A production line capacity intelligent management method for pharmaceutical production

Abstract

The application relates to the technical field of intelligent management of medicine production, and discloses a production line capacity intelligent management method for medicine production. The method uses multiple sensing terminals to collect real-time data of the production line, and forms a clean data set through collection and quality evaluation. The clean data is imported into a dynamically maintained correlation mapping network, and a forward-looking capacity deduction is performed based on the current state of the network to obtain a conclusion of future production rhythm. The deduction conclusion and real-time resource state are combined to generate a preliminary job timing arrangement, and then the arrangement is checked for compliance and simulated for performance under multiple constraints, and a recommended job timing arrangement is formed through iterative correction. During execution, compliance deviation and performance fluctuation are monitored in real time, and immediate correction is triggered. The method improves the adaptability and accuracy of capacity prediction through dynamic correlation deduction, and simultaneously optimizes compliance and operation performance at the source through the verification simulation cycle.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent management of pharmaceutical production, in particular to a production line capacity intelligent management method for pharmaceutical production. BACKGROUND

[0002] In the field of pharmaceutical production, production line capacity management is crucial for ensuring drug supply, controlling production costs, and meeting stringent quality regulations. Existing capacity management techniques typically rely on static models based on historical data or empirical formulas for capacity prediction. These methods treat the production system as a relatively fixed framework, and their deduction logic cannot fully absorb the massive, multi-source data generated during real-time operation of the production line, making it difficult to capture the dynamic interactions between device status, material flow, environmental parameters, and other dynamic elements. When the production line faces order changes, equipment abnormalities, or process adjustments, the prediction accuracy of static models decreases significantly, leading to a disconnect between predicted production rhythm and reality.

[0003] Existing production operation scheduling schemes focus on meeting delivery dates, optimizing equipment utilization, and other single or few goals. After scheduling is generated, only simple consistency checks are usually performed, or post-audit is relied on human experience to review the complex compliance constraints that must be followed in pharmaceutical production. This mode cannot systematically and automatically integrate multiple hard and flexible constraints such as Good Manufacturing Practice, equipment cleaning and disinfection cycle, material shelf life, personnel operation qualifications, environmental control standards, and energy consumption safety thresholds into the scheduling phase. Lack of full-process performance simulation of scheduling schemes under real constraints may result in risks in compliance or failure to achieve optimal allocation of resources and efficiency. SUMMARY

[0004] The present application aims to provide a production line capacity intelligent management method for pharmaceutical production to solve the problems raised in the background.

[0005] To achieve the above-mentioned purpose, the present application provides a production line capacity intelligent management method for pharmaceutical production, which comprises:

[0006] By deploying multiple types of sensing terminals at different physical locations in the production line, raw signal streams reflecting real-time operation are captured synchronously, and initial collection is performed according to a pre-defined pharmaceutical production data classification system to form a source data set with category labels;

[0007] Call the built-in data quality evaluation routine to score the credibility and mark the integrity of each data in the source data set with category labels, and perform screening and reconstruction operations accordingly to generate a clean data set;

[0008] Importing the clean data set into a dynamically maintained correlation mapping network, running a forward-looking capacity deduction process based on the current state of the correlation mapping network and the latest content of the clean data set, to obtain deduction conclusions about future production rhythm;

[0009] Combining the deduction conclusions about future production rhythm with real-time feedback resource availability state, constructing a preliminary job timing arrangement;

[0010] Performing compliance verification and performance simulation under multiple constraints on the preliminary job timing arrangement, and forming a recommended job timing arrangement after iterative correction;

[0011] In the actual execution process of the recommended job timing arrangement, a continuous compliance deviation monitoring and performance fluctuation tracking mechanism is started, and immediate correction logic is triggered according to the monitoring and tracking results, and coordination instructions are sent to the production line control system.

[0012] Preferably, the original signal stream reflecting the running situation is synchronously captured by deploying multiple types of sensing terminals at different physical locations of the production line, specifically including:

[0013] Reading device state original signals from vibration, temperature, and current sensors on production equipment;

[0014] Reading production results and consumption original signals from batch record terminals, material weighing systems, and quality inspection ports;

[0015] Reading temperature, humidity, and particle count original signals from environmental monitoring probes;

[0016] Reading personnel presence and operation original signals from access control, attendance, and electronic batch recording systems;

[0017] Packing and time stamping all original signals according to a unified time reference to form an original signal stream.

[0018] Preferably, the built-in data quality evaluation routine is called to score the credibility and mark the integrity of each data in the source data set with category labels, including:

[0019] Setting normal value range and reasonable logical relationship rules for each type of data;

[0020] Comparing each data in the source data set with the corresponding normal value range, and assigning a lower credibility score to data outside the range;

[0021] Checking whether the logical relationship between associated data items in the source data set conforms to the reasonable logical relationship rules, and assigning an integrity loss mark to data sets that do not conform to the rules;

[0022] According to a preset data credibility threshold, data items with credibility scores lower than the data credibility threshold are marked as to-be-verified data;

[0023] For the data set marked as integrity loss, record the missing data elements and their context information.

[0024] Preferably, the filtering and reconstruction operations are performed according to the data to generate a clean data set, including:

[0025] Eliminate data items in the source data set that are marked as to-be-verified and cannot be corrected by auxiliary information sources;

[0026] For the data set marked as integrity loss, according to the historical association patterns stored in the association mapping network, infer and supplement the most likely missing data elements;

[0027] Convert all retained and supplemented data to a unified unit of measurement, time format, and naming specification;

[0028] After the elimination, supplementation, and conversion processing, the data is reorganized in chronological order and by category to form a clean data set that can be directly called by subsequent links.

[0029] Preferably, the clean data set is imported into the dynamically maintained association mapping network, including:

[0030] The association mapping network is used to represent the internal relationship between equipment, materials, procedures, and personnel in the drug production process;

[0031] The association mapping network exists in the form of nodes and edges, with nodes representing entities in drug production, including specific equipment units, material batches, process step documents, and qualified personnel, and edges representing known relationships between entities;

[0032] When a new clean data set is imported, identify the entities involved and the newly generated or changed relationships between entities;

[0033] Update or add corresponding nodes and edges in the association mapping network to reflect the latest configuration and state of the production line;

[0034] Establish an index link from a specific data item in the clean data set to the corresponding node in the association mapping network.

[0035] Preferably, based on the current state of the association mapping network and the latest content of the clean data set, a forward-looking capacity deduction process is run to obtain deduction conclusions about future production rhythm, including:

[0036] Extract currently active device nodes, available material nodes, and on-duty personnel nodes and their connection relationships from the association mapping network;

[0037] Extract recent production efficiency, equipment failure interval history, and material consumption rate data from the clean data set;

[0038] Construct a virtual timeline starting from the current time, using the active device nodes, available material nodes, and on-duty personnel nodes as resources, and simulate the task execution and resource consumption process over a future period based on historical efficiency and consumption rate;

[0039] During the simulation, resource conflicts and bottlenecks are identified based on the constraints defined in the associated mapping network.

[0040] The virtual timeline state at the end of the simulation is output as the deduction conclusion, which includes the predicted production completion time and the expected remaining resources.

[0041] Preferably, the preliminary job sequence arrangement, which combines the projected conclusions about future production rhythm with real-time feedback on resource availability, includes:

[0042] Analyze the deduction conclusions to obtain the suggested task initiation order and the estimated duration of each task;

[0043] Get real-time equipment availability status from the equipment monitoring system, real-time material inventory quantity from the inventory management system, and real-time personnel on-duty status from the personnel management system.

[0044] Based on the suggested task initiation sequence, each task is matched with the specific equipment, materials, and personnel actually available at the current moment;

[0045] When resource shortage occurs during the matching process, the task order is adjusted or the task is marked as waiting for resources according to the task priority rules.

[0046] All tasks that have been successfully matched and have a specific start time are arranged in chronological order to form a preliminary job sequence arrangement.

[0047] Preferably, the step of performing compliance verification and performance simulation under multiple constraints on the initial job timing arrangement, and iteratively correcting it to form a recommended job timing arrangement, includes:

[0048] Multiple constraints are set, including production environment conditions required by Good Manufacturing Practice (GMP) for pharmaceuticals, continuous operating limits of equipment, maximum working hours of personnel, and the first-in-first-out (FIFO) principle for materials.

[0049] The initial job timing arrangement is placed in a simulation environment and executed step by step according to its timeline, with each step checking whether any of the multiple constraints are violated.

[0050] For any step that violates the constraints, a strategy is selected from a pre-set adjustment strategy library for correction. The strategies include postponing the task, changing equipment or personnel, or splitting the task.

[0051] Re-simulate and verify using the revised job timing arrangement, repeating this process until the job timing arrangement fully satisfies all multiple constraints in the simulation.

[0052] The timing arrangements that pass the final verification will be marked as recommended job timing arrangements.

[0053] Preferably, during the actual execution of the recommended job sequence arrangement, a continuous compliance deviation monitoring and performance fluctuation tracking mechanism is initiated, including:

[0054] When the recommended work sequence is actually executed on the production line, real data related to task execution is continuously collected, including the actual start and end times of the task, the actual materials consumed, and the actual environmental data generated.

[0055] The collected actual data is compared in real time with the pre-defined planned data in the recommended job timing arrangement and the thresholds specified in the multiple constraints.

[0056] When actual data is found to deviate from planned data by more than the allowable tolerance, or when actual data approaches or exceeds the compliance threshold, it is recorded as a compliance deviation event.

[0057] Calculate the ratio of the actual execution efficiency to the planned efficiency of the critical task. When the ratio is consistently lower than a set level, it is recorded as a performance fluctuation event.

[0058] Generate detailed logs containing the aforementioned compliance deviation events and performance fluctuation events.

[0059] Preferably, the step of triggering real-time correction logic based on monitoring and tracking results and sending coordination instructions to the production line control system includes:

[0060] Receive detailed logs generated by the compliance deviation monitoring and performance fluctuation tracking mechanism;

[0061] Analyze the event types, severity, and root causes in the detailed logs;

[0062] Based on preset real-time correction rules, it is determined whether intervention is needed on the currently executing recommended job sequence arrangement. The real-time correction rules specify the intervention levels corresponding to different event types and severity.

[0063] When intervention is deemed necessary, a local time-series adjustment plan or resource reallocation plan for the affected task is quickly generated based on the aforementioned correlation mapping network and the current clean data set.

[0064] The local timing adjustment scheme or resource reallocation scheme is converted into specific equipment start-up and shutdown, material flow or personnel allocation instructions, and sent to the production line control system as coordination instructions.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] By constructing a dynamically maintained associative mapping network, clean data sets obtained through real-time processing are used as input, and extrapolation calculations are performed based on the latest association states between network nodes. This network can characterize the real-time dependencies and influence paths of material flow, information flow, and control flow among entities such as equipment, processes, and the environment. The extrapolation process simulates the propagation effects of data and events in the network, ensuring that conclusions about future production rhythms are not only based on historical patterns but also on the immediate and dynamic interactions between the various components of the system. This gives capacity extrapolation adaptive evolution capabilities, enabling it to more sensitively perceive micro-changes in the internal state of the production line and their potential chain reactions. Consequently, it can update predictions more quickly in the face of disturbances, providing a more realistic and forward-looking situational assessment basis for subsequent scheduling decisions.

[0067] After generating the initial job sequence arrangement, an iterative correction process is introduced that integrates real-time verification of multiple constraints and full-process performance simulation. The system automatically and synchronously reviews the arrangement scheme using various compliance rules and business constraints related to pharmaceutical production. Then, within the feasible solution space that satisfies the constraints, the scheme is digitally simulated to quantitatively evaluate its performance indicators such as capacity output, resource consumption, and time span under specific constraints. Based on the feedback from verification and simulation, the system automatically adjusts the process sequence, resource allocation parameters, or time windows, and re-enters the verification-simulation cycle. This process continues until a recommended job sequence arrangement is output that strictly meets all preset complex compliance conditions and has been verified as highly efficient in the simulation environment. This ensures that the final production scheduling scheme has both compliance and efficiency guarantees from the outset, reducing adjustments and compliance risks during the execution phase, and directly improving the executability and overall economic benefits of the production scheduling scheme. Attached Figure Description

[0068] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent production line capacity management method for pharmaceutical manufacturing described in this invention.

[0069] Figure 2 A flowchart for the execution of data quality assessment routines;

[0070] Figure 3 A flowchart for generating a clean dataset for filtering and refactoring operations;

[0071] Figure 4 A bar chart for monitoring task completion time deviation;

[0072] Figure 5 A stacked bar chart for classifying and statistically analyzing production anomalies. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Please see Figure 1 This invention provides an intelligent management method for production line capacity in pharmaceutical manufacturing. The method includes: multiple sensing terminals deployed at different physical locations on the production line synchronously capturing raw signal streams reflecting the actual operation, and initially aggregating them according to a predefined pharmaceutical production data classification system to form a source data set with category labels; calling a built-in data quality assessment routine to perform credibility scoring and integrity marking on each data item in the source data set, and performing filtering and reconstruction operations accordingly to generate a clean data set; importing the clean data set into a dynamically maintained association mapping network, and running a forward-looking capacity projection process based on the current state of the association mapping network and the latest content of the clean data set to obtain projection conclusions about the future production rhythm; combining the projection conclusions with the real-time feedback of resource availability status to construct a preliminary work sequence arrangement; performing compliance verification and performance simulation under multiple constraints on the preliminary work sequence arrangement, and iteratively correcting it to form a recommended work sequence arrangement; during the actual execution of the recommended work sequence arrangement, initiating a continuous compliance deviation monitoring and performance fluctuation tracking mechanism, and triggering immediate correction logic based on the monitoring and tracking results to send coordination instructions to the production line control system.

[0075] In one embodiment of the present invention, see [reference] Figure 2Multiple types of sensing terminals are deployed at different physical locations on the production line to synchronously capture raw signal streams reflecting the actual operation. This process specifically includes reading raw signals of equipment status from vibration, temperature, and current sensors on production equipment; reading raw signals of production results and consumption from batch recording terminals, material weighing systems, and quality inspection ports; reading raw signals of temperature, humidity, and particle count from environmental monitoring probes; and reading raw signals of personnel presence and operation from access control, attendance, and electronic batch recording systems. All raw signals are packaged and timestamped according to a unified time base to form a raw signal stream. The built-in data quality assessment routine is invoked to perform credibility scoring and integrity marking on each data item in the source dataset with category labels. This process sets normal numerical ranges and reasonable logical relationship rules for each type of data, compares each data item in the source dataset with the corresponding normal numerical range, assigns a lower credibility score to data that exceeds the range, checks whether the logical relationship between related data items in the source dataset conforms to the reasonable logical relationship rules, assigns an integrity missing mark to data sets that do not conform to the rules, marks data items with credibility scores below the preset data credibility threshold as data to be verified, and records the missing data elements and their context information for data sets marked as integrity missing.

[0076] The raw signal stream is simultaneously captured by multiple sensing terminals deployed at different physical locations on the production line, and the source data set with category labels is processed by calling built-in data quality assessment routines. Taking a tablet production line as an example, the deployment and signal capture operations of multiple sensing terminals are specifically executed as follows: Vibration sensors on the production equipment are installed on the main drive shaft of the tablet press, temperature sensors are embedded in the coating pan jacket, and current sensors are connected to the power supply circuit of the mixer motor. These sensors continuously read the raw signals of equipment status. Batch recording terminals are located at the end of the packaging line, material weighing systems are installed at the feeding port of the raw material storage room, and quality inspection ports are connected to online near-infrared spectrometers. These terminals read the raw signals of production results and consumption. Environmental monitoring probes are distributed in key functional rooms of the clean area, continuously reading the raw signals of temperature, humidity, and suspended particulate count. Access control card readers are installed at the entrance of the changing room, attendance terminals are located in the workshop office, and the operation terminals of the electronic batch recording system are distributed at various positions. These systems read the raw signals of personnel presence and operation. All raw signals captured from the aforementioned sensing terminals are transmitted to a unified data acquisition gateway. Based on a time reference provided by a high-precision network time protocol server, the data acquisition gateway binds each signal data packet to a timestamp accurate to the millisecond level, forming a raw signal stream with a unified time sequence marker. This raw signal stream is then sent to a classification and aggregation module, which automatically attaches category labels to each data packet according to a predefined pharmaceutical production data classification system.

[0077] In some embodiments, a built-in data quality assessment routine is activated to process a source dataset labeled with categories. The routine first accesses a predefined rule base, which sets normal numerical ranges and reasonable logical relationship rules for each data category. For example, the normal numerical range for the vibration frequency of a tablet press spindle is set to 45 Hz to 55 Hz, and the reasonable logical relationship rule for the coating pan temperature requires that it and the jacket steam pressure reading meet a positive correlation constraint at a specific process stage. The routine compares each data item in the source dataset with its corresponding normal numerical range, assigning a lower confidence score to data items outside the range. For example, when a tablet press signal with a vibration frequency of 65 Hz is read, the routine assigns a score below a preset data confidence threshold. Simultaneously, the routine checks the logical relationships between related data items in the source dataset, such as checking whether the mixer current value and the material weighing system's feed value at the same time point conform to a predetermined power consumption-load model, assigning an integrity missing flag to data sets that do not conform to the reasonable logical relationship rules. Based on a preset data credibility threshold, the data quality assessment routine marks all data items with credibility scores below the threshold as data to be verified. For datasets marked as lacking completeness, the data quality assessment routine records the missing data elements and their specific contextual information.

[0078] It is understandable that the calculation of data credibility scores can be quantified based on the degree of bias. In one optional implementation, the credibility score of an individual data item is calculated using the following formula:

[0079] ;

[0080] in: Represents the credibility score of a single data item. This represents the raw values ​​measured by the sensing terminal. This represents the median value of the normal range for this type of data. This represents the preset allowable deviation radius. When the measured value... With median The absolute deviation within the allowable deviation radius Credibility score It decreases linearly between 0 and 1; when the absolute deviation exceeds the allowable deviation radius... At that time, credibility score The value is directly assigned to 0. The data quality assessment routine performs a marking operation based on this score.

[0081] In one embodiment of the present invention, see [reference] Figure 3Based on the data quality assessment results, a filtering and reconstruction operation is performed to generate a clean dataset. This operation removes data items from the source dataset that are marked as unverifiable and cannot be corrected by auxiliary information sources. For datasets marked as lacking completeness, the most likely missing data elements are inferred and added based on the historical association patterns stored in the association mapping network. All retained and added data are transformed according to a unified unit of measurement, time format, and naming convention. The data after the removal, addition, and transformation processes are then reorganized according to time order and category. The clean data set is imported into a dynamically maintained association mapping network, which is used to characterize the inherent relationships between equipment, materials, procedures, and personnel in the pharmaceutical manufacturing process. This network exists in the form of nodes and edges. Nodes represent entities in pharmaceutical manufacturing, including specific equipment units, material batches, process step documents, and qualified personnel. Edges represent known relationships between entities. When a new clean data set is imported, the entities involved and the newly generated or changed relationships between entities are identified. The corresponding nodes and edges are updated or added in the association mapping network to reflect the latest configuration and status of the production line. An index link is established from specific data items in the clean data set to the corresponding nodes in the association mapping network.

[0082] Based on the data quality assessment results, screening and reconstruction operations are performed to generate a clean dataset and import the clean dataset into a dynamically maintained association mapping network. Taking a wet granulation production line as an example, the screening and reconstruction operations are specifically executed as follows: The built-in data processing module receives the source dataset with confidence scores and integrity tags from the data quality assessment routine. The data processing module first performs a rejection operation, scanning all data items marked as pending verification in the source dataset. For these data items, the data processing module attempts to access auxiliary information sources for correction; auxiliary information sources include equipment maintenance logs, manual confirmation records, or backup readings of redundant sensors. For example, a peak current data of a granulator motor marked as pending verification may be corrected and retained if a momentary power fluctuation event is recorded in the equipment maintenance log at the same time point; if no valid correction basis can be obtained from any auxiliary information source, the data item is directly rejected. Next, the data processing module processes datasets marked as lacking completeness. For example, a dataset might show an operation record for "adhesive solution addition" but lack the associated "material weighing system reading." The data processing module then infers the missing data based on historical association patterns stored in the association mapping network. This network records a quantitative relationship model between "adhesive solution addition" operations and "material weighing system readings" from the past 100 batches. Using this model, combined with the process parameters of the current batch, the data processing module infers the most likely missing material weight data element and adds it. After the removal and addition are completed, the data processing module performs a transformation operation on all retained and added data, converting the data according to a unified unit of measurement, time format, and naming convention. For example, it converts all Fahrenheit readings from temperature sensors from different suppliers to Celsius, unifies all timestamps to the format "YYYY-MM-DDHH:MM:SS.fff," and unifies aliases for the same equipment in different systems to a standard equipment number. The data processing module reorganizes the data, which has undergone filtering, filling, and transformation, according to time sequence and data category, forming a clean data set with a well-structured structure that can be directly called by subsequent analysis processes.

[0083] In some embodiments, the generated clean data set is imported into a dynamically maintained association mapping network. The association mapping network is a mesh structure persistently stored in memory and a graph database, used to characterize the intrinsic relationships between equipment, materials, procedures, and personnel in the pharmaceutical manufacturing process. The association mapping network exists in the form of nodes and edges; nodes represent entities in pharmaceutical manufacturing, and edges represent known relationships between entities. When a new clean data set is imported, the update engine of the association mapping network begins operation; the update engine identifies the entities involved in the clean data set and the newly created or changed relationships between entities. For example, if the clean data set contains a new record showing that “Material Batch ACTIVE-BATCH-2024-005” is issued to “Tablet Compressor P-301”, the update engine checks in the association mapping network whether the nodes “Material Batch ACTIVE-BATCH-2024-005” and “Tablet Compressor P-301” exist. If they do not exist, the corresponding new nodes are created, and then a “Material-Used-Equipment” relationship edge is created between them. In this way, nodes and edges in the association mapping network are continuously updated or added to reflect the latest configuration and status of the production line in real time. At the same time, the update engine establishes index links from specific data items in the clean data set to corresponding nodes in the association mapping network. For example, it establishes an index link between the data record of "current speed of tablet press P-301" in the clean data set and the "tablet press P-301" equipment node in the association mapping network, so that all relevant real-time and historical data can be quickly traced back through the nodes.

[0084] It is understandable that when imputing data from a dataset marked as incomplete, the inference process can be based on a probabilistic model. In one optional implementation, the missing data elements are calculated using the following formula. The inferred value is based on the formula for historical association patterns stored in the association mapping network:

[0085] ;

[0086] in: The value representing the inferred missing data element. This represents the number of historical association patterns in the association mapping network that are related to the currently missing context. This represents the similarity coefficient between the k-th historical association pattern and the current missing context. The weight representing the k-th historical association pattern. This represents the historical data value corresponding to the k-th historical association pattern. Represents the process parameters of the current batch The adjustment factor. The association mapping network uses this formula to infer the most likely missing data element value from historical association patterns.

[0087] In practice, the dynamic maintenance of the association mapping network is ongoing. Each import of a clean data set may trigger a minor update to the network structure. The association mapping network not only stores static configuration relationships but also records dynamic relationship states through edge attribute updates. For example, the attributes of the edge "material-used-for-equipment" can record information such as the cumulative usage time and consumption of this batch of materials on the current equipment. In some embodiments, the association mapping network supports version management. When process step documents are changed or equipment undergoes major modifications, the association mapping network retains snapshots of the old version's nodes and relationships and links to the new version to support accurate traceability and analysis of historical production data. It can be understood that the index link from the clean data set to the association mapping network nodes is bidirectional, allowing both querying all associated data from network nodes and locating the entity node and associated entities to which any data item belongs in the network from any data item. This bidirectional link forms the basis for subsequent complex relationship analysis and deduction.

[0088] In one embodiment of the present invention, a forward-looking capacity projection process is run based on the current state of the association mapping network and the latest content of the clean data set to obtain projection conclusions about the future production rhythm. This process extracts currently active equipment nodes, available material nodes, and on-duty personnel nodes and their connection relationships from the association mapping network, and extracts recent production efficiency, historical equipment failure intervals, and material consumption rate data from the clean data set to construct a virtual timeline starting from the current time. The active equipment nodes, available material nodes, and on-duty personnel nodes are used as resources. Based on historical efficiency and consumption rates, the task execution and resource consumption process over a future period is simulated. During the simulation, resource conflicts and bottlenecks are determined according to the constraints defined in the association mapping network. The virtual timeline state at the end of the simulation is output as the projection conclusion. The projection conclusion includes the predicted production completion time and the expected remaining resource amount. Combining the simulation results with real-time feedback on resource availability, a preliminary work sequence arrangement is constructed. This process analyzes the simulation results to obtain the suggested task start order and the estimated duration of each task. It also obtains the real-time availability status of equipment from the equipment monitoring system, the real-time inventory quantity of materials from the inventory management system, and the real-time on-duty status of personnel from the personnel management system. Based on the suggested task start order, each task is matched with the specific equipment, materials, and personnel actually available at the current moment. When resource shortages occur during the matching process, the task order is adjusted or marked as waiting for resources according to the task priority rules. All tasks that are successfully matched and have a specific start time determined are arranged in chronological order to form a preliminary work sequence arrangement.

[0089] Based on the current state of the correlation mapping network and the latest content of the clean data set, a forward-looking capacity projection process is run to obtain projection conclusions about the future production rhythm, and a preliminary work sequence arrangement is constructed by combining the projection conclusions with the resource availability status in real time. Taking a tablet production line that includes tableting, coating, and aluminum-plastic packaging processes as an example, the execution of the forward-looking capacity projection process is as follows: After the capacity projection engine is started, it first accesses the dynamically maintained association mapping network and extracts the currently active equipment nodes, available material nodes, and on-duty personnel nodes and their connection relationships from the association mapping network. For example, the extracted nodes include "Tablet Compressor PT-101" (status: running), "Coating Machine CC-201" (status: standby), "Aluminum-Plastic Packaging Machine BL-301" (status: running), "Raw Material Batch API-LOT-0501", "Coating Powder Batch COAT-LOT-0302", "Operator Zhang San", and "Operator Li Si". The connection relationship defined in the association mapping network shows that there is a serial dependency between "Tablet Compressor PT-101" and "Aluminum-Plastic Packaging Machine BL-301" because the uncoated tablets completed in the tableting process are the input material for the packaging process. The capacity simulation engine simultaneously accesses the latest cleanroom dataset, extracting recent production efficiency, historical equipment failure intervals, and material consumption rate data. For example, it extracts the average production speed of the "PT-101 tablet press" as 500 tablets per minute, the average speed of the "BL-301 aluminum-plastic packaging machine" as 40 plates per minute, and the average interval between the last three failures of the "PT-101 tablet press" as 45 hours. Based on this information, the capacity simulation engine constructs a virtual timeline starting from the current time T0, using the extracted active equipment nodes, available material nodes, and on-duty personnel nodes as virtual resources. Based on the historical efficiency and consumption rate data extracted from the cleanroom dataset, it simulates the task execution and resource consumption process over a future period. During the simulation, the capacity simulation engine identifies resource conflicts and bottlenecks based on the constraints defined in the correlation mapping network. For example, the simulation finds that when the accumulated tablets produced by the "PT-101 tablet press" exceed the buffer tank capacity, the feeding of the "BL-301 aluminum-plastic packaging machine" will experience delays, forming a bottleneck. At the end of the simulation, the capacity projection engine outputs the final state of the virtual timeline as the projection conclusion, which includes the predicted production completion time and the expected remaining resources.

[0090] In some embodiments, the process of constructing a preliminary job sequence arrangement is then initiated. The arrangement engine parses the projection conclusions output by the capacity projection process, and obtains the suggested task start order and the estimated duration of each task from the projection conclusions; for example, the parsed suggested order is "start the coating machine CC-201 for preheating (duration: 30 minutes) → tablet press PT-101 continues to produce batch BATCH-X (duration: 7.5 hours) → coating machine CC-201 starts coating (duration: 2 hours)". At the same time, the arrangement engine obtains the real-time equipment availability status fed back by the equipment monitoring system, the real-time material inventory quantity fed back by the inventory management system, and the real-time personnel on-duty status fed back by the personnel management system through the interface; for example, the obtained real-time status shows that "coating machine CC-201" has completed self-check and is ready, and the inventory of "coating powder batch COAT-LOT-0302" is 100 kg, but "operator Li Si" is away from his post due to a temporary task and is expected to return in 1 hour. Based on the task start order suggested by the deduction conclusions, the orchestration engine matches each task with the specific equipment, materials, and personnel actually available at the current moment. When resource shortages occur during the matching process, the orchestration engine adjusts according to preset task priority rules. For example, the "tableting" task has a higher priority than "coating." Therefore, when "Operator Li Si" is temporarily absent, resulting in insufficient personnel for the coating task, the orchestration engine chooses to postpone the "start coating machine CC-201 for preheating" task by 1 hour instead of adjusting the tableting task, and marks this coating task as "waiting for resources." The orchestration engine arranges all successfully matched tasks with determined specific start times in chronological order, forming a structured preliminary job sequence orchestration document.

[0091] It is understandable that when estimating task duration in the capacity projection process, historical baseline efficiency and current equipment status factors can be comprehensively considered. In an optional implementation, the task is calculated using the following formula. Estimated duration :

[0092] ;

[0093] in, Representative task The estimated duration, Representative task The target output to be achieved This represents the historical baseline production efficiency of the equipment corresponding to this task, extracted from the cleanroom dataset. This represents an efficiency factor calculated based on the real-time health status of device nodes. Efficiency factor The calculation will associate the maintenance records of the device node in the mapping network with attributes such as the deviation of the current operating parameters from the baseline parameters.

[0094] In one embodiment of the present invention, the preliminary job sequence arrangement is subjected to compliance verification and performance simulation under multiple constraints. After iterative correction, a recommended job sequence arrangement is formed. This process sets multiple constraints, including production environment conditions required by Good Manufacturing Practices (GMP), continuous operating limits of equipment, maximum working hours of personnel, and the first-in-first-out (FIFO) principle of materials. The preliminary job sequence arrangement is placed in a simulation environment and executed step by step according to its timeline. At each step, it is checked whether any of the multiple constraints are violated. For the steps that violate the constraints, a strategy is selected from a preset adjustment strategy library for correction. These strategies include postponing tasks, changing equipment or personnel, and splitting tasks. The corrected job sequence arrangement is used to re-simulate and verify. This process is repeated until the job sequence arrangement fully meets all the multiple constraints in the simulation. The sequence arrangement that passes the final verification is marked as the recommended job sequence arrangement. During the actual execution of the recommended work sequence arrangement, a continuous compliance deviation monitoring and performance fluctuation tracking mechanism is activated. This mechanism continuously collects actual data related to task execution when the production line actually executes the recommended work sequence arrangement, including the actual start and end times of the task, the actual materials consumed, and the actual environmental data generated. The collected actual data is compared in real time with the planned data and thresholds specified in the multiple constraints in the recommended work sequence arrangement. When the actual data deviates from the planned data beyond the allowable tolerance, or when the actual data approaches or exceeds the compliance threshold, it is recorded as a compliance deviation event. The ratio of the actual execution efficiency to the planned efficiency of the key task is calculated. When this ratio is continuously lower than the set level, it is recorded as a performance fluctuation event, and a detailed log containing compliance deviation events and performance fluctuation events is generated.

[0095] The initial job sequence arrangement undergoes compliance verification and performance simulation under multiple constraints to generate a recommended job sequence arrangement. During the actual execution of the recommended job sequence arrangement, a continuous compliance deviation monitoring and performance fluctuation tracking mechanism is initiated. Taking the operation plan of an aseptic filling production line as an example, the compliance verification and performance simulation process is executed as follows: The verification and simulation engine first sets multiple constraints, which explicitly include the production environment conditions required by Good Manufacturing Practices (GMP), continuous equipment operating time limits, maximum working hours for personnel, and the first-in, first-out (FIFO) principle for materials. The verification and simulation engine places the received initial job sequence arrangement into a discrete event simulation environment. The simulation environment executes the simulation step by step according to the task timeline defined in the initial job sequence arrangement, checking for violations of any multiple constraints at each step of the simulation. For example, when the simulation reaches the 5th hour of the "Product A Filling" task plan, it checks whether the accumulated running time of the "Filling Machine" on the virtual timeline has exceeded the 12-hour continuous operation limit, and whether the simulated reading of the virtual "Level A Laminar Flow Velocity" is still within the range of 0.36-0.54 m / s. For any violations of multiple constraints found in the simulation, the verification and simulation engine selects a strategy from the preset adjustment strategy library for correction. For example, when the simulation finds that the continuous operation of the "Filling Machine" will time out, the strategy "Delay Task" in the adjustment strategy library may be triggered, which means inserting a "Pre-Clean Equipment" task of preset duration for the subsequent "Product B Filling" task. If the simulation finds that the simulated laminar flow velocity value is lower than the lower limit for a certain period, the strategy "Replace Equipment" in the adjustment strategy library may be triggered, which means adjusting the production task planned for "Filling Line 1" for that period to "Filling Line 2" where the wind speed meets the standard. The verification and simulation engine re-performs the entire simulation process and compliance verification using the revised job timing orchestration. This iterative process is repeated until the job timing orchestration fully meets all the set multi-constraints in the simulation. The timing orchestration that passes the final verification is marked as the recommended job timing orchestration and locked for release.

[0096] In some embodiments, the specific thresholds and rules for multiple constraints are stored in a configurable parameter table, facilitating adjustments based on different product process specifications or regulatory updates. See Table 1, Multiple Constraint Parameter Table.

[0097] Table 1: Parameter Table for Multiple Constraints

[0098]

[0099] During the actual execution of the recommended work sequence, a continuous compliance deviation monitoring and performance fluctuation tracking mechanism is activated. This monitoring and tracking mechanism continuously collects actual data related to task execution through sensing terminals as the recommended work sequence is actually executed on the production line. This data includes the actual start and end times of the tasks, the actual amount of materials consumed, and the actual environmental data generated. The monitoring and tracking mechanism compares the collected actual data in real time with the planned data and thresholds specified in the multiple constraints within the recommended work sequence; for example, it compares the actual collected "real-time particle count in the filling area" with the maximum particle count threshold allowed for that time period in the plan, and compares the "actual filling end time" with the "planned filling end time." When actual data deviates from planned data beyond the allowable tolerance, or when actual data approaches or exceeds compliance thresholds, the monitoring and tracking mechanism records it as a compliance deviation event. For example, if the actual filling completion time is delayed by 30 minutes compared to the plan (allowable tolerance is ±10 minutes), a "progress delay" compliance deviation event is recorded. If the real-time particle count reading exceeds the dynamic warning line three times consecutively but does not exceed the standard, an "environmental parameter approaching threshold" compliance deviation event is recorded. The monitoring and tracking mechanism also calculates the ratio of actual execution efficiency to planned efficiency for critical tasks. When the ratio of actual execution efficiency to planned efficiency is consistently lower than the set level, it is recorded as a performance fluctuation event. For example, if the planned efficiency for a "cut-in" task is 300 bottles per minute, but the average actual efficiency over five consecutive minutes is only 270 bottles per minute, and the ratio of 0.9 is consistently lower than the set level of 0.95, a "cut-in efficiency fluctuation" event is recorded. The monitoring and tracking mechanism generates detailed logs containing all compliance deviation events and performance fluctuation event types, timestamps, specific values, and contextual information.

[0100] It is understandable that a comprehensive index can be used for the quantitative assessment of compliance deviations. In one optional implementation, this refers to the instantaneous compliance deviation index for a specific monitoring parameter at time point t. The calculation formula is as follows:

[0101] ;

[0102] in: This represents the immediate compliance deviation index at time point t. The actual measured value at time point t. The planned value representing time point t. The allowable tolerance for this parameter is... The upper limit threshold representing the compliance requirement at time point t. This represents the lower limit threshold for compliance at time point t. When When the set threshold is exceeded, a compliance deviation event will be logged.

[0103] In practical implementation, the simulation environment of the verification and simulation engine incorporates a dynamic model of the production process, capable of simulating the inertia of equipment start-up and shutdown, material delivery delays, and changes in personnel operation speed, making performance simulation more realistic. In some embodiments, the adjustment strategy library is an extensible set of rules, allowing production engineers to add new adjustment strategies and their applicable conditions for newly emerging constraint conflict scenarios. It can be understood that the compliance deviation monitoring and performance fluctuation tracking mechanisms operate in parallel, with detailed logs stored in a time-series database, each event record bearing a precise timestamp and an associated production task identifier.

[0104] See Figure 4 This is a bar chart monitoring task completion time deviation. The data trend shows that the deviation from the actual task completion time consistently exceeds the allowable tolerance (10 minutes), especially after 12:00, indicating a significant increase in deviation. This suggests that the production rhythm deviates from the plan during this period, requiring attention to process bottlenecks or resource conflicts. This type of chart is typically used for compliance monitoring of production processes. By comparing actual and planned progress, it promptly identifies schedule deviation risks and assists in adjusting subsequent production strategies. It visually presents the degree of deviation between actual progress and the plan, helping managers quickly identify "schedule delay" risks and prevent delays from escalating and impacting overall capacity. The time trend of the deviation can pinpoint concentrated periods of schedule delays, allowing for reverse tracing of the status of equipment, personnel, materials, and other resources during those periods, providing data support for optimizing production processes and improving efficiency.

[0105] In one embodiment of the present invention, an instant correction logic is triggered based on monitoring and tracking results to send a coordination instruction to the production line control system. This logic receives detailed logs generated by the compliance deviation monitoring and performance fluctuation tracking mechanism, analyzes the event types, severity, and root causes in the detailed logs, and determines whether intervention is needed for the currently executing recommended work sequence arrangement according to preset instant correction rules. The instant correction rules specify the intervention levels corresponding to different event types and severity. When intervention is determined to be necessary, a local timing adjustment plan or resource reallocation plan for the affected tasks is quickly generated based on the association mapping network and the current clean data set. The local timing adjustment plan or resource reallocation plan is converted into specific equipment start / stop, material flow, or personnel allocation instructions and sent to the production line control system as a coordination instruction.

[0106] Based on the monitoring and tracking results generated by the compliance deviation monitoring and performance fluctuation tracking mechanism, the immediate correction logic is triggered and a coordination instruction is sent to the production line control system. Taking the operation of a freeze-dried powder injection production line as an example, the triggering and execution of the immediate correction logic are as follows: The immediate correction logic module continuously receives detailed logs generated by the compliance deviation monitoring and performance fluctuation tracking mechanism. The detailed logs record events such as "filling speed decrease performance fluctuation event" and "aseptic isolator pressure difference abnormal compliance deviation event". The analysis unit of the immediate correction logic module analyzes the event type, severity, and root cause in the detailed logs; for example, the analysis unit analyzes the event type of "filling speed decrease performance fluctuation event" as "equipment performance fluctuation", the severity is rated as "medium" based on the percentage and duration of continuous performance below the planned efficiency, and the root cause is initially associated with possible minor blockage in the filling needle. Based on preset real-time correction rules, it is determined whether intervention is needed in the currently executing recommended job sequence arrangement. These rules clearly define the intervention levels corresponding to different event types and severity levels. For example, the rules stipulate that "medium" or higher severity "equipment performance fluctuations" require a "scheduling intervention" level response, while "low" severity "instantaneous environmental parameter fluctuations" may only trigger an "alarm recording" level response. When it is determined that "scheduling intervention" is necessary, the planning unit of the real-time correction logic module is activated.

[0107] In some embodiments, the planning unit quickly generates local timing adjustment schemes or resource reallocation schemes for affected tasks based on a dynamic association mapping network and the latest cleanroom data set. The planning unit first accesses the dynamic association mapping network to query entity nodes related to the affected tasks and their connections; for example, for the "vial filling" task affected by a decrease in filling speed, the planning unit queries nodes such as "filling machine F-01," "spare filling needle assembly," and "operator Zhao," as well as the serial dependency relationship between "filling machine F-01" and "capping machine C-01." Simultaneously, the planning unit obtains the latest equipment status, material inventory, and personnel location information from the current cleanroom data set. Based on this real-time information, the planning unit quickly generates a local adjustment scheme, such as pausing the "vial filling" task for 15 minutes to execute an online needle cleaning procedure; simultaneously, postponing the subsequent "semi-capping" and "freeze dryer" tasks, originally scheduled to follow filling, by 15 minutes, and notifying "operator Zhao" to prepare for cleaning operations. This solution only adjusts the local timing and resource allocation of the affected task chains, without changing the plans for other unrelated tasks.

[0108] It is understandable that the process of determining the intervention level in the immediate correction rules can be quantified. In one optional implementation, the comprehensive intervention urgency index for a given event is calculated using the following formula. :

[0109] ;

[0110] in: The comprehensive intervention urgency index representing the event, A quantitative score representing the severity of the event (e.g., low = 1, medium = 2, high = 3). Represents the duration of the event (in minutes). The critical path priority weight represents the impact of the event on the task. These are the preset coefficients for each component. When the comprehensive intervention urgency index... Exceeding the preset intervention threshold When it is determined that proactive intervention is necessary.

[0111] In practice, once a local timing adjustment plan or resource reallocation plan is generated, the instruction conversion unit transforms it into specific instructions that the production line control system can recognize and execute. For example, the instruction conversion unit converts "pause the filling task for 15 minutes" into a "pause" instruction sent to the programmable logic controller of the filling machine F-01; "execute the online needle cleaning procedure" into a sequence of instructions that triggers the cleaning workstation to start a specific cleaning cycle; and "postpone the entire task for 15 minutes" into a plan timestamp update instruction sent to the scheduling module of the manufacturing execution system. These specific equipment start / stop, material flow, or personnel allocation instructions are encapsulated into coordination instruction packages and sent to the production line control system in real time via standard industrial communication protocols. The production line control system receives and executes these coordination instructions, thereby keeping the actual operation of the production line synchronized with the decisions made by the real-time correction logic.

[0112] See Figure 5 This is a stacked bar chart summarizing the statistics of production anomalies. Events involving equipment performance fluctuations and abnormal environmental parameters are the most frequent, reflecting that equipment status and environmental stability are the main risks in production. Low-level events generally account for a high proportion, while high-level events only occur in small numbers in equipment failures and equipment performance fluctuations, indicating that the overall risk is mainly mild anomalies. Equipment-related events (performance fluctuations, failures) contain both high quantity and high-level risk, requiring close attention to equipment maintenance and monitoring. This type of chart is commonly used in production risk management scenarios to identify high-risk event types and assist in resource allocation; quantify the proportion of anomalies at different levels to support risk level classification and response strategy formulation; and track changes in anomaly trends to assess the effectiveness of improvements in production stability.

[0113] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0114] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent capacity management of a pharmaceutical production line, characterized in that, The method includes: By deploying multiple types of sensing terminals at different physical locations on the production line, the raw signal streams reflecting the actual operation are captured synchronously and initially collected according to a predefined drug production data classification system to form a set of source data with category labels. The built-in data quality assessment routine is invoked to perform a credibility score and integrity mark on each data item in the source data set with category labels, and then a filtering and reconstruction operation is performed to generate a clean data set. The cleanroom dataset is imported into a dynamically maintained association mapping network. This network characterizes the intrinsic relationships between equipment, materials, procedures, and personnel in the pharmaceutical manufacturing process. The network exists in the form of nodes and edges. Nodes represent entities in pharmaceutical production, including specific equipment units, material batches, process step documents, and qualified personnel. Edges represent known relationships between entities. Based on the current state of the association mapping network and the latest content of the cleanroom dataset, a forward-looking capacity projection process is run to obtain projection conclusions about future production schedules, including: Extract currently active device nodes, available material nodes, and on-duty personnel nodes and their connection relationships from the association mapping network; Extract recent production efficiency, equipment failure interval history, and material consumption rate data from the clean data set; Construct a virtual timeline starting from the current time, using the active device nodes, available material nodes, and on-duty personnel nodes as resources, and simulate the task execution and resource consumption process over a future period based on historical efficiency and consumption rate; During the simulation, resource conflicts and bottlenecks are identified based on the constraints defined in the associated mapping network. The virtual timeline state at the end of the simulation is output as the deduction conclusion, which includes the predicted production completion time and the expected remaining resources. Based on the aforementioned projections of future production rhythms and the real-time feedback on resource availability, a preliminary work sequence arrangement is constructed. The preliminary job sequence arrangement is subjected to compliance verification and performance simulation under multiple constraints, and a recommended job sequence arrangement is formed after iterative correction. During the actual execution of the recommended work sequence arrangement, a continuous compliance deviation monitoring and performance fluctuation tracking mechanism is initiated, and an immediate correction logic is triggered based on the monitoring and tracking results to send coordination instructions to the production line control system.

2. The intelligent capacity management method for pharmaceutical production lines as described in claim 1, characterized in that, The method of synchronously capturing raw signal streams reflecting the real-time operation by deploying multiple types of sensing terminals at different physical locations on the production line specifically includes: Read raw signals of equipment status from vibration, temperature, and current sensors on the production equipment; Read the original signals of production results and consumption from the batch recording terminal, material weighing system and quality inspection port; Raw signals of temperature, humidity, and particle count are read from environmental monitoring probes; Read personnel presence and operation records from access control, attendance, and electronic batch record systems; All raw signals are packaged and timestamped according to a unified time base to form a raw signal stream.

3. The intelligent capacity management method for pharmaceutical production lines as described in claim 1, characterized in that, The invocation of the built-in data quality assessment routine performs credibility scoring and integrity marking on each data item in the source dataset with category labels, including: Define normal numerical ranges and reasonable logical relationship rules for each type of data; Each data item in the source dataset is compared with its corresponding normal value range, and data that exceeds the range is assigned a lower confidence score. Check whether the logical relationship between related data items in the source data set conforms to the reasonable logical relationship rules, and assign an integrity missing mark to data sets that do not conform to the rules; According to the preset data credibility threshold, data items with credibility scores lower than the data credibility threshold are marked as data to be verified; For a dataset marked as having incomplete data, record the missing data elements and their context information.

4. The intelligent capacity management method for pharmaceutical production lines as described in claim 1, characterized in that, The process of performing filtering and reconstruction operations to generate a clean dataset includes: Remove data items from the source data set that are marked as pending verification and cannot be corrected by auxiliary information sources; For data sets marked as lacking integrity, the most likely missing data elements are inferred and added based on the historical association patterns stored in the association mapping network; All retained and supplemented data are converted according to a unified unit of measurement, time format, and naming convention. The data, after being removed, added, and transformed, is reorganized according to time sequence and category to form a clean data set that can be directly accessed in subsequent processes.

5. The intelligent capacity management method for pharmaceutical production lines as described in claim 1, characterized in that, The step of importing the clean data set into a dynamically maintained association mapping network includes: When a new clean data set is imported, identify the entities involved and the new or changed relationships between entities. Update or add corresponding nodes and edges in the associated mapping network to reflect the latest configuration and status of the production line; Establish index links from specific data items in the clean data set to corresponding nodes in the association mapping network.

6. The intelligent capacity management method for pharmaceutical production lines as described in claim 1, characterized in that, The preliminary job sequence arrangement is constructed by combining the projected conclusions about future production rhythm with the real-time feedback on resource availability, including: Analyze the deduction conclusions to obtain the suggested task initiation order and the estimated duration of each task; Get real-time equipment availability status from the equipment monitoring system, real-time material inventory from the inventory management system, and real-time personnel on-duty status from the personnel management system. Based on the suggested task initiation sequence, each task is matched with the specific equipment, materials, and personnel actually available at the current moment; When resource shortage occurs during the matching process, the task order is adjusted or the task is marked as waiting for resources according to the task priority rules. All tasks that have been successfully matched and have a specific start time are arranged in chronological order to form a preliminary job sequence arrangement.

7. The intelligent capacity management method for pharmaceutical production lines as described in claim 1, characterized in that, The process of performing compliance checks and performance simulations under multiple constraints on the initial job timing arrangement, and iteratively refining it to form a recommended job timing arrangement, includes: Multiple constraints are set, including production environment conditions required by Good Manufacturing Practice (GMP) for pharmaceuticals, continuous operating limits of equipment, maximum working hours of personnel, and the first-in-first-out (FIFO) principle for materials. The initial job timing arrangement is placed in a simulation environment and executed step by step according to its timeline, with each step checking whether any of the multiple constraints are violated. For any step that violates the constraints, a strategy is selected from a pre-set adjustment strategy library for correction. The strategies include postponing the task, changing equipment or personnel, or splitting the task. Re-simulate and verify using the revised job timing arrangement, repeating this process until the job timing arrangement fully satisfies all multiple constraints in the simulation. The timing arrangements that pass the final verification will be marked as recommended job timing arrangements.

8. The intelligent capacity management method for pharmaceutical production lines as described in claim 1, characterized in that, During the actual execution of the recommended job sequence arrangement, a continuous compliance deviation monitoring and performance fluctuation tracking mechanism is initiated, including: When the recommended work sequence is actually executed on the production line, real data related to task execution is continuously collected, including the actual start and end times of the task, the actual materials consumed, and the actual environmental data generated. The collected actual data is compared in real time with the pre-defined planned data in the recommended job timing arrangement and the thresholds specified in the multiple constraints. When actual data is found to deviate from planned data by more than the allowable tolerance, or when actual data approaches or exceeds the compliance threshold, it is recorded as a compliance deviation event. Calculate the ratio of the actual execution efficiency to the planned efficiency of the critical task. When the ratio is consistently lower than a set level, it is recorded as a performance fluctuation event. Generate detailed logs containing the aforementioned compliance deviation events and performance fluctuation events.

9. The intelligent capacity management method for pharmaceutical production lines as described in claim 1, characterized in that, The real-time correction logic triggered based on monitoring and tracking results sends coordination instructions to the production line control system, including: Receive detailed logs generated by the compliance deviation monitoring and performance fluctuation tracking mechanism; Analyze the event types, severity, and root causes in the detailed logs; Based on preset real-time correction rules, it is determined whether intervention is needed on the currently executing recommended job sequence arrangement. The real-time correction rules specify the intervention levels corresponding to different event types and severity. When intervention is deemed necessary, a local time-series adjustment plan or resource reallocation plan for the affected task is quickly generated based on the aforementioned correlation mapping network and the current clean data set. The local timing adjustment scheme or resource reallocation scheme is converted into specific equipment start-up and shutdown, material flow or personnel allocation instructions, and sent to the production line control system as coordination instructions.

Images