Factory operation intelligent prediction and resource optimization method and system based on simulation data

Abstract

The application relates to the technical field of factory operation optimization, and discloses a factory operation intelligent prediction and resource optimization method and system based on simulation data, which comprises the following steps: step 1, acquiring a multi-source measured data set and an initial simulation parameter vector to obtain a trusted digital twin; step 2, acquiring a key disturbance factor to obtain a simulation data warehouse; step 3, setting a business core key index, and training a key index prediction model; step 4, determining a business acceptable substandard probability and a key index initial threshold value; step 5, constructing an operation optimization model to generate a candidate operation scheme set; step 6, obtaining an actual substandard probability and updating the key index threshold value to determine a publishable operation scheme; and step 7, publishing an operation execution package, collecting actual operation data and deviation data, and writing the actual operation data and the deviation data into the multi-source measured data set and the simulation data warehouse. The application realizes closed-loop management of intelligent prediction and resource optimization of factory operation key indexes.

Description

Technical Field

[0001] This invention belongs to the field of factory operation optimization technology, specifically relating to a method and system for intelligent prediction and resource optimization of factory operations based on simulation data. Background Technology

[0002] As manufacturing evolves towards lean, flexible, and multi-variety, small-batch production, factory operations management faces the combined impact of multiple factors, including order fluctuations, supply uncertainties, changes in equipment status, and personnel organizational constraints. This leads to significant randomness and coupling in operational indicators such as throughput, work-in-process inventory, and cost and energy consumption. In existing technologies, factory operation planning and resource allocation largely rely on empirical rules or single-indicator optimization, making it difficult to simultaneously consider delivery risks and resource utilization efficiency under multi-disturbance scenarios. Even with simulation analysis or digital twins, inconsistencies between simulation parameters and measured data, and the continuous accumulation of deviations between simulation outputs and actual data, often result in simulation data being unreliable as a stable basis for prediction and optimization. Furthermore, traditional optimization methods typically fix constraint boundaries, lacking risk tolerance expressions based on probabilistic perspectives, making it difficult to jointly control the risks of multiple key indicators. Simultaneously, in rolling scheduling scenarios, frequent updates to solutions lack a mechanism for re-verifying candidate solutions, leading to insufficient reproducibility and risk consistency. Summary of the Invention

[0003] This invention provides a method and system for intelligent prediction and resource optimization of factory operations based on simulation data. It solves the technical problems in related technologies, such as factory operation prediction and resource optimization relying on empirical rules or single indicator modeling, difficulty in covering multiple disturbance scenarios and lack of unified probability risk constraints, resulting in cumulative deviations between simulation output and measured data, lack of re-verification mechanism for candidate operation schemes, and insufficient consistency between the releaseability of schemes and risks during the rolling optimization process.

[0004] This invention provides a method for intelligent prediction and resource optimization of factory operations based on simulation data, comprising the following steps:

[0005] Step 1: Obtain the multi-source measured dataset and the initial simulation parameter vector. Based on the multi-source measured dataset, calibrate the initial simulation parameter vector to obtain a reliable digital twin.

[0006] Step 2: Obtain key disturbance factors, generate a scenario set and configure scenario weights, input the scenario set into a trusted digital twin to run the simulation, and obtain the simulation data warehouse;

[0007] Step 3: Set the core key indicators for the business, train the key indicator prediction model based on the simulation data warehouse, and obtain the key indicator prediction results.

[0008] Step 4: Determine the acceptable failure probability of the business based on the simulation data warehouse, construct joint opportunity constraint rules based on the acceptable failure probability of the business, and determine the initial threshold of key indicators based on the simulation data warehouse.

[0009] Step 5: Obtain the basic parameters of factory operation and the rolling window parameters, and build an operation optimization model based on the joint opportunity constraint rules within the rolling time window to generate a set of candidate operation solutions;

[0010] Step 6: Input the candidate operation plan set into the trusted digital twin for re-simulation stress test to obtain the actual failure probability; compare the actual failure probability with the business acceptable failure probability, update the safety margin and key indicator thresholds based on the comparison results, and determine the release operation plan based on the updated key indicator thresholds.

[0011] Step 7: Publish the operation execution package, collect actual operation data and obtain deviation data, and write the actual operation data and deviation data into the multi-source measured dataset and simulation data warehouse respectively.

[0012] This invention provides a factory operation intelligent prediction and resource optimization system based on simulation data, comprising:

[0013] The twin calibration module is used to acquire multi-source measured datasets and initial simulation parameter vectors, and calibrate the initial simulation parameter vectors based on the multi-source measured datasets to obtain a reliable digital twin.

[0014] The scenario simulation module is used to acquire key disturbance factors, generate a scenario set and configure scenario weights, input the scenario set into a trusted digital twin to run the simulation, and obtain a simulation data warehouse.

[0015] The indicator prediction module is used to set core key business indicators, train the key indicator prediction model based on the simulation data warehouse, and obtain the key indicator prediction results.

[0016] The risk constraint module is used to determine the acceptable failure probability of business based on the simulation data warehouse, construct joint opportunity constraint rules based on the acceptable failure probability of business, and determine the initial threshold of key indicators based on the simulation data warehouse.

[0017] The operation optimization module is used to obtain basic factory operation parameters and rolling window parameters, construct an operation optimization model based on joint opportunity constraint rules within the rolling time window, and generate a set of candidate operation solutions.

[0018] The stress calibration module is used to input the set of candidate operation plans into a trusted digital twin for re-simulation stress testing to obtain the actual failure probability; compare the actual failure probability with the business-acceptable failure probability, update the safety margin and key indicator thresholds based on the comparison results, and determine the release operation plan based on the updated key indicator thresholds.

[0019] The execution write-back module is used to publish the operation execution package, collect actual operation data and obtain deviation data, and write the actual operation data and deviation data into the multi-source measured dataset and simulation data warehouse, respectively.

[0020] The beneficial effects of this invention are as follows: Based on multi-source measured datasets, the invention calibrates the initial simulation parameter vector and forms a reliable digital twin, ensuring consistency between the simulation output and actual operational data in terms of core business indicators, providing a stable data foundation for subsequent simulation data warehouse construction; Based on key disturbance factors, a scenario set is generated and scenario weights are configured, enabling the simulation data warehouse to simultaneously possess disturbance coverage and occurrence frequency, thereby supporting the training and inference of key indicator prediction models under a unified probability caliber; By determining the acceptable probability of non-compliance and constructing joint opportunity constraint rules, a multi-indicator joint risk constraint boundary is formed; An operational optimization model is constructed within a rolling time window, generating a set of candidate operational solutions, allowing solutions under different comprehensive optimization target weight configurations to be compared under the same risk caliber; Through re-simulation stress testing, the actual probability of non-compliance is calculated and the safety margin and key indicator thresholds are updated in conjunction, determining the releasable operational solutions; Through operational execution package version identification, the attribution and write-back of actual operational data and deviation data is achieved, forming a traceable closed-loop iteration mechanism, thereby improving the stability, consistency, and verifiability of factory operational forecasting and resource optimization. Attached Figure Description

[0021] Figure 1 This is a flowchart of the intelligent prediction and resource optimization method for factory operations based on simulation data according to the present invention. Detailed Implementation

[0022] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0023] like Figure 1 As shown, the intelligent prediction and resource optimization method for factory operations based on simulation data includes the following steps:

[0024] Step 1: Obtain the multi-source measured dataset and the initial simulation parameter vector. Based on the multi-source measured dataset, calibrate the initial simulation parameter vector to obtain a reliable digital twin.

[0025] Step 2: Obtain key disturbance factors, generate a scenario set and configure scenario weights, input the scenario set into a trusted digital twin to run the simulation, and obtain the simulation data warehouse;

[0026] Step 3: Set the core key indicators for the business, train the key indicator prediction model based on the simulation data warehouse, and obtain the key indicator prediction results.

[0027] Step 4: Determine the acceptable failure probability of the business based on the simulation data warehouse, construct joint opportunity constraint rules based on the acceptable failure probability of the business, and determine the initial threshold of key indicators based on the simulation data warehouse.

[0028] Step 5: Obtain the basic parameters of factory operation and the rolling window parameters, and build an operation optimization model based on the joint opportunity constraint rules within the rolling time window to generate a set of candidate operation solutions;

[0029] Step 6: Input the candidate operation plan set into the trusted digital twin for re-simulation stress test to obtain the actual failure probability; compare the actual failure probability with the business acceptable failure probability, update the safety margin and key indicator thresholds based on the comparison results, and determine the release operation plan based on the updated key indicator thresholds.

[0030] Step 7: Publish the operation execution package, collect actual operation data and obtain deviation data, and write the actual operation data and deviation data into the multi-source measured dataset and simulation data warehouse respectively.

[0031] In one embodiment of the present invention, a multi-source measured dataset and an initial simulation parameter vector are acquired, and the initial simulation parameter vector is calibrated based on the multi-source measured dataset to obtain a reliable digital twin, including:

[0032] Step 11: Obtain the multi-source measured dataset and the initial simulation parameter vector. The multi-source measured dataset is used to represent real observation data of the factory operation process, including at least production execution data, enterprise resource planning data, warehouse management data, and cost and energy consumption statistics. Production execution data provides information such as process execution, equipment status, production cycle time, and completion events. Completion events include at least: order identifier, process identifier, and completion event. Enterprise resource planning data provides information such as orders, delivery dates, production plans, and resource requirements. Warehouse management data provides information such as inventory, inbound / outbound, batch tracking, and material flow events. Cost and energy consumption statistics provide energy consumption, cost accounting, and related statistical information during the production process. The initial simulation parameter vector represents the initial configuration parameter set of the simulation model, including the initial value set of parameters such as process cycle time, equipment availability, changeover rules, handling delay, personnel shift capacity, and inventory replenishment delay. This parameter vector is used to drive the simulation model and generate simulation output.

[0033] Step 12: During the calibration phase, a calibration time window is first determined from the multi-source measured dataset. This calibration time window is a statistically significant time interval selected from the multi-source measured dataset to ensure the consistency of the comparison benchmark during the calibration process. Subsequently, the multi-source measured dataset is aligned according to a unified primary key mapping rule. This unified primary key mapping rule is used to establish the correspondence between the same business objects across cross-source data, and includes at least order identifiers, process identifiers, equipment identifiers, shift identifiers, and material batch identifiers. Specifically, the order identifier uniquely identifies the order and its order-level operational records; the process identifier uniquely identifies the processing process and process execution records of the order; the equipment identifier uniquely identifies the processing equipment or production resource entity; the shift identifier uniquely identifies the organizational unit of human resources in the time dimension; and the material batch identifier uniquely identifies the batch-level flow records of raw materials and work-in-process. Through this primary key mapping rule, production execution data, enterprise resource planning data, warehouse management data, and cost and energy consumption statistics are uniformly aligned within the calibration time window, thus forming a data foundation that can be directly compared and statistically analyzed under the same caliber.

[0034] Step 13: After data alignment, calibration input data is generated based on the calibration time window. This calibration input data is a set of simulation model inputs extracted and structured from multi-source measured datasets within the calibration time window. It includes at least order arrival information, process routing information, equipment availability information, personnel shift and skill information, material inventory and replenishment information, and relevant latency statistics to ensure the completeness and traceability of the input data required for subsequent simulation model operation. Simultaneously, key calibration indicator observations are calculated and solidified based on the multi-source measured datasets. These key calibration indicator observations serve as a benchmark for simulation output and include throughput-related indicator observations, work-in-process quantity-related indicator observations, and cost and energy consumption-related indicator observations. Specifically, throughput-related indicator observations represent the statistical level of output completed per unit time; work-in-process quantity-related indicator observations represent the quantity distribution level of work-in-process within the calibration time window; and cost and energy consumption-related indicator observations represent the statistical level of cost and energy consumption per unit time or unit output. "Fixed" means that the observed values ​​of the above-mentioned key calibration indicators remain unchanged as a fixed reference throughout the entire calibration iteration process, so as to avoid the calibration target from drifting during the iteration process, thereby ensuring that the calibration process has determinism and verifiability.

[0035] Step 14: Input the calibration input data into the simulation model configured with the initial simulation parameter vector, run the simulation within the calibration time window, and output the simulated values ​​of key calibration indicators. The simulation model is a discrete event simulation or equivalent simulation calculation model used to characterize the factory operation process. Its output simulated values ​​of key calibration indicators are consistent with the observed values, corresponding to simulated values ​​of throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators, respectively. Then, subtract the simulated values ​​of key calibration indicators from the observed values ​​of key calibration indicators one by one to obtain the calibration deviations for each item. These calibration deviations are used to quantify the degree of inconsistency between the simulation output and the measured observations, and include at least the deviations of throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators, thus providing a clear error-driven basis for updating the simulation model parameters.

[0036] In the parameter calibration optimization phase, each calibration deviation is multiplied by a preset weight to obtain a weighted deviation. Each weighted deviation is then multiplied by itself to obtain a squared deviation. The squared deviations are summed to obtain the calibration target value. This calibration target value is a unified optimization metric used to drive the iterative update of the initial simulation parameter vector. The initial simulation parameter vector is iteratively updated, and the simulation output, deviation calculation, and calibration target value calculation are repeatedly executed until the calibration deviation meets the preset allowable deviation range. After the stopping condition is met, the iteratively updated initial simulation parameter vector is solidified to obtain the calibration simulation parameter vector, and a trusted digital twin is obtained based on the calibration simulation parameter vector. The trusted digital twin is a simulation entity configured with the calibration simulation parameter vector as the only parameter. Within the calibration time window, its output of throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators remains consistent with actual observations, providing a stable and reliable simulation foundation for the subsequent scenario simulation to generate a simulation data warehouse.

[0037] Through the above implementation process, this invention forms a closed loop between the multi-source measured dataset of factory operation and the simulation parameter calibration without introducing equipment control commands or involving the underlying control logic of production equipment. This makes the simulation output of the trusted digital twin consistent with the actual operation statistics, thereby improving the training data quality of the key indicator prediction model based on simulation data and reducing the risk of distribution offset in the simulation data warehouse.

[0038] In one embodiment of the present invention, key perturbation factors are obtained, a scenario set is generated and scenario weights are configured, the scenario set is input into a trusted digital twin to run simulation, and a simulation data warehouse is obtained, including:

[0039] Step 21: Obtain key disturbance factors. These key disturbance factors are a set of operational disturbance variables that have a significant impact on factory operation results and exhibit random or fluctuating characteristics in actual operation. They are used to represent the differences in results caused by external or internal fluctuations under the same operational strategy. In this embodiment, key disturbance factors include incoming material delivery delay, product rework rate, equipment availability, and production area congestion level. Incoming material delivery delay is a statistical measure of the time difference between the planned arrival time and the actual arrival time of materials; the product rework rate is a statistical measure of the proportion of the number of items entering the rework process within the statistical window to the number of completed items; the equipment availability rate is a statistical measure of the proportion of the time the equipment is in a usable state to the total time of the statistical window; and the production area congestion level is a statistical measure reflecting the level of congestion in logistics and work-in-process flow within the production area, which can be calculated from operational statistical data such as the number of work-in-process items remaining, handling waiting time, or area access capacity occupancy rate. To ensure the traceability and consistency of the scenario value range, the value range of each key disturbance factor is determined based on multi-source measured datasets within the calibration time window to constrain the scenario construction to not exceed the observable operational statistical range.

[0040] After determining the value range, a scenario set is generated based on the combination of key disturbance factor values. This scenario set is a collection of multiple key disturbance factor value combinations, used to represent the operating environment under different disturbance conditions. To ensure scenario traceability and indexability, this embodiment assigns a scenario identifier to each scenario and ensures a one-to-one correspondence between the scenario identifier and the key disturbance factor value combination, thereby guaranteeing that subsequent simulation outputs can be back-tracked to the corresponding disturbance condition combination.

[0041] Step 22: Based on the multi-source measured dataset, the occurrence frequency of each scenario is statistically analyzed as the initial weight value. This initial weight value is a statistical measure of the probability level of a scenario's occurrence in historical operational samples, reflecting the relative frequency of each scenario's occurrence. Further, the initial weight value of each scenario is divided by the sum of the initial weight values ​​of all scenarios to obtain the scenario weight, ensuring the sum of all scenario weights is one. The scenario weights are then bound one-to-one with the scenario identifiers. The scenario weight is a unified weighting caliber for subsequent weighted statistics and probability measurements based on simulation output. On one hand, it is used for weighted distribution statistics of the simulation output; on the other hand, it is used to transform the set of scenarios that meet or do not meet the threshold into a calculable probability quantity, thus maintaining the same probability caliber as the subsequent joint chance constraint rules and actual non-compliance probability calculation.

[0042] Step 23: After configuring the scenario set and scenario weights, input the scenario set into the trusted digital twin and run the simulation. Specifically, for each scenario, read its key perturbation factor values ​​as the scenario simulation input, run the simulation while keeping the trusted digital twin unchanged, and output the simulation output data. Keeping the trusted digital twin unchanged means that the parameter configuration of the trusted digital twin remains the configuration determined by the calibration simulation parameter vector, and does not change with scenario switching, so as to ensure that the difference in simulation results of different scenarios only comes from the difference in key perturbation factor values ​​rather than model parameter drift. The simulation output data includes the simulation output corresponding to the core key indicators of the business. By simulating and outputting the scenario set one by one, simulation output data corresponding one-to-one with scenario identifiers is obtained, and the scenario identifier, key perturbation factor values, scenario weights, and simulation output data are written into the simulation data warehouse. The simulation data warehouse is a structured storage unit used to store scenario-driven simulation samples and solidify the correspondence between scenario identifiers, key perturbation factor values, scenario weights, and simulation output data.

[0043] Through the above implementation process, this invention constructs a scenario set by key disturbance factors and introduces scenario weights, enabling the simulation data warehouse to simultaneously express the disturbance coverage and the probability of disturbance occurrence; the trusted digital twin runs simulations on different scenarios under the condition that the parameter configuration remains unchanged, so that the difference in simulation output can be attributed to the change in the value of key disturbance factors, thereby enhancing the interpretability and traceability of the simulation data warehouse.

[0044] In one embodiment of the present invention, core key business indicators are defined, and a key indicator prediction model is trained based on a simulation data warehouse to obtain key indicator prediction results, including:

[0045] Step 31: Based on the simulation data warehouse, set the core key business indicators and solidify the definition of these indicators. The core key business indicators are a set of indicators used to characterize the factory's operational status and results. In this embodiment, they include throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators. Throughput-related indicators reflect the level of completed output per unit time; work-in-process quantity-related indicators reflect the quantity distribution level of work-in-process within the production system; and cost and energy consumption-related indicators reflect the cost and energy consumption levels during the production process. The definition of the core key business indicators is a fixed calculation method used in the simulation data warehouse based on simulation output data. A fixed calculation method means that the simulation output data is calculated using a preset statistical method, time scale, and aggregation rules, and remains consistent to avoid drift in the definition of indicators between different steps.

[0046] Step 32: Construct a training sample set based on the simulation data warehouse. Specifically, use the values ​​of key disturbance factors and scenario weights in the simulation data warehouse as input-side features, and use the core business indicators calculated according to the business core key indicators as output-side labels. Align the input-side features and output-side labels with scenario identifiers to ensure a one-to-one correspondence between the input and output sides for each training sample. The training sample set also retains scenario weights as sample weights. Sample weights are used to characterize the relative importance or frequency of different scenarios in historical operational statistics, so that the training process can perform weighted fitting of different scenario samples under a unified probability caliber, avoiding the problem of low-frequency scenarios having an excessive impact on the prediction model or high-frequency scenarios being ignored.

[0047] Step 33: Determine the input-output structure of the key indicator prediction model based on the input-side features and output-side labels of the training sample set, and train the key indicator prediction model based on the training sample set. The key indicator prediction model maps scenario inputs to core business key indicator outputs. The input-output structure means that the input end of the key indicator prediction model receives input-side features formed by the values ​​of key perturbation factors and scenario weights, and the output end outputs predicted values ​​corresponding to the core business key indicators, thus ensuring that the model structure is consistent with the field structure of the training sample set. Subsequently, the key indicator prediction model is trained using sample weights for parameter fitting. Using sample weights for parameter fitting means that during model training, the fitting contribution of each sample is weighted according to the sample weights stored in the training sample set, so that the model parameter updates can reflect the overall distribution characteristics under the scenario weight caliber. After training is completed, the trained key indicator prediction model is used to infer the key indicator prediction results onto the target input set. The target input set can be the set of key disturbance factor values ​​corresponding to the current scrolling window. The key indicator prediction results obtained from the inference output establish a one-to-one correspondence with the key disturbance factor values, while retaining the scenario identifier so that statistics, constraint verification and scheme comparison can be performed according to the scenario identifier in subsequent steps.

[0048] Through the above implementation process, this invention solidifies the definition of core key business indicators, enabling the simulation output data of the simulation data warehouse to form core key business indicators according to a unified definition, ensuring that the prediction results of key indicators can be consistently aligned with joint opportunity constraint rules and operational optimization models; it constructs a training sample set by taking the values ​​of key disturbance factors and scenario weights, and uses the sample weights to participate in parameter fitting, so that the key indicator prediction model can characterize the relationship between multiple disturbance scenarios and core key business indicators under the scenario weight definition; by retaining scenario identifiers and corresponding them one-to-one with the values ​​of key disturbance factors, it supports subsequent calls and traceability.

[0049] In one embodiment of the present invention, the acceptable failure probability of business operations is determined based on a simulation data warehouse, joint opportunity constraint rules are constructed based on the acceptable failure probability of business operations, and initial thresholds for key indicators are determined based on the simulation data warehouse, including:

[0050] Step 41: Calculate the core business key indicator sample set based on the simulation data warehouse and according to the business core key indicator caliber. The core business key indicator sample set refers to the set of core business key indicator values ​​calculated from the simulation output data corresponding to each scenario identifier in the simulation data warehouse, according to the business core key indicator caliber. Through the above calculation, the core business key indicator sample set is made to correspond one-to-one with the scenario identifier, and the scenario weights corresponding to each scenario identifier are read synchronously and the scenario weight weighting statistical caliber is solidified. The scenario weight weighting statistical caliber means that when performing distribution statistics and probability measurement on the core business key indicator sample set, the scenario weight is used as the only weight field and remains unchanged, so that the statistical results can reflect the overall distribution characteristics under the scenario occurrence frequency caliber.

[0051] Step 42: Read the risk tolerance configuration and determine the acceptable failure probability for the business. The risk tolerance configuration is a preset configuration item used to express the operational level's tolerance for failure, which can be maintained in a fixed manner in the system parameters. The acceptable failure probability is the upper limit of tolerance for core key business indicators not meeting the threshold under a scenario-weighted statistical approach. The acceptable failure probability is solidified as the probability benchmark for subsequent joint opportunity constraint rule verification and comparison with the actual failure probability. This probability benchmark will not be redefined or adjusted in subsequent steps to ensure a consistent comparison benchmark for risk assessment at different stages.

[0052] Step 43: Based on the simulation data warehouse, perform scenario-weighted distribution statistics on throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators, and determine the initial threshold for key indicators accordingly. Specifically, for each type of indicator, sort its indicator value in the core key indicator sample set according to size, and accumulate the corresponding scenario weight according to the sorting result; when the accumulated scenario weight reaches the target accumulated weight position determined by the acceptable failure probability of the business, the indicator value corresponding to that position is determined as the initial threshold for the key indicator. The target accumulated weight position is used to convert the acceptable failure probability of the business into a threshold extraction position, which is consistent with the scenario weighted statistical caliber, thus giving the threshold extraction process a defined calculation rule.

[0053] After determining the initial thresholds for key indicators, joint opportunity constraint rules are constructed based on these thresholds. These joint opportunity constraint rules are rule expressions that jointly constrain multiple core business key indicators under the same probabilistic caliber. Specifically, they are constructed as follows: for each scenario in the simulation data warehouse, it is determined whether throughput-related indicators, work-in-process quantity-related indicators, and cost / energy consumption-related indicators simultaneously meet their respective initial key indicator thresholds; the scenario weights corresponding to those simultaneously meeting the thresholds are summed to obtain a total scenario weight; this total scenario weight is then compared with a target weight lower limit determined by the business's acceptable failure probability to obtain the satisfaction determination result of the joint opportunity constraint rules. The target weight lower limit is a threshold determination lower limit converted from the business's acceptable failure probability, used to enable the joint opportunity constraint rules to be calculated and verified in a scenario weighted form.

[0054] Through the above implementation process, this invention forms a set of indicator samples with scenario weights based on the simulation data warehouse and the core key indicators of the business, and solidifies the acceptable probability of non-compliance with the business by configuring risk tolerance, so that the risk measurement has a unified probability benchmark; the initial threshold of the key indicators is determined by the scenario weighted distribution statistics, and joint opportunity constraint rules are constructed based on the initial threshold of the key indicators, so that multiple core key indicators of the business can be jointly constrained and calculated and verified under the same scenario weight, thereby providing a definite risk constraint boundary for the subsequent operation optimization model, and providing a consistent probability caliber for comparing the actual non-compliance probability of re-simulation stress test.

[0055] In one embodiment of the present invention, basic factory operation parameters and rolling window parameters are obtained, and an operation optimization model is constructed based on joint opportunity constraint rules within the rolling time window to generate a set of candidate operation solutions, including:

[0056] Step 51: Obtain the factory operation basic parameters and rolling window parameters. The factory operation basic parameters characterize the operational objects and resource status within the rolling time window, including order information, equipment information, personnel information, and material information. Order information includes at least the order quantity, delivery time requirements, process route, and priority; equipment information includes at least equipment capacity, available time periods, processing cycle time, and changeover constraints; personnel information includes at least shift arrangements, skill matrix, and attendance availability; and material information includes at least inventory levels, batch status, replenishment arrival time, and requisition constraints. The rolling window parameters characterize the time boundary and recalculation frequency of the rolling solution, including the current moment, the optimization time window length, and the scheme recalculation interval. The optimization time window length is the time span covered by the current round of solution; the scheme recalculation interval is the time interval between two adjacent rounds of rolling solution. Based on the current moment and the optimization time window length, the rolling time window is determined by using the current moment as the starting point and extending the optimization time window length backward from the current moment as the ending point.

[0057] Step 52: Construct an operational optimization model within the rolling time window. This operational optimization model is an optimization calculation model for factory operation management and resource allocation, used to solve for the optimal values ​​of operational decision variables under operational constraints. To ensure that the risk constraint boundary remains consistent during the rolling solution process, the joint opportunity constraint rules and the initial thresholds of key indicators are written into the constraint set of the operational optimization model and remain unchanged. The constraint set is a set of constraints in the operational optimization model used to limit the range of values ​​for operational decision variables. The joint opportunity constraint rules are used to limit the joint risk boundary of core business indicators under scenario weighting, and the initial thresholds of key indicators are used as threshold parameters for the joint opportunity constraint rules.

[0058] Furthermore, operational decision variables are defined within the rolling time window. These variables represent adjustable decision-making content of the operational plan, including time slot assignment for orders on equipment, skill matching for shifts, and material preparation and handling plans. Specifically, time slot assignment for orders on equipment determines the equipment allocation and execution time slot corresponding to the order within the rolling time window; skill matching for shifts determines the personnel configuration and skill coverage for each shift within the rolling time window; and the material preparation and handling plan determines the time arrangement for material preparation and handling, as well as the corresponding resource allocation, within the rolling time window. Through these definitions, the solution output of the operational optimization model can be directly solidified into subsequent executable candidate operational plans.

[0059] Step 53: While keeping the joint opportunity constraint rules unchanged, construct a comprehensive optimization objective and solve the operation optimization model to generate a set of candidate operation schemes. The comprehensive optimization objective is used to quantitatively evaluate the merits of different values ​​of operation decision variables, and is composed of a weighted sum of the average level of total order delays, the fluctuation level of total order delays, the average level of work-in-process inventory, and the average level of cost and energy consumption, along with the weights of the comprehensive optimization objective.

[0060] Specifically, the average level of total order delays is used to characterize the overall delay level under scenario weights. It is obtained by multiplying the total order delays for each scenario by their corresponding scenario weights and then summing the results. The total order delays are the summed statistics of order delays derived from candidate operational plans within the rolling time window in the order dimension. The volatility level of total order delays is obtained by multiplying the difference between the total order delays for each scenario and the average level of total order delays by itself to obtain the squared difference. This squared difference is then multiplied by the corresponding scenario weights and summed. The squared difference measures the deviation of the total delays for each scenario from the average level, thus enabling the volatility level to characterize the dispersion of total order delays in the scenario dimension. The average level of work-in-process inventory is used to characterize the system's work-in-process load level under multiple scenarios. It is obtained by multiplying the work-in-process inventory for each scenario by their corresponding scenario weights and then summing the results. The work-in-process inventory is the scenario statistic of the work-in-process inventory within the rolling time window. The average cost and energy consumption level is used to characterize the comprehensive cost and energy consumption level under multiple scenarios. It is obtained by multiplying the cost and energy consumption of each scenario by the corresponding scenario weight and summing them. The cost and energy consumption is the scenario statistics of cost and energy consumption derived from candidate operation plans within the rolling time window.

[0061] For different comprehensive optimization objective weight configurations, the operation optimization model is solved to obtain multiple sets of operation decision variable values. Each set of operation decision variable values ​​is then mapped to a rolling time window and solidified as a candidate operation plan. Furthermore, all candidate operation plans are aggregated to form a candidate operation plan set. This candidate operation plan set serves as the input for subsequent re-simulation stress testing, ensuring that each candidate operation plan is comparable under the same joint opportunity constraint rule boundary and enabling consistent risk testing and plan selection within a trusted digital twin.

[0062] Through the above implementation process, this invention uses scenario weights as a unified weighting caliber to uniformly map the statistics of total order delays, work-in-process inventory, and cost and energy consumption under multiple scenario disturbances into calculable target components. This enables the operation optimization model to quantitatively compare candidate operation schemes under the same probability caliber. At the same time, it introduces the fluctuation level of total order delays as a dispersion index, so that the comprehensive optimization objective not only represents the average level, but also reflects the stability differences of operation results under different scenarios. This supports a unified evaluation of the risk sensitivity of the schemes during the rolling resource optimization process.

[0063] In one embodiment of the present invention, a set of candidate operational plans is input into a trusted digital twin for re-simulation stress testing to obtain the actual failure probability; the actual failure probability is compared with the business-acceptable failure probability, and the safety margin and key indicator thresholds are updated based on the comparison result; and an operational plan that can be released is determined based on the updated key indicator thresholds, including:

[0064] Step 61 involves inputting the candidate operation plan set into a trusted digital twin for re-simulation stress testing. This re-simulation stress test involves repeatedly running simulations under multiple scenario disturbances while keeping the trusted digital twin's parameter configuration unchanged, using the candidate operation plans as inputs to evaluate the distribution of indicators and the risk of non-compliance under different scenarios. Specifically, for each candidate operation plan in the candidate operation plan set, the values ​​of its operational decision variables and rolling time window are read; scenario identifiers, key disturbance factor values, and scenario weights are read from the simulation data warehouse and used as the re-simulation scenario input set. This re-simulation scenario input set ensures that the scenario caliber used in the re-simulation is consistent with the statistical caliber of the previous scenario. Subsequently, the operational decision variable values ​​corresponding to the candidate operation plans and the re-simulation scenario input set are input into the trusted digital twin, and the simulation is run while keeping the trusted digital twin's parameter configuration unchanged, outputting the results of the core business key indicators. The results of the core business key indicators are the set of indicator results calculated according to the core business key indicator caliber under each scenario, and each corresponds one-to-one with the scenario identifier.

[0065] Step 62: In the actual non-compliance probability calculation stage, the satisfaction of the core key business indicators for each scenario is determined based on the joint opportunity constraint rules and the initial threshold of key indicators. The satisfaction determination refers to determining whether the throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators for each scenario simultaneously meet the threshold conditions defined by the initial threshold of key indicators. These threshold conditions are provided by the initial threshold of key indicators and their joint satisfaction relationship is limited by the joint opportunity constraint rules. For scenarios that do not meet the initial threshold of key indicators, their corresponding scenario weights are read and accumulated to obtain the sum of non-compliance weights; this sum of non-compliance weights is determined as the actual non-compliance probability. The actual non-compliance probability is a probability measure of the risk of non-compliance of the candidate operating plan under the scenario weight caliber.

[0066] Step 63: In the threshold update and scheme determination stage, the actual failure probability is compared with the business-acceptable failure probability. The business-acceptable failure probability is the probability benchmark corresponding to the risk tolerance configuration, used to limit the upper limit of failure acceptable at the operational level. When the actual failure probability is higher than the business-acceptable failure probability, the safety margin is updated and the key indicator threshold is updated based on the initial key indicator threshold; when the actual failure probability is not higher than the business-acceptable failure probability, the safety margin and the key indicator threshold remain unchanged. The safety margin is the threshold adjustment amount or adjustment rule used to adjust the risk margin of the key indicator threshold, and it maintains the same direction as the key indicator threshold update to ensure the uniformity of the threshold update. After completing the key indicator threshold update, based on the updated key indicator threshold and according to the joint opportunity constraint rule, the candidate operation schemes are judged for satisfaction to determine the releasable operation scheme. The releasable operation scheme is the candidate operation scheme judged to be satisfied under the updated key indicator threshold and the joint opportunity constraint rule, which serves as the sole source of schemes for subsequent operation execution package generation and execution data write-back.

[0067] Through the above implementation process, this invention performs re-simulation stress tests on the candidate operation plan set and outputs the results of core key business indicators, making the indicator distribution of candidate operation plans under multi-scenario disturbance conditions calculable; by accumulating scenario weights to obtain the actual failure probability and comparing it with the acceptable failure probability under the same probability caliber, the risk of the plan can be quantitatively judged with a unified probability benchmark; furthermore, the comparison results drive the update of safety margin and key indicator thresholds, and determine the releasable operation plan under the updated key indicator thresholds, so that the risk threshold boundary and plan selection form a closed-loop calibration mechanism, thereby enhancing the stability and consistency of plan release during the rolling resource optimization process.

[0068] In one embodiment of the present invention, an operation execution package is released, actual operation data is collected and deviation data is obtained, and the actual operation data and deviation data are respectively written into a multi-source measured dataset and a simulation data warehouse, including:

[0069] Step 71: Generate an operational execution package based on the publishable operational plan. The operational execution package is a structured and solidified result of the publishable operational plan, used to express the execution arrangements of the publishable operational plan within a rolling time window, and serving as the attribution benchmark for subsequent actual operational data collection. Specifically, the values ​​of operational decision variables in the publishable operational plan are bound to the rolling time window and solidified into an operational execution package, ensuring that the operational execution package uniquely corresponds to the operational arrangements within the current rolling time window. A version identifier is generated for each operational execution package, ensuring a one-to-one correspondence between the version identifier and the rolling time window. The version identifier is a unique identifier used to identify the version of the operational execution package and its corresponding execution scope, ensuring that subsequently collected actual operational data and deviation data can establish a definite association with the corresponding operational execution package.

[0070] Step 72: Collect actual operational data within the rolling time window corresponding to the operational execution package and generate deviation data. The actual operational data refers to the operational result data collected during the execution of the operational execution package, which includes at least the actual order completion time, actual work-in-process quantity, and actual cost / energy consumption. The actual work-in-process quantity is used to characterize the actual statistical level of work-in-process within the rolling time window. Further, generate scheme prediction data based on the publishable operational scheme. The scheme prediction data is the prediction output data corresponding to the publishable operational scheme within the same rolling time window, and its scope is consistent with the actual operational data, used to form an alignable comparison benchmark. Subsequently, align the actual operational data and the scheme prediction data using the same primary key. The same primary key is a primary key system used to achieve consistent indexing across dimensions such as orders, processes, equipment, shifts, and material batches, and can follow the primary key scope formed by the aforementioned unified primary key mapping rules. Calculate the differences for the actual order completion time, actual work-in-process quantity, and actual cost / energy consumption to obtain deviation data, and bind the deviation data to a version identifier. The deviation data is used to characterize the degree of deviation between the operational execution results and the scheme prediction data.

[0071] Step 73: Write actual operational data into the multi-source measured dataset and deviation data into the simulation data warehouse, and solidify the writing rules. The writing rules are as follows: actual operational data corresponding to the same version identifier is only written into the multi-source measured dataset, and deviation data corresponding to the same version identifier is only written into the simulation data warehouse. Through this partitioned writing method, the multi-source measured dataset continuously accumulates real operational observation data, and the simulation data warehouse continuously accumulates deviation data related to scenario statistics, risk measurement, and model calibration. This ensures that the two types of data are independent and traceable in terms of storage objects, statistical definitions, and usage, avoiding data definition conflicts caused by mixed writing.

[0072] Through the above implementation process, this invention establishes a definite correspondence between the publishable operation plan, actual operation data, and deviation data by identifying the operation execution package version, forming a traceable attribution link; by aligning the actual operation data and the plan prediction data with the same primary key and calculating the deviation data, the quantification of the deviation between prediction and execution is achieved; by writing the actual operation data and deviation data into the multi-source measured dataset and the simulation data warehouse respectively and solidifying the partitioning and writing rules, the next round of trusted digital twin calibration and scenario statistics are updated under a consistent caliber, forming a closed-loop iteration.

[0073] This invention provides a factory operation intelligent prediction and resource optimization system based on simulation data, comprising:

[0074] The twin calibration module is used to acquire multi-source measured datasets and initial simulation parameter vectors, and calibrate the initial simulation parameter vectors based on the multi-source measured datasets to obtain a reliable digital twin.

[0075] The scenario simulation module is used to acquire key disturbance factors, generate a scenario set and configure scenario weights, input the scenario set into a trusted digital twin to run the simulation, and obtain a simulation data warehouse.

[0076] The indicator prediction module is used to set core key business indicators, train the key indicator prediction model based on the simulation data warehouse, and obtain the key indicator prediction results.

[0077] The risk constraint module is used to determine the acceptable failure probability of business based on the simulation data warehouse, construct joint opportunity constraint rules based on the acceptable failure probability of business, and determine the initial threshold of key indicators based on the simulation data warehouse.

[0078] The operation optimization module is used to obtain basic factory operation parameters and rolling window parameters, construct an operation optimization model based on joint opportunity constraint rules within the rolling time window, and generate a set of candidate operation solutions.

[0079] The stress calibration module is used to input the set of candidate operation plans into a trusted digital twin for re-simulation stress testing to obtain the actual failure probability; compare the actual failure probability with the business-acceptable failure probability, update the safety margin and key indicator thresholds based on the comparison results, and determine the release operation plan based on the updated key indicator thresholds.

[0080] The execution write-back module is used to publish the operation execution package, collect actual operation data and obtain deviation data, and write the actual operation data and deviation data into the multi-source measured dataset and simulation data warehouse, respectively.

[0081] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0082] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of the present embodiments, all of which are within the protection scope of the present embodiments.

Claims

1. A method for intelligent prediction and resource optimization of factory operations based on simulation data, characterized in that, Includes the following steps: Step 1: Obtain the multi-source measured dataset and the initial simulation parameter vector. Based on the multi-source measured dataset, calibrate the initial simulation parameter vector to obtain a reliable digital twin. Step 2: Obtain key perturbation factors, generate a scenario set and configure scenario weights, input the scenario set into a trusted digital twin to run simulation, and obtain a simulation data warehouse, including: Step 21: Obtain key disturbance factors, including material delivery delay, product return rate, equipment availability, and production area congestion. Within the calibration time window, determine the value range of each key disturbance factor based on multi-source measured datasets, and generate a scenario set based on the combination of key disturbance factor values, so that the scenario identifier corresponds one-to-one with the combination of key disturbance factor values. Step 22: Based on the multi-source measured dataset, the occurrence frequency of each scenario is statistically analyzed as the initial weight value. The initial weight value of each scenario is divided by the sum of the initial weight values ​​of all scenarios to obtain the scenario weight. The sum of the weights of all scenarios is equal to one, and the scenario weights are bound to the scenario identifiers one by one. Step 23: Input the scenario set into the trusted digital twin and run the simulation. For each scenario, read the value of its key disturbance factor as the scenario simulation input. Under the condition that the trusted digital twin remains unchanged, output the simulation output data. The simulation output data includes the simulation output corresponding to the core key indicators of the business. Write the scenario identifier, key disturbance factor value, scenario weight and simulation output data into the simulation data warehouse. Step 3: Set the core key indicators for the business, train the key indicator prediction model based on the simulation data warehouse, and obtain the key indicator prediction results. Step 4: Determine the acceptable failure probability of the business based on the simulation data warehouse, construct joint opportunity constraint rules based on the acceptable failure probability, and determine the initial thresholds of key indicators based on the simulation data warehouse, including: Step 41: Based on the simulation data warehouse and according to the business core key indicator caliber, calculate the business core key indicator sample set, so that the business core key indicator sample set corresponds one-to-one with the scenario identifier; synchronously read the scenario weight corresponding one-to-one with the scenario identifier and solidify the scenario weight weighted statistical caliber. Step 42: Read the risk tolerance configuration and determine the acceptable failure probability of the business. Then, solidify the acceptable failure probability of the business into a probability benchmark for subsequent joint opportunity constraint rule verification and comparison with the actual failure probability. Step 43: Based on the simulation data warehouse, perform scenario-weighted distribution statistics on throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators respectively; sort the indicator values ​​and accumulate the corresponding scenario weights; when the accumulated scenario weights reach the target accumulated weight position determined by the business-acceptable non-compliance probability, determine the corresponding indicator value as the initial threshold of the key indicator; construct joint opportunity constraint rules based on the initial thresholds of the key indicators, calculate the sum of scenario weights that simultaneously satisfy the initial thresholds of each key indicator, and compare the sum of scenario weights with the target weight lower limit determined by the business-acceptable non-compliance probability to obtain the satisfaction judgment result of the joint opportunity constraint rules; Step 5: Obtain the basic parameters of factory operation and the rolling window parameters, and build an operation optimization model based on the joint opportunity constraint rules within the rolling time window to generate a set of candidate operation solutions; Step 6: Input the candidate operation plan set into the trusted digital twin for re-simulation stress test to obtain the actual failure probability; compare the actual failure probability with the business acceptable failure probability, update the safety margin and key indicator thresholds based on the comparison results, and determine the release operation plan based on the updated key indicator thresholds. Step 7: Publish the operation execution package, collect actual operation data and obtain deviation data, and write the actual operation data and deviation data into the multi-source measured dataset and simulation data warehouse respectively.

2. The intelligent prediction and resource optimization method for factory operations based on simulation data according to claim 1, characterized in that, Obtain a multi-source measured dataset and an initial simulation parameter vector. Based on the multi-source measured dataset, calibrate the initial simulation parameter vector to obtain a reliable digital twin, including: Step 11: Determine the calibration time window in the multi-source measured dataset and align the multi-source measured dataset according to the unified primary key mapping rule. The unified primary key mapping rule includes order identifier, process identifier, equipment identifier, shift identifier, and material batch identifier. The multi-source measured dataset includes production execution data, enterprise resource planning data, warehouse management data, and cost and energy consumption statistics. Step 12: Generate calibration input data based on the calibration time window, and calculate and solidify the observed values ​​of key calibration indicators based on the multi-source measured dataset; the observed values ​​of key calibration indicators include: throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators. Step 13: Input the calibration input data into the simulation model configured with the initial simulation parameter vector, and output the simulation values ​​of the key calibration indicators within the calibration time window; subtract the simulation values ​​of the key calibration indicators from the observed values ​​of the key calibration indicators one by one to obtain the calibration deviations of each item; Step 14: Multiply each calibration deviation by a preset weight to obtain a weighted deviation, multiply each weighted deviation by itself to obtain a squared deviation, and sum the squared deviations to obtain the calibration target value; iteratively update the initial simulation parameter vector and repeatedly output the simulation values ​​of key calibration indicators and the calibration target value until the calibration deviation meets the preset allowable deviation range, solidify the iteratively updated initial simulation parameter vector to obtain the calibration simulation parameter vector, and obtain a reliable digital twin based on the calibration simulation parameter vector.

3. The intelligent prediction and resource optimization method for factory operations based on simulation data according to claim 1, characterized in that, Define core key business metrics, train a key metric prediction model based on a simulation data warehouse, and obtain the key metric prediction results, including: Step 31: Set core key business indicators based on the simulation data warehouse and solidify the definition of core key business indicators; the core key business indicators include throughput-related indicators, work-in-process quantity-related indicators, and cost and energy consumption-related indicators; the definition of core key business indicators is a fixed calculation method for calculating core key business indicators based on simulation output data in the simulation data warehouse. Step 32: Construct a training sample set based on the simulation data warehouse, using the values ​​of key disturbance factors and scenario weights as input-side features, the core key indicators of the business as output-side labels, and aligning the input-side features and output-side labels with scenario identifiers, while retaining the scenario weights as sample weights in the training sample set. Step 33: Determine the input-output structure of the key indicator prediction model based on the input-side features and output-side labels of the training sample set. Train the key indicator prediction model based on the training sample set and with sample weights participating in parameter fitting. Use the trained key indicator prediction model to infer the key indicator prediction results on the target input set and establish a one-to-one correspondence between the key indicator prediction results and the key perturbation factor values, while retaining the scenario label.

4. The intelligent prediction and resource optimization method for factory operations based on simulation data according to claim 1, characterized in that, Obtain basic factory operation parameters and rolling window parameters, construct an operation optimization model based on joint opportunity constraint rules within the rolling time window, and generate a set of candidate operation solutions, including: Step 51: Obtain basic factory operation parameters and rolling window parameters. The basic factory operation parameters include order information, equipment information, personnel information, and material information. The rolling window parameters include the current time, the length of the optimization time window, and the recalculation interval. The rolling time window is determined by taking the current time as the starting point and extending the optimization time window length backward from the current time as the ending point. Step 52: Construct an operation optimization model within the rolling time window, write the joint opportunity constraint rules and the initial threshold of key indicators into the constraint set of the operation optimization model and keep them unchanged; define operation decision variables within the rolling time window, including the time period assignment of orders on the equipment, the skill matching of shifts, and the material preparation and handling plan. Step 53: Under the condition of keeping the joint opportunity constraint rules unchanged, construct a comprehensive optimization objective. The comprehensive optimization objective consists of the average level of total order delay, the fluctuation level of total order delay, the average level of work-in-process quantity, and the average level of cost and energy consumption. Solve the operation optimization model for different comprehensive optimization objective weight configurations to obtain multiple sets of operation decision variable values. Solidify each set of operation decision variable values ​​and the corresponding rolling time window as candidate operation schemes, and summarize them to form a set of candidate operation schemes.

5. The intelligent prediction and resource optimization method for factory operations based on simulation data according to claim 4, characterized in that, The comprehensive optimization objective is composed of a weighted sum of the average level of total order delays, the fluctuation level of total order delays, the average level of work-in-process inventory, and the average level of cost and energy consumption, along with the weights assigned to the comprehensive optimization objective. The average level of total order delays is obtained by multiplying the total order delays for each scenario by the corresponding scenario weight and summing the results. The fluctuation level of total order delays is obtained by multiplying the difference between the total order delays for each scenario and the average level of total order delays by itself, obtaining the squared difference, multiplying it by the corresponding scenario weight, and summing the results. The average level of work-in-process inventory is obtained by multiplying the work-in-process inventory for each scenario by the corresponding scenario weight and summing the results. The average level of cost and energy consumption is obtained by multiplying the cost and energy consumption for each scenario by the corresponding scenario weight and summing the results.

6. The intelligent prediction and resource optimization method for factory operations based on simulation data according to claim 1, characterized in that, The candidate operational plan set is input into a trusted digital twin for re-simulation stress testing to obtain the actual failure probability. This actual failure probability is compared with the business-acceptable failure probability. Based on the comparison results, the safety margin and key indicator thresholds are updated. Finally, based on the updated key indicator thresholds, the deployable operational plans are determined, including: Step 61: Input the candidate operation plan set into the trusted digital twin for re-simulation stress test; read the values ​​of operation decision variables and rolling time window for each candidate operation plan, and read the scenario identifier, key disturbance factor values ​​and scenario weight from the simulation data warehouse as the re-simulation scenario input set; output the core key indicators of the business under the condition that the parameter configuration of the trusted digital twin remains unchanged. Step 62: Based on the joint opportunity constraint rules and the initial threshold of key indicators, determine the satisfaction of the business core key indicator results for each scenario, accumulate the scenario weights corresponding to the scenarios that do not meet the initial threshold of key indicators to obtain the sum of non-compliance weights, and determine the sum of non-compliance weights as the actual non-compliance probability. Step 63: Compare the actual failure probability with the business-acceptable failure probability; when the actual failure probability is higher than the business-acceptable failure probability, update the safety margin and update the key indicator threshold based on the initial key indicator threshold; when the actual failure probability is not higher than the business-acceptable failure probability, keep the safety margin and the key indicator threshold unchanged; based on the updated key indicator threshold and according to the joint opportunity constraint rules, determine the satisfaction of candidate operation plans and identify the operation plan that can be released.

7. The intelligent prediction and resource optimization method for factory operations based on simulation data according to claim 1, characterized in that, Release the operational execution package, collect actual operational data and obtain deviation data, and write the actual operational data and deviation data into the multi-source measured dataset and simulation data warehouse, respectively, including: Step 71: Generate an operation execution package based on the publishable operation plan, bind the values ​​of the operation decision variables in the publishable operation plan with the rolling time window and solidify them into the operation execution package, and generate a version identifier for the operation execution package so that the version identifier corresponds one-to-one with the rolling time window; Step 72: Collect actual operational data within the rolling time window corresponding to the operational execution package, and bind the actual operational data with the version identifier; generate plan prediction data based on the publishable operational plan, align the actual operational data and the plan prediction data with the same primary key, calculate the difference for the actual completion time of the order, the actual quantity of work-in-process, and the actual cost energy consumption, respectively, to obtain the deviation data, and bind the deviation data with the version identifier; Step 73: Write the actual operation data into the multi-source test dataset and write the deviation data into the simulation data warehouse, and solidify the writing rules so that the actual operation data corresponding to the same version identifier is only written into the multi-source test dataset and the deviation data corresponding to the same version identifier is only written into the simulation data warehouse.

8. A factory operation intelligent prediction and resource optimization system based on simulation data, characterized in that, The method for intelligent prediction and resource optimization of factory operations based on simulation data, as described in any one of claims 1-7, includes: The twin calibration module is used to acquire multi-source measured datasets and initial simulation parameter vectors, and calibrate the initial simulation parameter vectors based on the multi-source measured datasets to obtain a reliable digital twin. The scenario simulation module is used to acquire key disturbance factors, generate a scenario set and configure scenario weights, input the scenario set into a trusted digital twin to run the simulation, and obtain a simulation data warehouse. The indicator prediction module is used to set core key business indicators, train the key indicator prediction model based on the simulation data warehouse, and obtain the key indicator prediction results. The risk constraint module is used to determine the acceptable failure probability of business based on the simulation data warehouse, construct joint opportunity constraint rules based on the acceptable failure probability of business, and determine the initial threshold of key indicators based on the simulation data warehouse. The operation optimization module is used to obtain basic factory operation parameters and rolling window parameters, construct an operation optimization model based on joint opportunity constraint rules within the rolling time window, and generate a set of candidate operation solutions. The stress calibration module is used to input the set of candidate operation plans into a trusted digital twin for re-simulation stress testing to obtain the actual failure probability; compare the actual failure probability with the business-acceptable failure probability, update the safety margin and key indicator thresholds based on the comparison results, and determine the release operation plan based on the updated key indicator thresholds. The execution write-back module is used to publish the operation execution package, collect actual operation data and obtain deviation data, and write the actual operation data and deviation data into the multi-source measured dataset and simulation data warehouse, respectively.

Images