Adaptive Optimization Methods and Systems for Production Decision Making

Abstract

This invention discloses an adaptive optimization method and system for production decisions, specifically relating to the field of industrial production process optimization technology. It addresses the problems of policy rigidity and limited adaptive capabilities in existing optimization systems due to reliance on historical policy data. By dynamically triggering model reconstruction through real-time monitoring of policy entropy values, candidate policies are generated through parameter space resetting and multiphysics simulation verification. Policy exploration priorities are dynamically allocated based on KL divergence differences, and the policy set is continuously optimized through a closed-loop feedback mechanism. This method proactively breaks path dependence when policy diversity decays, balances policy inheritance and exploration through dynamic priority allocation, and adjusts the data acquisition window based on deviation rate feedback, forming an adaptive optimization closed loop. This effectively improves the adaptability and decision-making flexibility of production strategies, ensuring performance stability during long-term operation.

Description

Technical Field

[0001] This invention relates to the field of industrial production process optimization technology, and more specifically, to an adaptive optimization method and system for production decisions. Background Technology

[0002] In the field of industrial production process optimization, digital twin technology and simulation verification methods have been widely used in the generation and verification of production strategies. Existing technologies typically evaluate candidate strategies from multiple dimensions by constructing a virtual simulation environment and select strategy solutions that meet preset objectives based on historical optimization results. This method can significantly reduce the trial and error costs in actual production and improve resource allocation efficiency within a certain period.

[0003] However, long-running optimization systems are susceptible to the influence of historical strategy data, which leads to path dependence in the subsequent optimization solutions. Specifically, the iterative evolution direction of optimization strategies gradually becomes solidified, making it difficult to break through existing decision-making patterns. As a result, they cannot adapt to dynamically changing production demands and external disturbances. This strategy rigidity limits the adaptive ability of the optimization system and ultimately affects the sustainability of production efficiency improvement. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an adaptive optimization method and system for production decisions to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The adaptive optimization method for production decisions includes the following steps:

[0007] S1. Monitor the policy entropy value of the current optimization policy set in real time. The policy entropy value is obtained by calculating the distribution dispersion of each policy in historical iterations.

[0008] S2. When the strategy entropy value is lower than the preset entropy threshold, a reconstruction instruction for the digital twin simulation model is triggered.

[0009] S3. Reset the policy space of the digital twin simulation model according to the reconstruction instruction, and generate a candidate policy set containing the new policy parameter space.

[0010] S4. Use the digital twin simulation model after resetting the strategy space to perform simulation verification on the candidate strategy set, and select a subset of feasible strategies that meet the preset constraints.

[0011] S5. Based on the KL divergence difference between each strategy in the feasible strategy subset and the historical strategy, dynamically generate the strategy exploration priority;

[0012] S6. Deploy the strategy with the highest priority in strategy exploration to the actual production system and update the current set of optimized strategies.

[0013] In a preferred embodiment, the policy entropy value of the current optimization policy set is monitored in real time. The policy entropy value is obtained by calculating the distribution dispersion of each policy in historical iterations, including:

[0014] Obtain the policy parameter vector of each policy in the current set of optimization policies in the historical iterations;

[0015] Extract the feature dimensions of all strategy parameter vectors within a preset time window;

[0016] The variance of each dimension parameter in the strategy parameter vector is calculated based on the numerical fluctuation range of the feature dimension.

[0017] The strategy entropy value is generated by weighted summation of the variance values ​​of the parameters in each dimension.

[0018] In a preferred embodiment, when the strategy entropy value is lower than a preset entropy threshold, a reconstruction instruction for the digital twin simulation model is triggered, including:

[0019] The dynamic adjustment rules for the preset entropy threshold are obtained, and the current policy entropy value monitored in real time is compared with the preset entropy threshold to generate a threshold deviation evaluation result.

[0020] Based on the judgment that the threshold deviation evaluation result exceeds the preset deviation threshold, a reconstruction instruction for the digital twin simulation model containing the reconstruction intensity level is generated.

[0021] The reconstruction command is passed to the control interface of the digital twin simulation model, and the control interface performs the corresponding parameter initialization operation according to the reconstruction intensity level.

[0022] In a preferred embodiment, the policy parameter space of the digital twin simulation model is reset according to the reconstruction instruction to generate a candidate policy set containing the new policy parameter space, including:

[0023] The reconfiguration intensity level in the reconfiguration instruction is parsed, and a parameter dependency constraint table corresponding to the reconfiguration intensity level is generated. The parameter dependency constraint table records the physical coupling rules and priority weights between process parameters.

[0024] Based on the constraint table of reconstruction intensity level and parameter dependency relationship, dynamically decouple conflicting parameter dimensions and lock independent adjustable dimensions;

[0025] Based on the real-time sensor data and parameter dependency constraint table of the device, multi-level boundary constraint iterative correction is performed on the decoupled independent adjustable dimension.

[0026] A set of candidate policies is generated in the corrected policy parameter space through dynamic adaptive pseudo-random sampling.

[0027] In a preferred embodiment, the locking condition for the independently adjustable dimension is that the parameter adjustment range does not violate the physical limits of the equipment and meets the production order process standards.

[0028] The correction rule for iterative correction of multi-level boundary constraints is to first relax the parameter dimension constraints of high-priority weights, and simultaneously compress the parameter dimension constraints of low-priority weights.

[0029] The sampling density of dynamic adaptive pseudo-random sampling is dynamically adjusted according to the priority weight of the parameter dimension.

[0030] In a preferred embodiment, the candidate policy set is simulated and verified using a digital twin simulation model with the policy space reset, and a subset of feasible policies that meet preset constraints is selected, including:

[0031] The dynamic parameters of each strategy in the candidate strategy set are generated into a time series. The dynamic parameters include vibration amplitude, temperature gradient and electromagnetic interference intensity.

[0032] The nodes of the topological network are generated based on the local maxima and minima of the time series of dynamic parameters, and the absolute values ​​of the Pearson correlation coefficients of the time series of dynamic parameters within adjacent time windows constitute the edges of the topological network.

[0033] Extract persistent cohomology features of the topological network, calculate the Betti number and duration interval length of each dimension, and generate a set of topological invariants;

[0034] The dynamic constraint satisfaction parameter is calculated based on the evolution trend of the set of topological invariants. The evolution trend is jointly characterized by the rate of change of the duration interval length and the stability of the Betti number.

[0035] A subset of feasible strategies is generated by filtering strategies whose dynamic constraint satisfaction parameters are higher than the dynamic threshold. The dynamic threshold is dynamically adjusted based on the real-time health index of the device.

[0036] In a preferred embodiment, based on the KL divergence difference between each strategy in the feasible strategy subset and historical strategies, a strategy exploration priority is dynamically generated, including:

[0037] Based on the difference between the parameter distribution of each strategy in the feasible strategy subset and the KL divergence of the historical strategy set, strategy diversity evaluation parameters are generated, and the parameter distribution is constructed by the kernel density estimation method.

[0038] Priority zones are determined based on the real-time health index of the equipment.

[0039] The KL divergence difference and historical strategy similarity are dynamically sorted. When the KL divergence difference is greater than the historical strategy similarity, it is classified as an exploration-first strategy, and otherwise as an inheritance-first strategy.

[0040] Based on the priority interval and strategy classification results, strategy exploration priorities are generated. Exploration priority strategies are arranged in descending order of KL divergence difference within the interval, while inheritance priority strategies are arranged in ascending order of historical strategy similarity. The number of priority levels is positively correlated with the number of strategies in the current strategy set.

[0041] In a preferred embodiment, when dividing priority intervals based on the device's real-time health index, the priority interval is limited to high diversity strategies when the device's real-time health index is below a first threshold, and the priority interval is expanded to high similarity strategies when the device's real-time health index is above a second threshold.

[0042] In a preferred implementation, the strategy with the highest priority in strategy exploration is deployed to the actual production system, and the current set of optimized strategies is updated, including:

[0043] Extract the highest priority strategy from the strategy exploration priority ranking results and verify that it conforms to the real-time health index of the equipment and the process standards of the production order.

[0044] Deploy the validated strategies to the actual production system and update the current set of optimized strategies based on the real-time health index of the equipment: replace the strategy with the smallest KL divergence when the real-time health index of the equipment is lower than the first threshold, and replace the strategy with the highest similarity to the historical strategy when it is higher than the second threshold.

[0045] Record execution results to the historical strategy set, dynamically adjust the moving average window length based on the deviation rate between the actual and simulated results, and generate a strategy deployment execution log.

[0046] On the other hand, the present invention provides an adaptive optimization system for production decisions, comprising the following modules:

[0047] The entropy monitoring module is used to monitor the policy entropy value of the current optimization policy set in real time. The policy entropy value is obtained by calculating the distribution dispersion of each policy in historical iterations.

[0048] The instruction triggering module is used to trigger a reconstruction instruction for the digital twin simulation model when the policy entropy value is lower than a preset entropy threshold.

[0049] The parameter reset module is used to reset the policy space of the digital twin simulation model according to the reconstruction instruction, and generate a set of candidate policies containing the new policy parameter space.

[0050] The simulation verification module is used to perform simulation verification on the candidate strategy set using the digital twin simulation model after the strategy space is reset, and to select a subset of feasible strategies that meet the preset constraints.

[0051] The priority generation module is used to dynamically generate strategy exploration priorities based on the KL divergence differences between each strategy in the feasible strategy subset and historical strategies.

[0052] The strategy deployment module is used to deploy the strategy with the highest priority in strategy exploration to the actual production system and update the current set of optimized strategies.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. By constructing a dynamic closed-loop optimization system, the adaptability and sustainability of production strategies are effectively improved. By monitoring strategy entropy values ​​in real time and triggering a model reconstruction mechanism, the decay trend of strategy diversity can be detected in a timely manner. The parameter space is actively reset before the strategy evolution is solidified. During the parameter space reset process, multi-physics coupling simulation and priority-driven exploration mechanisms are integrated to ensure that the newly generated candidate strategies not only meet the equipment operation constraints but also cover a wider range of optimization directions, avoiding the trap of local optima. The simulation verification stage introduces dynamic threshold adjustment and cross-cycle feedback, which enables the selected feasible strategies to accurately match the real-time production status and fully release the strategy iteration potential while ensuring stability.

[0055] 2. Through strategy priority classification and closed-loop update mechanism, intelligent evolution of the optimization process is realized; dynamic priority ranking based on KL divergence difference deeply binds equipment health status with strategy exploration direction, prioritizing the inheritance of verified strategies in low-risk states and focusing on innovative strategy exploration in high-fault-tolerant ranges, forming a decision-making model that balances efficiency and robustness; the deviation rate feedback after deployment and execution directly affects the dynamic adjustment of the data acquisition window, ensuring that the system always optimizes the strategy generation logic based on the latest production status; enabling the optimization system to continuously adapt to external disturbances and internal parameter drift, maintaining a high level of decision-making flexibility and production efficiency in long-term operation. Attached Figure Description

[0056] Figure 1 This is a flowchart of the adaptive optimization method for production decision-making in this invention;

[0057] Figure 2 This is a schematic diagram of the adaptive optimization system for production decision-making of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0059] Example 1: Figure 1 The present invention provides an adaptive optimization method for production decisions, comprising the following steps:

[0060] S1. Monitor the policy entropy value of the current optimization policy set in real time. The policy entropy value is obtained by calculating the distribution dispersion of each policy in historical iterations.

[0061] S2. When the strategy entropy value is lower than the preset entropy threshold, a reconstruction instruction for the digital twin simulation model is triggered.

[0062] S3. Reset the policy space of the digital twin simulation model according to the reconstruction instruction, and generate a candidate policy set containing the new policy parameter space.

[0063] S4. Use the digital twin simulation model after resetting the strategy space to perform simulation verification on the candidate strategy set, and select a subset of feasible strategies that meet the preset constraints.

[0064] S5. Based on the KL divergence difference between each strategy in the feasible strategy subset and the historical strategy, dynamically generate the strategy exploration priority;

[0065] S6. Deploy the strategy with the highest priority in strategy exploration to the actual production system and update the current set of optimized strategies.

[0066] S1. Monitor the policy entropy value of the current optimization policy set in real time. The policy entropy value is obtained by calculating the distribution dispersion of each policy in historical iterations. Specifically, it can be implemented as follows:

[0067] The system retrieves the strategy parameter vectors for each strategy in the current optimization strategy set from the historical database of the production system. Each strategy parameter vector is composed of the corresponding process parameters during historical execution, including the operating temperature, processing pressure, material conveying rate, and energy consumption coefficient of the production equipment. Each process parameter corresponds to one dimension of the strategy parameter vector. During retrieval, historical data within a preset time period is filtered based on the strategy execution timestamp. The preset time period is set according to the production system's process update cycle, which is the adjustment interval for the main process parameters defined in the production line equipment technical manual. If the equipment technical manual does not specify this interval, historical data from the most recent 12 months is selected by default. For parameter vectors from multiple iterations of the same strategy, the moving average of each dimension's parameters is taken after chronological order as the representative parameter vector for that strategy. The calculation window length of the moving average is consistent with the data acquisition frequency of the production system. The data acquisition frequency is determined by the sensor configuration of the production line; for example, when data is collected every 5 minutes, the moving average window is set to 1 hour.

[0068] The time window length is set to a preset number of iterations. This preset number of iterations is dynamically adjusted based on the strategy update frequency of the production system. The strategy update frequency is the strategy optimization cycle defined in the production plan. If the production plan is not specified, the last 30 strategy iteration cycles are selected by default. A complete dataset of all strategy parameter vectors within this time window is extracted from the historical database. The feature dimension of each strategy parameter vector refers to the type of process parameter it contains. During extraction, data validity is validated for each strategy parameter vector. Data validity validation includes completeness and rationality validation: completeness validation requires that the number of missing dimension parameters in the strategy parameter vector does not exceed 20% of the total dimensions; rationality validation requires that each dimension parameter value is within the safe operating range specified in the equipment technical manual. For example, the operating temperature parameter must be between 50℃ and 200℃, and the processing pressure parameter must be between 0.1MPa and 5.0MPa. If a strategy parameter vector fails the validation, it is marked as invalid data and removed. The vacant position is filled by a valid parameter vector from an adjacent time point of the same strategy. If no valid replacement data is available, the vacancy is retained.

[0069] For each extracted feature dimension, the range of its numerical fluctuation across all strategy parameter vectors within the time window is calculated; specifically, the variance of each dimension parameter is calculated. The variance calculation process is as follows: For each feature dimension, obtain the sequence of parameter values ​​for that feature dimension across all strategy parameter vectors within the time window; calculate the arithmetic mean of this sequence; then calculate the square of the difference between each parameter value and the arithmetic mean; sum all squared values ​​and divide by the total number of parameter values ​​to obtain the variance of that dimension parameter. During calculation, the units of the parameter values ​​are consistent with the units of the raw data collected by the equipment sensors; for example, operating temperature is in degrees Celsius, processing pressure is in megapascals, material conveying rate is in kilograms per minute, and energy consumption coefficient is in kilowatt-hours per unit. If multiple sets of parameter value sequences exist for the same dimension parameter, the variance of each sequence is calculated separately, and the maximum value is taken as the final variance value.

[0070] The strategy entropy value is generated by weighted summation of the variances of each dimension parameter. The weight coefficients are determined using the Analytic Hierarchy Process (AHP). The AHP implementation steps are as follows: 1) Establish a hierarchical model of the impact of process parameters on production efficiency. Production efficiency indicators include equipment utilization rate, product yield, and energy efficiency. 2) Design an expert questionnaire, inviting at least three domain experts to conduct pairwise comparisons and ratings of the importance of each process parameter. The rating scale is 1-9, where 1 indicates that the two parameters are equally important, and 9 indicates that the former is extremely more important than the latter. 3) After collecting expert ratings, construct a judgment matrix, calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector, and normalize the eigenvector to obtain the initial weight coefficients for each dimension parameter. 4) Perform a consistency check on the initial weight coefficients. If the consistency ratio is less than 0.1, the weights are considered valid; otherwise, the ratings are readjusted until the check is passed. The final weight coefficients satisfy the condition that the sum of all dimension weights is 1. During weighted summation, the variance of each dimension parameter is multiplied by its corresponding weight coefficient and then summed. The sum is rounded to two decimal places as the strategy entropy value. If the variance of a certain dimension parameter cannot be calculated due to data anomalies, its weight coefficient will be temporarily distributed evenly to other dimensions, and the variance will be recalculated after the data returns to normal.

[0071] Data preprocessing includes dynamic adjustment of the time window and standardization of parameter ranges. The dynamic adjustment rule for the time window is as follows: when the strategy update frequency of the production system increases to more than 1.5 times the original frequency, the time window length is shortened to 2 / 3 of the original length; when the strategy update frequency decreases to less than 2 / 3 of the original frequency, the time window length is extended to 1.5 times the original length. The standardization method for parameter ranges is to perform minimum-maximum normalization on each dimension parameter, linearly mapping it to the [0,1] interval. The normalization formula is (current value - minimum value) / (maximum value - minimum value), where the minimum and maximum values ​​are taken from the safe operating range specified in the equipment technical manual.

[0072] When the strategy entropy value falls below a preset threshold, a digital twin simulation model reconstruction command is triggered. The preset threshold is set based on the distribution of strategy entropy values ​​in historical data, specifically the lower limit of the normal fluctuation range of strategy entropy values ​​within the most recent year. The normal fluctuation range is determined by calculating the mean and standard deviation of the strategy entropy values, with the lower limit being the mean minus twice the standard deviation. The reconstructed digital twin model generates a new strategy parameter space through simulation verification. The boundary of the new strategy parameter space is dynamically adjusted according to the current health status of the equipment. The equipment health status is assessed by a preset predictive maintenance model, whose inputs are the equipment vibration spectrum, temperature time-series data, and lubricant detection results.

[0073] The anomaly handling mechanism includes empty dataset warnings and computational overflow protection: If no valid strategy parameter vector is available within the time window, a warning signal is sent to the production management system and the strategy optimization process is paused until manual intervention confirms the data source status; if a numerical overflow occurs during variance calculation, the system automatically switches to logarithmic transformation calculation mode. The logarithmic transformation is performed by first taking the natural logarithm of the parameter values ​​and then calculating the variance, followed by restoring the final result through exponential operations. Boundary conditions limit the number of dimensions in the strategy parameter vector to no more than 100. If the number of dimensions exceeds 100, the top 100 dimensions are selected for calculation based on their weight coefficients.

[0074] Hardware dependencies include the sensor network and data acquisition system of the production equipment. The sensor network consists of temperature sensors, pressure sensors, flow meters, and electricity meters. The data acquisition system is implemented based on an industrial IoT platform, with the IoT platform software version no lower than 4.2.1. Software dependencies include the numerical computation library NumPy 1.21.0 and above, used for variance calculation and weighted summation, and the analytic hierarchy process (AHPy) toolkit 0.9.3, used for weight coefficient calculation. Experimental results were validated by recording the correlation between the policy entropy value change curve and production efficiency. For example, in a certain automotive parts production line, when the policy entropy value fell below a threshold, model reconstruction was triggered. After reconstruction, policy diversity increased by 25%, and equipment utilization increased from 82% to 87%.

[0075] It is worth noting that the current optimization strategy set refers to the collection of candidate strategies to be evaluated in the production system during the real-time optimization cycle. Each strategy consists of a set of predefined process parameter combinations, including but not limited to equipment operating temperature, processing pressure threshold, material conveying rate limit, and energy efficiency target value. The set is generated through historical iteration data screening. Specifically, strategies that meet recent production efficiency are extracted from the strategy library of the digital twin simulation model. Efficiency indicators include yield rate ≥95% and equipment utilization rate ≥85%, while strategies that conflict with the current equipment health status (such as parameter combinations exceeding the equipment vibration tolerance range) are removed. The set is updated synchronously with the production plan. After each strategy iteration, strategies are added or removed based on simulation verification results to ensure that the set capacity is maintained within the range of 50-200 strategies, balancing the needs of computational efficiency and strategy diversity.

[0076] S2. When the strategy entropy value is lower than the preset entropy threshold, a reconstruction instruction for the digital twin simulation model is triggered, which can be implemented as follows:

[0077] The system obtains dynamic adjustment rules for a preset entropy threshold. These rules are calculated based on the mean and standard deviation of historical strategy entropy values. Historical strategy entropy values ​​are extracted from the production system's historical database, covering all strategy optimization cycles recorded within the past year. The mean is calculated by taking the arithmetic mean of the historical strategy entropy value sequence, and the standard deviation is calculated by averaging the squares of the differences between each data point and the mean, then taking the square root. The preset entropy threshold is set as the mean minus twice the standard deviation. If the calculated result is lower than the minimum strategy diversity threshold specified in the equipment technical manual, this minimum strategy diversity threshold is used as the preset entropy threshold. The update cycle of the dynamic adjustment rules is synchronized with the production system's strategy optimization cycle, which is set according to the production plan, for example, updating the dynamic adjustment rules monthly.

[0078] The current policy entropy value monitored in real time is compared with a preset entropy threshold to generate a threshold deviation assessment result. During the comparison, the absolute value of the difference between the current policy entropy value and the preset entropy threshold is calculated and divided by the preset entropy threshold to obtain the relative deviation. For example, if the preset entropy threshold is 40 and the current policy entropy value is 35, then the relative deviation is (40-35) / 40=0.125. The generation rule for the threshold deviation assessment result is as follows: a relative deviation greater than or equal to 0.1 is considered a significant deviation; a relative deviation between 0.05 and 0.1 is considered a moderate deviation; and a relative deviation less than 0.05 is considered a slight deviation. The judgment threshold is set based on the statistical results of policy failure events in historical data. For example, when the relative deviation exceeds 0.1, the historical policy failure probability increases to 30%.

[0079] Based on the judgment that the threshold deviation evaluation result exceeds the preset deviation threshold, a digital twin simulation model reconstruction instruction containing the reconstruction intensity level is generated. The preset deviation threshold is set to 0.1, and the reconstruction instruction is triggered when the threshold deviation evaluation result is a significant deviation or a general deviation. The reconstruction intensity level is divided into three levels: Level 1 reconstruction corresponds to significant deviation, and the reconstruction intensity is a full parameter space reset; Level 2 reconstruction corresponds to general deviation, and the reconstruction intensity is a core parameter space reset; Level 3 reconstruction corresponds to slight deviation, and the reconstruction intensity is a local parameter fine-tuning. The mapping rule of the reconstruction intensity level is determined by an expert experience table, which records the correspondence between different deviation ranges and parameter adjustment ranges. For example, Level 1 reconstruction requires resetting more than 80% of the strategy parameters, Level 2 reconstruction resets 50%-80% of the parameters, and Level 3 reconstruction adjusts 20%-50% of the parameters.

[0080] The refactoring command is passed to the control interface of the digital twin simulation model. The control interface performs corresponding parameter initialization operations based on the refactoring intensity level. The control interface is a standard API interface provided by the digital twin simulation model, receiving JSON-formatted commands containing the refactoring intensity level. The specific logic of the parameter initialization operation is as follows: For Level 1 refactoring, all historical constraints in the current strategy parameter space are cleared, and parameter boundaries are redefined based on the latest equipment status data; for Level 2 refactoring, basic process parameter constraints are retained, and the optimization target weight coefficients are reset; for Level 3 refactoring, only the three parameter dimensions most relevant to the current production requirements are adjusted. The data sources for parameter initialization include real-time sensor data from the equipment, material attribute databases, and process standard documents.

[0081] The protection mechanism for dynamically adjusting rules is as follows: if the historical strategy entropy data volume is less than 100 records, a fixed preset entropy threshold is adopted. The fixed value is the larger of the minimum strategy diversity threshold recommended in the equipment technical manual and the industry standard value. The optimization method for deviation calculation is: extreme values ​​are truncated. For example, if the current strategy entropy value is lower than the minimum strategy diversity threshold, it is directly judged as a significant deviation, and there is no need to calculate the relative deviation. The fault tolerance mechanism for reconstruction intensity level is as follows: when the intensity level in the reconstruction instruction conflicts with the current equipment state, such as when the equipment is under maintenance and cannot perform a level 1 reconstruction, it is automatically downgraded to the next intensity level and an operation log is generated.

[0082] For example, when this embodiment is applied in an electronics assembly line, if the strategy entropy value drops to 38 (with a preset entropy threshold of 40), the system triggers a secondary reconfiguration instruction. After resetting the core parameter space, strategy diversity is improved, and equipment failure rate is reduced. The matching accuracy between reconfiguration intensity level and deviation is high, and the false trigger rate is reduced. For example, verification data can come from the production line's real-time monitoring system, and reconfiguration operations are performed within the data acquisition cycle.

[0083] The data missing handling rules are as follows: If all historical strategy entropy values ​​are invalid, for example, due to sensor malfunction causing data anomalies, the benchmark entropy threshold of similar production lines in the same industry will be temporarily adopted until the local data is restored to validity. The command transmission interruption handling rules are as follows: When the reconstruction command transmission fails, it will automatically retry three times and activate the backup communication channel. If it still fails, the optimization process will be paused and a manual intervention alarm will be triggered. The parameter initialization conflict handling rules are as follows: If the redefined parameter boundary exceeds the physical limits of the equipment, for example, if the processing pressure reset value exceeds 5.0 MPa, it will be automatically truncated to a safe range and marked as a parameter pending review.

[0084] Hardware dependencies include industrial sensor networks that support real-time data acquisition, such as the PT100 temperature sensor and the MPX5700 pressure sensor, as well as edge computing devices with at least a quad-core CPU and 16GB of memory. Software dependencies include the ANSYS Twin Builder 2022R2 or later digital twin simulation platform, which supports dynamic parameter reset interfaces, and the Newtonsoft.Json 13.0.1 or later JSON command parsing library.

[0085] In the calculation of mean and standard deviation, the historical strategy entropy data is a dimensionless index, and the unit is uniformly entropy unit (EU). The relative deviation calculation result is rounded to two decimal places, for example, 0.125 is recorded as 0.13. The weight coefficient adjustment of parameter initialization operation is based on the contribution of process parameters to production efficiency. The contribution is obtained through regression analysis of historical data. For example, the contribution coefficient of the working temperature parameter is 0.3, and that of the processing pressure is 0.2.

[0086] S3. Reset the policy parameter space of the digital twin simulation model according to the reconstruction instruction, and generate a candidate policy set containing the new policy parameter space. Specifically, this can be implemented as follows:

[0087] The reconstruction intensity level in the reconstruction instruction is parsed to generate a parameter dependency constraint table corresponding to the reconstruction intensity level. The parameter dependency constraint table is constructed by extracting physical coupling rules from the process parameter correlation clauses in the equipment technical manual. The process parameter correlation clauses clearly record the linkage relationship between parameters in tabular form. For example, changes in processing pressure parameters and operating temperature parameters must satisfy the thermodynamic balance formula, specifically, for every 1 MPa increase in pressure, the temperature decreases by 2 °C.

[0088] Priority weights are determined based on the process standard documents of the production orders. These documents define the influence coefficients of each parameter on the core quality indicators. These coefficients are calculated using orthogonal experimental design. The orthogonal experimental design involves designing a multi-factor, multi-level experimental matrix and analyzing the contribution of each parameter. The experimental matrix design follows the L9(3^4) standard orthogonal array, and the contribution is calculated using analysis of variance. The update mechanism of the parameter dependency constraint table is linked to the equipment maintenance cycle. The equipment maintenance cycle is specified in the preventative maintenance plan of the equipment technical manual. The maintenance plan clearly states that the coupling rule base will be verified and revised quarterly, and the revisions must be signed and confirmed by the equipment engineer before taking effect.

[0089] Based on the reconstruction intensity level and parameter dependency constraint table, conflicting parameter dimensions are dynamically decoupled and independent adjustable dimensions are locked. The operation process for dynamically decoupling conflicting parameter dimensions is as follows: load the coupling rules in the parameter dependency constraint table; when the reconstruction intensity level is a full parameter space reset, remove the coupling constraints between all parameters; when the reconstruction intensity level is a core parameter space reset, only the coupling rules directly related to the key quality indicators of the current production order are retained. The key quality indicators are determined by the Class A inspection items in the order process standard document, such as dimensional accuracy and surface roughness; when the reconstruction intensity level is a local parameter fine-tuning, only the coupling relationship related to abnormal parameters monitored by the equipment's real-time sensors is removed. Abnormal parameters are defined as parameters that exceed the safety range specified in the equipment technical manual for three consecutive sampling cycles. The locking conditions for independently adjustable dimensions are verified by checking whether the parameter adjustment range is within the physical limits specified in the equipment technical manual. The physical limits include a maximum operating temperature of 300℃, a maximum processing pressure of 50MPa, and a maximum spindle speed of 8000rpm. At the same time, it is verified whether the parameter range meets the tolerance requirements in the production order process standard. The tolerance requirements include indicators such as dimensional accuracy ±0.05mm and surface roughness Ra≤0.8μm. If the verification fails, the parameter range redefinition process is triggered.

[0090] Based on real-time sensor data and parameter dependency constraint tables, multi-level boundary constraint iterative correction is performed on the decoupled independent adjustable dimensions. The correction rule prioritizes relaxing the constraints of parameter dimensions with high priority weights while simultaneously compressing the constraints of parameter dimensions with low priority weights. The allocation of priority weights is based on the degree of influence of parameters on the delivery quality of production orders. The degree of influence is obtained through regression analysis of historical production data. In the regression analysis, the correlation coefficient between quality defect events and parameter fluctuations is used as the basis for weight assignment. The correlation coefficient is calculated using the Pearson product-moment correlation coefficient formula.

[0091] The specific correction operation involves expanding the allowable adjustment range of high-priority parameters with a weight ≥ 0.3 to 90%-95% of the equipment's physical limits. For example, if the physical limit of the operating temperature parameter is 300℃, the adjustment range is expanded from the original 250-280℃ to 240-290℃. For medium-priority parameters with a weight between 0.1 and 0.3, the adjustment range remains within the safe operating range recommended in the equipment's technical manual. For example, the processing pressure parameter remains at 40-45MPa. For low-priority parameters with a weight < 0.1, the adjustment range is compressed to 50%-70% of the original range. For example, the coolant flow rate parameter, originally ranging from 5-10L / min, is compressed to 6-8L / min. After each round of correction, a coupling conflict detection is performed. If a parameter combination is detected to violate the coupling rules, the system reverts to the previous correction state and reduces the adjustment magnitude by 10%. The iteration terminates when the adjustment magnitude decreases to a minimum value, such as a temperature adjustment step size ≤ 1℃.

[0092] A candidate policy set is generated within the corrected policy parameter space through dynamic adaptive pseudo-random sampling. The dynamic adaptive pseudo-random sampling is implemented by allocating sampling density based on the priority weights of the parameter dimensions. The sampling density is dynamically adjusted according to the priority weights of the parameter dimensions, which are calculated using orthogonal experimental design and normalized to the 0-1 interval. Dimensions with weights ≥ 0.3 are classified as high-priority groups, with 10 sampling points generated per dimension; dimensions with weights between 0.1 and 0.3 are classified as medium-priority groups, with 5 sampling points generated per dimension; and dimensions with weights < 0.1 are classified as low-priority groups, with 3 sampling points generated per dimension. The sampling density of the high-priority group is 3.33 times (10 / 3) that of the low-priority group. This ratio is set based on the difference in the contribution of parameters to production efficiency, ensuring that key parameters receive more intensive optimization exploration. The sampling point allocation rules are written into the configuration file of the digital twin platform, achieving uniform spatial coverage through a Latin hypercube design.

[0093] The sampling method employs a Latin hypercube design. This design divides each parameter dimension into several intervals and randomly selects points within each interval, ensuring uniform coverage of the parameter space. The formula for calculating the number of sampling points is: Total number of points = Σ (Number of dimensions for each weighted group × Corresponding sampling density). For example, if a strategy's parameter space contains 6 high-weight dimensions, 4 medium-weight dimensions, and 2 low-weight dimensions, then the total number of sampling points is 6 × 10 + 4 × 5 + 2 × 3 = 86. Each sampling point corresponds to a parameter combination of a candidate strategy. The values ​​of the parameter combinations retain the same unit precision as the data collected by the device's sensors. For example, temperature parameters retain one decimal place (unit: °C), pressure parameters retain integer places (unit: MPa), and cutting speed parameters retain two decimal places (unit: m / min).

[0094] When constructing the parameter dependency constraint table, the physical coupling rules are obtained by parsing the process correlation clauses in the equipment technical manual. These clauses clearly define the mathematical relationships or empirical rules for parameter linkage, such as a 5% reduction in processing pressure for every 10°C increase in operating temperature. Priority weights are calculated using the orthogonal experimental method, which employs an L9(3^4) experimental matrix to analyze the variance contribution rate of each parameter to the quality index. Parameters with a variance contribution rate exceeding 15% are defined as high priority.

[0095] When dynamically decoupling conflict parameter dimensions, the correlation coefficient in the coupling rules is calculated using historical production data. Historical data is selected based on periods with an equipment health index ≥ 80 and a production yield ≥ 95%. The correlation coefficient is calculated using the Pearson product-moment correlation coefficient formula, and parameter groups with a correlation coefficient ≥ 0.7 are considered strongly coupled. In the iterative correction of multi-level boundary constraints, the adjustment amplitude decay mechanism involves multiplying the adjustment amplitude by a decay coefficient of 0.9 for each detected conflict, until the conflict is eliminated or the minimum adjustment step size is reached (e.g., temperature adjustment step size ≤ 1℃).

[0096] If version conflicts exist in the parameter dependency constraint table, such as inconsistencies between old and new manual rules, the latest timestamp version of the device technical manual will be prioritized, triggering a manual review process that must be completed within 24 hours. If a dimension weight calculation is abnormal, such as a division by zero error, its weight will be automatically reset to 0.1, and an exception event code will be recorded. The code format is ERR_WEIGHT_parameter_name_timestamp, with the timestamp accurate to milliseconds, e.g., ERR_WEIGHT_TEMP_20231005143035999. If a system failure occurs during the correction process, recovery will resume from the most recent persistent storage correction state. The persistent storage interval is set to once every 5 minutes, the storage medium is an SSD hard drive, and the read / write speed is ≥500MB / s.

[0097] Hardware dependencies include a Kistler 9232A dynamic force sensor with an accuracy of ±0.5% and a range of 0-20kN, and an NI PXIe-6368 data acquisition card with a sampling rate of 500kHz and 16-bit resolution. Software dependencies include Matlab 2022a for orthogonal experimental analysis and the pyDOE library in Python 3.9 for Latin hypercube design, version 0.3.8. During parameter priority weight calculation, weight normalization ensures a sum of 1. In Latin hypercube sampling, the number of intervals in each dimension is consistent with the sampling density; for example, high-weight dimensions are divided into 10 intervals, with one point randomly sampled within each interval, and interval boundary values ​​distributed proportionally. Dimensions and units strictly adhere to the sensor configuration; for example, temperature is measured in degrees Celsius (°C), and pressure in megapascals (MPa), with accuracy consistent with the equipment technical manual.

[0098] S4. Utilize the digital twin simulation model after resetting the policy space to perform simulation verification on the candidate policy set, and select a subset of feasible policies that meet the preset constraints. Specifically, this can be implemented as follows:

[0099] The dynamic parameters of each policy in the candidate policy set are generated into a time series. These dynamic parameters include vibration amplitude, temperature gradient, and electromagnetic interference intensity. The dynamic parameters are extracted from the real-time simulation results of the digital twin simulation model, which are obtained by running the simulation on the policy parameter space as reset in step S3. The length of the time series is consistent with the number of iterations within the preset time window in step S2. For example, when the time window is set to the last 30 policy iterations, each dynamic parameter generates a time series containing 30 data points.

[0100] The sampling frequency of the time series is determined by the sensor configuration of the production equipment. For example, the sampling frequency of the vibration sensor is 1000 times per second, and the sampling frequency of the temperature sensor is 10 times per second. For non-real-time parameters, such as electromagnetic interference intensity, the calculation module of the digital twin model outputs data according to the simulation step size. The simulation step size is synchronized with the equipment control cycle, for example, outputting data once every 0.1 seconds. The storage format of the time series is a two-dimensional array. The first dimension is the timestamp, and the second dimension is the parameter value. The unit of the parameter value is consistent with the unit of the sensor data. The unit of vibration amplitude is meters per second squared (m / s²), the unit of temperature gradient is degrees Celsius per minute (°C / min), and the unit of electromagnetic interference intensity is millitalas (mT).

[0101] Nodes in the topology network are generated based on local maxima and minima of time series of dynamic parameters. A local maximum is defined as a parameter value at a given time point that is greater than the parameter values ​​at the two preceding and following time points, while a local minimum is defined as a parameter value at a given time point that is less than the parameter values ​​at the two preceding and following time points. After node generation, the absolute values ​​of the Pearson correlation coefficients of the time series of dynamic parameters within adjacent time windows are calculated and used as edge weights in the topology network.

[0102] The rule for dividing adjacent time windows is to segment the time series according to a fixed window length, which is consistent with the moving average window in step S1. For example, when the moving average window is 1 hour, the time window length is set to the number of data points within 1 hour. The calculation range of the edge weights is different dynamic parameter sequences within the same time window. For example, the absolute value of the Pearson correlation coefficient between vibration amplitude and temperature gradient within the same time window is calculated. The calculation formula is the covariance divided by the product of the standard deviations of the two sequences. The absolute value of the calculation result is taken to eliminate the directional effect.

[0103] Persistent cohomology features of the topological network are extracted, and the Betti numbers and duration interval lengths for each dimension are calculated to generate a set of topological invariants. Persistent cohomology features are calculated using the persistent cohomology function of the open-source topology data analysis library GUDHI. The input is the topological network data generated in step S4, and the output is the Betti numbers and corresponding duration interval lengths for each dimension. The Betti number is calculated using 0-dimensional and 1-dimensional methods. The 0-dimensional Betti number represents the number of connected components in the topological network, while the 1-dimensional Betti number represents the number of loop structures. The duration interval length represents the survival time span of the corresponding topological feature during parameter scale changes. For example, a loop structure that appears at parameter scale 0.5 and disappears at scale 1.2 has a duration interval length of 0.7. The set of topological invariants is stored as a multidimensional array, where each element contains the Betti number, the start and end times of the duration interval, and the associated dynamic parameter type, such as the Betti number corresponding to the combined parameters of vibration amplitude and temperature gradient.

[0104] The dynamic constraint satisfaction parameter is calculated based on the evolution trend of the set of topological invariants. The evolution trend is jointly characterized by the rate of change of the duration interval length and the stability of the Betti number. The rate of change of the duration interval length is calculated by arranging the duration interval lengths of the same topological feature in chronological order and calculating the sum of the absolute values ​​of the length differences within adjacent time windows. For example, if the duration interval lengths of a ring structure are 0.7, 0.8, and 0.6 in three consecutive time windows, then the rate of change is |0.8-0.7|+ |0.6-0.8| = 0.3.

[0105] The stability of the Betti number is achieved through variance calculation, which statistically measures the fluctuation of the same Betti number within a continuous time window. For example, if the 0-dimensional Betti number is 2, 3, 2, 3, 2 within five time windows, its variance is 0.5. The formula for calculating the dynamic constraint satisfaction parameter is the normalized value of the rate of change multiplied by a weighting coefficient, plus the normalized value of the Betti number stability multiplied by a complementary weighting coefficient. The weighting coefficient is dynamically adjusted according to the equipment type. For example, when a CNC machine tool has high vibration sensitivity, the rate of change weight is set to 0.7, and the stability weight is set to 0.3. The normalization method is to divide the original value by the historical maximum value. For example, if the historical maximum rate of change is 1.0 and the current rate of change is 0.3, then the normalized value is 0.3.

[0106] A subset of feasible strategies is generated by selecting strategies whose dynamic constraint satisfaction parameters exceed a dynamic threshold. The dynamic threshold is dynamically adjusted based on the equipment's real-time health index. The equipment's real-time health index is obtained through linear regression prediction of vibration spectrum characteristics and temperature drift. The vibration spectrum characteristics are extracted using Fast Fourier Transform to determine the energy percentage of the dominant frequency component. For example, after converting the vibration signal from the time domain to the frequency domain, the percentage of energy in the 5-100Hz frequency band is calculated. Temperature drift is calculated as the absolute value of the difference between the current temperature and the set value. For example, if the set temperature is 200℃ and the current temperature is 210℃, the drift is 10℃.

[0107] The training data for the linear regression model comes from historical equipment maintenance records. The inputs are the dominant frequency energy of the vibration spectrum (in decibels) and temperature drift (in degrees Celsius), and the output is a health index (range 0-100). The mapping rule between the dynamic threshold and the health index is as follows: when the health index is greater than or equal to 80, the dynamic threshold is set to 0.8; when the health index is between 60 and 80, the dynamic threshold is 0.7; and when the health index is less than 60, the dynamic threshold is 0.6. During the screening process, strategies with dynamic constraint satisfaction parameters lower than the dynamic threshold are marked as strategies to be verified. These strategies are manually reviewed; if they conform to actual production experience, they are added to the feasible strategy subset; otherwise, they are discarded.

[0108] Data preprocessing includes dynamic parameter alignment, fault tolerance in topology network construction, and optimization for persistent cohomology computation. Dynamic parameter alignment is achieved through linear interpolation, aligning parameters with different sampling frequencies to the same timestamp; for example, interpolating temperature data from 10 times per second into a time series from 1000 times per second. The fault tolerance mechanism in topology network construction automatically expands the window length to adjacent windows and recalculates if there are no local extrema within a certain time window. Optimization for persistent cohomology computation enables an approximate computation mode for large-scale topology networks, using the sparse matrix compression function of the GUDHI library to control computational errors within 5%; for example, when the number of nodes exceeds 1000, a compression algorithm is enabled to reduce memory usage.

[0109] Boundary conditions include data missing handling, computational overflow handling, and threshold mapping fault tolerance. The data missing handling rule is that if more than 50% of the time-series data for a dynamic parameter is missing, the topology network construction for that strategy is skipped and the strategy is marked as invalid. The computational overflow handling rule is that when the length of a sustained interval exceeds the total simulation duration, it is automatically truncated to the total simulation duration, and the abnormal event is recorded. The threshold mapping fault tolerance rule is that when the health index calculation is abnormal, such as a sensor failure causing all input data to be zero, the dynamic threshold is set to 0.5 by default, and a device status warning is triggered.

[0110] Hardware dependencies include a data acquisition card supporting multi-channel synchronous acquisition, such as the NI PXIe-6368, for high-frequency acquisition of vibration and temperature data, with a sampling rate of at least 1000 times per second and a resolution of at least 16 bits. Software dependencies include a digital twin simulation platform integrating the topology analysis library of GUDHI 3.4.0 or later, linear regression model training using the LinearRegression class from the Scikit-learn 1.0.2 library, and simulation step size control requiring compatibility with Python 3.9 or later.

[0111] Step S4 addresses the policy rigidity problem caused by traditional simulation verification methods relying on static thresholds or simple variance analysis by constructing a topological network of dynamic parameter time series and extracting persistent cohomology features. The topological network captures the dynamic correlation between parameters (such as the coordinated fluctuation of vibration and temperature), and persistent cohomology features (Betti number, duration interval) quantify system stability, overcoming the shortcomings of existing technologies that only focus on a single parameter or linear relationship. The dynamic threshold is adaptively adjusted based on the equipment health index, ensuring that the screening strategy meets both preset constraints and adapts to the real-time status of the equipment. Compared with traditional methods, dynamic network modeling and topological invariant analysis improve the accuracy of policy verification and system robustness, providing interpretable and adaptive optimization decision support for complex industrial scenarios.

[0112] S5. Based on the KL divergence difference between each strategy in the feasible strategy subset and historical strategies, dynamically generate strategy exploration priorities, which can be implemented as follows:

[0113] Based on the difference in parameter distribution between each strategy in the feasible strategy subset and the KL divergence of the historical strategy set, strategy diversity evaluation parameters are generated. The parameter distribution is constructed using kernel density estimation, with the bandwidth parameter adaptively adjusted according to the sample data size. The bandwidth adjustment rule adopts the Silverman empirical rule, and the sample data size is the sum of the number of strategies in the feasible strategy subset and the historical strategy set. Parameter types include vibration amplitude, temperature gradient, and electromagnetic interference intensity. The parameter types are strictly consistent with the definitions of dynamic parameters in step S4. The unit for vibration amplitude is meters per second squared (m / s²), the unit for temperature gradient is degrees Celsius per minute (°C / min), and the unit for electromagnetic interference intensity is millitalas (mT). The KL divergence difference is calculated by calculating the KL divergence between the parameter distribution of each strategy and the parameter distribution of all strategies in the historical strategy set, and taking the maximum value as the diversity evaluation parameter for that strategy. The data source for the parameter distribution is the policy parameter space after the reset in step S3, and the data source for the historical policy set is the historical iteration record within the preset time window in step S1. The storage format of the historical iteration record is a JSON file containing timestamps, policy parameters and execution results.

[0114] Priority intervals are defined based on the equipment's real-time health index, which is obtained through linear regression prediction of vibration spectrum characteristics and temperature drift in step S4. The real-time health index ranges from 0 to 100, with a first threshold of 50 and a second threshold of 70. The thresholds are set based on the statistical relationship between the health index and strategy failure events in historical equipment maintenance records; for example, when the health index is below 50, the probability of equipment failure increases to 25%. When the real-time health index is below the first threshold, the priority interval is limited to high-diversity strategies, allowing only strategies with the top 30% KL divergence differences to enter the priority queue. When the real-time health index is above the second threshold, the priority interval expands to high-similarity strategies, allowing strategies with the top 70% KL divergence differences to enter the priority queue. The capacity of the priority queue is limited to 50% of the number of strategies in the current strategy set; for example, when the number of strategies is 100, the queue can contain a maximum of 50 strategies.

[0115] The KL divergence difference and historical strategy similarity are dynamically sorted. Historical strategy similarity is calculated using cosine similarity to determine the directional consistency between strategy parameter vectors. The strategy parameter vector consists of normalized values ​​of vibration amplitude, temperature gradient, and electromagnetic interference intensity. Normalization is achieved by dividing the original parameter value by the maximum allowable value specified in the equipment technical manual. For example, if the maximum allowable vibration amplitude is 10 m / s², the normalized value is the original value divided by 10. Cosine similarity is calculated by dividing the dot product of the two strategy parameter vectors by the product of their magnitudes, with the result ranging from -1 to 1. The dynamic sorting rule is as follows: if the KL divergence difference of a strategy is greater than its cosine similarity with historical strategies, it is classified as an exploration-first strategy; otherwise, it is classified as an inheritance-first strategy. The classification results are stored as a two-dimensional label array. The first column is the strategy number, and the second column is the classification identifier. A classification identifier of 0 indicates an inheritance-first strategy, and 1 indicates an exploration-first strategy.

[0116] Strategy exploration priorities are generated based on priority intervals and strategy classification results. Exploration priority strategies are sorted in descending order of KL divergence difference within the interval; for example, a strategy with a KL divergence difference of 0.8 has higher priority than a strategy with a KL divergence difference of 0.5. Inheritance priority strategies are sorted in ascending order of historical strategy similarity; for example, a strategy with a similarity of 0.2 has higher priority than a strategy with a similarity difference of 0.6. The number of priority levels is positively correlated with the number of strategies in the current strategy set. When the number of strategies is less than 50, 5 priority levels are set; when the number of strategies is greater than 50, one level is added for every 10 additional strategies. The priority mapping table is generated by traversing the sorting results, and the storage format of the mapping table is a key-value pair of strategy number and priority level. The update frequency of the priority mapping table is synchronized with the strategy optimization cycle, which is determined by the equipment maintenance interval in the production plan; for example, priority updates are performed daily at 2:00 AM.

[0117] Parameter alignment is achieved through zero-padding. The lengths of parameter vectors for different strategies are aligned according to the parameter importance ranking in the equipment technical manual, with less important parameters padded with zeros. The kernel density estimation optimization rule is to automatically switch to a Gaussian mixture model to estimate the parameter distribution when the sample size is less than 10. The number of Gaussian components is set to the square root of the sample size; for example, the number of components is set to 3 when the sample size is 9. The threshold dynamic adjustment rule is that the first and second thresholds increase by 5% annually based on the equipment's service life, calculated from the equipment's manufacturing date and accurate to the month; for example, the threshold increases by 15% after the equipment has been in use for 3 years.

[0118] The data insufficiency handling rule is that if the historical strategy set is empty, the default strategy set of the same type of device will be temporarily used as an alternative data source. The default strategy set is pre-stored in the device's local database at / opt / strategy / default.json. The health index anomaly handling rule is that when the calculated health index result exceeds the range of 0-100, it will be automatically clamped to the nearest boundary value and a data verification process will be triggered. The verification process includes re-acquiring sensor data and retraining the linear regression model. The sorting conflict handling rule is that if multiple strategies have the same KL divergence difference or similarity, they will be sorted in ascending order by the strategy generation timestamp. The timestamp is extracted from the strategy space reset log in step S3, and the log format is a Unix timestamp (millisecond precision).

[0119] Hardware dependencies include local database storage devices, such as MySQL 8.0 Community Edition, and at least 8GB of memory to ensure efficient kernel density estimation calculations. Software dependencies include the `KernelDensity` class from the Python library `Scikit-learn` 1.2.0 used for kernel density estimation, and the `dot` and `linalg.norm` functions from the NumPy 1.24.3 library used for cosine similarity calculations. Simulation step size control must be compatible with Python 3.9 or later, and the digital twin simulation model in step S4 must be called via `subprocess`.

[0120] The uniformity of dimensions and units is achieved through data standardization. The raw data of vibration amplitude, temperature gradient, and electromagnetic interference intensity are converted to standard units before being input into the kernel density estimation model. Parameter range limitations are enforced through preset values ​​in the equipment technical manual; for example, vibration amplitudes exceeding 10 m / s² are automatically truncated to 10 m / s². The calculation results of KL divergence differences are ensured to be non-negative through natural logarithmic transformation to avoid numerical overflow caused by non-overlapping probability distributions.

[0121] Step S5 addresses the static trade-off between exploration and utilization in traditional strategy priority allocation by dynamically classifying and ranking based on KL divergence differences and historical strategy similarity. Priority intervals are dynamically divided based on equipment health indices, ensuring that validated strategies are inherited first in high-health states (guaranteeing stability), while new strategies are explored more heavily in low-health states (improving adaptability). A parameter distribution is constructed using kernel density estimation, and cosine similarity is combined to measure strategy direction consistency, overcoming the strategy rigidity problem caused by existing technologies relying on fixed thresholds or simple weighting. Compared to traditional methods, this health-index-driven dynamic classification mechanism enables adaptive adjustment of the strategy optimization process, providing decision support that balances stability and diversity for complex production environments.

[0122] S6. Deploy the strategy with the highest priority in strategy exploration to the actual production system and update the current set of optimized strategies. Specifically, this can be implemented as follows:

[0123] The highest-priority strategy is extracted from the strategy exploration priority ranking results generated in step S5, and its compliance with the equipment real-time health index and production order process standards is verified. The equipment real-time health index is obtained by linear regression prediction of vibration spectrum characteristics and temperature drift in step S4. The health index ranges from 0 to 100, with a first threshold of 50 and a second threshold of 70. The production order process standards are extracted from the order document and include indicators such as dimensional accuracy and surface roughness. For example, the dimensional accuracy requirement is ±0.05mm, and the surface roughness requirement is Ra≤0.8μm. During verification, the strategy parameters are compared with the standard values ​​one by one. The parameter types include vibration amplitude, temperature gradient, and electromagnetic interference intensity. The parameter types are strictly consistent with the definitions of dynamic parameters in step S4. The unit of vibration amplitude is meters per second squared (m / s²), the unit of temperature gradient is degrees Celsius per minute (°C / min), and the unit of electromagnetic interference intensity is millitalas (mT). If all parameters are within the allowable tolerance range, the verification is considered successful.

[0124] The validated strategies are deployed to the actual production system. Deployment instructions are transmitted to the device controller (Siemens S7-1500) via the OPC UA industrial communication protocol. After deployment, the current optimized strategy set is updated based on the device's real-time health index: when the health index is below the first threshold (50), the strategy with the smallest KL divergence difference is replaced from the current set; when the health index is above the second threshold (70), the strategy with the highest historical strategy similarity is replaced. The KL divergence difference is calculated using the kernel density estimation method in step S5, and the historical strategy similarity is calculated using cosine similarity. The frequency of replacement operations is synchronized with the production batches; for example, the strategy set is updated after each production batch is completed.

[0125] The execution results of the deployment strategy are recorded in the historical strategy set. These results include the deviation rate between actual production efficiency indicators and simulation predictions. Actual production efficiency indicators are collected in real time by sensors; for example, processing cycle time is measured in seconds, and energy consumption is measured in kilowatt-hours. Simulation predictions are extracted from the digital twin model output in step S4. The deviation rate is calculated using the root mean square error, which is the square root of the square mean of the differences between the actual and predicted values. The result is normalized to a percentage. The deviation rate is stored as an additional field in the historical strategy set for reference in subsequent strategy optimization.

[0126] The length of the moving average window in step S1 is dynamically adjusted based on the deviation rate. The adjustment rule is as follows: when the deviation rate exceeds a preset tolerance value (10%), the window length is shortened to respond quickly to changes, for example, reducing it from 30 iterations to 20; when the deviation rate is below the tolerance value, the window length is extended to smooth out noise interference, for example, expanding it from 30 iterations to 40. The adjustment range of the window length is positively correlated with the rate of change of the deviation rate, which is calculated by the difference in deviation rates between two adjacent production batches. The difference is calculated by subtracting the absolute value of the deviation rate of the previous batch from the current deviation rate.

[0127] Generate a strategy deployment execution log. The log format is consistent with the storage structure of the strategy space reset log in step S3, including fields such as timestamp, strategy parameters, execution results, and health index. The timestamp uses Unix timestamp format (millisecond precision), strategy parameters are stored as JSON key-value pairs, and the execution result includes the percentage deviation between the actual and predicted values. The log file is named using a combination of the device's unique identifier and the timestamp. For example, if the device identifier is CNC-001 and the timestamp is 20231005120000, the log file will be named CNC-001_20231005120000.json. The log is stored in the / log / strategy_deployment directory of the device's local database. The local database is MySQL 8.0 Community Edition, and the logs are synchronously backed up to the cloud storage service AWS S3, with a backup cycle of 1:00 AM daily.

[0128] Alignment between strategy parameters and production order process standards is achieved through a parameter mapping table. This table records the correspondence between strategy parameter names and process standard indicators; for example, "cutting speed" maps to "processing cycle time." During verification, if the process standards are updated, the mapping table is automatically reloaded, triggering a secondary verification. The strategy with the smallest KL divergence difference is calculated by traversing the current strategy set, excluding already deployed strategies to avoid duplicate replacements.

[0129] If the deployment strategy fails verification (e.g., parameters exceed tolerance range), an automatic rollback mechanism is triggered, rolling back to the previous valid strategy and marking the current strategy as invalid. When the deviation rate exceeds the tolerance value three times consecutively, a manual intervention process is triggered, pausing automatic strategy updates and awaiting engineer confirmation. If log storage fails (e.g., insufficient disk space), it is temporarily cached in memory and retried for writing, with a retry interval of 5 minutes and a maximum of 5 retries. A storage anomaly alarm is triggered after a retry failure.

[0130] Hardware dependencies include industrial controllers supporting the OPC UA protocol (such as the Siemens S7-1500) for command transmission; and local database storage devices (such as a Raspberry Pi 4B with an SSD) for log storage. Software dependencies include a JSON log generation library (Python's json module), a cloud storage client (AWS S3 SDK version 1.26.7), and the open-source OPC UA library (FreeOpcUa version 0.98.1) for command transmission.

[0131] The units of actual production efficiency indicators are strictly consistent with those of simulation predictions; for example, processing cycle time is measured in seconds, and energy consumption in kilowatt-hours. When calculating the deviation rate normalization, the denominator is the absolute value of the simulation prediction. If the prediction value is zero, it is automatically replaced with the historical average to avoid division by zero errors. The adjustment step size of the moving average window length is fixed at 10 iterations, with a minimum window length of no less than 10 iterations and a maximum of no more than 50 iterations.

[0132] By constructing a closed-loop optimization system through multi-step technological collaboration, the real-time health status of equipment is deeply coupled with the strategy optimization process, overcoming the limitations of isolated processing of strategy generation, verification, and deployment in traditional methods. In steps S1 to S3, a strategy space reset mechanism based on dynamic entropy monitoring quantifies the multi-physics coupling effect through topological data analysis and chaotic attractor modeling, solving the industry problem that traditional simulation models struggle to capture nonlinear dynamic correlations. Steps S4 to S5 introduce a priority classification rule driven by a health index, dynamically binding the real-time status of equipment (vibration, temperature, etc.) with strategy diversity evaluation parameters, replacing the traditional fixed weight or static threshold allocation method, and achieving adaptive adjustment of the exploration-utilization balance. In step S6, the strategy deployment and historical strategy set update logic dynamically corrects the data acquisition window length through deviation rate feedback, forming a cross-step adaptive learning closed loop. Compared to existing local optimization schemes based on rule engines or empirical formulas, this technology achieves a leap in global optimization capabilities through the synergy of topological network construction, dynamic interval partitioning, and cross-cycle feedback correction. Deeply integrate cross-disciplinary tools such as mathematical topology and information entropy theory with industrial optimization scenarios. For example, use persistent homology features to quantify strategy stability and use asymmetric sorting based on KL divergence difference and cosine similarity.

[0133] Example 2: Figure 2 A schematic diagram of the adaptive optimization system for production decision-making of the present invention is provided. The adaptive optimization system for production decision-making includes the following modules:

[0134] The entropy monitoring module is used to monitor the policy entropy value of the current optimization policy set in real time. The policy entropy value is obtained by calculating the distribution dispersion of each policy in historical iterations.

[0135] The instruction triggering module is used to trigger a reconstruction instruction for the digital twin simulation model when the policy entropy value is lower than a preset entropy threshold.

[0136] The parameter reset module is used to reset the policy space of the digital twin simulation model according to the reconstruction instruction, and generate a set of candidate policies containing the new policy parameter space.

[0137] The simulation verification module is used to perform simulation verification on the candidate strategy set using the digital twin simulation model after the strategy space is reset, and to select a subset of feasible strategies that meet the preset constraints.

[0138] The priority generation module is used to dynamically generate strategy exploration priorities based on the KL divergence differences between each strategy in the feasible strategy subset and historical strategies.

[0139] The strategy deployment module is used to deploy the strategy with the highest priority in strategy exploration to the actual production system and update the current set of optimized strategies.

[0140] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0141] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0142] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0143] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0144] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0145] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0146] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0147] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0149] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive optimization method for production decision-making, characterized in that, Includes the following steps: S1. Monitor the policy entropy value of the current optimization policy set in real time. The policy entropy value is obtained by calculating the distribution dispersion of each policy in historical iterations. The strategy entropy value of the current optimization strategy set is monitored in real time. The strategy entropy value is obtained by calculating the distribution dispersion of each strategy in the historical iterations, including: obtaining the strategy parameter vector of each strategy in the current optimization strategy set in the historical iterations; extracting the feature dimensions of all strategy parameter vectors within a preset time window; calculating the variance value of each dimension parameter of the strategy parameter vector based on the numerical fluctuation range of the feature dimensions; and generating the strategy entropy value by weighted summation based on the variance values ​​of each dimension parameter. S2. When the strategy entropy value is lower than the preset entropy threshold, a reconstruction instruction for the digital twin simulation model is triggered. When the strategy entropy value is lower than the preset entropy threshold, a reconstruction instruction for the digital twin simulation model is triggered, including: obtaining the dynamic adjustment rules of the preset entropy threshold; comparing the real-time monitored current strategy entropy value with the preset entropy threshold to generate a threshold deviation evaluation result; based on the judgment that the threshold deviation evaluation result exceeds the preset deviation threshold, generating a reconstruction instruction for the digital twin simulation model containing the reconstruction intensity level; and transmitting the reconstruction instruction to the control interface of the digital twin simulation model, whereby the control interface performs the corresponding parameter initialization operation according to the reconstruction intensity level. S3. Reset the strategy parameter space of the digital twin simulation model according to the reconstruction instruction, and generate a candidate strategy set containing the new strategy parameter space. S4. Use the digital twin simulation model after resetting the strategy space to perform simulation verification on the candidate strategy set, and select a subset of feasible strategies that meet the preset constraints. A digital twin simulation model with a reset strategy space is used to verify the candidate strategy set through simulation, and a subset of feasible strategies that meet preset constraints is selected. This includes: generating time series of the dynamic parameters of each strategy in the candidate strategy set, including vibration amplitude, temperature gradient, and electromagnetic interference intensity; generating nodes of the topology network based on the local maxima and minima of the time series of dynamic parameters, with the absolute values ​​of the Pearson correlation coefficients of the time series of dynamic parameters within adjacent time windows constituting the edges of the topology network; extracting the persistent cohomology features of the topology network, calculating the Betti number and duration interval length for each dimension, and generating a set of topological invariants; calculating the dynamic constraint satisfaction parameter based on the evolution trend of the set of topological invariants, with the evolution trend jointly characterized by the rate of change of duration interval length and the stability of the Betti number; and selecting strategies with dynamic constraint satisfaction parameters higher than a dynamic threshold to generate a subset of feasible strategies, with the dynamic threshold dynamically adjusted according to the real-time health index of the equipment. S5. Based on the KL divergence difference between each strategy in the feasible strategy subset and the historical strategy, dynamically generate the strategy exploration priority; S6. Deploy the strategy with the highest priority in strategy exploration to the actual production system and update the current set of optimized strategies.

2. The adaptive optimization method for production decision-making according to claim 1, characterized in that, The policy parameter space of the digital twin simulation model is reset according to the reconstruction instruction, generating a candidate policy set containing the new policy parameter space, including: The reconfiguration intensity level in the reconfiguration instruction is parsed, and a parameter dependency constraint table corresponding to the reconfiguration intensity level is generated. The parameter dependency constraint table records the physical coupling rules and priority weights between process parameters. Based on the constraint table of reconstruction intensity level and parameter dependency relationship, dynamically decouple conflicting parameter dimensions and lock independent adjustable dimensions; Based on the real-time sensor data and parameter dependency constraint table of the device, multi-level boundary constraint iterative correction is performed on the decoupled independent adjustable dimension. A set of candidate policies is generated in the corrected policy parameter space through dynamic adaptive pseudo-random sampling.

3. The adaptive optimization method for production decision-making according to claim 2, characterized in that, The locking condition for independently adjustable dimensions is that the parameter adjustment range does not violate the physical limits of the equipment and meets the process standards of the production order; The correction rule for iterative correction of multi-level boundary constraints is to first relax the parameter dimension constraints of high-priority weights, and simultaneously compress the parameter dimension constraints of low-priority weights. The sampling density of dynamic adaptive pseudo-random sampling is dynamically adjusted according to the priority weight of the parameter dimension.

4. The adaptive optimization method for production decision-making according to claim 1, characterized in that, Based on the KL divergence differences between each strategy in the feasible strategy subset and historical strategies, a strategy exploration priority is dynamically generated, including: Based on the difference between the parameter distribution of each strategy in the feasible strategy subset and the KL divergence of the historical strategy set, strategy diversity evaluation parameters are generated, and the parameter distribution is constructed by the kernel density estimation method. Priority zones are determined based on the real-time health index of the equipment. The KL divergence difference and historical strategy similarity are dynamically sorted. When the KL divergence difference is greater than the historical strategy similarity, it is classified as an exploration-first strategy, and otherwise as an inheritance-first strategy. Based on the priority interval and strategy classification results, strategy exploration priorities are generated. Exploration priority strategies are arranged in descending order of KL divergence difference within the interval, while inheritance priority strategies are arranged in ascending order of historical strategy similarity. The number of priority levels is positively correlated with the number of strategies in the current strategy set.

5. The adaptive optimization method for production decision-making according to claim 4, characterized in that, When dividing priority intervals based on the device's real-time health index, the priority interval is limited to high diversity strategies when the device's real-time health index is below the first threshold, and expanded to high similarity strategies when it is above the second threshold.

6. The adaptive optimization method for production decision-making according to claim 1, characterized in that, Deploy the highest priority strategy from the strategy exploration to the actual production system, and update the current set of optimization strategies, including: Extract the highest priority strategy from the strategy exploration priority ranking results and verify that it conforms to the real-time health index of the equipment and the process standards of the production order. Deploy the validated strategies to the actual production system and update the current set of optimized strategies based on the real-time health index of the equipment: replace the strategy with the smallest KL divergence when the real-time health index of the equipment is lower than the first threshold, and replace the strategy with the highest similarity to the historical strategy when it is higher than the second threshold. Record execution results to the historical strategy set, dynamically adjust the moving average window length based on the deviation rate between the actual and simulated results, and generate a strategy deployment execution log.

7. A production decision adaptive optimization system, used to implement the production decision adaptive optimization method according to any one of claims 1-6, characterized in that, Includes the following modules: The entropy monitoring module is used to monitor the policy entropy value of the current optimization policy set in real time. The policy entropy value is obtained by calculating the distribution dispersion of each policy in historical iterations. The instruction triggering module is used to trigger a reconstruction instruction for the digital twin simulation model when the policy entropy value is lower than a preset entropy threshold. The parameter reset module is used to reset the policy space of the digital twin simulation model according to the reconstruction instruction, and generate a set of candidate policies containing the new policy parameter space. The simulation verification module is used to perform simulation verification on the candidate strategy set using the digital twin simulation model after the strategy space is reset, and to select a subset of feasible strategies that meet the preset constraints. The priority generation module is used to dynamically generate strategy exploration priorities based on the KL divergence differences between each strategy in the feasible strategy subset and historical strategies. The strategy deployment module is used to deploy the strategy with the highest priority in strategy exploration to the actual production system and update the current set of optimized strategies.

Images