AI-based intelligent scheduling and coordination methods and systems for multi-reactor systems

Abstract

This invention relates to the field of intelligent scheduling technology, and more particularly to a multi-reactor AI intelligent scheduling and coordination method and system. The method acquires reactor state data and task sets, identifies conflicts, traces the root cause task chain backward to establish a cross-mapping between the task dependency tree and the state trajectory, and generates constraint propagation rules. Based on these rules, tasks and reactors are associated to obtain an initial mapping, and constraint relaxation is calculated to construct scheduling instructions containing a state fluctuation tolerance interval. During allocation, deviations within the tolerance interval do not trigger rescheduling, and the constraint rules are adaptively corrected using real-time deviations. This invention achieves efficient coordination and dynamic fault tolerance among multiple reactors, significantly improving scheduling stability and resource utilization.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology, and in particular to a multi-reactor AI intelligent scheduling and coordination method and system. Background Technology

[0002] In the field of multi-reactor collaborative scheduling, such as in process industries like chemical, pharmaceutical, and bio-fermentation, multiple reactors often share limited resources (e.g., utilities, raw materials, material pipelines). Existing methods have significant shortcomings. When multiple reactors simultaneously request the same resource, conflicts are resolved through post-event negotiation or priority allocation, lacking a systematic diagnosis of the root causes of the conflict. Conflicts may originate from delays in upstream tasks or the cascading effects of resource allocation, and conventional approaches struggle to trace the complete task chain leading to the conflict, failing to identify the propagation path. This makes scheduling systems prone to getting trapped in local optima when dealing with complex constraints, and can even lead to global scheduling failure due to unforeseen cascading conflicts.

[0003] The existing scheduling requirements stipulate that each reactor must strictly adhere to preset time and state parameters, making it overly sensitive to minor fluctuations in actual operation (such as temperature deviations or extended reaction times). Once an unacceptable deviation occurs, the system often needs to trigger a full rescheduling, which not only consumes significant computational resources but may also lead to production interruptions. Frequent rescheduling increases system instability and reduces overall operational efficiency. Summary of the Invention

[0004] This invention provides a multi-reactor AI intelligent scheduling and coordination method and system, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a multi-reactor AI intelligent scheduling and coordination method, comprising:

[0006] The system acquires current operating status data and a set of tasks to be executed from multiple reactors, identifies resource sharing conflicts among reactors, traces the root cause task chain leading to the resource sharing conflict in reverse, establishes a cross-mapping relationship between the task dependency tree and the reactor state change trajectory, and dynamically generates constraint propagation rules for predicting potential cascading conflicts based on the cross-mapping relationship.

[0007] Based on the constraint propagation rules, each task in the set of tasks to be executed is associated with multiple reactors to obtain an initial mapping relationship;

[0008] The relaxation degree of each constraint is calculated based on the constraint propagation rule. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value. A scheduling elasticity space is constructed for each reactor in the initial mapping relationship based on the relaxation degree, and a scheduling instruction containing the state fluctuation tolerance interval is generated.

[0009] Based on the scheduling instructions, task allocation operations are performed on the multiple reactors. When the actual operating state of the reactor deviates from the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data.

[0010] Acquire current operating status data and a set of tasks to be executed from multiple reactors, identify reactor resource sharing conflicts, and establish a cross-mapping relationship between a task dependency tree and the reactor state change trajectory by tracing the root cause task chain leading to the resource sharing conflict.

[0011] For each task in the set of tasks to be executed, extract the type of demand for reactor resources and the time period of occupation; parse the current operating status data into the resource availability time window of each reactor;

[0012] By comparing the task resource requirement type with the resource availability time window, task combinations with overlapping time windows and the same resource type are identified and marked as potential resource sharing conflict pairs.

[0013] For each task in the potential resource sharing conflict pair, backtrack upwards layer by layer along its set of preceding dependent tasks to the starting task node, and record the ordered sequence of all tasks on the backtracking path as the root cause task chain; construct the task dependency tree according to the dependency relationship of each task in the root cause task chain.

[0014] For each task in the root cause task chain, extract the sequence of state changes of the relevant reactors during the execution period of the task from the current running status data to form a reactor state change trajectory;

[0015] By establishing an association mapping between each task node in the task dependency tree and its corresponding reactor state change trajectory, a cross-mapping relationship is obtained.

[0016] Based on the aforementioned cross-mapping relationship, constraint propagation rules for predicting potential cascading conflicts are dynamically generated, including:

[0017] Extract the strength of the dependency relationship between each task node in the task dependency tree and the state transition pattern of each reactor in the reactor state change trajectory from the cross mapping relationship;

[0018] For each dependency path in the task dependency tree, forward propagation analysis is performed to analyze the chain reaction of changes in the execution time and resource requirements of subsequent task nodes when the starting task of the path is delayed or resource consumption increases. Task nodes whose chain reaction exceeds a preset impact threshold are marked as potential cascading conflict node groups.

[0019] For each task node in the potential cascading conflict node group, combined with the reactor state transition mode associated with the task node, the condition boundary that causes the associated reactor to fail to complete the state transition as expected when the execution period of the task is offset is derived, and the condition boundary is transformed into a constraint expression.

[0020] The constraint expressions corresponding to all task nodes in the task dependency tree are combined with the dependency strength to construct a constraint propagation rule.

[0021] Based on the constraint propagation rules, each task in the set of tasks to be executed is associated with multiple reactors to obtain an initial mapping relationship, including:

[0022] Extract the constraint expressions corresponding to each task node from the constraint propagation rules;

[0023] For each task in the set of tasks to be executed, a set of candidate reactors that meet the constraints is selected based on the constraint expression corresponding to the task.

[0024] For each reactor in the candidate reactor set, the constraint propagation impact value on subsequent task nodes in the task dependency tree is calculated based on the dependency strength in the constraint propagation rule when the task is assigned to that reactor. The constraint propagation impact value characterizes the degree to which the range of selectable reactors for subsequent tasks is reduced after the assignment decision is propagated to subsequent tasks through the dependency path.

[0025] The reactor with the smallest constraint propagation impact value is selected as the target reactor for the task, and the task is associated with the target reactor. All tasks in the set of tasks to be executed are traversed to complete the association between the task and the reactor, and the initial mapping relationship is obtained.

[0026] Based on the constraint propagation rules, the relaxation degree of each constraint is calculated. A scheduling elasticity space is constructed for each reactor in the initial mapping relationship based on the relaxation degree, and scheduling instructions containing state fluctuation tolerance intervals are generated, including:

[0027] For each reactor in the initial mapping relationship, extract the relaxation degree of the constraints corresponding to each task assigned to that reactor. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value.

[0028] Based on the relaxation, three threshold levels are set to divide the scheduling elasticity space into a core elasticity region, an extended elasticity region, and a limit elasticity region. The core elasticity region corresponds to the state fluctuation range where the relaxation is greater than the first threshold, the extended elasticity region corresponds to the state fluctuation range where the relaxation is between the first threshold and the second threshold, and the limit elasticity region corresponds to the state fluctuation range where the relaxation is between the second threshold and zero.

[0029] For each task assigned to the reactor, based on the difference in the probability of triggering cascading conflicts when the state deviates positively or negatively according to the corresponding constraints of the task, different widths of tolerance interval boundaries are set for the positive deviation direction and the negative deviation direction to form an asymmetric tolerance interval.

[0030] The scheduling instructions are generated by combining the hierarchical region boundary of the scheduling elastic space with the asymmetric tolerance interval boundary. The scheduling instructions specify the response strategy for the reactor state fluctuation in different regions.

[0031] Forming an asymmetric tolerance interval includes:

[0032] Extract all dependency paths of the task in the task dependency tree from the constraint propagation rules, and count the path length and number of nodes of each dependency path;

[0033] For the positive and negative deviations of the task, the constraint tightening amount propagated to subsequent task nodes along each dependency path is simulated when the task state deviates by a unit. Dependency paths that cause the constraint tightening amount of subsequent task nodes to exceed the trigger threshold are marked as trigger paths. The cascading conflict trigger probability is calculated based on the proportion of the number of trigger paths to the total number of dependency paths, the average path length corresponding to the trigger path, and the average number of nodes.

[0034] Calculate the ratio of the probability of cascading conflict triggering in a positive deviation to the probability of cascading conflict triggering in a negative deviation. Based on the ratio, assign tolerance interval width weights to the positive and negative deviation directions. The larger the ratio, the smaller the tolerance interval width weight assigned to the deviation direction.

[0035] Based on the relaxation degree and the tolerance interval width weight, the tolerance interval boundary in the positive deviation direction and the tolerance interval boundary in the negative deviation direction are calculated to form an asymmetric tolerance interval.

[0036] When the actual operating state of the reactor deviates from the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data, including:

[0037] Based on the scheduling instructions, task allocation operations are performed on the multiple reactors; actual operating status data of each reactor during the task allocation operation are collected, and the deviation between the actual operating status data and the expected operating status in the scheduling instructions is calculated;

[0038] The deviation value is compared with the boundary of the state fluctuation tolerance range of the corresponding reactor. When the deviation value is within the tolerance range, it is marked as an absorbable deviation. When the deviation value exceeds the tolerance range, it is marked as a trigger rescheduling deviation.

[0039] For reactors marked as triggering rescheduling deviations, the task corresponding to the reactor is unbound from the current reactor. Based on the updated reactor operating status data and constraint propagation rules, the candidate reactor set and constraint propagation impact value of the task are recalculated, and the task is reassigned to the reactor with the smallest constraint propagation impact value.

[0040] For reactors marked as absorbable deviations, record the position of the corresponding task in the task dependency tree and the actual impact of the deviation value on subsequent task nodes after it is absorbed by the elastic space.

[0041] The distribution characteristics of absorbable deviations and the propagation attenuation law of actual impact in multiple task allocation operations are statistically analyzed. Based on the distribution characteristics and propagation attenuation law, the calculation parameters of dependency strength and constraint tightening in the constraint propagation rules are corrected.

[0042] A second aspect of the present invention provides a multi-reactor AI intelligent scheduling and coordination system, comprising:

[0043] The conflict identification unit is used to acquire the current operating status data and the set of tasks to be executed of multiple reactors, identify reactor resource sharing conflicts, trace the root cause task chain that leads to the resource sharing conflict in reverse, establish a cross-mapping relationship between the task dependency tree and the reactor state change trajectory, and dynamically generate constraint propagation rules for predicting potential cascading conflicts based on the cross-mapping relationship.

[0044] The task mapping unit is used to associate each task in the set of tasks to be executed with multiple reactors according to the constraint propagation rules to obtain an initial mapping relationship;

[0045] The elastic scheduling unit is used to calculate the relaxation degree of each constraint based on the constraint propagation rule. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value. Based on the relaxation degree, a scheduling elastic space is constructed for each reactor in the initial mapping relationship, and a scheduling instruction containing the state fluctuation tolerance range is generated.

[0046] An adaptive correction unit is used to perform task allocation operations on the multiple reactors based on the scheduling instructions. When the actual operating state of the reactor deviates within the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data.

[0047] A third aspect of the present invention provides an electronic device, comprising:

[0048] processor;

[0049] Memory used to store processor-executable instructions;

[0050] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0051] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0052] This invention traces the root cause task chain leading to resource-sharing conflicts in reverse, establishes a cross-mapping relationship between the task dependency tree and the reactor state change trajectory, and generates constraint propagation rules to predict potential cascading conflicts. This deep tracing and prediction mechanism fundamentally suppresses the chain reaction amplification effect of conflicts, transforming multi-reactor scheduling from a passive response to an active defense, significantly reducing unplanned downtime and material losses caused by conflicts, and greatly improving the stability and reliability of system operation.

[0053] A reactor scheduling flexibility space is constructed based on the relaxation of each constraint. This space clearly defines the tolerance range for state fluctuations, and deviations within this range do not require rescheduling. This design significantly reduces the frequency of scheduling calculations, lowers system energy consumption and communication overhead, while ensuring continuous and stable operation of the production process under disturbances. The size of the flexibility space can be dynamically adjusted according to actual needs, further enhancing the flexibility and adaptability of scheduling, enabling the reactor to utilize resources and capacity more efficiently.

[0054] By adaptively correcting constraint propagation rules using real-time performance deviation data, this closed-loop evolutionary capability ensures that the scheduling strategy maintains high performance throughout long-term operation, effectively addressing dynamic factors such as equipment performance degradation and changes in raw material composition, and reducing the workload of manual monitoring and maintenance. Combined with these measures, multi-reactor collaborative scheduling significantly improves task completion rate, resource utilization, response speed, and robustness. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the AI-powered intelligent scheduling and coordination method for multiple reactors. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0057] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0058] Figure 1 This is a flowchart illustrating the multi-reactor AI intelligent scheduling and coordination method according to an embodiment of the present invention.

[0059] Multi-reactor AI intelligent scheduling and coordination methods include:

[0060] The system acquires current operating status data and a set of tasks to be executed from multiple reactors, identifies resource sharing conflicts among reactors, traces the root cause task chain leading to the resource sharing conflict in reverse, establishes a cross-mapping relationship between the task dependency tree and the reactor state change trajectory, and dynamically generates constraint propagation rules for predicting potential cascading conflicts based on the cross-mapping relationship.

[0061] Based on the constraint propagation rules, each task in the set of tasks to be executed is associated with multiple reactors to obtain an initial mapping relationship;

[0062] The relaxation degree of each constraint is calculated based on the constraint propagation rule. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value. A scheduling elasticity space is constructed for each reactor in the initial mapping relationship based on the relaxation degree, and a scheduling instruction containing the state fluctuation tolerance interval is generated.

[0063] Based on the scheduling instructions, task allocation operations are performed on the multiple reactors. When the actual operating state of the reactor deviates from the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data.

[0064] Acquire current operating status data and a set of tasks to be executed from multiple reactors, identify reactor resource sharing conflicts, and establish a cross-mapping relationship between a task dependency tree and the reactor state change trajectory by tracing the root cause task chain leading to the resource sharing conflict.

[0065] For each task in the set of tasks to be executed, extract the type of demand for reactor resources and the time period of occupation; parse the current operating status data into the resource availability time window of each reactor;

[0066] By comparing the task resource requirement type with the resource availability time window, task combinations with overlapping time windows and the same resource type are identified and marked as potential resource sharing conflict pairs.

[0067] For each task in the potential resource sharing conflict pair, backtrack upwards layer by layer along its set of preceding dependent tasks to the starting task node, and record the ordered sequence of all tasks on the backtracking path as the root cause task chain; construct the task dependency tree according to the dependency relationship of each task in the root cause task chain.

[0068] For each task in the root cause task chain, extract the sequence of state changes of the relevant reactors during the execution period of the task from the current running status data to form a reactor state change trajectory;

[0069] By establishing an association mapping between each task node in the task dependency tree and its corresponding reactor state change trajectory, a cross-mapping relationship is obtained.

[0070] For example, in real-world production scenarios with multiple reactors, several chemical reactors typically need to share limited resources such as material transport pipelines, heating steam systems, cooling water circulation devices, and analytical testing equipment. When multiple tasks simultaneously request the use of the same resources, it is essential to accurately identify these resource-sharing conflicts and trace their root causes in order to develop effective scheduling strategies.

[0071] For each task in the set of tasks to be executed, it is first necessary to extract its specific requirements for reactor resources from the task description information. These requirements include, but are not limited to, material conveying capacity, heating power range, cooling water flow rate, type of stirring equipment, pressure control system, and online analytical instruments. Simultaneously, the time period during which each task occupies specific resources is recorded. This time period typically includes the planned start time and expected end time of the task, forming a time interval representation. For example, if a polymerization reaction task requires the use of a high-pressure steam heating system from 8:00 to 11:30, and another oxidation reaction task requires the use of the same steam system from 9:15 to 12:00, then the two overlap in time.

[0072] Deeply analyze the current operational status data to extract the availability time windows for each reactor and its associated public resources. This status data includes the reactor's current operating mode (idle, running, maintenance, cleaning, etc.), remaining available time, reserved resource usage periods, and equipment health status assessment results. By analyzing the real-time data stream of the reactor control system, precise information can be obtained about when a reactor is running from 7:00 to 10:00, requires cleaning from 10:00 to 10:45, and is available for new tasks after 10:45. For shared resources such as the cooling water circulation system, it is necessary to calculate the currently occupied flow share and the remaining allocable capacity, and convert this information into availability windows in the time dimension to determine the time periods during which the resource can respond to new usage requests.

[0073] After extracting task requirements and constructing resource availability windows, a fine-grained comparison operation is performed. For each task's resource requirement type, entries with the same resource type are retrieved from the availability time windows of all reactors and public resources. It is determined whether there is a temporal overlap between the task request's occupied period and the resource availability time window; the overlap determination method is to check whether the two time intervals intersect. When task A needs to... A steam heating system is used during certain periods, and this steam system is in... The time slot is already occupied by other tasks; if it is the planned start time for task A... and If the conditions are met, it is determined that there is a time window overlap. Further checks are made to ensure that the resource types are completely identical; for example, both tasks must use a steam heating device with a rated power of 500 kilowatts, rather than devices of different power ratings. Task pairs that meet both the conditions of overlapping time windows and identical resource types are marked as potential resource-sharing conflict pairs, and the specific resource name, conflict time period, and involved task identifiers are recorded.

[0074] For identified potential resource-sharing conflicts, it is necessary to trace the root cause of each conflicting task. Tasks in chemical production are often not isolated but have strict dependencies on other tasks. For example, a distillation task can only start after a reaction task is completed and produces products, while the reaction task requires the pretreatment of raw materials to be completed first. These dependencies form complex task chains, and a resource conflict in a downstream task may stem from an abnormal delay or advancement in the execution time of an upstream task.

[0075] When performing the reverse tracing operation, a task is selected from the conflict pair as the starting point, and the set of its preceding dependent tasks is queried. Preceding dependent tasks refer to all tasks that must be completed before the current task begins; these dependencies are usually explicitly defined in the production process. For example, if task E's preceding dependent set includes tasks C and D, it means that task E must wait for both C and D to complete before it can start. For each task in the preceding dependent set, its respective preceding dependent tasks are traced upwards, forming a recursive backtracking process. This backtracking operation continues until the starting task node with no preceding dependent tasks is reached—the source task of the entire production process. All tasks encountered along the backtracking path are arranged in the execution order from the starting node to the current conflicting task, forming an ordered task sequence. This sequence completely records the entire task chain from the source of production to the point of resource conflict, and is defined as the root cause task chain.

[0076] The root cause task chain reveals the complete causal path leading to resource-sharing conflicts, but to more clearly express the dependency structure between tasks, the chain structure needs to be transformed into a tree structure. Each task in the root cause task chain is abstracted as a tree node, with node attributes including task identifier, resource requirements, and planned execution time. Parent-child node connections are established based on the dependencies between tasks: if task B depends on task A, then A is set as the parent node of B, and B becomes a child node of A. The starting task node serves as the root node of the tree, and the task that ultimately causes the conflict serves as a leaf node or intermediate node. For complex cases with multiple dependency paths, the constructed task dependency tree may exhibit a multi-branch structure, with each branch representing a task execution path that may affect the final conflicting task.

[0077] While constructing the task dependency tree, it is necessary to extract and record the state change information of relevant reactors during the execution of each task. From the current operating status data source, historical operating data of all reactors involved in the planned execution period of the task are filtered out. This data includes multi-dimensional state parameters such as reactor temperature curves, pressure change trends, material flow fluctuations, stirring speed adjustment records, and control valve opening changes. These discrete state data points are organized in chronological order to form a continuous trajectory reflecting the evolution of the reactor state over time. For example, during the execution of a heating reaction task, the reactor temperature gradually increases from room temperature (25 degrees Celsius) to 180 degrees Celsius, experiencing a heating phase, a isothermal holding phase, and a cooling phase. The rate of temperature change and duration of each phase constitute the temperature state change trajectory of the reactor during this task. Similarly, other state parameters such as pressure and flow rate also form independent change trajectories. Multiple trajectories together depict the comprehensive state evolution of the reactor during task execution.

[0078] This approach establishes a mapping between each task node in the task dependency tree and its corresponding reactor state change trajectory, achieving information fusion between the task logic level and the physical equipment state level. Specifically, a state trajectory reference field is added to the task node's data structure, storing indexes or pointers to all relevant reactor state change trajectories during the task's execution. This association allows analysis of a task node to not only obtain its logical attributes such as task type, duration, and resource requirements, but also to directly access the actual state evolution of the reactor during task execution. For example, the state trajectory associated with task node "polymerization reaction A" shows that reactor R3 experienced temperature fluctuations exceeding the normal range during the task's execution. This abnormal state information could be a key factor causing subsequent task delays and resource conflicts.

[0079] By establishing a cross-mapping relationship between the task dependency tree and the reactor state change trajectory, a comprehensive information model is formed that tightly integrates task logical dependencies with equipment physical states. This mapping relationship supports bidirectional queries: it can trace the actual state performance of the reactor during the execution of a task node, and it can also reverse the process from an abnormal reactor state to locate potentially affected task nodes. This cross-mapping provides a solid data foundation for the subsequent dynamic generation of constraint propagation rules, enabling the system to predict which task combinations may trigger cascading resource conflicts based on the task dependency structure and equipment state evolution history. This allows the system to pre-set corresponding constraints during scheduling decisions, preventing conflicts from manifesting in actual execution.

[0080] Based on the aforementioned cross-mapping relationship, constraint propagation rules for predicting potential cascading conflicts are dynamically generated, including:

[0081] Extract the strength of the dependency relationship between each task node in the task dependency tree and the state transition pattern of each reactor in the reactor state change trajectory from the cross mapping relationship;

[0082] For each dependency path in the task dependency tree, forward propagation analysis is performed to analyze the chain reaction of changes in the execution time and resource requirements of subsequent task nodes when the starting task of the path is delayed or resource consumption increases. Task nodes whose chain reaction exceeds a preset impact threshold are marked as potential cascading conflict node groups.

[0083] For each task node in the potential cascading conflict node group, combined with the reactor state transition mode associated with the task node, the condition boundary that causes the associated reactor to fail to complete the state transition as expected when the execution period of the task is offset is derived, and the condition boundary is transformed into a constraint expression.

[0084] The constraint expressions corresponding to all task nodes in the task dependency tree are combined with the dependency strength to construct a constraint propagation rule.

[0085] For example, in a chemical production system, there are complex task dependencies and resource competition relationships among multiple reactors. When a task is delayed or its resource consumption increases, it often triggers a chain reaction, causing multiple subsequent tasks to fail to execute as planned. To effectively predict and constrain such cascading conflicts, it is necessary to extract key features from the cross-mapping relationships and construct constraint propagation rules through forward propagation analysis.

[0086] The dependency strength between task nodes in the task dependency tree is extracted from the cross-mapping relationship. The task dependency tree records the sequential dependencies between tasks; for example, task B can only start after task A is completed, and task C requires intermediate products from task B as raw materials. The dependency strength reflects the degree of constraint of the upstream task on the downstream task, represented by a value between 0 and 1. The specific quantification method is as follows: if task B requires all the products of task A to start, the dependency strength is set to 0.9; if task B only needs 50% of the products of task A to start, the dependency strength is set to 0.5; if task B can obtain alternative raw materials from other sources and does not completely depend on task A, the dependency strength is set to 0.3. The larger the dependency strength value, the tighter the dependency, and the more sensitive the subsequent task is to the time offset and resource changes of the preceding task.

[0087] Simultaneously, the state transition patterns of each reactor in the reactor state change trajectory are extracted from the cross-mapping relationship. Reactors undergo different state transition processes when performing different tasks, such as transitioning from an idle state to a preheating state, then to a reaction state, and finally to a cooling and cleaning state. The state transition patterns describe the state evolution path of the reactor driven by a specific task and the duration range of each state. Through historical operating data analysis, typical time windows for each state transition can be identified. For example, a reactor typically requires 30 to 45 minutes of preheating time to transition from an idle state to a reaction state, and 20 to 30 minutes to transition from a reaction state to a cooling state. The state transition patterns also include the triggering conditions for state switching, such as the reactor entering the reaction state only after the temperature reaches a set value, and cooling only starting after the reaction is complete and the pressure drops to a safe range.

[0088] For each dependency path in the task dependency tree, a forward propagation analysis is performed. A dependency path is a series of task nodes connected along dependency relationships, starting from a given initial task. The initial task of the path is selected as the starting point of the analysis. It is assumed that this task is delayed (i.e., the actual start time is later than the planned start time) or that resource consumption increases (i.e., the usage of shared resources such as steam and cooling water exceeds expectations). Under this assumption, the analysis propagates downwards layer by layer, examining how the execution time and resource requirements of subsequent task nodes in the path will change.

[0089] For each subsequent task node in the path, calculate the cascading change in its execution time. If the starting task is delayed... The time unit, and the strength of the dependency between this task and the next task. In that case, the planned start time for the next task will be postponed. Time unit. Continuing along the path, delays in second-level tasks will further affect third-level tasks, and so on. During propagation, consider whether there is a buffer time between tasks. If there is a buffer time between two tasks... ,and Then subsequent tasks need not be delayed; if Then the start time of subsequent tasks needs to be postponed. Units of time.

[0090] Simultaneously, analyze the cascading changes in resource demand. If the resource consumption of the initial task increases... If the unit is a resource pool that is shared with subsequent tasks, then the available resources in the resource pool will decrease. Subsequent tasks that were originally planned to use resources from this resource pool may need to wait or switch to other resources due to insufficient resources. The change in resource demand is represented as the difference between the actual available resources for subsequent tasks and the planned demand. By accumulating the time offsets and resource changes of all task nodes on the path, the cascading change for each task node is obtained.

[0091] The cascading changes are compared with preset impact thresholds. These thresholds are set based on the flexibility of the production plan and the urgency of the tasks; for example, a time offset threshold of 10 minutes and a resource change threshold of 15% of the original planned demand. If the time offset or resource change of a task node exceeds the corresponding threshold, that task node is considered significantly affected and may trigger cascading conflicts. All task nodes exceeding the threshold are grouped together as potential cascading conflict node groups.

[0092] For each task node in a potential cascading conflict node group, and considering the reactor state transition mode associated with that task node, the conditional boundary that would prevent the associated reactor from completing the state transition as expected when the task execution period deviates is derived. For example, a task node is originally scheduled to start execution when the reactor is in the preheating completed state. If the start of this task is delayed due to a delay in an upstream task, the reactor needs to wait additionally in the preheating state. Excessive waiting time in the preheating state may lead to energy loss or equipment aging. By analyzing the maximum allowable duration of each state in the reactor state transition mode, the upper limit of the tolerance for task deviation is determined.

[0093] Specifically, assuming the maximum permissible duration of the reactor in the preheated state is The task node was originally scheduled to be at time Start execution, if the actual start time of the task is The actual duration of the reactor in the preheating state is from the moment the reactor enters the preheating state to... The time interval. If the time interval exceeds... If this happens, the reactor cannot complete the state transition as expected and may need to be reheated or enter a fault state. Therefore, the condition boundary can be expressed as the offset at the start of the task must satisfy... ,in This is the time interval between the reactor entering the preheating state and the start of the originally planned task.

[0094] The boundary conditions are transformed into constraint expressions. These constraint expressions describe the relationship between the task execution period and the reactor state transition in the form of mathematical inequalities. For example, in the scenario above, the constraint expression is that the difference between the actual start time of the task and the planned start time must not exceed the difference between the maximum allowable preheating time of the reactor and the preheated time. The constraint expressions also include resource constraints, such as the amount of resources required by the task must not exceed the upper limit of resources available in the reactor's current state.

[0095] The constraint expressions corresponding to all task nodes in the task dependency tree are summarized. Each task node corresponds to at least one constraint expression, describing the constraint relationship between the task and the state transition of the associated reactor. These constraint expressions are combined with the dependency strengths extracted from the task dependency tree to construct constraint propagation rules. Constraint propagation rules are represented in the form of a set of rules, each rule containing a constraint expression, dependency strength, and triggering condition. The triggering condition describes under what circumstances the constraint needs to be activated; for example, when the execution time offset of a task exceeds a certain threshold, it is necessary to check whether all downstream dependent tasks of that task satisfy the corresponding constraint expressions.

[0096] During actual scheduling, if a deviation from the expected execution of a task is detected, the set of affected task nodes is automatically calculated based on constraint propagation rules, and the constraints corresponding to these task nodes are checked for violations. If a constraint is violated, a corresponding scheduling adjustment strategy is triggered, such as reassigning tasks to other reactors or adjusting the task execution order. Through the dynamic generation and application of constraint propagation rules, the cascading conflicts in a multi-reactor system can be predicted and proactively controlled in advance, ensuring the stability and feasibility of the overall system scheduling plan.

[0097] Based on the constraint propagation rules, each task in the set of tasks to be executed is associated with multiple reactors to obtain an initial mapping relationship, including:

[0098] Extract the constraint expressions corresponding to each task node from the constraint propagation rules;

[0099] For each task in the set of tasks to be executed, a set of candidate reactors that meet the constraints is selected based on the constraint expression corresponding to the task.

[0100] For each reactor in the candidate reactor set, the constraint propagation impact value on subsequent task nodes in the task dependency tree is calculated based on the dependency strength in the constraint propagation rule when the task is assigned to that reactor. The constraint propagation impact value characterizes the degree to which the range of selectable reactors for subsequent tasks is reduced after the assignment decision is propagated to subsequent tasks through the dependency path.

[0101] The reactor with the smallest constraint propagation impact value is selected as the target reactor for the task, and the task is associated with the target reactor. All tasks in the set of tasks to be executed are traversed to complete the association between the task and the reactor, and the initial mapping relationship is obtained.

[0102] For example, the constraint propagation rules, during their establishment, record the prerequisite dependencies, resource requirement types, temporal constraint ranges, and cascading influence paths between each task node in the task dependency tree. For any task in the set of tasks to be executed, its constraint expression includes the task's start time window, required reactor capacity range, temperature and pressure condition range, auxiliary resource requirements, and time interval requirements with the predecessor task. When extracting constraint expressions, the symbolic constraint relationships need to be converted into calculable numerical ranges. For example, if a task requires starting within 15 to 30 minutes after the completion of the predecessor task, this temporal constraint is converted into a range of relative time offset values ​​to facilitate numerical comparison in subsequent filtering operations.

[0103] For each task in the set of tasks to be executed, all available reactors are examined one by one according to the constraint expression corresponding to the task, and a set of candidate reactors that meet all constraints is selected. The examination process first determines whether the physical parameters of the reactor meet the task requirements, including whether the reactor volume can accommodate the required material volume, whether the reactor's maximum operating temperature and pressure cover the process parameter range of the task, and whether the type of stirring device equipped in the reactor matches the mixing method required by the task. Secondly, the feasibility in the time dimension is examined. Based on the task's currently allocated time slots, it is determined whether there are idle periods for the reactor within the task's time window, and whether the length of the idle periods is sufficient to cover the expected execution time of the task plus the necessary preheating and cleaning time. Thirdly, the compatibility in the resource dimension is examined. It is determined whether the auxiliary resources on which the reactor depends (such as steam supply pipelines, cooling water circulation systems, and vacuum pump sets) are occupied by other tasks within the planned execution period of the task. If there are resource sharing conflicts, it is necessary to further determine whether peak usage can be achieved through time fine-tuning. After the above multi-level screening, each reactor in the resulting set of candidate reactors meets the basic execution requirements of the task in terms of physical conditions, time window, and resource availability.

[0104] For each reactor in the candidate reactor set, further evaluation is needed to assess the cascading effects of assigning a task to that reactor on the entire task dependency tree. The evaluation process involves quantitative calculations based on the dependency strengths recorded in the constraint propagation rules. Assuming the task... Distributed to reactor This allocation decision will determine the task. The actual start time The determination at this moment will be made through directed edges in the task dependency tree to all tasks. The propagation effect of subsequent missions of the precursor. For the mission... A direct follow-up task Its constraints and propagation effects are reflected in the task. The feasible start time window is narrowed, and the degree of narrowing depends on the task. With the task Dependency strength between and the minimum time interval requirement between the two. .Task The lower bound of the feasible start time window has been postponed to This delay in the lower bound will make it possible to meet the task requirements. The number of time-constrained reactors has been reduced. When calculating the impact of constraint propagation, the statistics for the task under the new time constraints are taken into account. The reduction ratio of the candidate reactor set to the original candidate set indicates that the allocation decision severely impairs the scheduling flexibility of subsequent tasks.

[0105] The calculation of the impact value of constraint propagation needs to consider multi-level propagation effects. (Task) The allocation decision not only affects its directly subsequent tasks, but also indirectly subsequent tasks through the dependency chain. For tasks in the task dependency tree... For all descendant task nodes, the constraint propagation effect is calculated level by level using a depth-first traversal approach. Let the task... For the task A descendant task that has propagated through several layers of dependencies, task To the mission There exists a dependency path containing several intermediate task nodes. As propagation proceeds forward along this dependency path, the narrowing effect of the time window at each intermediate node accumulates and propagates to the next level node. The computation propagates to the task... When considering the constraints, it is necessary to perform a composite operation on the strength of dependencies at each level along the path to obtain the task. For the task The overall impact intensity The overall impact strength is calculated as a product of the dependency strengths of each segment along the path, reflecting the characteristic of constraint propagation decaying step by step along the dependency chain. This applies to tasks with multiple dependent tasks within the dependency tree. To the mission In the case of dependent paths, the path with the greatest overall influence among all paths is selected as the dominant propagation path, and the influence strength of this path is used as the target propagation path. For the task The constraint propagation impact measurement.

[0106] The task Distributed to reactor The resulting constraint propagation effect value This is obtained by summarizing the reduction degree of the candidate set of all affected subsequent tasks in the task dependency tree. Specifically, the calculation method is to traverse all tasks in the task dependency tree... For the ancestor node's task, for each subsequent task Calculation in the task Distributed to reactor Under the premise of the task The reduction rate of the candidate reactor set compared to the original set The reduction rate is defined as the ratio of the number of reactors excluded to the size of the original candidate set. The reduction rate for each subsequent task is then calculated based on its ratio to the task's size. The combined influence intensity between A weighted summation is performed to obtain the comprehensive constraint propagation impact value. During the weighted summation process, subsequent tasks with greater comprehensive impact strength contribute more to the total impact value through their reduction rate, reflecting that closely dependent subsequent tasks have higher priority in scheduling decisions. For different reactors in the candidate reactor set, the tasks are... Assigning tasks to different reactors will result in different outcomes. The actual start time varies, resulting in different constraint propagation impact values.

[0107] After obtaining the task After assigning the constraint propagation impact values ​​to each reactor in the candidate reactor set, the reactor with the smallest impact value is selected as the task. The target reactor is determined by minimizing its impact. Minimizing the impact means that the allocation decision minimizes the impairment to the scheduling flexibility of subsequent tasks in the task dependency tree, preserving more reactor selection space for subsequent tasks and reducing the risk of encountering infeasible scheduling options. When multiple reactors have the same minimum impact value, auxiliary decision indicators are used for selection. These indicators include the reactor's current load balancing level, the expected execution efficiency of the task on that reactor, and the cleaning preparation time required for switching the reactor to different task types. By comprehensively considering the constraint propagation effect and auxiliary indicators, a unique target reactor is finally determined, and the task is assigned accordingly. Establish an association with the target reactor, and record the task identifier, reactor identifier, planned start time, expected end time, and a list of auxiliary resources used.

[0108] The process iterates through all tasks in the task set to be executed, completing the association operation between each task and reactor, ultimately obtaining the initial mapping relationship. The traversal order follows the topological sorting of the task dependency tree, prioritizing root node tasks without predecessor tasks, and then processing task nodes whose predecessor tasks have already been assigned. Assigning tasks according to the topological order ensures that the assignment decisions for all predecessor tasks are determined when processing a task, and the task's time window constraints and resource availability judgments are calculated based on the actual assignment results of the predecessor tasks. During the traversal, if the candidate reactor set for a task is found to be empty, it is necessary to backtrack to the predecessor tasks and readjust the assignment decisions, releasing more scheduling space by changing the reactor selection of the predecessor tasks until a feasible reactor assignment scheme is found for the task. The initial mapping relationship generation process gradually constructs a global mapping scheme by minimizing the impact value of constraint propagation, maximizing the adaptability of the overall scheduling scheme to future state fluctuations while ensuring that each task can obtain a feasible assignment.

[0109] Based on the constraint propagation rules, the relaxation degree of each constraint is calculated. A scheduling elasticity space is constructed for each reactor in the initial mapping relationship based on the relaxation degree, and scheduling instructions containing state fluctuation tolerance intervals are generated, including:

[0110] For each reactor in the initial mapping relationship, extract the relaxation degree of the constraints corresponding to each task assigned to that reactor. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value.

[0111] Based on the relaxation, three threshold levels are set to divide the scheduling elasticity space into a core elasticity region, an extended elasticity region, and a limit elasticity region. The core elasticity region corresponds to the state fluctuation range where the relaxation is greater than the first threshold, the extended elasticity region corresponds to the state fluctuation range where the relaxation is between the first threshold and the second threshold, and the limit elasticity region corresponds to the state fluctuation range where the relaxation is between the second threshold and zero.

[0112] For each task assigned to the reactor, based on the difference in the probability of triggering cascading conflicts when the state deviates positively or negatively according to the corresponding constraints of the task, different widths of tolerance interval boundaries are set for the positive deviation direction and the negative deviation direction to form an asymmetric tolerance interval.

[0113] The scheduling instructions are generated by combining the hierarchical region boundary of the scheduling elastic space with the asymmetric tolerance interval boundary. The scheduling instructions specify the response strategy for the reactor state fluctuation in different regions.

[0114] For example, after obtaining the initial mapping relationship between reactors and tasks, a scheduling elasticity space needs to be constructed for each reactor to cope with state fluctuations during actual operation. For each reactor in the initial mapping relationship, firstly, all task nodes assigned to that reactor are extracted, and the constraints corresponding to these task nodes are traversed to obtain the relaxation value of each constraint. The relaxation is calculated as follows: for a certain constraint, while keeping other constraints unchanged, the allowable deviation range of the constraint is gradually increased. After each increase, it is re-verified whether all subsequent tasks in the task dependency tree can still find a feasible reactor allocation scheme. When a certain deviation range causes at least one subsequent task to have no feasible scheme, the previous value of the deviation range is determined as the relaxation of the constraint. For example, if the expected value of a temperature constraint is 350 degrees Celsius, and the allowable deviation is gradually relaxed to ±10 degrees Celsius, ±15 degrees Celsius, and ±20 degrees Celsius for verification, and it is found that there is no feasible solution for subsequent tasks at ±20 degrees Celsius, then the slack of this constraint is determined to be 15 degrees Celsius. This means that when the actual temperature fluctuates within the range of 335 to 365 degrees Celsius, the constraint is still considered basically satisfied and will not immediately trigger a scheduling reconfiguration. For constraints involving the time dimension, the slack may be expressed as the allowed time lead or delay; for resource-related constraints, the slack is reflected in the fluctuation space of resource usage.

[0115] After obtaining the relaxation degree of each constraint, three threshold levels are set based on the magnitude of the relaxation degree to divide the scheduling elasticity space into three levels of regions. The first threshold... The second threshold is typically set at the 75th percentile of the relaxation distribution. Set as the 25th percentile value of the relaxation distribution. When the relaxation of a constraint exceeds the first threshold... When the constraint's corresponding state fluctuation range is within the core elastic region, state deviations within this region are considered normal operational fluctuations, and the system requires no adjustment measures. For example, if the feed rate constraint relaxation of a reactor is 8 kg / h and the first threshold is 6 kg / h, then this constraint belongs to the core elastic region. When the actual feed rate fluctuates within 6 kg / h above or below the expected value, the scheduling system maintains the current task allocation scheme unchanged. When the constraint relaxation is between the first threshold... With the second threshold When the actual feed rate deviates from the expected value by more than 6 kg per hour but less than 8 kg per hour, the system enters the extended elasticity zone to assess whether the time window for subsequent tasks needs to be fine-tuned. During this period, the state fluctuation range corresponding to the constraint is classified as an extended elasticity zone. While deviations within this zone are still within acceptable limits, they are approaching the constraint's boundary conditions. Therefore, the system needs to activate an early warning mechanism to closely monitor the constraint's changing trend and prepare for local adjustment contingency plans. Continuing with the feed rate example, if the relaxation is 8 kg per hour and the second threshold is 3 kg per hour, then when the actual feed rate deviates from the expected value by more than 6 kg per hour but less than 8 kg per hour, the scheduling system enters the extended elasticity zone response mode and begins to evaluate whether fine-tuning of the time window for subsequent tasks is necessary. When the constraint relaxation is between the second threshold... When the value is between zero and zero, the range of state fluctuations corresponding to the constraint is classified into the limit elastic region. At this point, the state deviation has approached the hard boundary of the constraint. Although cascading conflicts have not yet been triggered, the system must take immediate intervention measures, including temporarily adjusting the resource allocation ratio, starting backup equipment, or performing emergency task reordering, to prevent the deviation from expanding further and causing the scheduling scheme to fail.

[0116] Building upon the three-tiered stratified region, the probability difference of cascading conflicts triggered by constraints in positive and negative deviations is further considered. For each task assigned to a reactor, the cascading impact paths of the corresponding constraints are analyzed when the state deviates positively and negatively. Taking reaction time constraints as an example, if the planned reaction time for a task is 120 minutes, a positive deviation in the actual reaction time (i.e., an extended reaction time) may cause the start time of subsequent tasks to be shifted later, leading to conflicts in the steam system's occupancy period. In this case, the probability of cascading conflicts is high. However, when the actual reaction time deviates negatively (i.e., a shortened reaction time), although reactor resources may be released earlier, insufficient material conversion may affect product quality. While such effects exist, they often do not immediately trigger scheduling conflicts for other tasks, thus the probability of cascading conflicts is low. Based on this probability difference, the tolerance interval width is set for the positive deviation direction. Less than the tolerance interval width set in the negative deviation direction This creates an asymmetric tolerance interval. Specifically, if the relaxation degree of a certain constraint is... The probability of a positive deviation triggering a cascading conflict is The probability of triggering a cascading conflict with negative deviation is Then the positive tolerance boundary is set as the expected value plus... The negative tolerance boundary is set as the expected value minus Through this asymmetric design, the scheduling system can more accurately control risks while tolerating deviations from actual operating conditions, and impose stricter restrictions on deviations that are prone to cascading conflicts.

[0117] The scheduling instructions for the reactor are generated by combining the three-level hierarchical region boundaries of the scheduling elasticity space with the asymmetric tolerance interval boundaries. These instructions record the reactor's response strategies in different state fluctuation regions in the form of structured data. When the reactor's actual operating state is within the core elasticity region, the scheduling instructions instruct the system to maintain the current task execution plan, without initiating any adjustment mechanisms, and only perform routine data collection and logging. When the actual operating state enters the extended elasticity region, the scheduling instructions trigger a warning-level response, activating the real-time monitoring module, increasing the data sampling frequency, and simultaneously activating the preparation process for local adjustment contingency plans to assess whether fine-tuning of the time windows for subsequent tasks is necessary. When the actual operating state enters the extreme elasticity region, the scheduling instructions trigger an emergency-level response, and the system immediately performs resource reallocation operations, which may include temporarily calling redundant resources from other low-priority tasks, activating backup reactors to share the load, or reordering some branches in the task dependency tree. If the actual operating state exceeds the boundary of the extreme elasticity region, the current scheduling scheme is deemed unsuitable for the actual situation, and the system triggers a global rescheduling process to recalculate the mapping relationship between all reactors and tasks.

[0118] During the implementation of the asymmetric tolerance range, the scheduling instructions also include rules for determining the direction of deviation. The system continuously collects the actual operating parameters of the reactor, compares the actual parameter values ​​with the expected parameter values, and determines a positive deviation when the actual value is greater than the expected value and a negative deviation when the actual value is less than the expected value. Based on the direction of deviation, the corresponding tolerance range boundary is queried to determine whether the current deviation is within the tolerance range. Taking the pressure control of a certain reactor as an example, if the expected pressure is 2.5 MPa, the positive tolerance boundary is 2.8 MPa, and the negative tolerance boundary is 2.1 MPa, when the actual pressure is 2.7 MPa, the system identifies it as a positive deviation within the core elasticity region, and no adjustment is required; when the actual pressure rises to 2.85 MPa, although it is a positive deviation, it has exceeded the positive tolerance boundary. At this time, even if the pressure value is still within the total allowable relaxation range, the corresponding response strategy still needs to be activated because the cascading risk of positive deviation is higher.

[0119] After the scheduling instructions are generated, they are distributed to the control units of each reactor in the form of an instruction set. Each reactor control unit has an embedded status monitoring module and a decision execution module. The status monitoring module collects actual operating data according to the sampling frequency specified in the scheduling instructions, and the decision execution module automatically selects the corresponding response strategy based on the elastic space region in which the actual state is located. When the reactor state fluctuates within the core elastic region, the decision execution module does not interfere with the normal execution process of the task; when the state enters the extended elastic region or the extreme elastic region, the decision execution module performs operations such as early warning notification, resource allocation, or task rescheduling according to the preset operation sequence in the scheduling instructions. This hierarchical response mechanism significantly reduces the scheduling system's demand for real-time computing resources, avoids system oscillations caused by frequent rescheduling, and, through the design of the asymmetric tolerance interval, enables the scheduling system to tolerate normal operating fluctuations to the maximum extent while ensuring a safety margin, thereby improving the overall scheduling efficiency and robustness.

[0120] Forming an asymmetric tolerance interval includes:

[0121] Extract all dependency paths of the task in the task dependency tree from the constraint propagation rules, and count the path length and number of nodes of each dependency path;

[0122] For the positive and negative deviations of the task, the constraint tightening amount propagated to subsequent task nodes along each dependency path is simulated when the task state deviates by a unit. Dependency paths that cause the constraint tightening amount of subsequent task nodes to exceed the trigger threshold are marked as trigger paths. The cascading conflict trigger probability is calculated based on the proportion of the number of trigger paths to the total number of dependency paths, the average path length corresponding to the trigger path, and the average number of nodes.

[0123] Calculate the ratio of the probability of cascading conflict triggering in a positive deviation to the probability of cascading conflict triggering in a negative deviation. Based on the ratio, assign tolerance interval width weights to the positive and negative deviation directions. The larger the ratio, the smaller the tolerance interval width weight assigned to the deviation direction.

[0124] Based on the relaxation degree and the tolerance interval width weight, the tolerance interval boundary in the positive deviation direction and the tolerance interval boundary in the negative deviation direction are calculated to form an asymmetric tolerance interval.

[0125] For example, the design of the asymmetric tolerance interval fully considers the asymmetric impact of task state deviation on subsequent task chains. By finely analyzing the difference in behavior between positive and negative deviations in constraint propagation, the scheduling instructions can achieve precise tolerance control of state fluctuations in different directions.

[0126] For a task node that has completed its initial mapping, the first step is to extract all dependency paths involved in the task dependency tree from the constraint propagation rule base. A dependency path is a complete sequence of paths that starts from the current task node and extends along the dependency edges to the leaf nodes of the dependency tree. Each dependency path contains two basic attributes: path length and number of nodes. The path length is defined as the cumulative sum of the dependency strengths between adjacent task nodes in the path, and the number of nodes refers to the total number of task nodes contained in the path. By using a depth-first search algorithm to traverse the task dependency tree, all dependency paths for the task can be systematically enumerated. The number of dependency paths that can form in the task dependency tree of each subsequent task node depends on the branching structure of the dependency tree, and can reach up to [number missing] in the worst case. The order of magnitude. Calculate the path length for each path. With the number of nodes ,in Assigning path numbers can provide a quantitative basis for subsequent cascading conflict propagation analysis.

[0127] After obtaining the complete set of dependency paths for the task, cascading conflict propagation simulations need to be performed for both positive and negative deviations. A positive deviation refers to a task's actual start time being later than the planned start time or actual resource usage exceeding planned resource usage; a negative deviation corresponds to an earlier start time or reduced resource usage. For each dependency path, the simulation process assumes a unit deviation for the current task. This deviation will propagate along the dependency path to subsequent task nodes. During the propagation process, the first... Constraint tightening amount received by each subsequent task node This depends on the cumulative propagation of dependencies along the path and the absorption effect of buffer time. Specifically, if from the current task to the... The path of each node contains If there are intermediate nodes, the constraint tightening amount can be expressed as: ,in For the first on the path The strength transitivity coefficient of segment dependency. This represents the corresponding buffer time absorption amount. It also represents the amount of constraint tightening at a subsequent task node. Exceeding the preset trigger threshold of this node When the constraint state of the node is triggered, it is determined that the node enters the warning state, and the dependency path is marked as the trigger path.

[0128] After simulating all dependent paths for both positive and negative deviations, the number of trigger paths in each direction is counted. The number of trigger paths for positive deviation is denoted as... The number of trigger paths for negative deviation is denoted as The total number of dependency paths is denoted as Trigger path percentage and This directly reflects the extent of cascading conflicts caused by deviations in different directions. In addition to the proportion of triggering paths, the average path length of each triggering path also needs to be calculated. and and the average number of nodes in the trigger path. and The average path length characterizes the cumulative strength of constraint propagation, while the average number of nodes reflects the scale of the affected task.

[0129] Cascade Conflict Trigger Probability The calculation comprehensively considers three factors: the proportion of triggering paths, the average path length, and the average number of nodes. The probability of triggering a cascading conflict with positive deviation is calculated as follows: ,in The path length influencing factor, These are factors influencing the number of nodes. These two factors are determined based on statistical analysis of historical scheduling data, and are typically... The value ranges from 0.05 to 0.15. The value ranges from 0.03 to 0.10. Similarly, the probability of triggering a cascading conflict with negative deviation is calculated as follows: When the proportion of triggering paths in a certain direction is higher, the average path length is longer, or the average number of nodes is greater, the probability of cascading conflicts in that direction increases accordingly, indicating that state deviations in that direction are more likely to trigger chain reactions in the scheduling system.

[0130] Calculate the ratio of the cascading conflict trigger probability of positive deviation to negative deviation. This ratio is the core basis for the design of the asymmetric tolerance interval. If This indicates that positive deviations are more likely to trigger cascading conflicts, in which case stricter tolerance limits should be imposed on the direction of positive deviation, i.e., a smaller tolerance interval should be allocated; conversely, if... In this case, stricter restrictions should be imposed on the negative deviation direction. The allocation of the tolerance interval width weight follows an inverse proportionality principle, with the tolerance interval width weight for the positive deviation direction being higher. Tolerance interval width weight relative to the negative deviation direction Satisfy normalization conditions ,and By solving the constraint equation, we obtain... , .when When significantly greater than 1, Approaching 0, Approaching 1 indicates that almost the entire tolerance interval width should be allocated to the negative deviation direction; when When it approaches 1, and They approach equality, and the tolerance intervals are symmetrically distributed.

[0131] Relaxation based on task correspondence constraints Calculate the specific tolerance interval boundary using the tolerance interval width weight. Relaxation The overall tolerance space characterizing this constraint is determined by the aforementioned method. The width of the tolerance interval for the positive deviation direction. Calculated as Tolerance range width for negative deviation direction Calculated as Let the standard execution status parameters of the task be... Then the tolerance interval boundary for the positive deviation direction is The tolerance interval boundary for negative deviation direction is The complete asymmetric tolerance interval is represented as: For time-related constraints, state parameters The representative start time of the plan is indicated by the tolerance interval boundaries, which represent the earliest and latest allowed start times. For resource-related constraints, The standard resource usage is represented by the tolerance interval boundary, which indicates the permissible range of resource usage variation.

[0132] In actual scheduling, as long as the reactor's real-time operating status remains within the asymmetric tolerance range, rescheduling is not required even if it deviates from the planned state. This design significantly reduces the computational burden of the scheduling system and the cost of frequent reactor adjustments. When the probability of a positive deviation from the task... It is the probability of negative deviation. When it is 3 times, that is positive tolerance interval width It accounts for only 25% of the total relaxation, while the negative tolerance range is wide. Achieving a total relaxation rate of 75% ensures strict control over high-risk deviations. In a real-world example from a chemical production system, a reaction task had a relaxation rate of 12 minutes. Dependency path analysis revealed that positive deviations triggered 4 out of 5 dependency paths, while negative deviations triggered only 1. Calculations... The resulting asymmetric tolerance interval extends forward by 3 minutes and backward by 9 minutes, which improves the success rate of cascading conflict prevention by about 40% compared to the symmetric interval.

[0133] During scheduling and execution, the distribution characteristics of actual state deviations and the actual triggering of cascading conflicts are continuously monitored. If the actual frequency of positive deviations for a task is found to be significantly higher than predicted, or the actual triggering rate of cascading conflicts caused by positive deviations exceeds the predicted value, the system will increase the impact factor of the positive deviation for that task. and , thereby improving , to make the ratio If the tolerance interval is increased, the positive tolerance interval width for that task will automatically narrow in the next scheduling cycle. This feedback correction mechanism ensures that the asymmetric tolerance interval design always matches the dynamic characteristics of the actual production scenario, achieving truly intelligent fault-tolerant scheduling.

[0134] When the actual operating state of the reactor deviates from the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data, including:

[0135] Based on the scheduling instructions, task allocation operations are performed on the multiple reactors;

[0136] Collect actual operating status data of each reactor during the task allocation process, and calculate the deviation between the actual operating status data and the expected operating status in the scheduling instructions;

[0137] The deviation value is compared with the boundary of the state fluctuation tolerance range of the corresponding reactor. When the deviation value is within the tolerance range, it is marked as an absorbable deviation. When the deviation value exceeds the tolerance range, it is marked as a trigger rescheduling deviation.

[0138] For reactors marked as triggering rescheduling deviations, the task corresponding to the reactor is unbound from the current reactor. Based on the updated reactor operating status data and constraint propagation rules, the candidate reactor set and constraint propagation impact value of the task are recalculated, and the task is reassigned to the reactor with the smallest constraint propagation impact value.

[0139] For reactors marked as absorbable deviations, record the position of the corresponding task in the task dependency tree and the actual impact of the deviation value on subsequent task nodes after it is absorbed by the elastic space.

[0140] The distribution characteristics of absorbable deviations and the propagation attenuation law of actual impact in multiple task allocation operations are statistically analyzed. Based on the distribution characteristics and propagation attenuation law, the calculation parameters of dependency strength and constraint tightening in the constraint propagation rules are corrected.

[0141] For example, after the scheduling system issues scheduling instructions containing tolerance ranges for state fluctuations to multiple reactors, each reactor begins to perform actual production operations according to its assigned tasks. During task execution, the actual operating state of the reactors may deviate from the expected values ​​due to factors such as differences in raw material batches, fluctuations in ambient temperature, or equipment aging. To avoid frequent global rescheduling that could cause system oscillations, a real-time performance deviation acquisition and adaptive correction mechanism is needed to balance scheduling stability and resource utilization.

[0142] Data on the actual operating status of each reactor during the task allocation process is collected, including measured reaction temperature, catalyst activity indicators, material flow rate, and reaction time, among other key parameters. Taking a fine chemical reactor as an example, when performing an oxidation reaction, the expected operating status is as follows: reaction temperature... Catalyst conversion rate Material residence time Actual data collected showed the reaction temperature. Catalyst conversion rate Material residence time Calculate the deviation value for each parameter separately: temperature deviation. Conversion rate deviation Dwell time deviation .

[0143] The calculated deviation value is compared with the pre-constructed state fluctuation tolerance interval boundary in the scheduling instruction. The tolerance interval is divided into a core elastic region, an extended elastic region, and a limiting elastic region based on constraint relaxation. Assume the temperature parameter tolerance interval for this reactor is... The conversion rate tolerance range is The tolerable dwell time range is Comparing the actual deviation value with the tolerance range boundary revealed: temperature deviation... lie in Conversion rate deviation within the range lie in Within the interval, the deviation of the dwell time lie in Within the acceptable range. Since the deviations of all key parameters have not exceeded the tolerance range boundaries, the reactor is marked as being in an absorbable deviation state, and no rescheduling is required.

[0144] If a polymerization reactor is performing a polymerization task, the actual reaction temperature is... , and expected temperature The tolerance range is Calculate the temperature deviation. The deviation was found to exceed the upper boundary of the tolerance range. At this point, the reactor is marked as triggering a rescheduling deviation state, and the task unbinding and reallocation process is immediately initiated. The aggregation task is unbound from the current reactor, releasing the reactor's resources and updating its operational status label. Based on the latest collected operational status data of the reactors, the current load, equipment health, and resource availability of each candidate reactor are reassessed. Combining the updated constraint propagation rules, the constraint propagation impact value generated when the aggregation task is assigned to each candidate reactor is calculated. Assume the candidate reactor set contains reactors... , and Calculate the constraint propagation impact value respectively. , , Select the reactor with the least impact. As the target for reallocation, the aggregation task will be assigned to And generate new scheduling instructions.

[0145] For reactors marked as absorbable deviations, further analysis is needed to determine the actual impact of this deviation on subsequent task nodes after it is absorbed by the elastic space. This can be achieved by tracing the subsequent task chain of the reactor's currently executing task using a task dependency tree. Assume that the subsequent tasks of this oxidation reaction task in the task dependency tree include separation, purification, and packaging. Temperature deviation. It may lead to an increase in the content of byproducts in oxidation products. This necessitates extending the separation time for the separation task. To achieve the desired purity. Extended separation time consumes shared separation equipment resources, thus delaying the actual start time of the purification process. Real-time data collection revealed that delays in refining tasks were within their tolerable range. The effect was absorbed internally without triggering further cascading conflicts. The propagation chain of this effect was recorded: oxidation task temperature deviation. → Separation task extension → Refinement mission postponed → No cascading conflict was triggered.

[0146] Conversion rate deviation This leads to a decrease in the yield of oxidation products. This necessitates increasing the amount of raw materials fed into subsequent batches to meet total production demands. This adjustment is absorbed by the raw material inventory flexibility and will not affect the scheduling of subsequent tasks. (Dwell time deviation) This results in a slight increase in reactor occupancy time, which delays the actual start time of the next task to be performed. Because the time tolerance interval for this downstream task is... The delay was successfully absorbed. The aforementioned impact data was stored in a historical deviation database, including fields such as deviation value, deviation direction, absorbed location, propagation depth, and attenuation coefficient.

[0147] The distribution characteristics of absorbable bias in multiple task allocation operations were statistically analyzed. This was achieved through... Cluster analysis of historical data from the previous scheduling cycle revealed that temperature deviations exhibited a normal distribution, with a mean of [missing value]. Standard deviation The conversion rate deviation exhibits a left-skewed distribution, with a mean of [missing value]. Standard deviation The dwell time deviation exhibits a bimodal distribution, with the main peak located at... The secondary peak is located at The physical causes of the deviation distribution characteristics were analyzed: the positive bias of temperature deviation stemmed from the decline in cooling system efficiency, the negative bias of conversion rate deviation stemmed from the gradual decrease in catalyst activity over time, and the bimodal distribution of residence time reflected the switching between fast and slow reaction processes.

[0148] Extract the propagation decay law of the actual impact. Analysis of the impact propagation chain reveals that the impact exhibits an exponential decay characteristic as the deviation propagates from the initial task to subsequent task nodes. Define a decay factor. Characterizing the proportion of influence retained after propagation through one layer. Statistical data shows the attenuation factor of temperature deviation. That is, the influence is retained after propagating through one layer. Attenuation factor of conversion rate deviation Attenuation factor of residence time deviation Analysis of the sources of the difference in decay factors reveals the following: temperature deviation affects product quality, but quality issues are partially corrected in subsequent separation stages, resulting in slower decay; conversion rate deviation affects yield, but insufficient yield can be quickly corrected by supplementing raw materials, leading to faster decay; residence time deviation directly consumes time resources, and time delays accumulate in the task chain, resulting in the slowest decay.

[0149] The calculation parameters for dependency strength and constraint tightening in the constraint propagation rule are revised based on the characteristics of deviation distribution and propagation attenuation law. In the original constraint propagation rule, the dependency strength... Set to a fixed value This indicates the degree of influence of the upstream task on the downstream task. Based on actual data, when the upstream task is a temperature-sensitive task, the actual dependency strength should be increased to [value missing]. When the upstream task is a conversion rate-sensitive task, the actual dependency strength should be reduced to [value missing]. When correcting the calculation parameters for constraint tightening, a deviation direction weighting factor is introduced. The constraint tightening weight for positive deviations (such as increased temperature or prolonged time) The constraint tightening weight for negative deviations (such as decreased conversion rate or insufficient yield) This weighting factor reflects the different impacts of different deviation directions on subsequent task constraints: positive deviations often lead to longer resource usage and require stricter constraints; negative deviations can be compensated for through resource redundancy, and the degree of constraint tightening can be appropriately relaxed.

[0150] An adaptive learning mechanism is introduced to dynamically update the constraint propagation rule parameters. The parameter update period is defined as follows: In each scheduling cycle, upon reaching the update cycle, based on the most recent... The deviation distribution parameters and decay factor are recalculated using historical data. Parameter updates employ an exponentially weighted moving average method, with new parameter values... From historical parameter values Compared with current statistics Weighted result, weight coefficient The calculation formula is: This update mechanism both preserves historical experience to avoid drastic parameter fluctuations and responds promptly to changes in process trends. When a significant change in the deviation distribution parameter is detected (e.g., a change in standard deviation exceeding a certain threshold), it will update the mechanism. When this occurs, an early warning mechanism is triggered to prompt process engineers to check the equipment status and raw material quality, preventing scheduling failures due to equipment malfunctions or raw material abnormalities.

[0151] A second aspect of the present invention provides a multi-reactor AI intelligent scheduling and coordination system, comprising:

[0152] The conflict identification unit is used to acquire the current operating status data and the set of tasks to be executed of multiple reactors, identify reactor resource sharing conflicts, trace the root cause task chain that leads to the resource sharing conflict in reverse, establish a cross-mapping relationship between the task dependency tree and the reactor state change trajectory, and dynamically generate constraint propagation rules for predicting potential cascading conflicts based on the cross-mapping relationship.

[0153] The task mapping unit is used to associate each task in the set of tasks to be executed with multiple reactors according to the constraint propagation rules to obtain an initial mapping relationship;

[0154] The elastic scheduling unit is used to calculate the relaxation degree of each constraint based on the constraint propagation rule. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value. Based on the relaxation degree, a scheduling elastic space is constructed for each reactor in the initial mapping relationship, and a scheduling instruction containing the state fluctuation tolerance range is generated.

[0155] An adaptive correction unit is used to perform task allocation operations on the multiple reactors based on the scheduling instructions. When the actual operating state of the reactor deviates within the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data.

[0156] A third aspect of the present invention provides an electronic device, comprising:

[0157] processor;

[0158] Memory used to store processor-executable instructions;

[0159] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0160] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

Claims

1. A multi-reactor AI intelligent scheduling and coordination method, characterized in that, include: The system acquires current operating status data and a set of tasks to be executed from multiple reactors, identifies resource sharing conflicts among reactors, traces the root cause task chain leading to the resource sharing conflict in reverse, establishes a cross-mapping relationship between the task dependency tree and the reactor state change trajectory, and dynamically generates constraint propagation rules for predicting potential cascading conflicts based on the cross-mapping relationship. Based on the constraint propagation rules, each task in the set of tasks to be executed is associated with multiple reactors to obtain an initial mapping relationship; The relaxation degree of each constraint is calculated based on the constraint propagation rule. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value. A scheduling elasticity space is constructed for each reactor in the initial mapping relationship based on the relaxation degree, and a scheduling instruction containing the state fluctuation tolerance interval is generated. Based on the scheduling instructions, task allocation operations are performed on the multiple reactors. When the actual operating state of the reactor deviates within the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data. Identify reactor resource sharing conflicts, and establish a cross-mapping relationship between the task dependency tree and the reactor state change trajectory by tracing the root cause task chain leading to the resource sharing conflict in reverse, including: For each task in the set of tasks to be executed, extract the type of demand for reactor resources and the time period of occupation; parse the current operating status data into the resource availability time window of each reactor; By comparing the task resource requirement type with the resource availability time window, task combinations with overlapping time windows and the same resource type are identified and marked as potential resource sharing conflict pairs. For each task in the potential resource sharing conflict pair, backtrack upwards layer by layer along its set of preceding dependent tasks to the starting task node, and record the ordered sequence of all tasks on the backtracking path as the root cause task chain; construct the task dependency tree according to the dependency relationship of each task in the root cause task chain. For each task in the root cause task chain, extract the sequence of state changes of the relevant reactors during the execution period of the task from the current running status data to form a reactor state change trajectory; By establishing an association mapping between each task node in the task dependency tree and its corresponding reactor state change trajectory, a cross-mapping relationship is obtained.

2. The method of claim 1, wherein, Based on the aforementioned cross-mapping relationship, constraint propagation rules for predicting potential cascading conflicts are dynamically generated, including: Extract the strength of the dependency relationship between each task node in the task dependency tree and the state transition pattern of each reactor in the reactor state change trajectory from the cross mapping relationship; For each dependency path in the task dependency tree, forward propagation analysis is performed to analyze the chain reaction of changes in the execution time and resource requirements of subsequent task nodes when the starting task of the path is delayed or resource consumption increases. Task nodes whose chain reaction exceeds a preset impact threshold are marked as potential cascading conflict node groups. For each task node in the potential cascading conflict node group, combined with the reactor state transition mode associated with the task node, the condition boundary that causes the associated reactor to fail to complete the state transition as expected when the execution period of the task is offset is derived, and the condition boundary is transformed into a constraint expression. The constraint expressions corresponding to all task nodes in the task dependency tree are combined with the dependency strength to construct a constraint propagation rule.

3. The method of claim 1, wherein, Based on the constraint propagation rules, each task in the set of tasks to be executed is associated with multiple reactors to obtain an initial mapping relationship, including: Extract the constraint expressions corresponding to each task node from the constraint propagation rules; For each task in the set of tasks to be executed, a set of candidate reactors that meet the constraints is selected based on the constraint expression corresponding to the task. For each reactor in the candidate reactor set, the constraint propagation impact value on subsequent task nodes in the task dependency tree is calculated based on the dependency strength in the constraint propagation rule when the task is assigned to that reactor. The constraint propagation impact value characterizes the degree to which the range of selectable reactors for subsequent tasks is reduced after the assignment decision is propagated to subsequent tasks through the dependency path. The reactor with the smallest constraint propagation impact value is selected as the target reactor for the task, and the task is associated with the target reactor. All tasks in the set of tasks to be executed are traversed to complete the association between the task and the reactor, and the initial mapping relationship is obtained.

4. The method of claim 1, wherein, Based on the constraint propagation rules, the relaxation degree of each constraint is calculated. A scheduling elasticity space is constructed for each reactor in the initial mapping relationship based on the relaxation degree, and scheduling instructions containing state fluctuation tolerance intervals are generated, including: For each reactor in the initial mapping relationship, extract the relaxation degree of the constraints corresponding to each task assigned to that reactor. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value. Based on the relaxation, three threshold levels are set to divide the scheduling elasticity space into a core elasticity region, an extended elasticity region, and a limit elasticity region. The core elasticity region corresponds to the state fluctuation range where the relaxation is greater than the first threshold, the extended elasticity region corresponds to the state fluctuation range where the relaxation is between the first threshold and the second threshold, and the limit elasticity region corresponds to the state fluctuation range where the relaxation is between the second threshold and zero. For each task assigned to the reactor, based on the difference in the probability of triggering cascading conflicts when the state deviates positively or negatively according to the corresponding constraints of the task, different widths of tolerance interval boundaries are set for the positive deviation direction and the negative deviation direction to form an asymmetric tolerance interval. The scheduling instructions are generated by combining the hierarchical region boundary of the scheduling elastic space with the asymmetric tolerance interval boundary.

5. The method according to claim 4, characterized in that, Forming an asymmetric tolerance interval includes: Extract all dependency paths of the task in the task dependency tree from the constraint propagation rules, and count the path length and number of nodes of each dependency path; For the positive and negative deviations of the task, the constraint tightening amount propagated to subsequent task nodes along each dependency path is simulated when the task state deviates by a unit. Dependency paths that cause the constraint tightening amount of subsequent task nodes to exceed the trigger threshold are marked as trigger paths. The cascading conflict trigger probability is calculated based on the proportion of the number of trigger paths to the total number of dependency paths, the average path length corresponding to the trigger path, and the average number of nodes. Calculate the ratio of the probability of cascading conflict triggering in a positive deviation to the probability of cascading conflict triggering in a negative deviation. Based on the ratio, assign tolerance interval width weights to the positive and negative deviation directions. The larger the ratio, the smaller the tolerance interval width weight assigned to the deviation direction. Based on the relaxation degree and the tolerance interval width weight, the tolerance interval boundary in the positive deviation direction and the tolerance interval boundary in the negative deviation direction are calculated to form an asymmetric tolerance interval.

6. The method according to claim 1, characterized in that, When the actual operating state of the reactor deviates from the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data, including: Collect actual operating status data of each reactor during the task allocation process, and calculate the deviation between the actual operating status data and the expected operating status in the scheduling instructions; The deviation value is compared with the boundary of the state fluctuation tolerance range of the corresponding reactor. When the deviation value is within the tolerance range, it is marked as an absorbable deviation. When the deviation value exceeds the tolerance range, it is marked as a trigger rescheduling deviation. For reactors marked as absorbable deviations, record the position of the corresponding task in the task dependency tree and the actual impact of the deviation value on subsequent task nodes after it is absorbed by the elastic space. The distribution characteristics of absorbable deviations and the propagation attenuation law of actual impact in multiple task allocation operations are statistically analyzed. Based on the distribution characteristics and propagation attenuation law, the calculation parameters of dependency strength and constraint tightening in the constraint propagation rules are corrected.

7. A multi-reactor AI intelligent scheduling and coordination system for implementing the method of any one of claims 1-6, characterized in that, include: The conflict identification unit is used to acquire the current operating status data and the set of tasks to be executed of multiple reactors, identify reactor resource sharing conflicts, trace the root cause task chain that leads to the resource sharing conflict in reverse, establish a cross-mapping relationship between the task dependency tree and the reactor state change trajectory, and dynamically generate constraint propagation rules for predicting potential cascading conflicts based on the cross-mapping relationship. The task mapping unit is used to associate each task in the set of tasks to be executed with multiple reactors according to the constraint propagation rules to obtain an initial mapping relationship; The elastic scheduling unit is used to calculate the relaxation degree of each constraint based on the constraint propagation rule. The relaxation degree represents the tolerance range of the corresponding constraint from the expected value. Based on the relaxation degree, a scheduling elastic space is constructed for each reactor in the initial mapping relationship, and a scheduling instruction containing the state fluctuation tolerance range is generated. An adaptive correction unit is used to perform task allocation operations on the multiple reactors based on the scheduling instructions. When the actual operating state of the reactor deviates within the tolerance range, there is no need to trigger rescheduling. The constraint propagation rules are adaptively corrected using the collected real-time performance deviation data.

8. An electronic device, comprising: include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon computer program instructions, wherein, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images