constructed production process model, obtain the task set, resource set, dependency relationship set, and constraint conditions as the input of PSO;

[0038] The fitness function is to minimize the processing time and the total delay:

[0039] ;

[0040] where, is the delivery time of task T a ; C max is the total processing time; n represents the maximum number of tasks;

[0041] Each particle represents a scheduling scheme, including the task execution order and resource allocation; Particle X is represented by the vector [X T , X R , where X T is the task order and X R is the resource allocation scheme;

[0042] Initialize the particle swarm size N, the maximum number of iterations i ter , the speed range V min , V max , the learning factor c 1 , c 2 , and the inertia weight ω;

[0043] Randomly initialize the position X i (t) and the speed V i (t) to ensure that the initial values of the task order and resource allocation are reasonable;

[0044] Traverse each particle and calculate the fitness function;

[0045] Decode the particle position X i (t) into a scheduling scheme, including the task execution order and resource allocation scheme;

[0046] Convert the sorted X T into the task priority order;

[0047] Check whether the resource constraints and task time window constraints are satisfied;

[0048] Calculate the fitness function, and add a penalty term if the constraints are violated;

[0049] Update the state of each particle using the speed and position update formulas:

[0050] Speed update formula:

[0051] ;

[0052] Position update formula:

[0053] ;

[0054] Among them, is the historical optimal solution of particle i; is the global optimal solution of the population; r 1 , r 2 is a random factor;

[0055] Perform constraint checking on the updated particles for correction;

[0056] For the fitness of the current particle, if it is better than the historical optimal solution , then update the current particle to the historical optimal solution ; if the fitness of the current particle is better than the global optimal solution , update the current particle to the global optimal solution ;

[0057] If the maximum number of iterations is reached or the change in the global optimal value is less than the threshold, end the iteration and output the generated scheduling plan, that is, the initial scheduling plan.

[0058] Furthermore, S4 is specifically as follows:

[0059] Based on the initial scheduling plan, decompose the tasks to the specific workstations and equipment terminals according to the resource allocation results; ensure that the tasks meet the process requirements, equipment capabilities, and process constraints;

[0060] Use the MES system to decompose and send the scheduling plan to each workstation, equipment, or operator, and monitor the task progress, resource usage, and abnormal status through the MES system, and dynamically adjust the equipment and plan.

[0061] Furthermore, the said S5 is specifically as follows:

[0062] S51: Real-time track the execution status of tasks and collect process deviations as feedback data of the production system;

[0063] S52: Based on the real-time monitoring data, combined with the PPO algorithm, dynamically optimize the scheduling plan.

[0064] Furthermore, S51 is specifically as follows:

[0065] Through the MES system and factory Internet of Things devices, collect the following data in real time and real-time track the execution status of tasks:

[0066] Actual start time , actual end time , actual processing time , equipment M k 's real-time status S k (t), the said real-time status includes idle, running, and faulty, and the equipment failure probability ; remaining inventory Rk (t); Resource consumption rate ;

[0067] Collect process deviations as feedback data for the production system, including:

[0068] Processing time deviation , task delay and equipment failure time , where is the planned processing time, is the planned task start time.

[0069] Furthermore, S52 is specifically:

[0070] In scheduling optimization, the reinforcement learning problem is modeled as a Markov decision process, including the following elements:

[0071] State space S, including: the current task state , equipment state , resource state , the current task queue and dependencies ;

[0072] Action space A, including task assignment, task order adjustment, and handling of inserted tasks;

[0073] Reward function R: ;

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] where R 1 , R 2 , R 3 , R 4 are respectively minimizing the total project duration, minimizing task delay, maximizing equipment utilization, and penalizing equipment failures, w 1 , w 2 , w 3 , w 4 are weight coefficients; is the equipment working efficiency;

[0079] Initialize the policy network π θ (a∣s) and the value function ;

[0080] Take the initial scheduling plan as the basic input;

[0081] Under the current policy π θ Simulate the scheduling optimization process, and collect sequences of state s t , action a t and reward r t ; Update the policy network according to the following objective function:

[0082] ;

[0083] where θ represents the parameters of the current policy network; is the importance sampling ratio; is the advantage function; is the truncation threshold; is the clipping operation, indicating clipping the value of r t (θ) so that it is within the interval; is the expectation function; is the objective function;

[0084] ;

[0085] where Q(s t , a t ) is the total return after taking action a t in state s t ; V(s t ) is the state value in state s t ;

[0086] The loss function of the value function :

[0087] ;

[0088] where is the estimated value of the current value function network for state s t ; R t represents the actual target return value at time step t;

[0089] Repeat sampling, policy update, and value function update until the model converges to obtain the adjusted scheduling plan.

[0090] The present invention has the following beneficial effects:

[0091] 1. The present invention can generate an efficient scheduling plan in a complex multi-constraint environment, and realize dynamic adaptive scheduling optimization by monitoring task execution in real time and feedback process deviations, so as to effectively adapt to emergencies and rapidly changing production demands in the cable production line;

[0092] 2. The present invention constructs a data-driven production process model through a double-layer model, combining mathematical formulas and optimization objectives, which can dynamically adapt to real-time changes, improve production efficiency, provide scheduling optimization support for the cable production process, optimize the PSO scheduling algorithm based on the production process model, provide task logic (task flow model) and resource constraint (resource allocation model) information for PSO, define the legality and fitness evaluation criteria of PSO particle solutions, and ensure the real-time update of the production scheduling plan through dynamic feedback. Therefore, the PSO scheduling algorithm based on the production process model can generate efficient, flexible, and feasible scheduling plans.

[0093] 3. By real-time monitoring the task execution process and combining with the PPO algorithm, the present invention can dynamically adjust the production scheduling plan, solve problems such as processing time delay and equipment failure, and realize the intelligent and dynamic optimization of production scheduling by utilizing the self-adaptability of reinforcement learning and the real-time feedback ability of the MES system, effectively improving production efficiency and resource utilization rate, and enhancing the system's response ability to emergencies at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0095] The following further describes the present invention in detail with reference to the drawings and specific embodiments:

[0096] Refer to Figure 1 , in this embodiment, an AI-driven adaptive production scheduling method for a cable production line is provided, including the following steps:

[0097] S1: Collect production line data, including equipment operation data, production task data, inventory data, and environmental data, and preprocess it;

[0098] S2: Based on the preprocessed production line data, establish a data-driven production process model;

[0099] S3: Based on the constructed production process model, use the particle swarm optimization algorithm to generate an initial scheduling plan;

[0100] S4: Based on the initial scheduling plan, allocate tasks to specific workstations and equipment, and the production line executes the plan;

[0101] S5: Monitor the task execution process, collect process deviations (such as processing time delay or equipment status changes); combine with deep reinforcement learning (such as the PPO algorithm), and adjust the scheduling according to the real-time production status.

[0102] In this embodiment, the production line data is collected, including equipment operation data, production task data, inventory data, and environmental data, as follows:

[0103] Integrate the PLC interface through intelligent sensors to collect equipment operation data, including equipment status (startup, working, shutdown, fault), extrusion temperature, tension, line speed, equipment rotation speed, and power;

[0104] Integrate with the production management database through the MES system interface to obtain real-time production task data, including current order requirements (cable type, specification, length), progress (current completed quantity, time consumption), and priority;

[0105] Integrate the intelligent warehousing system with the MES system to synchronize inventory data regularly, including raw material inventory (such as copper wire, PVC sheath material, etc.) and remaining consumables (such as lubricating oil, cooling water);

[0106] Collect environmental data in real time through environmental Internet of Things sensors, including workshop temperature, humidity, and air data.

[0107] In this embodiment, the preprocessing is as follows: adopt moving average filtering to filter out instantaneous abnormal fluctuations (such as abnormal tension values), and introduce rule detection (such as temperature overlimit, extrusion line speed fluctuation) to find error values; use the mean filling interpolation method to repair missing values; align the periodic data collected by the PLC to avoid errors caused by different sampling rates, and use the NTP protocol to ensure time consistency, ensuring that various types of data are synchronized based on a unified timestamp.

[0108] In this embodiment, the production process model is a two-layer model architecture, including a task flow model and a resource status model, specifically as follows:

[0109] The task flow model dynamically describes the execution sequence, process requirements, and scheduling constraints of tasks, and is modeled using a directed acyclic graph (DAG). Each task is used as a node T of the graph a , and the dependency relationship between tasks is represented by an edge e ab ;

[0110] The resource status model describes the status of materials and equipment in the production process through resource consumption and allocation rules; regards resources as constraint conditions to limit the allocation and execution of tasks.

[0111] In this embodiment, the resource status model is specifically as follows:

[0112] Describe the dynamic status of resources: including the availability and consumption of equipment, materials, and manpower, etc., indicating the dynamic distribution and use of resources in the production process.

[0113] Constraint task allocation and execution:

[0114] Each task must meet its resource requirements, and tasks cannot be executed when resources are insufficient; ensure the optimal use of resource allocation;

[0115] Support dynamic adjustment and optimization:

[0116] Optimize resource allocation and task scheduling according to the real-time production status, and coordinate the execution order of processes.

[0117] In this embodiment, the task flow model is as follows:

[0118] Model all production tasks as a directed acyclic graph G=(V,E), where:

[0119] is the task set, and each task represents a node;

[0120] is task and task the set of dependency relationships between; represents task is task the prerequisite task of, ;

[0121] Acyclicity:

[0122]

[0123] That is, avoid circular dependencies and ensure the legality of the topological order of the task flow;

[0124] Among them, task T a corresponds to a node in the DAG graph, and its attribute set includes:

[0125] , where are respectively the task end time, task start time, task duration, task resource set and task priority of task ;

[0126] Task constraints include time constraints and resource constraints, as follows:

[0127] , that is, the start time of the subsequent task must be later than or equal to the end time of its prerequisite task;

[0128] , that is, the start time of task T a must be within its earliest start time and latest start time range;

[0129] , that is, at any time t, for the kth type of resource R k, the total resource usage of the running tasks cannot exceed the maximum resource capacity ; active(t) is the set of tasks being executed at time t; is the task 's consumption of resource R k ;

[0130] If the resource supply decreases due to real-time changes and the remaining inventory R k (t) is lower than the demand, then adjust the start time of the affected tasks:

[0131] ;

[0132] where t replenish (k) represents the time point when resource R k is replenished.

[0133] In this embodiment, the specific steps of S3 are as follows:

[0134] Based on the constructed production process model, obtain the task set, resource set, dependency relationship set, and constraint conditions as the input of PSO;

[0135] The fitness function is to minimize the makespan and minimize the total delay:

[0136] ;

[0137] Among them, is the delivery time of task T a ; C max is the total processing time; n represents the maximum number of tasks;

[0138] Each particle represents a scheduling scheme, including the task execution order and resource allocation; the particle X is represented by the vector [X T , X R , where X T is the task order and X R is the resource allocation scheme;

[0139] Initialize the particle swarm size N, the maximum number of iterations i ter , the speed range V min , V max , the learning factor c 1 , c 2 , and the inertia weight ω;

[0140] Randomly initialize the position X i (t) and the speed V i (t) to ensure that the initial values of the task order and resource allocation are reasonable;

[0141] Traverse each particle and calculate the fitness function;

[0142] Decode the particle position X i (t) is the scheduling scheme, including the task execution order and the resource allocation scheme;

[0143] Sort the X T Convert it into the task priority order;

[0144] Check whether the resource constraints and the task time window constraints are satisfied;

[0145] Calculate the fitness function, and add a penalty term if the constraints are violated;

[0146] Use the velocity and position update formulas to update the state of each particle:

[0147] Velocity update formula:

[0148] ;

[0149] Position update formula:

[0150] ;

[0151] Among them, is the historical optimal solution of particle i; is the global optimal solution of the population; r 1 , r 2 is a random factor;

[0152] Conduct constraint checks on the updated particles, and correct them if necessary (such as handling dependency contradictions in the task order and resource overloading);

[0153] For the fitness of the current particle, if it is better than the historical optimal solution , then update the current particle to the historical optimal solution ; if the fitness of the current particle is better than the global optimal solution , update the current particle to the global optimal solution .

[0154] If the maximum number of iterations is reached or the change in the global optimal value is less than the threshold, end the iteration and output the generated scheduling scheme, that is, the initial scheduling scheme.

[0155] In this embodiment, S4 is specifically:

[0156] Based on the initial scheduling scheme, decompose the tasks to the specific workstations and equipment terminals according to the resource allocation results; ensure that the tasks meet the process requirements, equipment capabilities, and process constraints;

[0157] Use the MES system to decompose the scheduling plan and send it to each workstation, equipment, or operator, and monitor the task progress, resource usage, and abnormal status through the MES system to dynamically adjust the equipment and plan.

[0158] In this embodiment, S5 is specifically as follows:

[0159] S51: Real-time track the execution status of the task and collect process deviations (such as processing time delay, equipment status change) as feedback data of the production system;

[0160] S52: Based on the real-time monitoring data, combined with the PPO algorithm, dynamically optimize the scheduling plan.

[0161] In this embodiment, S51 is specifically as follows:

[0162] Through the MES system and factory Internet of Things devices, collect the following data in real time to real-time track the execution status of the task:

[0163] Actual start time , actual end time , actual processing time , equipment M k 's real-time status S k (t), the real-time status includes idle, running, and faulty, and the equipment failure probability ; remaining inventory R k (t); resource consumption rate ;

[0164] Collect process deviations as feedback data of the production system, including:

[0165] Processing time deviation , task delay and equipment failure time , where is the planned processing time, is the planned task start time.

[0166] In this embodiment, S52 is specifically as follows:

[0167] In the scheduling optimization, the reinforcement learning problem is modeled as a Markov decision process, including the following elements:

[0168] State space S, including: current task status , equipment status , resource status , current task queue and dependencies ;

[0169] Action space A, including task assignment, task order adjustment, and rush order task processing;

[0170] Reward function R: ;

[0171] ;

[0172] ;

[0173] ;

[0174] ;

[0175] where R 1 , R 2 , R 3 , R 4 are minimizing the total project duration, minimizing task delays, maximizing equipment utilization, and penalizing equipment failures, respectively, and w 1 , w 2 , w 3 , w 4 are weight coefficients; is the equipment working efficiency;

[0176] Initialize the policy network π θ (a|s) and the value function ;

[0177] Use the initial scheduling plan as the basic input;

[0178] Under the current policy π θ , simulate the scheduling optimization process and collect sequences of states s t , actions a t and rewards r t ; Update the policy network according to the following objective function:

[0179] ;

[0180] where θ represents the parameters of the current policy network; is the importance sampling ratio; is the advantage function; is the truncation threshold; is the clipping operation, indicating clipping the value of r t (θ) to make it within the interval; is the expectation function; is the objective function;

[0181] ;

[0182] where Q(s t , a t)The total reward after taking action a in state s t Take action a t ; V(s t ) is the state value in state s t ;

[0183] The loss function of the value function :

[0184] ;

[0185] Wherein, is the estimated value of the current value function network for state s t ; R t represents the actual target return value at time step t.

[0186] Repeat sampling, policy update, and value function update until the model converges to obtain an adjusted scheduling scheme.

[0187] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0188] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one or more of the processes Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0189] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means for implementing the functions specified in one or more of the processes Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0190] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable apparatus provide steps for realizing the functions specified in one process Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps of the functions specified in one block or a plurality of blocks.

[0191] As mentioned above, the above are only the preferred embodiments of the present invention, and are not intended to limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An AI-driven adaptive production scheduling method for cable production lines, characterized in that: The following steps are involved: S1: Collect production line data, including equipment operation data, production task data, inventory data and environmental data, and pre-process them; S2: Establish a data-driven production process model based on the preprocessed production line data; The production process model is a two-layer model architecture, including a task flow model and a resource status model, as follows: The task flow model dynamically describes the execution order, process requirements and scheduling constraints of tasks. It uses a directed acyclic graph to model each task. Each task is a node T in the graph. a , through the edge e ab Represents dependencies between tasks; Resource status model, which describes the status of materials and equipment in the production process through resource consumption and allocation rules; it regards resources as constraints to limit the allocation and execution of tasks; S3: Based on the constructed production process model, the particle swarm optimization algorithm is used to generate the initial scheduling plan; S4: Based on the initial scheduling plan, tasks are assigned to specific workstations and equipment, and the production line execution plan; S5: Monitor the execution process of tasks and collect process deviations; combine deep reinforcement learning to adjust scheduling according to real-time production status; The task process model is as follows: Model all production tasks as a directed acyclic graph G=(V,E), where: is a set of tasks, each task represents a node; It's a task With the task The set of dependencies between them; Indicates the task It's a task Prerequisites ; Acyclicity: That is, avoid circular dependencies to ensure the topological order of the task flow is legal; Among them, task T a Corresponding to a node in the DAG graph, task T a The properties collection includes: ,in Respectively for tasks task end time, task start time, task duration, task resource set and task priority; Task constraints include time constraints and resource constraints, as follows: , the start time of the successor task must be later than or equal to the end time of its predecessor task; , Task T a The start time must be at its earliest start time and the latest start time within the scope; , at any time t, for the kth resource R k , the total resource usage of running tasks cannot exceed the maximum resource capacity ; active(t) is the set of tasks being executed at time t; For the task For resources R k consumption; If the resource supply decreases due to real-time changes, the inventory surplus R k (t) is lower than the demand, then adjust the start time of the affected tasks: ； where t replenish (k) represents resource R k Supplementary time point.

2. The AI-driven adaptive production scheduling method for cable production lines according to claim 1 is characterized in that: The collected production line data includes equipment operation data, production task data, inventory data and environmental data, as follows: Through the intelligent sensor integrated PLC interface, the equipment operation data is collected, including equipment status, extrusion temperature, tension, line speed, equipment speed and power; Integrate with the production management database through the MES system interface to obtain real-time production task data, including current order requirements, progress and priority; Integrate the smart warehousing system with the MES system to synchronize inventory data regularly, including raw material inventory and remaining consumables; Environmental data, including workshop temperature, humidity and air data, are collected in real time through environmental IoT sensors.

3. The AI-driven adaptive production scheduling method for cable production lines according to claim 2 is characterized in that: The preprocessing is specifically as follows: using sliding average filtering to filter out instantaneous abnormal fluctuations, and introducing rule detection to find erroneous values; using mean filling interpolation method to repair missing values; aligning the periodic data collected by PLC to avoid errors caused by different sampling rates, and using NTP protocol to ensure time consistency, ensuring that various types of data are synchronized based on a unified timestamp.

4. The AI-driven adaptive production scheduling method for cable production lines according to claim 1 is characterized in that: The S3 is specifically as follows: Based on the constructed production process model, the task set, resource set, dependency set and constraint conditions are obtained as the input of PSO; The fitness function is to minimize the duration and minimize the total delay: ； in, It is task T a Delivery time; C max is the total processing time; n represents the maximum task volume; Each particle represents a scheduling scheme, including task execution order and resource allocation; particle X is represented by the vector [X T ,X R ] indicates that X T is the task order, X R It is a resource allocation plan; Initialize particle swarm size N, maximum number of iterations i ter , speed range V min ,V max , learning factors c1, c2, and inertia weight ω; Randomly initialize the particle position X i (t) and speed V i (t) Ensure that the initial values ​​of task sequence and resource allocation are reasonable; Traverse each particle and calculate the fitness function; Decode particle position X i (t) Scheduling plan, including task execution order and resource allocation plan; After sorting X T Translated into task priority order; Check whether resource constraints and task time window constraints are met; Calculate the fitness function and add a penalty term if the constraint is violated; Update the state of each particle using the velocity and position update formula: Speed ​​update formula: ； Position update formula: ； in, is the historical optimal solution of particle i; is the global optimal solution of the population; r1, r2 are random factors; Constraint checks are performed on the updated particles for correction; For the fitness of the current particle, if it is better than the historical optimal solution , then update the current particle to the historical optimal solution ; If the fitness of the current particle is better than the global optimal solution , update the current particle to the global optimal solution ; If the maximum number of iterations is reached or the change in the global optimal value is less than the threshold, the iteration ends and the generated scheduling plan, i.e., the initial scheduling plan, is output.

5. The AI-driven adaptive production scheduling method for cable production lines according to claim 4 is characterized in that: The S4 is specifically: Based on the initial scheduling plan, tasks are decomposed into specific workstations and equipment according to resource allocation results; ensuring that tasks meet process requirements, equipment capabilities, and process constraints; The MES system is used to decompose the scheduling plan and send it to each workstation, equipment or operator. The MES system is also used to monitor task progress, resource usage, abnormal status, and dynamically adjust equipment and plans.

6. The AI-driven adaptive production scheduling method for cable production lines according to claim 1 is characterized in that: The S5 is specifically: S51: Track the execution status of tasks in real time and collect process deviations as feedback data for the production system; S52: Based on real-time monitoring data and combined with the PPO algorithm, dynamically optimize the scheduling plan.

7. The AI-driven adaptive production scheduling method for cable production lines according to claim 6 is characterized in that: The S51 is specifically: Through the MES system and factory IoT devices, the following data is collected in real time to track the execution status of tasks in real time: Actual start time , actual end time , actual processing time , device M k Real-time status S k (t), the real-time status includes idle, running, fault, and equipment failure probability ; Inventory remaining R k (t); Resource consumption rate ; Collect process deviations as feedback data for the production system, including: Processing time deviation , Task Delay And equipment failure time ,in To plan the processing time, The start time of the scheduled task.

8. The AI-driven adaptive production scheduling method for cable production lines according to claim 7 is characterized in that: The S52 is specifically: In scheduling optimization, the reinforcement learning problem is modeled as a Markov decision process, which includes the following elements: State space S, including: current task state , Device Status , Resource Status , current task queue and dependencies ; Action space A, including task allocation, task sequence adjustment and task insertion processing; Reward function R: ; ； ； ； ； Among them, R1, R2, R3, and R4 are minimizing the total construction period, minimizing task delays, maximizing equipment utilization, and punishing equipment failures, respectively, and w1, w2, w3, and w4 are weight coefficients; For equipment working efficiency; Initialize the policy network π θ (a|s) and value function V ϕ (s); Take the initial scheduling plan as the basic input; In the current strategy π θ Under the simulation scheduling optimization process, the state s is collected t 、Action a t and reward r t The policy network is updated according to the following objective function: ； Among them, θ represents the parameters of the current policy network; is the importance sampling ratio; is the advantage function; is the cutoff threshold; is the cropping operation, indicating cropping r t The value of (θ) is such that within the interval; is the expected function; is the objective function; ； Among them, Q(s t ,a t ) is in state s t Take action a t Total return after t ) is in state s t The state value of the following; Loss function of value function : ； in, is the current value function network for state s t The estimated value of R t represents the actual target return value at time step t; Repeat sampling, policy updating, and value function updating until the model converges and obtains the adjusted scheduling plan.

Images