Intelligent digital printing management method based on data processing

Abstract

The invention relates to the technical field of intelligent printing production management, and discloses an intelligent digital printing management method based on data processing. According to the method, before a printing task enters actual production, a virtual task sandbox is created for pre-scheduling deduction; in the sandbox, an order is disassembled into atomic processes through a task decomposition model, and a virtual execution agent binding real-time states of equipment and materials is created for each process. A cross-process constraint network is constructed based on proxy simulation to quantify dependency, competition and space-time conflicts between processes. And operating the constraint to meet the solving process, performing dynamic relaxation and adjustment on the network constraint, and outputting a feasible pre-scheduling scheme. And finally, synchronizing the scheme to a real production queue to start production, and collecting actual data to continuously correct the model and the network. According to the method, the risk-free simulation optimization before production and the dynamic quantitative coordination of the production process are realized, and the scheduling precision, the resource utilization rate and the production delivery reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent printing production management technology, specifically to an intelligent digital printing management method based on data processing. Background Technology

[0002] Current digital printing production management primarily employs scheduling methods based on fixed process templates or manual experience-based rules. Orders enter the production queue directly, and scheduling relies on static sorting and resource allocation based on manual experience. This model is ill-suited to complex production scenarios with multiple orders, multiple processes, and multiple constraints. The scheduling results fail to accurately reflect the dynamic changes on the production floor, leading to a disconnect between planning and execution, and making it difficult to pre-assess the actual impact of different options on equipment, materials, and delivery timelines.

[0003] The core flaw in existing technologies lies in the lack of a simulation environment capable of pre-validating the feasibility of production plans. Plan adjustments must rely on physical production trial and error, which easily leads to resource conflicts and delivery delays, resulting in high trial and error costs. Furthermore, existing methods lack effective quantitative analysis and modeling tools for the dynamic relationships between processes, allowing only rough estimates. This results in a lack of precise data for scheduling optimization, hindering improvements in production efficiency and resource utilization.

[0004] The purpose of this invention is to address the aforementioned deficiencies, focusing on two key issues: how to conduct risk-free simulation verification and optimization of scheduling schemes before actual production to reduce trial-and-error costs; and how to accurately quantify and dynamically manage the complex constraints between production processes to provide reliable data-driven decision support for intelligent scheduling. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent digital printing management method based on data processing to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent digital printing management method based on data processing, the method comprising: Before a printing task officially enters the production queue, a virtual task sandbox is created. Order information, including printing quantity, color mode, binding method, and delivery time limit, is imported into the virtual task sandbox for pre-scheduling simulation. In the virtual task sandbox, a task decomposition model is established to decompose the printing order into several atomic processes, and an independent virtual execution agent is created for each atomic process. The virtual execution agent is bound to equipment status information and material inventory information. Based on the execution process simulation of each virtual execution agent, a cross-process constraint network is constructed, and the dependency strength, resource competition relationship and spatiotemporal conflict value between atomic processes are quantified in the cross-process constraint network; The process of running constraints satisfies the solution process, dynamically relaxes and adjusts the constraints applied to the cross-process constraint network in the virtual task sandbox, and outputs a feasible pre-scheduling scheme. The pre-scheduling scheme is synchronized from the virtual task sandbox to the real production queue, the physical production process is started, and actual production data is continuously collected to correct the task decomposition model and the cross-process constraint network.

[0007] Preferably, the process of creating a virtual task sandbox and importing order information into the virtual task sandbox for pre-scheduling simulation specifically includes: An independent computing space isolated from the physical printing environment is established, and the operating speed of the independent computing space can be accelerated to simulate multiple production cycles within a physical unit of time. In the independent computing space, a unique sandbox copy is generated for each received print order, the sandbox copy containing a complete operational data mirror of the order information; Set simulation termination conditions, including meeting the delivery time constraints of all orders, or the total simulated resource consumption exceeding a preset threshold. Load real-time equipment status snapshots and material inventory snapshots collected from the printing workshop into the virtual task sandbox as the initial environment parameters for the simulation; The simulation controller is activated, which drives the task decomposition model and the virtual execution agent to simulate the production process forward in the independent computing space based on the initial environment parameters until the simulation termination condition is met.

[0008] Preferably, the process of establishing a task decomposition model, decomposing the printing order into several atomic processes, and creating an independent virtual execution agent for each atomic process specifically includes: Define the criteria for dividing atomic processes, which are based on the smallest independently operable unit of equipment and the smallest convertible state of materials, so that a single atomic process does not cross different physical equipment or cause a fundamental change in the form of materials. The aforementioned classification criteria are used to perform structured analysis on the order information in the sandbox copy, identifying the categories of color management, layout assembly, digital imaging, media processing, surface finishing, finished product slitting, and packaging preparation processes; For each identified process category instance, a virtual execution agent is instantiated. The virtual execution agent is a software object that encapsulates state, rules, and behavior. Configure a parameter set for each virtual execution agent. The parameter set includes the standard working hours corresponding to the process, the energy consumption baseline, the type and quantity of required materials, the available alternative equipment models, and the tolerance range for the output quality of the preceding process. All instantiated virtual execution agents are logically linked according to the inherent order of the process flow to form an agent network that can be independently scheduled and simulated in the virtual task sandbox.

[0009] Preferably, the process of constructing a cross-process constraint network and quantifying the dependency strength, resource competition relationship, and spatiotemporal conflict value between atomic processes in the cross-process constraint network specifically includes: Scan the parameter sets of all virtual execution agents in the agent network and extract agent pairs with input-output relationships, where the input-output relationship means that the output material or semi-finished product of one agent is a necessary input of another agent; For each agent pair with an input-output relationship, the dependency strength is calculated based on the process complexity and color accuracy requirements of the order. High-precision color replication and complex process superposition will result in a higher dependency strength value. The identification parameters collectively declare that multiple virtual execution agents require the same physical equipment or the same material. These agents form a resource competition relationship and calculate a resource competition index based on their respective standard working hours and task urgency. Analyze the simulated execution time windows and physical locations of each virtual execution agent in the agent network. If the time windows overlap and the physical locations are in the same workshop's logistics bottleneck area, then a spatiotemporal conflict is determined, and the spatiotemporal conflict value is calculated. By using dependency strength, resource competition relationship, and spatiotemporal conflict value as weighted edges, and attaching them between the corresponding nodes of the proxy network, the logically linked proxy network is transformed into a concrete cross-process constraint network with quantitative constraints.

[0010] Preferably, the process of dynamically relaxing and adjusting the constraints applied to the cross-process constraint network in the virtual task sandbox and outputting a feasible pre-scheduling plan, which satisfies the operational constraints, specifically includes: The initialization process satisfies the constraints by randomly assigning a simulation start time within its process time interval to each virtual execution agent in the cross-process constraint network, thus forming an initial solution. Define a constraint violation cost function to calculate the overall degree of violation of the current solution of three types of constraints: dependency strength, resource competition relationship, and spatiotemporal conflict value. The higher the dependency strength, the higher the cost penalty when the constraint is violated. An iterative optimization search strategy is adopted to find the arrangement scheme that minimizes the cost function value of the constraint violation in the solution space. The iterative optimization search strategy is explored by swapping the execution order of adjacent agents, fine-tuning the simulation start time of agents, or allocating different virtual device copies to resource-competing agents. A dynamic relaxation trigger is set up. When the constraint violates the cost function and cannot be reduced in multiple consecutive iterations, the trigger is activated and some constraints are relaxed according to preset rules. The preset rules include allowing minor delays in delivery time, allowing the use of slightly more expensive alternative materials, or allowing equipment to operate under short-term overload within a safety threshold. When the constraint violates a cost function value that is lower than an acceptable threshold or reaches the maximum number of iterations, the search is terminated, and the currently obtained optimal solution is decoded into a specific process scheduling table, equipment allocation table, and material requirements plan. The three together constitute the pre-scheduling scheme.

[0011] Preferably, the process of synchronizing the pre-scheduling scheme from the virtual task sandbox to the real production queue and starting the physical production process specifically includes: The process scheduling table output from the virtual task sandbox is converted into an instruction sequence that can be recognized by the physical production control system. The conversion includes mapping the unique identifier of the virtual execution agent to the control code of the real equipment and converting the simulated timestamp into the time base of the physical system. According to the equipment allocation table, the corresponding printing press, binding machine and slitting machine are issued with preset equipment parameters. The preset parameters include color curve, pressure value, temperature setting and speed level. Based on the material requirements plan, a material requisition and distribution list is generated, triggering the warehouse management system to execute the outbound operation and deliver the materials to the workstations specified in the process schedule. A real-time mapping channel between the sandbox and physical production is established. Through this real-time mapping channel, the actual start time, end time, resource consumption, and abnormal events of each process in the physical production process are captured in real time and transmitted back to the virtual task sandbox. In the virtual task sandbox, the state of the corresponding virtual execution agent is updated based on the actual data returned, and the simulation parameters are overwritten with the actual data to make the agent network state in the virtual task sandbox as synchronized as possible with the physical world.

[0012] Preferably, the process of continuously collecting actual production data to revise the task decomposition model and cross-process constraint network specifically includes: Data acquisition probes are deployed at each key process node in the physical production process. These probes record the actual running time of the equipment, the actual quantity of materials consumed, the time spent transferring work-in-process between processes, and the quality inspection results. The collected actual production data is cleaned and formatted to remove obvious abnormal records and noisy data, and the time, quantity, and time consumption information are unified into a data structure consistent with the internal representation of the virtual task sandbox. The formatted actual data is compared and analyzed with the predicted data of the corresponding process in the virtual task sandbox to calculate the prediction deviation, which includes the deviation of working hours, material consumption, and waiting time. The calculated prediction bias is used to calibrate the standard working hours and energy consumption baseline parameters of the corresponding atomic processes in the task decomposition model. The calibration adopts the exponential weighted moving average method so that the model parameters slowly track the average level of actual production. Using the calculated prediction bias, especially the inter-process waiting time bias, the weight coefficients of the dependency strength and spatiotemporal conflict values ​​of the corresponding edges in the cross-process constraint network are adjusted by backpropagation to reduce the systematic overestimation or underestimation of constraint strength.

[0013] Preferably, the process of determining a spatiotemporal conflict and calculating the spatiotemporal conflict value specifically includes: The simulation execution time windows and physical locations of each virtual execution agent in the analysis agent network overlap, and if the time windows overlap and the physical locations are located in the same logistics bottleneck area of ​​the workshop, the process of determining a spatiotemporal conflict specifically includes: Extract the simulation execution time window data and the bound physical location coordinate data of each virtual execution agent from the simulation environment database of the virtual task sandbox; The simulated execution time window of each virtual execution agent is compared pairwise with the simulated execution time windows of all other virtual execution agents in the agent network to detect whether there is any overlap in time windows. For virtual execution agent pairs with overlapping time windows, query their bound physical location coordinates and calculate the actual logistics path distance between the locations of the two agents. The calculated actual logistics path distance is compared with the preset workshop logistics bottleneck area radius threshold. If the actual logistics path distance is less than or equal to the workshop logistics bottleneck area radius threshold, it is determined that there is a spatiotemporal conflict between the two virtual execution agents. Based on the overlap ratio of time windows and the proximity of the actual logistics path distance to the bottleneck area radius, a weighted calculation is performed to obtain a quantitative value of the spatiotemporal conflict, which is then used as the spatiotemporal conflict value.

[0014] Preferably, the defined constraint violation cost function is used to calculate the overall degree of violation of the three types of constraints—dependency strength, resource competition relationship, and spatiotemporal conflict value—by the current solution. The process of incurring a higher cost penalty when a constraint with higher dependency strength is violated specifically includes: Basic cost weighting coefficients are set for dependence intensity constraints, resource competition constraints, and spatiotemporal conflict value constraints, respectively. For dependency strength constraints, a corresponding penalty coefficient is set based on the dependency strength value recorded on each edge in the cross-process constraint network. The higher the dependency strength value, the larger the penalty coefficient. To address the constraints of resource competition, a piecewise linear penalty function is set based on the magnitude of the resource competition index. The higher the resource competition index, the steeper the slope of the penalty cost growth per unit index. To address the spatiotemporal conflict value constraint, an exponentially increasing penalty term is set based on the size of the conflict value. The larger the conflict value, the more exponentially the penalty cost increases. Calculate the penalty cost for each of the violated constraints in the current solution according to the calculation rules corresponding to its type, and then add the penalty costs of the three types of constraints together to obtain the overall constraint violation cost function value.

[0015] Preferably, the process of using the calculated prediction deviation to calibrate the standard working hours and energy consumption baseline parameters of the corresponding atomic processes in the task decomposition model, wherein the calibration employs an exponentially weighted moving average method to slowly track the average level of actual production, specifically includes: Extract the time prediction deviation sequence and energy consumption prediction deviation sequence of a specific atomic process in the most recent production cycle from the comparative analysis results; A smoothing coefficient is set for the exponentially weighted moving average method, which determines the rate at which the influence of historical data on the current calibration decays; The prediction deviation value of the current period is weighted and calculated with the model parameters calibrated in the previous period. The new standard working hour parameter is equal to the standard working hour parameter of the previous period plus the product of the smoothing coefficient and the prediction deviation of the current working hour. The prediction deviation value of the current period is weighted and calculated with the model parameters calibrated in the previous period. The new energy consumption baseline parameter is equal to the energy consumption baseline parameter of the previous period plus the product of the smoothing coefficient and the current energy consumption prediction deviation. The calculated new standard working time parameters and new energy consumption baseline parameters are updated to the parameter set of the atomic process described in the task decomposition model to complete this round of calibration.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By creating an independent virtual task sandbox environment before the physical production process begins, order information containing specific parameters is imported into this environment for complete pre-scheduling simulation. This process establishes an isolated digital simulation space outside the production system, enabling the simulation and prediction of task flows, resource allocation, and delivery deadlines before actual equipment and materials are used. Managers can identify potential production bottlenecks and resource conflicts based on the simulation results, and test, compare, and adjust scheduling schemes in the virtual environment to select feasible production plans. This reduces equipment idleness, material waiting, and delivery delays caused by improper planning leading directly to production, improving the foresight and reliability of production scheduling.

[0017] By decomposing printing orders into atomic processes and creating independent virtual execution agents for each process, each agent dynamically binds to real-time equipment status and material inventory information during simulation. The interaction of multiple agents during simulation automatically constructs a cross-process constraint network, which quantifies the specific dependency strength between processes, precise resource competition relationships, and calculable spatiotemporal conflict values. This analysis method based on intelligent agent simulation and quantified constraint networks achieves a precise characterization of the complex dynamic relationships within the production system. It provides real-time, structured data input regarding process coordination and resource competition for subsequent intelligent scheduling processes such as constraint satisfaction solving, enabling production planning to shift from arrangements based on static templates and experience to coordinated optimization based on dynamic data and precise models. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent digital printing management method based on data processing described in this invention. Figure 2 A flowchart for creating a virtual task sandbox and pre-scheduling simulation; Figure 3 A flowchart for constructing a cross-process constraint network; Figure 4 Trend chart of calibration effect for intelligent digital printing model; Figure 5 A comparison chart of resource consumption in the atomic processes of digital printing. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1This invention provides an intelligent digital printing management method based on data processing. The method includes creating a virtual task sandbox for any printing order before it is allowed to enter the physical production line. This sandbox is a simulated computing space isolated from the real production environment. All order information, including print quantity, color mode, binding method, and delivery deadline, is imported into this virtual task sandbox. Inside the sandbox, a task decomposition model first breaks down the complex printing order into multiple indivisible atomic processes. Each atomic process is represented by an independent virtual execution agent, which carries the status information of the equipment required for that process and the material inventory information. Based on the simulated operation of all these virtual execution agents, the system constructs a cross-process constraint network, which quantitatively characterizes the complex relationships between atomic processes in terms of dependency strength, resource competition, and time and space occupancy. The system runs a constraint satisfaction solution process, which intelligently and dynamically relaxes and adjusts various constraints imposed on the cross-process constraint network within the virtual task sandbox, ultimately searching for and outputting a pre-scheduling scheme feasible in the simulated environment. Once generated, this scheme is synchronized from the virtual task sandbox to the real production management system, triggering the actual physical production process. Throughout the entire physical production process, the system continuously collects actual production data from each stage and uses this real data to correct and optimize the task decomposition model and cross-process constraint network in the virtual task sandbox, thereby continuously improving its simulation capabilities.

[0021] In one embodiment of the present invention, see [reference] Figure 2 In practical implementation, taking a corporate brochure order of 5000 copies, using CMYK four-color binding, requiring perfect binding, and to be delivered in 72 hours as an example, before the order officially enters the physical queue in the production workshop, the system immediately creates an independent computing space, i.e., a virtual task sandbox, isolated from the physical printing environment. The operating speed of the independent computing space is set to be accelerated; for example, the simulation time base is set to 5 times physical time. This means that within 1 hour of physical time, the independent computing space can simulate a 5-hour production cycle. The system generates a unique sandbox copy for the received corporate brochure order, which fully contains all operable data mirrors such as "Print Quantity: 5000", "Color Mode: CMYK", "Binding Method: Perfect Binding", and "Delivery Deadline: 72 hours". The simulation termination condition is set to meet the delivery deadline constraints of all orders, and a threshold for the total simulated resource consumption is also set. When the simulated power or paper consumption exceeds the preset threshold, the simulation will also terminate.

[0022] In practice, before initiating the simulation, the virtual task sandbox loads real-time snapshots of equipment status and material inventory from the printing workshop. The equipment status snapshots include information that one digital printing press is in "standby" mode and one perfect binding machine is in "maintenance" mode; the material inventory snapshots show that there are 15,000 sheets of coated paper of a specific weight and sufficient stock of a specific type of hot melt adhesive. This information together constitutes the initial environmental parameters for the simulation. The simulation controller is then activated, driving the task decomposition model and subsequently generated virtual execution agents to simulate the production process forward in an independent computational space based on the initial environmental parameters.

[0023] In some embodiments, the task decomposition model performs structured parsing of order information in the sandbox copy based on predefined atomic process partitioning criteria. The partitioning criteria are based on the smallest independently operable unit of equipment and the smallest convertible state of materials. For example, "digital printing" is defined as an atomic process because it is completed independently by a digital printing press, with blank paper as input and printed paper as output, and the material form does not fundamentally change; while "lamination" is defined as another independent atomic process. Through parsing, process categories such as color management, layout assembly, digital imaging, folding, perfect binding, and cutting are identified from this corporate brochure order.

[0024] In some embodiments, for each identified process category instance, the system instantiates an independent virtual execution agent. One virtual execution agent is instantiated for the "Digital Imaging" process, and another for the "Perfect Binding" process. Each virtual execution agent is a software object encapsulating state, rules, and behaviors. The parameter set configured for the "Digital Imaging" virtual execution agent includes: standard working time of 0.05 hours per thousand prints, energy consumption baseline of 5 kWh, required material of 5000 sheets of coated paper of a specific weight, a list of available alternative equipment models, and a color difference tolerance range of ΔE < 3 for the output file of the preceding "Color Management" process. All instantiated virtual execution agents are logically linked according to the inherent sequence of the process flow: "Color Management → Page Layout → Digital Imaging → Folding → Perfect Binding → Cutting," ultimately forming an agent network that can be independently scheduled and simulated in a virtual task sandbox.

[0025] It is understandable that the simulation controller drives the entire agent network to run in an accelerated, independent computing space. The simulation begins with the "color management" virtual execution agent and continues until the "tailoring" virtual execution agent completes its work. The simulation process continuously checks whether the delivery time constraint is met, i.e., whether the completion time of the simulated "tailoring" process is earlier than the order creation time plus 72 hours, and cumulatively calculates the total amount of materials and energy consumed by each virtual execution agent in the simulation. The acceleration factor of the virtual task sandbox is determined by a coefficient, and their relationship can be expressed by the following formula:

[0026] Where: symbol Represents the physical time consumed in the simulation process, denoted by [symbol]. Represents the total time span of the simulation within an independent computational space, symbol This represents the acceleration factor during the simulation. (When the total simulation time span...) For 10 hours and running at accelerated speed When set to 5, the physical time consumed in the simulation process The simulation lasts for 2 hours. It continues until the simulation results meet the delivery time constraints of all orders or the total resource consumption of the simulation exceeds a preset threshold. At that time, the simulation termination condition is triggered, and the simulation controller stops working.

[0027] In one embodiment of the present invention, see [reference] Figure 3 The process of constructing a cross-process constraint network begins with a comprehensive scan of the agent network, where the system extracts all virtual execution agent pairs with input-output relationships. For example, the output of the "folding" virtual execution agent in a "saddle stitching manual" order is a necessary input for the "saddle stitching" virtual execution agent; these two virtual execution agents constitute an agent pair with an input-output relationship. For each agent pair with an input-output relationship, the system calculates the dependency strength based on the order's process complexity and color accuracy requirements. For the "folding" to "saddle stitching" process, which has strict process connection requirements and low error tolerance, the dependency strength is assigned a higher value, such as 0.9, because folding accuracy directly affects the binding quality of saddle stitching. For the "color management" to "digital imaging" process, due to the existence of a color correction tolerance range, the dependency strength is assigned a slightly lower value.

[0028] In practical implementation, analyzing the simulated execution time windows and physical locations of each virtual execution agent in the agent network is a crucial step in constructing a cross-process constraint network. The simulated execution time window data and bound physical location coordinate data of each virtual execution agent are extracted from the simulation environment database of the virtual task sandbox. The system compares the simulated execution time window of each virtual execution agent with the simulated execution time windows of all other virtual execution agents in the agent network pairwise to detect any overlapping time windows. For example, the simulated execution time window of the "saddle ordering" virtual execution agent is scheduled from 10:00 AM to 10:30 AM, while the simulated execution time window of a "finished product handling" virtual execution agent is scheduled from 10:15 AM to 10:45 AM; these two time windows partially overlap.

[0029] For virtual execution agent pairs with overlapping time windows, the system queries their bound physical location coordinates and calculates the actual logistics path distance between the locations of the two virtual execution agents. The calculated actual logistics path distance is compared with a preset threshold for the radius of the workshop logistics bottleneck area, set at 5 meters. If the actual logistics path distance is less than or equal to 5 meters, it is determined that the two virtual execution agents have a spatiotemporal conflict. The spatiotemporal conflict value is calculated using a weighted average based on the proportion of overlapping time windows and the proximity of the actual logistics path distance to the bottleneck area radius. The specific relationship for calculating the spatiotemporal conflict value can be expressed by the following formula:

[0030] Where: symbol The quantified value representing a spacetime conflict is the spacetime conflict value, with the symbol... The symbol represents the overlap duration of the execution time windows of the two virtual execution agents. The average of the total duration of the two time windows, with the sign... Represents the actual logistics path distance between the locations of two virtual execution agents, symbol Represents the preset threshold radius of the workshop logistics bottleneck area, symbol and This is a weighting coefficient used to balance the factors of temporal overlap and spatial distance. It represents the overlap duration between the "saddle order" virtual execution agent and the "finished product handling" virtual execution agent. It lasts 15 minutes, with an average total duration of [missing information]. The journey takes 30 minutes and the distance is [distance missing]. The bottleneck radius threshold is 3 meters. It is 5 meters, and and When all values ​​are set to 0.5, the calculated spatiotemporal conflict value is... It is 0.65.

[0031] In one embodiment of the present invention, each virtual execution agent in the cross-process constraint network is randomly assigned a simulation start time within its process time interval to form an initial solution. A constraint violation cost function is defined to calculate the overall degree of violation of the current solution by the three types of constraints: dependency strength, resource competition, and spatiotemporal conflict value. The higher the dependency strength, the higher the cost penalty when the constraint is violated. When defining this function, basic cost weight coefficients are set for dependency strength constraints, resource competition constraints, and spatiotemporal conflict value constraints. For dependency strength constraints, a corresponding penalty coefficient multiplier is set according to the dependency strength value recorded by each edge in the cross-process constraint network; the higher the dependency strength value, the larger the penalty coefficient multiplier. For resource competition constraints, a piecewise linear penalty function is set according to the size of the resource competition index; the higher the resource competition index, the steeper the slope of the penalty cost growth per unit index. For spatiotemporal conflict value constraints, an exponentially increasing penalty term based on the conflict value is set; the larger the conflict value, the exponentially increasing the penalty cost. The penalty cost for each violated constraint in the current solution is calculated according to its corresponding calculation rule. The penalty costs of the three types of constraints are then added together to obtain the overall constraint violation cost function value. The system employs an iterative optimization search strategy to find the arrangement scheme that minimizes the constraint violation cost function value in the solution space. This strategy involves exploratory searches by swapping the execution order of adjacent agents, fine-tuning the simulation start time of agents, or allocating different virtual device replicas to resource-competing agents. A dynamic relaxation trigger is set. When the constraint violation cost function fails to decrease in consecutive iterations, the trigger is activated, relaxing some constraints according to preset rules. These preset rules include allowing minor delays in delivery deadlines, allowing the use of slightly more expensive alternative materials, or allowing equipment to operate under short-term overload within a safe threshold. When the constraint violation cost function value falls below an acceptable threshold or reaches the maximum number of iterations, the search terminates. The currently obtained optimal solution is decoded into a specific process scheduling table, equipment allocation table, and material requirements plan, which together constitute a pre-scheduling scheme.

[0032] The initialization constraint satisfies the solution process by randomly assigning a simulation start time within its process time interval to each virtual execution agent in the cross-process constraint network, forming an initial solution. For example, the simulation start time for the "Digital Imaging" virtual execution agent of the "Saddle Stitching Manual" order is randomly assigned to 08:00, the simulation start time for the "Digital Imaging" virtual execution agent of the "Laminated Poster" order is randomly assigned to 08:05, and the simulation start time for the "Saddle Stitching" virtual execution agent is randomly assigned to 10:00. All assignments must satisfy the process sequence constraints of each virtual execution agent. A constraint violation cost function is defined to calculate the overall degree of violation of the three types of constraints: dependency strength, resource competition, and spatiotemporal conflict value. Basic cost weight coefficients are set for dependency strength constraints, resource competition constraints, and spatiotemporal conflict value constraints, respectively. For example, the basic cost weight coefficient for dependency strength constraints is set to 10, the basic cost weight coefficient for resource competition constraints is set to 8, and the basic cost weight coefficient for spatiotemporal conflict value constraints is set to 6. For dependency strength constraints, a corresponding penalty coefficient is set based on the dependency strength value recorded for each edge in the cross-process constraint network. The higher the dependency strength value, the larger the penalty coefficient. For example, when an edge with a dependency strength of 0.9 is violated, its penalty coefficient is 2.0; when an edge with a dependency strength of 0.7 is violated, its penalty coefficient is 1.4. For resource competition constraints, a piecewise linear penalty function is set based on the magnitude of the resource competition index. When the resource competition index is between 0 and 1, the penalty cost growth slope per unit index is 5; when the resource competition index is between 1 and 2, the penalty cost growth slope per unit index is 10. For spatiotemporal conflict value constraints, an exponential growth penalty term based on the conflict value magnitude is set. The penalty cost for all violated constraints in the current solution is calculated separately according to the calculation rules corresponding to their types. Then, the penalty costs of the three types of constraints are added together to obtain the overall constraint violation cost function value. The specific relationship for calculating the overall constraint violation cost function value can be expressed by the following formula:

[0033] Where: symbol The value of the cost function representing the violation of the overall constraint is indicated by the symbol. The basic cost weighting coefficient representing the dependency strength constraint, symbol The basic cost weighting coefficient representing the resource competition constraint, symbol The basic cost weighting coefficient representing the spatiotemporal conflict value constraint. (Symbol) Representing the The strength value of a violated dependency strength constraint, function This represents the penalty calculation corresponding to the value of the dependency strength, symbol This represents the numerical value corresponding to the dependency strength. The penalty multiplier. (Symbol) Representing the A resource competition index, a function of a resource competition relationship constraint that has been violated. Represents the resource competition index based on piecewise linear rules. The penalty calculation. (Symbol) Representing the A conflict value of a violated spatiotemporal conflict value constraint, function Represents conflict values ​​based on the exponential growth rule. The penalty calculation.

[0034] In some embodiments, an iterative optimization search strategy is employed to find the arrangement in the solution space that minimizes the constraint violation cost function. The iterative optimization search strategy explores the possibility of swapping the execution order of adjacent virtual execution agents, for example, swapping the simulation start time order of the "Digital Imaging" virtual execution agent for "Laminated Poster" and the "Digital Imaging" virtual execution agent for "Saddle Stitching Manual". The iterative optimization search strategy explores the possibility of fine-tuning the simulation start time of virtual execution agents, for example, fine-tuning the simulation start time of the "Saddle Stitching" virtual execution agent from 10:00 to 10:10. The iterative optimization search strategy explores the possibility of assigning different virtual device copies to resource-competing virtual execution agents, for example, assigning a high-speed laminator virtual copy A and a high-speed laminator virtual copy B to two competing "Laminated" virtual execution agents respectively.

[0035] In some embodiments, a dynamic relaxation trigger is configured. When the constraint violation cost function fails to decrease in 20 consecutive iterations, the dynamic relaxation trigger is activated, relaxing some constraints according to preset rules. Preset rules include allowing minor delays in delivery deadlines, such as allowing a delay of no more than 15 minutes in the delivery deadline of a "laminated poster" order. Preset rules include allowing the use of slightly more expensive alternative materials, such as allowing the use of a spare brand of hot melt adhesive that costs 5% more. Preset rules include allowing short-term overload operation of equipment within a safe threshold, such as allowing a digital printer to operate at 110% of its rated speed for no more than 30 minutes. Activation of the dynamic relaxation trigger modifies the boundary conditions of the relevant constraints, thereby expanding the search range of the solution space.

[0036] Understandably, the search process terminates when the constraint violation cost function value falls below an acceptable threshold or the maximum number of iterations is reached. The acceptable threshold is set to 50 cost units, and the maximum number of iterations is set to 1000. The currently obtained optimal solution is decoded into a specific process scheduling table, equipment allocation table, and material requirements plan, which together constitute the pre-scheduling plan. The process scheduling table clearly lists the start and end times of the physical process corresponding to each virtual execution agent; the equipment allocation table specifies when each physical device executes which process; and the material requirements plan details when each material needs to be delivered to which workstation.

[0037] In one embodiment of the present invention, the process scheduling table output from the virtual task sandbox is converted into an instruction sequence that can be recognized by the physical production control system. This conversion process includes mapping key information. The unique identifier of the virtual execution agent is mapped to the control code of the real device; for example, the virtual execution agent "Print_Agent_01" is mapped to the physical device "DigitalPress_Line2_Controller". The simulated timestamp in the virtual task sandbox is converted to the time base of the physical production control system, where the physical system's time base is the number of milliseconds since the epoch. The simulated timestamp "Day110:00:00" is converted into a specific millisecond timestamp based on the physical start time of the virtual task sandbox simulation. In a specific implementation, based on the equipment allocation table in the pre-scheduling scheme, equipment parameter preset instructions are issued to the corresponding physical devices. These instructions are directly issued to the programmable logic controller or production management system of the corresponding device via the workshop's industrial network. The parameter preset instructions contain all preset parameters required for the device to execute a specific order task. Refer to Table 1, which shows a list of equipment parameter preset instructions.

[0038] Table 1: Equipment Parameter Preset Instruction Table

[0039] In some embodiments, a material requisition and delivery list is generated based on the material requirements plan (MRP) in the pre-scheduled plan. The MRP lists an order for "saddle stitching manuals" requiring 5000 sheets of 157g coated paper and 5 kg of hot melt adhesive, and an order for "laminated posters" requiring 200 sheets of 200g photo fabric and 2 rolls of pre-coated film. The material requisition and delivery list triggers the warehouse management system (WMS), which executes outbound operations based on the list and transports the outbound materials to the specific workstations specified in the process schedule via automated guided vehicles (AGVs) or manual delivery. In some embodiments, a real-time mapping channel is established between the virtual task sandbox and the physical production process. This real-time mapping channel is implemented through a middleware service deployed in the shop floor network, which monitors status events and data reports from the physical production management system. Through the real-time mapping channel, the actual start time, end time, resource consumption, and abnormal events of each process in the physical production process are captured in real time and transmitted back to the virtual task sandbox.

[0040] It's understandable that in the virtual task sandbox, the state of the corresponding virtual execution agent is updated based on the actual data returned. When the "Start Printing" event from the DigitalPress_A printer is received, the virtual task sandbox updates the state of the virtual execution agent identified as "Print_Agent_01" from "Planned" to "Executing". Using actual data to overwrite the simulated parameters, when the actual data "Printing complete, power consumption 18.5 kWh" is received, the energy consumption value recorded in the virtual task sandbox for this virtual execution agent is updated from the simulated 20 kWh to 18.5 kWh. Through continuous data return and update operations, the agent network state in the virtual task sandbox remains synchronized with the production progress in the physical world. The time synchronization relationship in the virtual task sandbox can be expressed by the following formula:

[0041] Where: symbol Represents the actual simulated time point after the update in the virtual task sandbox, symbol Represents the physical reference time point at which the initial simulation scenario begins in the virtual task sandbox, symbol The physical timestamp represents the event report returned from the physical world, symbolized by... This represents the system's physical timestamp recorded when the physical production process starts. It is the physical reference time point of the virtual task sandbox. It is "2024-04-05 08:00:00", the system physical timestamp recorded when the physical production process started. It is 1712340000000, while the returned "printing complete" event physical timestamp The time point 1712349278901 corresponds to the actual simulated time point in the virtual task sandbox. It was updated to "2024-04-05 10:15:00".

[0042] See Figure 4 This is a trend chart showing the calibration effect of the intelligent digital printing model, illustrating the decreasing trend of prediction deviations in working hours and energy consumption after multiple rounds of calibration. With each calibration round, both types of prediction deviations show a significant decreasing trend, eventually stabilizing in the low deviation range of 1%-1.2%, indicating that model calibration effectively improves prediction accuracy. The chart also quantifies the optimization effect of model iteration, verifying the rationality of the "virtual sandbox + data feedback" mechanism and providing data support for the accuracy of subsequent production scheduling. When the deviation rate drops to around 1%, frequent calibration can be stopped, reducing computational resource consumption.

[0043] In one embodiment of the invention, data acquisition probes are deployed at each key process node in the physical production process. For the "digital imaging" process, the data acquisition probes deployed on the digital printing press record the actual operating time of the equipment, the actual quantity of materials consumed, the time spent transferring work-in-process between processes, and the quality inspection results. The data acquisition probes record the actual operating time of the digital printing press as 2.8 hours, the actual quantity of coated paper consumed as 5020 sheets, the time spent transferring work-in-process from the completion of the previous process "page assembly" to the start of this process as 0.2 hours, and the quality inspection result as an average color difference ΔE of 2.1. In specific implementation, the collected actual production data is cleaned and formatted. The system removes obvious abnormal records and noisy data, such as filtering out the extreme time consumption record of 0.01 hours caused by a momentary sensor failure. The cleaned data is formatted, unifying the time, quantity, and consumption information into a data structure consistent with the internal representation of the virtual task sandbox. The actual runtime of 2.8 hours was converted to a floating-point number of 2.8 in "hours", the actual number of 5020 consumed was recorded as an integer of 5020, and the waiting time of 0.2 hours was recorded as a floating-point number of 0.2.

[0044] In some embodiments, the formatted actual data is compared and analyzed with the predicted data of the corresponding process in the virtual task sandbox to calculate the prediction deviation. The predicted data for the "Digital Imaging" virtual execution agent in the virtual task sandbox are: standard working time 3.0 hours, standard material consumption 5000 sheets, and predicted waiting time 0.1 hours. The working time prediction deviation is the actual running time of 2.8 hours minus the predicted standard working time of 3.0 hours, resulting in -0.2 hours. The material consumption deviation is the actual consumption quantity of 5020 sheets minus the predicted standard consumption of 5000 sheets, resulting in 20 sheets. The waiting time deviation is the actual waiting time of 0.2 hours minus the predicted waiting time of 0.1 hours, resulting in 0.1 hours.

[0045] The calculated prediction bias is used to calibrate the standard working hours and energy consumption baseline parameters of the corresponding atomic processes in the task decomposition model. The calibration process employs an exponentially weighted moving average method. The predicted working hours and energy consumption bias sequences for the "digital imaging" atomic processes in the most recent production cycle are extracted from the comparative analysis results. It is assumed that the predicted working hours bias sequences for the most recent three cycles are [-0.1, 0.0, -0.2] hours. A smoothing coefficient is set for the exponentially weighted moving average method. This smoothing coefficient determines the rate at which the influence of historical data on the current calibration decays; it is set to 0.3. The prediction bias value of the current cycle is weighted and calculated with the model parameters calibrated in the previous cycle. The new standard working hour parameter is equal to the standard working hour parameter of the previous cycle plus the product of the smoothing coefficient and the current predicted working hour bias. The specific relationship of the calibration calculation can be expressed by the following formula:

[0046] Where: symbol This represents the new standard working time parameters obtained after this round of calibration, with the symbol... Represents the standard working time parameters after calibration in the previous cycle, symbol The smoothing coefficient set in the exponentially weighted moving average method is represented by the symbol. This represents the current work hour prediction deviation calculated from a comparative analysis of actual production data and predicted data for the current period. It is calculated based on the standard work hour parameters calibrated in the previous period. The smoothing coefficient is 3.0 hours. The current work hour prediction deviation is 0.3. The new standard working hour parameter is calculated based on -0.2 hours. The time is 2.94 hours. The calibration of the energy consumption baseline parameters follows the same logic: the new energy consumption baseline parameters are equal to the energy consumption baseline parameters calibrated in the previous cycle plus the product of the smoothing coefficient and the current energy consumption prediction deviation. The calculated new standard working hours parameters and new energy consumption baseline parameters are then updated to the parameter set of the "Digital Imaging" atomic process in the task decomposition model to complete this round of calibration.

[0047] See Figure 5 This is a resource consumption comparison chart for the atomic processes of digital printing, used to show the differences in standard working hours, material consumption, and equipment occupancy rates among different processes. Standard working hours are 0 for all processes; media processing has the highest material consumption, and digital imaging / media processing has the highest equipment occupancy rate (95%), making it the core process in terms of resource consumption; finished product slitting has the lowest material consumption and equipment occupancy rate, indicating less resource pressure. Identifying high-resource-consuming processes provides a basis for resource scheduling in production planning; identifying resource bottleneck processes helps optimize the production process.

[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0049] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart digital printing management method based on data processing, characterized in that, The method includes: Before a printing task officially enters the production queue, a virtual task sandbox is created. Order information, including printing quantity, color mode, binding method, and delivery time limit, is imported into the virtual task sandbox for pre-scheduling simulation. In the virtual task sandbox, a task decomposition model is established to decompose the printing order into several atomic processes, and an independent virtual execution agent is created for each atomic process. The virtual execution agent is bound to equipment status information and material inventory information. Based on the execution process simulation of each virtual execution agent, a cross-process constraint network is constructed, and the dependency strength, resource competition relationship and spatiotemporal conflict value between atomic processes are quantified in the cross-process constraint network; The process of running constraints satisfies the solution process, dynamically relaxes and adjusts the constraints applied to the cross-process constraint network in the virtual task sandbox, and outputs a feasible pre-scheduling scheme. The pre-scheduling scheme is synchronized from the virtual task sandbox to the real production queue, the physical production process is started, and actual production data is continuously collected to correct the task decomposition model and the cross-process constraint network.

2. The intelligent digital printing management method based on data processing according to claim 1, characterized in that, The process of creating a virtual task sandbox and importing order information into the virtual task sandbox for pre-scheduling simulation specifically includes: An independent computing space isolated from the physical printing environment is established, and the operating speed of the independent computing space can be accelerated to simulate multiple production cycles within a physical unit of time. In the independent computing space, a unique sandbox copy is generated for each received print order, the sandbox copy containing a complete operational data mirror of the order information; Set simulation termination conditions, including meeting the delivery time constraints of all orders, or the total simulated resource consumption exceeding a preset threshold. Load real-time equipment status snapshots and material inventory snapshots collected from the printing workshop into the virtual task sandbox as the initial environment parameters for the simulation; The simulation controller is activated, which drives the task decomposition model and the virtual execution agent to simulate the production process forward in the independent computing space based on the initial environment parameters until the simulation termination condition is met.

3. The intelligent digital printing management method based on data processing according to claim 2, characterized in that, The process of establishing a task decomposition model, which breaks down printing orders into several atomic processes, and creating an independent virtual execution agent for each atomic process, specifically includes: Define the criteria for dividing atomic processes, which are based on the smallest independently operable unit of equipment and the smallest convertible state of materials, so that a single atomic process does not cross different physical equipment or cause a fundamental change in the form of materials. The aforementioned classification criteria are used to perform structured analysis on the order information in the sandbox copy, identifying the categories of color management, layout assembly, digital imaging, media processing, surface finishing, finished product slitting, and packaging preparation processes; For each identified process category instance, a virtual execution agent is instantiated. The virtual execution agent is a software object that encapsulates state, rules, and behavior. Configure a parameter set for each virtual execution agent. The parameter set includes the standard working hours corresponding to the process, the energy consumption baseline, the type and quantity of required materials, the available alternative equipment models, and the tolerance range for the output quality of the preceding process. All instantiated virtual execution agents are logically linked according to the inherent order of the process flow to form an agent network that can be independently scheduled and simulated in the virtual task sandbox.

4. The intelligent digital printing management method based on data processing according to claim 3, characterized in that, The process of constructing a cross-process constraint network and quantifying the dependency strength, resource competition relationship, and spatiotemporal conflict value between atomic processes in the cross-process constraint network specifically includes: Scan the parameter sets of all virtual execution agents in the agent network and extract agent pairs with input-output relationships, where the input-output relationship means that the output material or semi-finished product of one agent is a necessary input of another agent; For each agent pair with an input-output relationship, the dependency strength is calculated based on the order's process complexity and color accuracy requirements. High-precision color replication and complex process superposition will result in a higher dependency strength value. The identification parameters collectively declare that multiple virtual execution agents require the same physical equipment or the same material. These agents form a resource competition relationship and calculate a resource competition index based on their respective standard working hours and task urgency. Analyze the simulated execution time windows and physical locations of each virtual execution agent in the agent network. If the time windows overlap and the physical locations are in the same workshop's logistics bottleneck area, then a spatiotemporal conflict is determined, and the spatiotemporal conflict value is calculated. By using dependency strength, resource competition relationship, and spatiotemporal conflict value as weighted edges, and attaching them between the corresponding nodes of the proxy network, the logically linked proxy network is transformed into a concrete, quantified cross-process constraint network.

5. The intelligent digital printing management method based on data processing according to claim 4, characterized in that, The process of dynamically relaxing and adjusting the constraints applied to the cross-process constraint network in the virtual task sandbox to output a feasible pre-scheduling plan, which satisfies the operational constraints, specifically includes: The initialization process satisfies the constraints by randomly assigning a simulation start time within its process time interval to each virtual execution agent in the cross-process constraint network, thus forming an initial solution. Define a constraint violation cost function to calculate the overall degree of violation of the current solution by three types of constraints: dependency strength, resource competition relationship, and spatiotemporal conflict value. The higher the dependency strength, the higher the cost penalty when the constraint is violated. An iterative optimization search strategy is adopted to find the arrangement scheme that minimizes the cost function value of the constraint violation in the solution space. The iterative optimization search strategy is explored by swapping the execution order of adjacent agents, fine-tuning the simulation start time of agents, or allocating different virtual device copies to resource-competing agents. A dynamic relaxation trigger is set up. When the constraint violates the cost function and cannot be reduced in multiple consecutive iterations, the trigger is activated and some constraints are relaxed according to preset rules. The preset rules include allowing minor delays in delivery time, allowing the use of slightly more expensive alternative materials, or allowing equipment to operate under short-term overload within a safety threshold. When the constraint violates a cost function value that is lower than an acceptable threshold or reaches the maximum number of iterations, the search is terminated, and the currently obtained optimal solution is decoded into a specific process scheduling table, equipment allocation table, and material requirements plan. The three together constitute the pre-scheduling scheme.

6. The intelligent digital printing management method based on data processing according to claim 5, characterized in that, The process of synchronizing the pre-scheduling plan from the virtual task sandbox to the real production queue and starting the physical production process specifically includes: The process scheduling table output from the virtual task sandbox is converted into an instruction sequence that can be recognized by the physical production control system. The conversion includes mapping the unique identifier of the virtual execution agent to the control code of the real equipment and converting the simulated timestamp into the time base of the physical system. According to the equipment allocation table, the corresponding printing press, binding machine and slitting machine are issued with preset equipment parameters. The preset parameters include color curve, pressure value, temperature setting and speed level. Based on the material requirements plan, a material requisition and distribution list is generated, triggering the warehouse management system to execute the outbound operation and deliver the materials to the workstations specified in the process schedule. A real-time mapping channel between the sandbox and physical production is established. Through this real-time mapping channel, the actual start time, end time, resource consumption, and abnormal events of each process in the physical production process are captured in real time and transmitted back to the virtual task sandbox. In the virtual task sandbox, the state of the corresponding virtual execution agent is updated based on the actual data returned, and the simulation parameters are overwritten with the actual data to make the agent network state in the virtual task sandbox as synchronized as possible with the physical world.

7. The intelligent digital printing management method based on data processing according to claim 6, characterized in that, The process of continuously collecting actual production data to revise the task decomposition model and cross-process constraint network specifically includes: Data acquisition probes are deployed at each key process node in the physical production process. These probes record the actual running time of the equipment, the actual quantity of materials consumed, the time spent transferring work-in-process between processes, and the quality inspection results. The collected actual production data is cleaned and formatted to remove obvious abnormal records and noisy data, and the time, quantity, and time consumption information are unified into a data structure consistent with the internal representation of the virtual task sandbox. The formatted actual data is compared and analyzed with the predicted data of the corresponding process in the virtual task sandbox to calculate the prediction deviation, which includes the deviation of working hours, material consumption, and waiting time. The calculated prediction bias is used to calibrate the standard working hours and energy consumption baseline parameters of the corresponding atomic processes in the task decomposition model. The calibration adopts the exponential weighted moving average method so that the model parameters slowly track the average level of actual production. Using the calculated prediction bias, especially the inter-process waiting time bias, the weight coefficients of the dependency strength and spatiotemporal conflict values ​​of the corresponding edges in the cross-process constraint network are adjusted by backpropagation to reduce the systematic overestimation or underestimation of constraint strength.

8. The intelligent digital printing management method based on data processing according to claim 4, characterized in that, The analysis process involves determining the simulated execution time windows and physical locations of each virtual execution agent in the agent network. If the time windows overlap and the physical locations are within the same logistics bottleneck area of ​​the same workshop, a spatiotemporal conflict is identified. The specific steps for calculating the spatiotemporal conflict value include: Extract the simulation execution time window data and the bound physical location coordinate data of each virtual execution agent from the simulation environment database of the virtual task sandbox; The simulated execution time window of each virtual execution agent is compared pairwise with the simulated execution time windows of all other virtual execution agents in the agent network to detect whether there is any overlap in time windows. For virtual execution agent pairs with overlapping time windows, query their bound physical location coordinates and calculate the actual logistics path distance between the locations of the two agents. The calculated actual logistics path distance is compared with the preset workshop logistics bottleneck area radius threshold. If the actual logistics path distance is less than or equal to the workshop logistics bottleneck area radius threshold, it is determined that there is a spatiotemporal conflict between the two virtual execution agents. Based on the overlap ratio of time windows and the proximity of the actual logistics path distance to the bottleneck area radius, a weighted calculation is performed to obtain a quantitative value of the spatiotemporal conflict, which is then used as the spatiotemporal conflict value.

9. The intelligent digital printing management method based on data processing according to claim 5, characterized in that, The defined constraint violation cost function is used to calculate the overall degree of violation of the three types of constraints—dependency strength, resource competition, and spatiotemporal conflict value—by the current solution. The process of incurring a higher cost penalty when a constraint with higher dependency strength is violated specifically includes: Basic cost weighting coefficients are set for dependence intensity constraints, resource competition constraints, and spatiotemporal conflict value constraints, respectively. For dependency strength constraints, a corresponding penalty coefficient is set based on the dependency strength value recorded on each edge in the cross-process constraint network. The higher the dependency strength value, the larger the penalty coefficient. To address the constraints of resource competition, a piecewise linear penalty function is set based on the magnitude of the resource competition index. The higher the resource competition index, the steeper the slope of the penalty cost growth per unit index. To address the spatiotemporal conflict value constraint, an exponentially increasing penalty term is set based on the size of the conflict value. The larger the conflict value, the more exponentially the penalty cost increases. Calculate the penalty cost for each of the violated constraints in the current solution according to the calculation rules corresponding to its type, and then add the penalty costs of the three types of constraints together to obtain the overall constraint violation cost function value.

10. The intelligent digital printing management method based on data processing according to claim 7, characterized in that, The process of using the calculated prediction deviation to calibrate the standard working hours and energy consumption baseline parameters of the corresponding atomic processes in the task decomposition model, wherein the calibration adopts an exponentially weighted moving average method to make the model parameters slowly track the average level of actual production, specifically includes: Extract the time prediction deviation sequence and energy consumption prediction deviation sequence of a specific atomic process in the most recent production cycle from the comparative analysis results; A smoothing coefficient is set for the exponentially weighted moving average method, which determines the rate at which the influence of historical data on the current calibration decays; The prediction deviation value of the current period is weighted and calculated with the model parameters calibrated in the previous period. The new standard working hour parameter is equal to the standard working hour parameter of the previous period plus the product of the smoothing coefficient and the prediction deviation of the current working hour. The prediction deviation value of the current period is weighted and calculated with the model parameters calibrated in the previous period. The new energy consumption baseline parameter is equal to the energy consumption baseline parameter of the previous period plus the product of the smoothing coefficient and the current energy consumption prediction deviation. The calculated new standard working time parameters and new energy consumption baseline parameters are updated to the parameter set of the atomic process described in the task decomposition model to complete this round of calibration.

Images