A Method and System for Assessing Enterprise Production Line Control Potential Based on Production Scheduling Strategies

Abstract

This invention relates to the field of production scheduling, and more particularly to a method and system for assessing the control potential of enterprise production lines based on production scheduling strategies. The method extracts multi-dimensional dynamic features based on production line configuration, assessment objectives, and real-time operational data. Candidate strategies are screened from a rule base using feature matching, and a benchmark strategy is determined. An indicator system covering efficiency, cost, flexibility, and stability is constructed, and simulation models are used to obtain the indicator values ​​for the benchmark and candidate strategies. An improved weighted Euclidean distance is used to calculate the control potential value of each candidate strategy relative to the benchmark strategy. A structured report is generated based on the assessment results, and optimal strategy recommendations are output. Simultaneously, the rule base is self-evolved and updated based on performance feedback. This invention achieves dynamic perception, simulation evaluation, and optimization decision-making for production line scheduling strategies, providing decision support for the overall optimized operation of the production line.

Description

Technical Field

[0001] This invention relates to the field of production scheduling, and in particular to a method and system for assessing the production line control potential of enterprises based on production scheduling strategies. Background Technology

[0002] Modern manufacturing enterprises face increasingly fierce market competition, and the operational efficiency and resource utilization of their production lines directly determine their profitability and market responsiveness. Production scheduling strategies, as the core guiding the daily operation of the production line, have a decisive impact on key performance indicators such as production efficiency, cost control, and order delivery.

[0003] Existing technologies have proposed many constructive solutions for production line scheduling. For example, patent application CN121979135A divides production tasks into rigid and flexible tasks and formulates different scheduling strategies. Another example is CN121616016A, which determines a multi-objective production scheduling model and constraints based on the actual production scenario, and uses an improved particle swarm optimization algorithm with adjusted particle encoding and fitness functions to solve the problem, generating a multi-objective production scheduling strategy for the production line to guide actual production arrangements.

[0004] However, existing scheduling evaluation methods still have the following main shortcomings: (1) Static limitations: It cannot effectively capture the nonlinear impact of discrete random disturbances (such as sudden equipment failures and material supply delays) on scheduling potential in the real-time operation of the production line; (2) Data silos and distortion: Traditional evaluation relies on historical average data, which leads to a significant “virtual-real mapping” error between the simulation model and the physical production line; (3) Insufficient handling of dimensional coupling: The weighted summation method commonly used in existing methods ignores the mutual constraints and coupling relationships between indicators such as efficiency, cost, and flexibility. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for assessing the production line control potential of enterprises based on production scheduling strategies.

[0006] This application discloses a method for assessing the production line control potential of enterprises based on production scheduling strategies, and adopts the following technical solution: Multidimensional dynamic features are extracted from the real-time operation data stream of the production line, including bottleneck workstation load rate features, work-in-process fluctuation gradient features, material arrival dispersion features, and material supply quantity deviation features; based on the production line configuration and evaluation objectives, and combined with the multidimensional dynamic features, candidate scheduling strategies are selected from the pre-set scheduling rule base. Establish a multi-dimensional evaluation index system that includes efficiency, cost, flexibility and stability. Load candidate scheduling strategies and preset benchmark scheduling strategies into a unified production line performance simulation model and output the multi-dimensional evaluation index values ​​for each strategy. For each candidate scheduling strategy, calculate its control potential value relative to the benchmark scheduling strategy based on the multi-dimensional evaluation index values. The optimal scheduling strategy is selected from the candidate scheduling strategies according to predefined logic. The predefined logic includes a first-level screening and a second-level screening. The first-level screening selects the candidate scheduling strategy with the highest regulation potential value as the optimal scheduling strategy. The second-level screening selects the optimal scheduling strategy by comparing the evaluation index values ​​of the candidate scheduling strategies if there are parallel candidate scheduling strategies with the same regulation potential value. The optimal scheduling strategy is executed in the physical production line and the actual performance feedback data is obtained. The evaluation index value of the actual operation is calculated. The evaluation index value of the optimal scheduling strategy in the simulation stage is extracted and combined with the evaluation index value of the actual operation to calculate the prediction residual. If the prediction residual exceeds the preset tolerance threshold, the rule adaptation weight of the optimal scheduling strategy in the preset scheduling rule base is adjusted to optimize the next round of evaluation.

[0007] Preferably, the bottleneck workstation load rate characteristic is obtained by the following steps: Set a time period and extract the equipment operation status data of each production line station within that period; calculate the cumulative time of equipment in working state and abnormal shutdown state respectively, and calculate the dynamic load rate of each station; the dynamic load rate is a ratio, the numerator is the cumulative time of equipment in working state, and the denominator is the time period minus the cumulative time of abnormal shutdown state. Traverse all workstations, select the workstation corresponding to the maximum dynamic load rate as the bottleneck workstation for the current cycle, and use the maximum dynamic load rate as the bottleneck workstation load rate characteristic.

[0008] Preferably, the step of obtaining the work-in-process fluctuation gradient characteristics includes: Along the spatial flow of the process, the difference in work-in-process queue length between adjacent workstations is calculated to obtain the spatial fluctuation gradient; the feature extraction sliding time step is set, and the rate of change of the work-in-process queue length at the bottleneck workstation over time is calculated to obtain the temporal fluctuation gradient. The standard deviation of the spatial fluctuation gradient of all workstations and the temporal fluctuation gradient of the bottleneck workstation are concatenated to output the work-in-process fluctuation gradient feature.

[0009] Preferably, the steps for obtaining the material arrival dispersion characteristics and material supply quantity deviation characteristics include: The standard delivery interval is obtained based on the delivery plan, and the actual delivery interval is calculated based on the actual delivery time. The root mean square error between the actual delivery interval and the standard planned interval is calculated to obtain the material delivery time dispersion characteristics. The material supply quantity deviation characteristics are obtained by calculating the average absolute deviation between the actual delivered quantity and the standard planned delivered quantity.

[0010] Preferably, the step of filtering candidate scheduling strategies from a pre-set scheduling rule base includes: Calculate the correlation coefficient between the multidimensional dynamic features and each scheduling rule in the pre-set scheduling rule base; match the business objective tendency score of each scheduling rule according to the evaluation objective of the current business; The process of obtaining the business objective tendency score is described as follows: a rule objective tendency mapping matrix is ​​preset, where each element of the matrix represents the basic tendency coefficient of a scheduling rule for an evaluation objective; the evaluation objective of the current business is transformed into an objective weight vector; for each scheduling rule, the weighted dot product of the objective weight vector and the basic tendency coefficient corresponding to the scheduling rule is calculated to generate the business objective tendency score of the scheduling rule. The correlation coefficients and business objective tendency scores are linearly weighted to calculate the comprehensive fit index of each rule, and candidate scheduling strategies are selected by thresholding.

[0011] Preferably, the multi-dimensional evaluation index system includes: Efficiency is expressed as the overall utilization rate of equipment, which is calculated by combining the equipment's time utilization rate, performance utilization rate, and qualified product rate. Cost is expressed as energy consumption per unit of product, which is obtained by statistically analyzing the total energy consumption within a set period and dividing it by the number of qualified products. The flexibility dimension is represented by the average order delay rate, which is calculated by averaging the delay time between the actual completion time and the promised delivery time for all orders. The stability dimension is represented by the standard deviation of daily output, which is derived by calculating the standard deviation of total daily output.

[0012] Preferably, the calculation process of the regulation potential value includes: A directional gain coefficient is introduced. When the evaluation index is positively improved, the directional gain coefficient is 1; when the evaluation index is negatively controlled, the directional gain coefficient is -1. First, calculate the net improvement index of a single evaluation indicator, which is described as multiplying the difference between the indicator value and the corresponding benchmark scheduling strategy evaluation indicator value, the indicator direction gain coefficient, and the indicator weight. The global net improvement index for each candidate scheduling strategy is obtained by summing the net improvement indices of all evaluation metrics.

[0013] Preferably, the calculation process of the regulation potential value further includes: For the current candidate scheduling strategy, first calculate the product of the corresponding sign function of the global net improvement index and the weighted average sum of the evaluation index, and then subtract the performance standard deviation term of the current candidate scheduling strategy in the simulation operation to obtain the regulation potential value. When the global net improvement index is greater than zero, the sign function takes the value of 1; when the global net improvement index is less than zero, the sign function takes the value of -1. The performance standard deviation term is the product of a preset robustness penalty coefficient and the performance standard deviation of the current candidate scheduling strategy in the simulation operation; The calculation process of the performance standard deviation is described as follows: introduce random perturbation in each simulation, record the multidimensional evaluation index of each simulation output, and calculate the arithmetic mean of each evaluation index in all simulations. The standard deviation of each evaluation index in all simulations is calculated based on the arithmetic mean; the standard deviations of all evaluation indices are weighted and summed to obtain the performance standard deviation.

[0014] Preferably, adjusting the rule adaptation weights of the optimal scheduling strategy in the preset scheduling rule base includes: Subtract the evolution term from the rule adaptation weight of the optimal scheduling strategy in the preset scheduling rule base; obtain the adjusted rule adaptation weight, and write it back to the preset scheduling rule base; The calculation process of the evolution term is described as follows: first, the prediction residuals of all evaluation indicators are weighted, and then multiplied by the preset evolution learning rate.

[0015] The beneficial effects of this invention are that, compared with the prior art, 1. This solution constructs a unified evaluation index system integrating multiple dimensions such as efficiency, cost, flexibility, and stability. Based on a high-fidelity production line performance model with confidence verification, quantitative simulation is performed. By introducing an improved weighted Euclidean distance algorithm with a robustness penalty term, it not only quantifies the comprehensive performance gap between candidate strategies and the benchmark strategy but also fully considers the stability risk of strategies in response to production fluctuations. This achieves a paradigm shift from simply pursuing "theoretical optimality" to considering "executive robustness." It calculates the comprehensive gap between candidate and benchmark strategies across various indicators to quantify the control potential value, thereby realizing a paradigm shift from local optimization of a single indicator to multi-dimensional system-level overall optimality. This effectively avoids the one-sidedness of strategy selection caused by missing dimensions, enabling the evaluation results to comprehensively and balancedly reflect the comprehensive performance of the scheduling strategy. This significantly improves the overall optimality of the production line scheduling scheme and the global perspective of management decisions, ensuring the systematic and comprehensive nature of the evaluation results and providing a decision-making basis for achieving overall optimal production line operation.

[0016] 2. By configuring indicator weights based on the Analytic Hierarchy Process (AHP) and combining this with normalization of indicators across various dimensions, the method effectively solves the problems of inconsistent evaluation of multi-dimensional indicators and the disconnect between strategy selection and business objectives. This approach can transform users' fuzzy, multi-objective evaluation needs (such as prioritizing cost or delivery speed) into precise weight coefficients in the model, thereby ensuring that the final recommended strategy is highly consistent with the company's core strategic needs, achieving customized evaluation and recommendation.

[0017] 3. By establishing a fully automated closed-loop process that evolves from potential assessment and report generation to strategy recommendation and performance feedback, a structured report containing rankings, radar charts, and grade assessments is ultimately generated. The S5 process automatically adjusts the matching logic of the rule base based on performance data. This not only significantly lowers the technical barrier to entry but also endows the system with the ability to self-iterate and continuously optimize. As operational data accumulates, the accuracy of the assessment continuously improves, greatly enhancing the intelligence level of enterprise production management and decision-making efficiency.

[0018] 4. By introducing a benchmark scheduling strategy as a unified performance comparison benchmark and employing digital twin high-fidelity simulation technology that integrates real-time data streams and residual compensation mechanisms, simulations were run under the same production line configuration data by loading different strategy parameters. This enabled accurate prediction and quantitative comparison of production line operating states under various scheduling strategies. This dynamically verified simulation method effectively isolates the influence of scheduling strategies, eliminates evaluation biases caused by the static nature of simulation model parameters and their disconnect from actual operating conditions in traditional assessments, and ensures the accuracy, reliability, and comparability of potential assessment results in real-time changing environments, providing more credible data support for decision-making.

[0019] 5. Through a state-driven intelligent strategy matching mechanism, multi-dimensional dynamic features are extracted from real-time operational data. Based on a feature correlation algorithm, a candidate strategy set is automatically selected from a pre-set scheduling rule base, overcoming the limitations of traditional methods, such as limited strategy selection range and difficulty in handling bottleneck drift and random disturbances. This system can quickly locate the most suitable candidate strategy from various classic scheduling rules based on the real-time load status and bottleneck distribution of the production line. This not only greatly expands the coverage of strategy evaluation but also significantly improves the efficiency and real-time targeting of strategy selection, laying a solid foundation for subsequent accurate evaluation. Attached Figure Description

[0020] Figure 1 The overall flowchart provided for this invention; Figure 2 The system architecture diagram of the evaluation index provided by this invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0022] As an example, see Figure 1 This paper discloses a specific implementation method for assessing the production line control potential of enterprises based on production scheduling strategies.

[0023] S1. By analyzing the static configuration, dynamic operating characteristics, and enterprise optimization goals of the production line, the system intelligently selects the candidate scheduling rules that best fit the current working conditions from the pre-set rule base, thereby achieving rapid adaptation of the scheduling strategy to the actual state of the production line.

[0024] 1.1: Obtain production line configuration data, evaluation targets, and real-time production line operation data stream; (1) Production line configuration data covers all basic information describing the static resources and production constraints of the production line; including but not limited to: the number and layout of equipment, the processing capacity and speed of each piece of equipment, the complete production process flow, the bill of materials (BOM), personnel arrangements, and inventory levels; among which: The number and layout of equipment refers to the total number of physical entities within the production line and their spatial topology and material flow relationship within the workshop. For example, in a combined automotive stamping and welding production line, the number of equipment is specified as 4 large hydraulic stamping units, 12 six-axis welding robots, and 3 continuous roller conveyor lines; the physical layout is defined as a series L-shaped line, specifying the spatial flow trajectory of workpieces between equipment and the length of conveyor slides, thereby determining the material handling time delay boundary in discrete event simulation.

[0025] The processing capacity and speed of each piece of equipment refer to the inherent technical parameter limits and standard operating cycle time of a single piece of equipment. For example, the processing capacity refers to the maximum workpiece clamping size or cutting hardness range supported by a specific CNC machine tool, while the processing speed is specifically quantified as the core time for the machine tool to process a single spindle bearing under standard processes being 45 seconds / piece, and the robot's grasping and stacking cycle being 8 times / minute, which is used as the deterministic step size for subsequent simulation clock advancement.

[0026] The personnel arrangement specifically refers to the number of operators and shift schedules allocated to each process or key manual control link in the complete production process. For example, on a multi-station mixed assembly production line, it is clearly stipulated that the front-end material loading station is configured with 2 people / shift, the middle-end manual assembly and quality inspection station is configured with 3 people / shift, and the rear-end finished product packaging station is configured with 1 person / shift, rather than just referring to the total number of people on the entire production line. This serves as the boundary condition for capacity constraints caused by personnel shift changes, absences, or shift switching in the simulation model.

[0027] (2) The evaluation objective refers to the current core needs of the manufacturing enterprise; for example, cost minimization, on-time delivery rate maximization, capacity utilization rate maximization, etc. (3) The real-time operation data stream of the production line reflects the instantaneous dynamic load of the production line, including the real-time queue length of each workstation, equipment operating status, instantaneous output rate, and random fluctuations in material arrival. Among them: The real-time queue length for each workstation specifically refers to the actual number of work-in-process (WIP) items waiting to be processed or stuck in a conveyor line blockage at the current data collection timestamp within the physical buffer or temporary work-in-process yard at the front end of a particular process. For example, when a precision grinding machine enters a brief unplanned downtime due to periodic wheel dressing, the number of gears waiting to be ground on its upstream connecting conveyor line reaches 15. This queue number directly reflects the severity and congestion gradient of the local dynamic bottleneck in the production line.

[0028] The random fluctuation deviation of material arrival refers to the discrete uncertainty of the actual arrival time and quantity compared to the original planned schedule when upstream suppliers or previous preparation processes deliver materials to the production line entrance. For example, the planned schedule stipulates that an AGV will transport a pallet (50 pieces) of machine casing blanks to the line preparation area every 30 minutes. However, due to scheduling conflicts or warehouse sorting delays, the actual arrival time exhibits random disturbances with a normal or Poisson distribution (e.g., the actual arrival interval fluctuates between 25 and 38 minutes, and the arrival quantity oscillates between 48 and 52 pieces due to losses). This deviation data is used to activate the random driving source in discrete event simulation to evaluate the resilience of the scheduling strategy in the face of material supply chain disturbances.

[0029] 1.2: Multidimensional dynamic feature extraction based on the data obtained in 1.1.

[0030] To transform the heterogeneous, massive, and unstructured production line configuration data and real-time operational data streams obtained in step 1.1 into control inputs that can be directly recognized by the scheduling rule base, the following specific technical extraction steps are performed to construct a multi-dimensional dynamic feature vector describing the current real-time operating conditions of the production line: (1) Bottleneck workstation load rate feature extraction: The bottleneck workstation load rate is used to characterize the dynamic busyness of key constraint nodes in the production line. The specific technical extraction steps are as follows: 1) Based on the complete production process described in 1.1, establish a set of production line workstations. ; This is the workstation index.

[0031] 2) Set the feature extraction rolling window period to (For example, in the past 2 hours), continuously retrieve data from each workstation within that period via the on-site industrial interface. j Equipment operating status data; 3) The cumulative time that the equipment is in working ("busy / processing") state. The cumulative time of "abnormal downtime" caused by unplanned downtime, tool pruning, or fault maintenance. ; 4) Calculate the value of each workstation according to the formula. j Dynamic load factor :

[0032] 5) Traverse the workstation set J The workstation corresponding to the maximum dynamic load rate is selected as the dynamic bottleneck workstation for the current cycle, and this maximum load rate is set... This is the extracted bottleneck workstation load rate feature. As a scalar, this feature directly reflects the congestion intensity of the core bottleneck node of the production line.

[0033] (2) Extraction of Work-in-Process Fluctuation Gradient Features: The work-in-process (WIP) fluctuation gradient is used to characterize the imbalance and accumulation trend of production line logistics in spatial distribution. The specific extraction steps are as follows: 1) Real-time capture of the front-end physical buffers of each workstation as described in section 1.1 at the current sampling time. t The real-time queue length (i.e., the actual number of work-in-process items). ; 2) Calculate the adjacent workstations (e.g., workstations) along the spatial flow of the process. j The downstream work station adjacent to it j The difference in queue length between +1) yields the spatial fluctuation gradient. : ; 3) Set the sliding time step for feature extraction (For example, every 5 minutes), calculate the rate of change of the bottleneck workstation queue length over time to obtain the time fluctuation gradient:

[0034] 4) The standard deviation of the spatial fluctuation gradient of all workstations. and the time fluctuation gradient of the bottleneck workstation The data is concatenated, and the output is the work-in-process fluctuation gradient feature vector. This feature is used to quantitatively assess whether there is a risk of localized material blockage or backlog in the production line, or a risk of material shortage and idling.

[0035] (3) Extraction of random deviation features of material arrival: The random deviation of material arrival is used to quantify the dynamic disturbance caused by the supply chain or upstream transportation to this production line. The specific technical extraction steps are as follows: 1) Read the Bill of Materials (BOM) and the corresponding AGV standard delivery schedule described in 1.1 to obtain the standard delivery time interval. (e.g., delivery every 30 minutes); 2) Record continuous operation during actual operation K The timestamp of the AGV cart's actual arrival at the line preparation area (e.g., the last 10 times). Calculate the time difference between two consecutive actual deliveries. ; 3) By calculating the root mean square error (RMSE) between the actual delivery interval and the standard planned interval, the dispersion characteristics of material delivery time are extracted. :

[0036] 4) Synchronous statistics K In this delivery, the actual quantity of a single pallet received... Compared with standard planned delivery quantity The absolute deviation average of (e.g., 50 pieces) is used to extract the material supply quantity deviation characteristics. :

[0037] 5) and The combination is a random deviation characteristic pair of material arrival, which is used to quantitatively characterize the impact level of external uncertainty on the stable operation of the production line.

[0038] (4) Feature vector construction: Finally, the bottleneck workstation load rate scalar and work-in-process fluctuation gradient feature vectors extracted above are concatenated and dynamically normalized to construct a multi-dimensional dynamic feature vector. This feature vector F This constitutes a "digital fingerprint" of the real-time production status of the physical production line within the current evaluation cycle, which is directly input into the subsequent feature correlation matching algorithm to activate the most suitable scheduling rule.

[0039] 1.3: Based on the multi-dimensional dynamic features extracted from configuration data, evaluation targets, and real-time operation data streams, intelligent matching and filtering are performed from a pre-set scheduling rule library to ultimately select one or more candidate scheduling strategies that are highly relevant to the current production line configuration and evaluation targets.

[0040] This rule base integrates various classic scheduling rules, such as "First-Come, First-Served (FIFO)," "Shortest Processing Time First (SPT)," "Earliest Delivery Date First (EDD)," and "Critical Ratio (CR)." Among them, the scheduling rule base predefines a baseline scheduling strategy, which serves as a unified comparison benchmark for all subsequent potential calculations.

[0041] The benchmark scheduling strategy serves as the zero-line benchmark for potential accounting, and its acquisition method specifically supports the following two types of industrial operation modes: (1) Historical performance benchmark mode: The scheduling logic of the production execution system (MES) that is actually running in the physical production line during the current evaluation period is directly read through the data interface (such as the existing manual experience scheduling rules on site). (2) Standard cold start baseline mode: If the physical production line is a newly built production line or the MES fails to export the current operating rules, the system will automatically activate the first-come-first-served (FIFO) rule specified in the rule base as the initial baseline scheduling strategy to objectively reflect the basic performance of the production line without active optimization intervention.

[0042] To improve the accuracy of the screening, this step implemented a two-dimensional matching architecture that combines "state feature-driven" and "business goal-oriented" approaches: (1) State Feature-Driven Dimension: This dimension aims to address the physical fit between scheduling rules and the real-time congestion status and bottleneck distribution of the current physical production line. In the pre-built scheduling rule library, each scheduling rule (such as FIFO, SPT, EDD, CR, etc.) is pre-encapsulated with its corresponding "performance envelope feature distribution area" (i.e., the working condition feature boundary that the rule is best at handling mathematically). Mathematically, the "performance envelope feature distribution area" is specifically represented as a standard feature vector template. This template is a reference benchmark vector composed of various multi-dimensional feature components (such as the center value of the bottleneck workstation load rate that is most suitable for the rule to play its role, the scalarized representation of the ideal range of work-in-process fluctuation gradient, etc.), used to accurately and quantitatively describe the specific working condition feature boundary that the scheduling rule is most suitable for.

[0043] This dimension is essentially a correlation analysis and ranking logic driven by state features, which extracts feature vectors from the production line's operating state. F (e.g., bottleneck workstation load rate, work-in-process fluctuation gradient), using the cosine similarity algorithm, calculate the current state feature vector. FThe cosine of the angle between the standard feature vector templates corresponding to each scheduling rule and the target feature vector template in the multidimensional space is used to quantitatively evaluate the fit between the current state features and the performance envelope of each scheduling rule.

[0044] For example, when an unplanned disturbance occurs on-site, such as a precision grinding machine station experiencing a brief unplanned downtime due to periodic wheel dressing, the feature vector extracted in step 1.2 is represented by: bottleneck station load rate. (Under extremely high load and congestion conditions), work-in-process time fluctuation gradient =4 pieces / 5 minutes (indicating that upstream gear components are rapidly accumulating). Since the "Shortest Processing Time First (SPT)" rule inherently tends to quickly clear congestion, its pre-defined standard feature vector template precisely corresponds to this type of high-load, high-backlog gradient operating condition. At this point, the correlation analysis algorithm, through matrix multiplication, calculates that the performance envelope correlation coefficients between this dynamic operating condition characteristic and the "Shortest Processing Time First (SPT)" and "Critical Ratio (CR)" rules are as high as 0.89 and 0.85, respectively, while the correlation coefficient with the "First-Come, First-Served (FIFO)" rule is only 0.21. The algorithm then dynamically sorts the rules from high to low based on the correlation coefficients, prioritizing scheduling rules that can quickly clear the buffer queue and release local bottleneck pressure.

[0045] (2) Business objective orientation dimension: This dimension aims to solve the strategic consistency problem between scheduling rules and the macro-management demands of senior management.

[0046] To provide a rigorous quantitative basis for subsequent "preference scores," the system utilizes a pre-defined mapping between evaluation targets and scheduling rule performance preferences (this mapping is constructed at the system's underlying level as a multi-dimensional "rule-target preference mapping matrix," i.e., a multi-dimensional data table that pre-defines the basic preference coefficients of various scheduling rules for each basic evaluation target; its row vectors correspond to various scheduling rules in the rule base, and its column vectors correspond to the standardized set of core enterprise evaluation targets). Combined with the dynamic bottleneck characteristics of the current production line, the evaluation targets are transformed into priority selections and specific quantitative scores for a particular set of rules. During the actual calculation, the system converts the user's evaluation targets into target weight vectors. (This reflects the degree of importance attached to cost, delivery time, and efficiency, and) Then, by calling the aforementioned mapping matrix, and calculating the weighted dot product between the target weight vector and the basic tendency coefficient vector corresponding to each scheduling rule, the business target tendency score of each rule is rigorously calculated. The specific calculation formula is as follows:

[0047] In the formula, For the first The business objective tendency score for each scheduling rule; For the first The weights of various evaluation objectives; The first element in the mapping matrix pre-calibrated using historical offline data statistics and expert experience The rule for the first The basic tendency coefficient of a target.

[0048] For example: When the evaluation objective is "cost minimization" (i.e., the cost objective weight is set to the highest), rules such as SPT and CR, which can reduce equipment switching costs, are prioritized for matching (these rules have the highest propensity scores calculated by the above formula). When the evaluation objective is "maximizing on-time delivery rate", priority should be given to matching rules that focus on delivery time, such as EDD and CR. When the evaluation objective is "maximizing capacity utilization", rules such as FIFO and SPT that improve equipment utilization should be prioritized.

[0049] After calculating the above two dimensions separately, a weighted fusion method is used to combine the "state feature correlation coefficient" (denoted as ) of each scheduling rule. ) and "Business Goal Tendency Score" (denoted as A linear combination is performed, and the comprehensive fitness index of each rule is calculated and denoted as . The specific weighted fusion formula is as follows:

[0050] In the formula and The aggregation weights for the preset feature dimensions and target dimensions are respectively (satisfying) ), The rule adaptation weights are used for self-evolutionary updates based on the feedback of physical output performance residuals.

[0051] The system sets a preset threshold (e.g., overall adaptability). The scheduling rules that exceed the threshold are filtered out, their abstract mathematical shells are stripped away, the corresponding constraint parameters are loaded, and they are encapsulated into an executable parameterized scheme, thus forming a set of candidate scheduling strategies.

[0052] S2. By establishing a multi-dimensional index system and using a high-fidelity simulation model, the improvement potential (P-value) of the candidate strategy relative to the benchmark strategy is quantitatively calculated.

[0053] The baseline scheduling strategy is the initial scheduling strategy for the first time the production line is introduced, or the scheduling strategy currently used by the production line during the evaluation period.

[0054] 2.1 Construct an evaluation index system that includes multiple key performance indicators; The evaluation index system is a multi-level, multi-dimensional comprehensive structure designed to comprehensively and balancedly reflect the overall performance of the production line. It includes four dimensions: efficiency, cost, flexibility, and stability, with specific, quantifiable key performance indicators (KPIs) under each dimension; among which: (1) The efficiency dimension is measured by the Equipment Overall Utilization Rate (OEE), which is a comprehensive indicator for measuring the degree of equipment effectiveness; its calculation formula is:

[0055] This formula comprehensively evaluates equipment production efficiency from three aspects: time, speed, and quality. It comprehensively reflects the equipment's availability, performance, and quality level, and is a core indicator for assessing production line efficiency. Specifically: Time utilization rate This calculation factor considers downtime losses within the planned production time and represents the proportion of time that equipment is actually operational during the planned production period, excluding time losses due to external interruptions such as equipment failures and process adjustments. The formula is as follows:

[0056] In the formula, Planned load time is defined as the total scheduled working hours within the current assessment window minus the planned downtime (such as statutory rest time and pre-shift routine inspection time). The actual operating time of the equipment is defined as the sum of planned load time minus various unplanned dynamic downtime.

[0057] Performance utilization rate This calculation takes into account the losses from equipment idling and speed reduction, and is used to characterize the extent to which the equipment's production cycle time reaches the ideal design level during actual operation. The calculation formula is as follows:

[0058] In the formula, The theoretical standard processing cycle for a single product (i.e., the inherent standard cycle time of the equipment). For the actual running time of the equipment The total number of products produced internally, including the sum of all qualified and defective products produced in the simulation environment; numerator This is the ideal net operating time required to theoretically complete the total output.

[0059] The pass rate takes into account the losses caused by quality defects. (2) The cost dimension is measured by the energy consumption per unit product. This indicator is obtained by statistically analyzing the total energy consumption within a specific period and dividing it by the number of qualified products. It directly reflects the energy utilization efficiency of the production process and is a key monitoring point for green manufacturing and cost control.

[0060] (3) The flexibility dimension is measured by the average order delay rate. This indicator is obtained by calculating the average delay time between the actual completion time and the promised delivery time of all orders. The lower the value, the stronger the production line's ability to adapt to changes and deliver on time.

[0061] The calculation method for this indicator is as follows: First, for each order, calculate the difference between its actual completion time and the promised delivery date, and only if the actual completion time is later than the promised delivery date, the difference is included in the delay time; otherwise, it is counted as zero. Then, calculate the arithmetic mean of the delay times for all orders.

[0062] (4) Stability dimension is measured by the standard deviation of daily output. This indicator is obtained by calculating the standard deviation of daily total output. The smaller the value, the smaller the fluctuation of production output, and the higher the feasibility of production plan and the accuracy of prediction.

[0063] It should be noted that, due to the significant differences in dimensions and numerical magnitudes among the four dimensions of indicators mentioned above (e.g., OEE is a percentage, while energy consumption is a larger energy unit), this step, while determining the indicator calculation standards, also includes normalization processing for each indicator (such as using Min-Max normalization or Z-score standardization). This process ensures that indicators of different natures can be weighted and compared under the same mathematical dimension, eliminating the misleading influence of dimensions on subsequent potential calculation results. The multi-level comprehensive structure of this evaluation indicator system can be found in [reference needed]. Figure 2 As shown.

[0064] 2.2 Running the model and outputting indicators: Based on the evaluation indicator system, run the preset production line performance model.

[0065] Discrete Event Simulation (DES) technology is used to simulate production line operation in a virtual environment. Its core working principle is to define and serialize discrete events such as "workpiece arrival," "equipment starts processing," "equipment completes processing," "workpiece transfer," and "order completion," and dynamically advance the simulation clock according to preset scheduling rules (i.e., loaded strategy parameters), thereby simulating the actual operation of the production line. The simulation is based on the same set of production line configuration data obtained in step S1. By loading different strategy parameters, the simulation is driven to simulate the dynamic state of the production line under different strategies.

[0066] A digital mapping is constructed based on the production line configuration data obtained from S1. Combined with the real-time data stream of S1, the model parameters (such as MTBF equipment failure interval and processing time fluctuation coefficient) are dynamically corrected through Kalman filtering or residual compensation mechanism to make the simulation approximate the real working conditions.

[0067] Furthermore, the baseline scheduling strategy parameters are loaded into the simulation model, and a set of quantitative values ​​for each indicator under the "baseline state" is output. ; Load the current candidate scheduling strategy parameters into the same simulation model (i.e., the same set of production line configuration data), and output a set of quantitative values ​​for each indicator under the candidate state. ; During simulation, the system automatically collects underlying data and calculates evaluation metrics across four dimensions: efficiency, cost, flexibility, and stability. By loading parameters for different scheduling strategies and running the simulation model under identical initial conditions (i.e., the same set of production line configuration data), it can accurately predict and output quantified metric values ​​for both the "baseline state" and "candidate state." This "controlled variable" simulation method ensures the accuracy and comparability of the potential assessment results.

[0068] 2.3: Calculate the regulation potential value using the improved weighted Euclidean distance formula, and comprehensively evaluate the merits of candidate strategies.

[0069] The adjustment potential value quantifies the potential ability of a candidate scheduling strategy to optimize overall production line performance compared to the baseline strategy. To accurately distinguish whether a candidate strategy leads to positive improvement or negative degradation, and to overcome the mathematical limitation of traditional Euclidean distance where the result is always positive due to squaring operations, this adjustment potential value is designed as a scalar value with a directional sign. A larger and positive value indicates a greater expected improvement in overall production line performance after adopting the candidate strategy; a negative value clearly indicates that the candidate strategy leads to a degradation in overall production line performance.

[0070] Before calculating the potential value, the index directional gain coefficient is first introduced. Calculate the global net improvement index before geometric mapping. If the first i If the evaluation indicator is a positive improvement indicator (such as efficiency, the higher the value, the better), then set... =1; If it is a negative control indicator (such as cost, the smaller the value, the better), then set it to 1. =-1. Then calculate the net improvement index:

[0071] in, The output of the candidate policy i The normalized quantitative value of each evaluation indicator. The first output of the baseline strategy i Normalized quantitative values ​​of each evaluation indicator; For the first The weights of each evaluation indicator. This index is the first to pinpoint the true direction of strategy evolution: when A value greater than 0 indicates overall positive improvement; when... When the value is less than 0, it indicates an overall negative degradation.

[0072] Based on the above direction verification logic, the formula for calculating the regulation potential value is as follows: ; To regulate potential values, For standard symbolic functions, when When >0, the value is 1; when When the value is less than 0, it takes the value of -1, thus assigning a positive or negative sign to the geometric distance term to represent the actual improvement or degradation. The total number of indicators; This represents the standard deviation of the candidate strategy's performance across multiple Monte Carlo simulation runs, used to characterize the strategy's stability risk; its specific calculation basis is as follows: (1) Set the total number of Monte Carlo simulation samples to be M Each simulation run introduces random dynamic disturbance flows, including unplanned equipment downtime and material arrival deviations. (2) Record the candidate scheduling strategy in the first... m ( m =1,2,..., M In the )th simulation run, the )th i The normalized quantitative output value of each evaluation indicator is denoted as... ; (3) Calculate the first step of the candidate scheduling strategy. i Each evaluation indicator in M Arithmetic mean in the second simulation run : (4) Calculate the first step of the candidate scheduling strategy. i Individual standard deviation of each evaluation indicator under dynamic disturbance This is used to measure the ability of a single indicator to withstand fluctuations.

[0073] (5) Finally, the index weights determined by the Analytic Hierarchy Process (AHP) are combined. The total performance standard deviation is obtained by weighted linear composite of the individual standard deviations. Closed-loop calculation value:

[0074] The preset robustness penalty coefficient In a further implementation, the indicator weight Configuration is achieved through the Analytic Hierarchy Process (AHP) based on the user's evaluation objectives (cost / delivery time / efficiency) to meet [the requirements]. .

[0075] S3: Transform the potential values ​​calculated by simulation into a visual basis for decision-making, select the optimal strategy through preset rules, and finally implement it on the physical production line for verification.

[0076] 3.1 Based on the regulation potential values ​​of all candidate strategies output by S2, sort and classify each candidate strategy.

[0077] The candidate strategies are sorted in descending order, and the potential level range is defined. All candidate strategies are divided into "high potential strategy", "medium potential strategy" and "low potential strategy" to facilitate the rapid identification of high-quality strategies.

[0078] In a feasible implementation, the differences between each candidate strategy and the benchmark strategy in four dimensions—efficiency, cost, flexibility, and stability—can be graphically displayed to intuitively expose the advantages and disadvantages of each strategy.

[0079] The potential values ​​of each candidate strategy are ranked, a difference graph with the benchmark strategy is provided, and the potential level of each strategy is assigned to form a structured evaluation report.

[0080] 3.2: Based on the structured evaluation report generated in 3.1, the optimal scheduling strategy is selected from the candidate strategies according to predefined logic.

[0081] The predefined logic includes primary filtering and secondary filtering; specifically: First-level screening: The candidate scheduling strategy with the highest regulation potential value is selected as the optimal scheduling strategy. Secondary screening: If there are ties in potential values, that is, if there are multiple highest values ​​with the same control potential, then according to the evaluation target in S1, the key indicators of the ties in candidate strategies are compared and the best one is selected for output; where, "key indicators" are defined as the core indicators corresponding to the user's evaluation target (e.g., when the evaluation target is "cost minimization", the key indicator is unit product energy consumption; when the evaluation target is "delivery on time rate maximization", the key indicator is average order delay rate).

[0082] If the evaluation objective is multiple (such as "cost + on-time delivery rate"), then the indicators are ranked by weight, and the indicator with the highest weight is compared first.

[0083] This hierarchical and progressive recommendation logic design ensures that suggestions can be provided quickly in most cases, while also possessing the ability to refine decisions when faced with multiple equally matched candidate strategies. Furthermore, after outputting the optimal scheduling strategy suggestion, the system enters the performance monitoring and feedback phase. Through an Industrial Internet of Things (IIoT) interface, the actual key performance indicators (performance data) of the physical production line after executing the strategy are acquired in real time. This process achieves a leap from "virtual simulation" to "physical execution," providing a data foundation for the closed-loop optimization of the entire system.

[0084] S4. Strategy Evolution: The scheduling rule base is updated self-evolvingly based on the deviation between actual feedback data and simulation predictions. This step aims to build a closed-loop feedback mechanism between simulation evaluation and on-site execution, enabling the system to continuously learn and self-calibrate. The specific technical execution logic is as follows: (1) After the optimal scheduling strategy output in step S3 is sent to the physical production line and completes a full evaluation cycle, the system captures the actual feedback data of that cycle in real time through the field industrial interface. To ensure comparability with the simulation prediction results in step S22, the system will strictly call the same set of evaluation index system accounting standards determined in step S21. Based on these actual underlying event data, the system will accurately calculate the quantitative values ​​of various actual operating indicators of the physical production line after executing the strategy. (For example, the actual equipment overall utilization rate (OEE), the actual unit energy consumption, etc.).

[0085] (2) Extract the quantified values ​​of the simulation prediction index corresponding to the initial output of the production line performance model for the optimal scheduling strategy in step S22. The system introduces predicted residuals. The concept of prediction residual is specifically defined as: the absolute difference or normalized relative error between the quantified value of the actual operating index obtained by the strategy in actual execution on the physical production line and the quantified value of the index originally predicted by the system through discrete event simulation. Its specific calculation formula is:

[0086] The residual vector It objectively quantifies the degree of fitting deviation between the current "theoretical simulation" and "physical reality" of the system.

[0087] (3) The aforementioned residual results will trigger an update of the rule base's bias logic. This "bias logic" is directly related to and corresponds to the rule adaptation weights used to calculate the comprehensive fit index in the "two-dimensional matching architecture" of step S1. In the initial state of step S1, when the system linearly combines the "state feature correlation coefficient" and the "business objective bias score," each rule by default enjoys the baseline adaptation weight. In the closed-loop evolution of this step, the system dynamically adjusts this weight using an adaptive penalty / reward mechanism based on residuals: If a scheduling strategy (such as the SPT rule) is actually executed, the predicted residual of its core indicator is... Exceeding the preset tolerance threshold (e.g.) If the rule's adaptation weight is greater than 15%, it indicates that the rule is not suitable for real-world physical conditions and its disturbance resistance is far lower than the simulation expectation. The system will automatically reduce the rule's adaptation weight in the rule base according to the following formula:

[0088] In the formula, The preset evolution learning rate (e.g., 0.05). The AHP weights of the indicators.

[0089] Conversely, if the residual is extremely small (e.g.) The percentage of cases with a percentage of less than 5% indicates that the theoretical prediction of the rule is highly consistent with the actual working conditions, and the system maintains or fine-tunes and amplifies its adaptation weight.

[0090] Updated rule adaptation weights This will be permanently written back to the system's pre-set scheduling rule base for self-learning and optimization of the next round of evaluation. When the system executes step S1 next time, this weight will act as a multiplier directly on the rule's "comprehensive fit index". This means that scheduling rules that previously performed poorly in the physical field (large residuals) will be automatically downgraded or filtered in future intelligent screening; while rules with stable performance (small residuals) will be more likely to be prioritized for matching.

[0091] As an embodiment of this application, a system for assessing the control potential of an enterprise production line based on a production scheduling strategy is disclosed. The system, implemented using the aforementioned control potential assessment method, includes: As an embodiment of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it employs the specific implementation described in the above-described method for evaluating regulatory potential.

[0092] As an embodiment of this application, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, employs the specific implementation described in the above-described method for assessing regulatory potential.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for assessing the production line control potential of enterprises based on production scheduling strategies, characterized in that, include: Multidimensional dynamic features are extracted from the real-time operation data stream of the production line, including bottleneck workstation load rate features, work-in-process fluctuation gradient features, material arrival dispersion features, and material supply quantity deviation features; based on the production line configuration and evaluation objectives, and combined with the multidimensional dynamic features, candidate scheduling strategies are selected from the pre-set scheduling rule base. Establish a multi-dimensional evaluation index system that includes efficiency, cost, flexibility and stability, load candidate scheduling strategies and preset benchmark scheduling strategies into a unified production line performance simulation model, and output the multi-dimensional evaluation index values ​​for each strategy. For each candidate scheduling strategy, its control potential value relative to the baseline scheduling strategy is calculated based on multi-dimensional evaluation index values. The optimal scheduling strategy is selected from the candidate scheduling strategies according to predefined logic. The predefined logic includes a first-level screening and a second-level screening. The first-level screening selects the candidate scheduling strategy with the highest regulation potential value as the optimal scheduling strategy. The second-level screening selects the optimal scheduling strategy by comparing the evaluation index values ​​of the candidate scheduling strategies if there are parallel candidate scheduling strategies with the same regulation potential value. The optimal scheduling strategy is executed in the physical production line and the actual performance feedback data is obtained. The evaluation index value of the actual operation is calculated. The evaluation index value of the optimal scheduling strategy in the simulation stage is extracted and combined with the evaluation index value of the actual operation to calculate the prediction residual. If the prediction residual exceeds the preset tolerance threshold, the rule adaptation weight of the optimal scheduling strategy in the preset scheduling rule base is adjusted to optimize the next round of evaluation.

2. The method for assessing the enterprise production line control potential based on production scheduling strategies according to claim 1, characterized in that, The steps for obtaining the bottleneck workstation load rate characteristics include: Set a time period and extract the equipment operation status data of each production line station within that period; calculate the cumulative time of equipment in working state and abnormal shutdown state respectively, and calculate the dynamic load rate of each station; the dynamic load rate is a ratio, the numerator is the cumulative time of equipment in working state, and the denominator is the time period minus the cumulative time of abnormal shutdown state. Traverse all workstations, select the workstation corresponding to the maximum dynamic load rate as the bottleneck workstation for the current cycle, and use the maximum dynamic load rate as the bottleneck workstation load rate characteristic.

3. The method for assessing the enterprise production line control potential based on production scheduling strategies according to claim 1, characterized in that, The steps for obtaining the fluctuation gradient characteristics of the work-in-process include: Along the spatial flow of the process, the difference in work-in-process queue length between adjacent workstations is calculated to obtain the spatial fluctuation gradient; the feature extraction sliding time step is set, and the rate of change of the work-in-process queue length at the bottleneck workstation over time is calculated to obtain the temporal fluctuation gradient. The standard deviation of the spatial fluctuation gradient of all workstations and the temporal fluctuation gradient of the bottleneck workstation are concatenated to output the work-in-process fluctuation gradient feature.

4. The method for assessing the enterprise production line control potential based on production scheduling strategies according to claim 1, characterized in that, The steps for obtaining the material arrival dispersion characteristics and material supply quantity deviation characteristics include: The standard delivery interval is obtained based on the delivery plan, and the actual delivery interval is calculated based on the actual delivery time. The root mean square error between the actual delivery interval and the standard planned interval is calculated to obtain the material delivery time dispersion characteristics. The material supply quantity deviation characteristics are obtained by calculating the average absolute deviation between the actual delivered quantity and the standard planned delivered quantity.

5. The method for assessing the enterprise production line control potential based on production scheduling strategies according to claim 1, characterized in that, The step of selecting candidate scheduling strategies from a pre-set scheduling rule base includes: Calculate the correlation coefficient between the multidimensional dynamic features and each scheduling rule in the pre-set scheduling rule base; match the business objective tendency score of each scheduling rule according to the evaluation objective of the current business; The process of obtaining the business objective tendency score is described as follows: a rule objective tendency mapping matrix is ​​preset, where each element of the matrix represents the basic tendency coefficient of a scheduling rule for an evaluation objective; the evaluation objective of the current business is transformed into an objective weight vector; for each scheduling rule, the weighted dot product of the objective weight vector and the basic tendency coefficient corresponding to the scheduling rule is calculated to generate the business objective tendency score of the scheduling rule. The correlation coefficients and business objective tendency scores are linearly weighted to calculate the comprehensive fit index of each rule, and candidate scheduling strategies are selected by thresholding.

6. The method for assessing the enterprise production line control potential based on production scheduling strategies according to claim 1, characterized in that, The multi-dimensional evaluation index system includes: Efficiency is expressed as the overall utilization rate of equipment, which is calculated by combining the equipment's time utilization rate, performance utilization rate, and qualified product rate. Cost is expressed as energy consumption per unit of product, which is obtained by statistically analyzing the total energy consumption within a set period and dividing it by the number of qualified products. The flexibility dimension is represented by the average order delay rate, which is calculated by averaging the delay time between the actual completion time and the promised delivery time for all orders. The stability dimension is represented by the standard deviation of daily output, which is derived by calculating the standard deviation of total daily output.

7. The method for assessing the enterprise production line control potential based on production scheduling strategies according to claim 1, characterized in that, The calculation process for the regulation potential value includes: A directional gain coefficient is introduced. When the evaluation index is positively improved, the directional gain coefficient is 1; when the evaluation index is negatively controlled, the directional gain coefficient is -1. First, calculate the net improvement index of a single evaluation indicator, which is described as multiplying the difference between the indicator value and the corresponding benchmark scheduling strategy evaluation indicator value, the indicator direction gain coefficient, and the indicator weight. The global net improvement index for each candidate scheduling strategy is obtained by summing the net improvement indices of all evaluation metrics.

8. The method for assessing the enterprise production line control potential based on production scheduling strategies according to claim 7, characterized in that, The calculation process for the regulation potential value also includes: For the current candidate scheduling strategy, first calculate the product of the corresponding sign function of the global net improvement index and the weighted average sum of the evaluation index, and then subtract the performance standard deviation term of the current candidate scheduling strategy in the simulation operation to obtain the regulation potential value. When the global net improvement index is greater than zero, the sign function takes the value of 1; when the global net improvement index is less than zero, the sign function takes the value of -1. The performance standard deviation term is the product of a preset robustness penalty coefficient and the performance standard deviation of the current candidate scheduling strategy in the simulation operation; The calculation process of the performance standard deviation is described as follows: introduce random perturbation in each simulation, record the multidimensional evaluation index of each simulation output, and calculate the arithmetic mean of each evaluation index in all simulations. The standard deviation of each evaluation index in all simulations is calculated based on the arithmetic mean; the standard deviations of all evaluation indices are weighted and summed to obtain the performance standard deviation.

9. The method for assessing the enterprise production line control potential based on production scheduling strategies according to claim 1, characterized in that, The adjustment of the rule adaptation weights of the optimal scheduling strategy in the preset scheduling rule base includes: Subtract the evolution term from the rule adaptation weight of the optimal scheduling strategy in the preset scheduling rule base; obtain the adjusted rule adaptation weight, and write it back to the preset scheduling rule base; The calculation process of the evolution term is described as follows: first, the prediction residuals of all evaluation indicators are weighted, and then multiplied by the preset evolution learning rate.

10. A production line control potential assessment system based on production scheduling strategies, executing the control potential assessment method as described in any one of claims 1-9, characterized in that, The system includes: The strategy extraction module is used to extract multi-dimensional dynamic features from the real-time operation data stream of the production line, including bottleneck workstation load rate features, work-in-process fluctuation gradient features, material arrival dispersion features, and material supply quantity deviation features; based on the production line configuration and evaluation objectives, combined with the multi-dimensional dynamic features, candidate scheduling strategies are selected from the pre-set scheduling rule base. The strategy evaluation module is used to establish a multi-dimensional evaluation index system including efficiency, cost, flexibility and stability. It loads candidate scheduling strategies and preset benchmark scheduling strategies into a unified production line performance simulation model and outputs the multi-dimensional evaluation index values ​​for each strategy. For each candidate scheduling strategy, it calculates its control potential value relative to the benchmark scheduling strategy based on the multi-dimensional evaluation index values. The strategy screening module is used to select the optimal scheduling strategy from candidate scheduling strategies according to predefined logic. The predefined logic includes primary screening and secondary screening. Primary screening selects the candidate scheduling strategy with the highest regulation potential value as the optimal scheduling strategy. Secondary screening selects the optimal scheduling strategy by comparing the evaluation index values ​​of the candidate scheduling strategies if there are parallel candidate scheduling strategies with the same regulation potential value. The strategy evolution module is used to execute the optimal scheduling strategy in the physical production line and obtain actual performance feedback data, calculate the evaluation index value of the actual operation; extract the evaluation index value of the optimal scheduling strategy in the simulation stage, and calculate the prediction residual by combining it with the evaluation index value of the actual operation; if the prediction residual exceeds the preset tolerance threshold, the rule adaptation weight of the optimal scheduling strategy in the preset scheduling rule base is adjusted to optimize the next round of evaluation.

Images