Order scheduling resource scheduling optimization method for flexible production of thermal fuses

Abstract

The invention relates to the technical field of intelligent manufacturing and production optimization scheduling, and particularly discloses a thermal fuse flexible production-oriented order scheduling resource scheduling optimization method, which comprises the following steps of: generating a structured vector by extracting inherent and dynamic characteristics of an order, and constructing a production scheduling dynamic environment model by fusing a device resource state time sequence; the order comprehensive process switching cost and the resource constraint tightness are calculated, order clusters are divided through multi-target constraint clustering, and a pre-arrangement sequence is generated; key resource conflict early warning is completed through time sequence prediction; constructing a multi-target scheduling model by taking the order cluster as a unit, and generating a look-ahead window scheme by combining a rolling optimization solidification plan baseline; and finally, on the basis of a real-time execution data updating model, through dynamic priority preemption and local rescheduling adjustment when abnormity occurs, the problems of high cost of small-batch mixed arrangement, conflict identification lagging, easy damage of a thermosensitive process steady state and poor compliance constraint adaptation are solved, and collaborative optimization of the scheduling efficiency, the production yield and the compliance is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and production optimization scheduling technology, specifically to an order scheduling resource scheduling optimization method for flexible production of temperature fuses. Background Technology

[0002] As a core thermistor component for circuit safety protection, thermal fuses possess unique characteristics in their production, including rigid thermistor manufacturing processes, safety certification isolation, mandatory 100% inspection, multi-SKU small-batch customization, and scarcity of key resources. Existing flexible production scheduling resource allocation methods suffer from the following technical deficiencies: The scheduling modeling does not take into account the specific characteristics of thermal fuses and does not include constraints such as rated operating temperature, rated voltage, rated current, safety certification, and aging of thermally sensitive materials in the model. Mixed scheduling of orders may lead to compliance risks and disruption of process steady state. When small-batch customer trial and trial production orders are mixed with large-batch orders after the customer's formal import, the cost of changeover, full inspection compliance, and traceability increases exponentially with the number of SKUs. Conventional clustering methods fail to achieve synergistic aggregation of process similarity and resource constraints, resulting in poor scheduling dimensionality reduction. The identification of critical resource conflicts relies on post-event statistics, lacking a predictive early warning mechanism. Abnormal rescheduling often involves complete reordering, frequently interrupting processes such as alloy smelting, organic material batching, stirring, briquetting, and epoxy resin curing, leading to a significant drop in product yield. Furthermore, it is extremely prone to causing mixed specifications. Thermal fuses are one-time-use components; only by testing their rated operating temperature can it be determined whether a thermal fuse has been mixed. For customers, thermal fuses are already installed in electrical equipment and shipped to end users; if they are mixed, it can easily lead to fires. Mixed specifications are one of the most fatal flaws of thermal fuses.

[0003] Scheduling optimization often focuses on a single delivery date, failing to balance multiple dimensions such as resource load, process steady state, and compliance costs. Hybrid evolutionary algorithms do not perform targeted optimization for conflicting resources, resulting in insufficient solution efficiency and feasibility. The abnormal response mechanism during the planning and execution phase is rigid, the resource preemption rules do not take into account the criticality of the process and the uninterruptible nature of thermally sensitive processes, and the logic for local rescheduling is missing, which further exacerbates production disturbances.

[0004] To address this, the present invention provides an order scheduling resource allocation optimization method for flexible production of temperature fuses. Summary of the Invention

[0005] The purpose of this invention is to provide an order scheduling resource allocation optimization method for flexible production of temperature fuses, in order to solve the aforementioned background problems.

[0006] The objective of this invention can be achieved through the following technical solution: an order scheduling resource allocation optimization method for flexible production of temperature fuses, comprising the following steps: Extract the inherent and dynamic features of orders to be scheduled, generate order feature vectors, and combine them with the time-series sequence of resource status of each device collected synchronously to construct a dynamic environment model for production scheduling and predict the key resource load in the next time unit. Calculate the fit between any two orders in the order to be scheduled. Using the fit as a constraint, use a multi-objective constrained clustering algorithm to divide the order to be scheduled into multiple order clusters and generate a pre-scheduled production sequence within each order cluster. Based on the pre-arranged production sequence within each order cluster and the time sequence of equipment resource status, the conflict probability of each key resource in the future period is calculated using the attention LSTM prediction model, potential conflict periods are identified, and conflict early warning information is generated. Using order clusters as scheduling units and combining conflict warning information, a multi-objective optimization scheduling model is constructed, and a forward-looking window scheduling scheme is generated through a rolling optimization algorithm. Real-time production data is collected to update the dynamic environment model of production scheduling. Dynamic priority resource preemption is triggered by the deviation analysis between the actual progress and the forward window scheduling plan, and local rescheduling updates are performed in combination with the rolling optimization algorithm.

[0007] Furthermore, the order feature vector is generated as follows: Inherent characteristics include rated operating temperature, rated voltage, rated current, size specifications, material code, process path, batch size, and delivery time; The dynamic characteristic is urgency, and the calculation process is as follows: To determine the maximum and minimum remaining delivery dates within a batch of all pending orders, a linear normalization mapping is used to normalize the remaining delivery date of the current order to its urgency level. By integrating inherent features with dynamic features based on urgency, a structured order feature vector is generated for each order.

[0008] Furthermore, the construction process of the production scheduling dynamic environment model is as follows: Based on a unified timestamp dimension, the operating status, current load rate, mold online status, estimated maintenance time, and estimated next available time of each device are sorted out and serialized into a single device-level resource status time sequence in chronological order; Establish the association constraints between the structured feature vector of orders and the time sequence of resource status, and clarify the matching rules between orders with different characteristics and corresponding compatible devices; The three types of data—order feature set, resource status time series set, and order-resource matching constraint—are integrated and encapsulated into a dynamic production scheduling environment model.

[0009] Furthermore, the method for predicting the critical resource load in the next time unit is as follows: Key resources refer to production equipment in a production line that has limited capacity and is irreplaceable; Set a fixed-duration time unit as the prediction unit; Iterate through the structured feature vectors of all pending orders and extract the demand parameters related to key resources within the prediction unit; Based on the process path, match the key resources required for the order; By combining order urgency and remaining delivery time, the planned production time of the process within the forecast unit is deduced, and then the actual occupation time of the process within the forecast unit is determined. Extract the resource unavailability duration within the prediction unit from the resource status time series of key resources, including the fault duration marked in the sequence, the planned maintenance duration, and the pre-occupancy duration of solidified orders; Effective availability of resources = Total duration of prediction units - Duration of unavailable resources; The total demand duration for the critical resource is aggregated from all orders within the future forecast unit. The ratio of the total demand duration within the future forecast unit to the effective usable duration of the resource is then calculated to obtain the critical resource load factor.

[0010] Furthermore, the adaptation degree is calculated as follows: Adaptability includes switching costs and resource constraints; Calculate the proportion of the average time spent on temperature switching, mold replacement, and equipment parameter resetting to the total changeover time within the historical cycle, and obtain the weights of temperature switching, mold replacement, and equipment parameter resetting. The switching cost is obtained by weighted summation of temperature switching, mold replacement, and equipment parameter resetting. The resource constraint tightness is calculated by proportionally dividing the intersection of the resource occupancy time of the two orders by the total occupancy time.

[0011] Furthermore, the method for generating pre-arranged production sequences within each order cluster is as follows: Traverse the structured feature vectors of all orders within the order cluster, extract the process path of each order, decompose the continuous process into atomic process nodes, and retain the order of the processes of each order; Perform an intersection operation on the atomic process chains of all orders within the order cluster, filter out the mandatory processes common to all orders, arrange them in the order of production processes, and form the cluster baseline process route; The priority rules are as follows: First, sort by urgency from high to low. If the urgency is the same, arrange the orders in ascending order based on the comprehensive process switching cost of each pair of orders. Combined with the short-term load forecast of key resources, the order processes are arranged to the low-load period of resources to avoid resource competition within the cluster. Orders within a cluster are sorted according to priority rules to form a pre-arranged production sequence within the cluster.

[0012] Furthermore, the probability of conflict for each key resource in the future time period is calculated as follows: Organize historical production data, including order cluster scheduling records, resource load data, and conflict event records, and encode input features, including: Process complexity vector: [Number of orders within the cluster, average number of processes, percentage of critical processes, number of melting point types]; Resource demand intensity vector: [occupancy duration of each key resource / duration of time unit, resource type weight]; Equipment historical status sequence: [load rate, number of failures, and number of maintenance visits for the same unit in the past 7 days]; Encode the scheduled order cluster data for the next L time periods into vectors, and then concatenate them with the historical status sequence of the equipment to form the model input; Set the prediction period, and the model outputs the probability distribution of the load rate. The conflict probability = P(load rate > safety threshold), and the safety threshold is set according to the equipment type. Iterate through all key resource and time period pairs. If the conflict probability meets the requirements, trigger an alert. Mark the key resource and time period pairs that trigger the alert as potential conflicts and generate alert information, including resource name, conflict time period, conflict probability, and the order clusters involved.

[0013] Furthermore, the process of generating the look-ahead window scheduling scheme is as follows: Define the look-ahead window length as the planning period that covers a fixed future duration, the rolling step size as the duration for each look-ahead window advance, and the fixed interval as the planning segment within the look-ahead window that is adjacent to the current time and needs to be locked for execution; Starting from the current time, the latest state of the production scheduling dynamic environment model and conflict warning information are input into the hybrid evolutionary algorithm, and the algorithm solves the complete scheduling solution within the look-ahead window. Only the process allocation, time arrangement, and resource allocation within the fixed interval of the forward window are locked as the planning baseline and issued to the production unit for execution. The planning baseline of the fixed interval within the forward window will no longer participate in subsequent optimization and adjustment. Within the look-ahead window, the remaining portion after subtracting the fixed interval from the look-ahead window length is the variable segment plan. The variable segment plan is only used as a pre-scheduling reference, reserving room for subsequent optimization and adjustment. Each time a rolling step is completed, the equipment status, order information and conflict warning information of the production scheduling dynamic environment model are updated synchronously, and the forward window is pushed forward by one rolling step, repeating the optimization-solidification process. Output the forward-looking window scheduling plan, including: the fixed planning baseline, which is the locked process plan issued to the production unit for execution, and the forward-looking window pre-scheduling plan, which is the pre-scheduling sequence of variable segments, resource allocation plan, and load forecast results.

[0014] Furthermore, the triggering method for the dynamic priority resource preemption is as follows: Collect data on process execution status and equipment status changes; Calculate the schedule deviation between actual progress and planned progress: Schedule Deviation = |Actual Time - Planned Time| / Planned Time; If the schedule deviation meets the requirements, dynamic priority resource preemption is triggered, and the affected time window is marked.

[0015] Furthermore, the process of the local rescheduling update is as follows: Based on the results of dynamic priority resource preemption, high-priority and low-priority processes are determined. After the resource preemption operation is executed, a new local resource allocation constraint is formed. Starting from the current time, for the affected time window that includes new constraints on local resource allocation, the rolling optimization algorithm is invoked to perform local rescheduling, and the adjusted process instructions are issued.

[0016] The beneficial effects of this invention are as follows: By embedding rigid constraints such as the thermal fuse's thermistor process, safety certification, and melting point compatibility into the entire process model, compliance risks and process steady-state disruptions are avoided from the source, and the problem of poor compatibility between conventional scheduling and special production is solved. By using multi-objective constrained clustering to aggregate similar orders, the unit cost of changeover, full inspection, and traceability is significantly reduced, balancing the scale effect of large batches with the flexible requirements of small batches, and improving the efficiency of scheduling dimensionality reduction. Attention LSTM is used to predict critical resource load and conflicts in advance, turning post-event remediation into pre-event warning, avoiding production interruptions caused by conflicting resources and improving the utilization rate of critical resources. Rolling optimization only solidifies short-term plans and limits the scope of impact by local rescheduling, avoiding full rearrangement from disrupting the melting and solidifying constant temperature field and reducing product yield loss; The scheduling model simultaneously considers delivery time, resource load, and delay costs, while the hybrid evolutionary algorithm resolves conflicts in a targeted manner, significantly improving solution efficiency and the feasibility of the solution. Based on dynamic priority resource preemption and local rescheduling, it can quickly adapt to unique anomalies such as equipment failure, urgent order insertion, and temperature control fluctuations, thereby reducing the plan execution deviation rate. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of the order scheduling resource allocation optimization method for flexible production of temperature fuses according to the present invention; Figure 2 This is a flowchart illustrating the calculation of the conflict probability of each key resource in the future time period in this invention. Detailed Implementation

[0019] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0020] Example: Please refer to Figure 1 As shown, the order scheduling resource scheduling optimization method for flexible production of temperature fuses according to the present invention specifically includes the following steps: Step 1: Extract the inherent and dynamic features of the orders to be scheduled, generate order feature vectors, and combine them with the time-series sequences of resource status of each device collected synchronously to build a dynamic environment model for production scheduling and predict the key resource load in the next time unit; In step one, the inherent and dynamic characteristics of the order to be scheduled include: Inherent characteristics include rated operating temperature, rated voltage, rated current, size specifications, material code, process path, batch size, and delivery time; Dynamic characteristics refer to urgency, and the calculation process is as follows: Iterate through all pending orders, determine the maximum and minimum remaining delivery dates within the batch, and use linear normalization mapping to convert the remaining delivery date of the current order into an urgency level in the range [0,1]. The formula is: Urgency level = 1 - (current remaining delivery date - minimum remaining delivery date) / (maximum remaining delivery date - minimum remaining delivery date). In step one, the process of generating the order feature vector includes: Receive a set of orders to be scheduled, and for each order, extract its inherent feature vector, including rated operating temperature, rated voltage, rated current, size specifications, material code, process path, batch size, and delivery date; The process path is a standardized process path. For example, the process path of different orders is broken down into standardized process nodes such as melting, pressing, packaging, testing and packaging. At the same time, dynamic features, namely urgency, are assigned. The inherent features are integrated with the dynamic urgency features to generate a unique structured feature vector for each order, forming an order feature set. In step one, the process of collecting the resource status time sequence of each device includes: Based on a unified timestamp dimension, the operating status, current load rate, mold online status, estimated maintenance time, and estimated next available time of each device are sorted out and serialized into a single device-level resource status time sequence in chronological order. All devices are then aggregated to form a full production line resource status time sequence set. In step one, the process of constructing the dynamic environment model for production scheduling includes: Establish the association constraints between the structured feature vector of the order and the time sequence of the resource status, such as matching the order process path with the corresponding processing equipment and the rated action temperature with the special tooling; The three types of data—order feature set, resource status time series set, and order-resource matching constraint set—are integrated and encapsulated into a dynamic production scheduling environment model. The model includes callable subsets of order features, resource status, and matching constraints. In step one, the prediction process for the critical resource load in the next time unit includes: Key resources refer to: core production equipment and special tooling in the production line that have limited capacity, are irreplaceable, have proprietary thermal / safety-compliant processes, and are prone to production conflicts, such as adjustable temperature melting furnaces, high-precision temperature testing platforms, special melting point molds, and safety-certified fixtures; Set a fixed-duration time unit as the prediction unit; The first point to clarify is that extracting resource requirements from the structured feature vector of the order is specifically as follows: Iterate through the structured feature vectors of all pending orders and extract the demand parameters related to key resources within the prediction unit: Lock in the process path and match the key resources required by the order (such as the smelting process - R1 smelting furnace, the inspection process - R3 test bench); Extract batch size, rated operating temperature / material type, and calculate the standard processing time per operation + changeover preparation time: Single process time = Standard working time per piece × Order batch size + Temperature / mold changeover time (temperature control reset and mold changeover time for orders with different melting points); By combining the urgency of the order with the remaining delivery period, the planned production period of the process within the forecast unit is extrapolated, and the actual time occupied by the process within the forecast unit is determined (processes spanning time units are proportionally calculated). Secondly, it should be noted that the resource availability is calculated from the resource status time series, specifically as follows: From the time series of resource status of key resources, extract the resource availability parameters within the prediction unit, including: Total duration within a resource prediction unit: equal to the fixed duration of the prediction unit; Resource unavailability duration: the duration of faults marked in the sequence, the planned maintenance duration, and the pre-occupancy duration of fixed orders; Effective availability of resources = Total duration of prediction units - Duration of unavailability; Thirdly, it should be noted that the calculation of critical resource load factor is as follows: Summarize the total demand duration of all orders within the future forecast unit for this critical resource, and calculate the load factor by combining it with the effective available time: Critical resource load factor = Total demand duration within the future forecast unit / Effective available time of the resource; Load factor ≤ 100%: Resource load is normal, no contention or conflict; Load factor > 100%: Resource overload, tight constraints; For example, if the production line has two orders for temperature fuses to be scheduled and three dedicated pieces of equipment as key production line resources, specific examples of the order feature subset, resource status subset, and matching constraint subset in the constructed dynamic production scheduling environment model include: The order feature subset consists of the structured feature vectors of each order: Order number O1: Inherent characteristics: Rated operating temperature 102℃, rated voltage 250V, rated current 15A, dimensions 8.3×7.5mm, material tin-bismuth-indium alloy, process path = melting - packaging - temperature detection, batch size 5000pcs, required delivery date 2026-02-06 12:00; Dynamic characteristics: Remaining delivery time 48 hours, urgency level 0.20; Order number O2: Inherent characteristics: Rated operating temperature 130℃, rated voltage 250V, rated current 15, dimensions 8.3×7.5mm, material indium tin alloy, process path = melting - packaging - temperature detection, batch size 2000pcs, required delivery date 2026-02-04 24:00; Dynamic characteristics: Remaining delivery time 12 hours, urgency level 0.90; The resource status subset consists of a time-series sequence of resource statuses for each device, serialized by timestamp: Timestamp - Status / Load / Mod / Maintenance / Next Available: Adjustable temperature melting furnace (R1): T0 (current) - running / 85% / 350℃ melting mold / no maintenance / after 30 minutes; T1 - idle / 0% / waiting for mold change / no maintenance / immediately; General Purpose Packaging Machine (R2): T0 (Current) - Idle / 0% / General Purpose Packaging Mold / No Maintenance / Immediately; T1 - Idle / 0% / General Purpose Packaging Mold / No Maintenance / Immediately; High-precision temperature test bench (R3): T0 (Current) - Running / 90% / General test fixture / Maintenance on 2026-02-05 00:00 / 60 min later; T1 - Fault / 0% / Pending repair / Maintenance in progress / 120 min later; The matching constraint subset is a hard constraint rule that matches order features to resource status: Process path constraints: O1 and O2 must be performed in the order of melting (R1) - packaging (R2) - temperature detection (R3), and the process cannot be reversed; Temperature specification constraints: O1 (390℃) can only use R1 390℃ melting mold, O2 (350℃) needs to be processed after changing to R1 350℃ melting mold; Resource exclusivity constraint: R1 / R2 / R3 can only handle one order's process at a time and cannot process them in parallel; Maintenance constraint: R3 will enter maintenance after 00:00 on 2026-02-05, and no testing procedures can be assigned during this period; Homogeneity constraint: The resource can be allocated only if the next available time of the resource is less than or equal to the scheduled start time of the order process; It should be noted that the role of order multi-dimensional feature extraction and dynamic environment modeling is to provide a unified data source that fits the production characteristics of temperature fuses for subsequent order aggregation, conflict prediction, and scheduling optimization, so as to avoid scheduling schemes from deviating from reality due to missing features (such as ignoring melting point attributes). Step 2: Calculate the fit between any two orders in the order to be scheduled. Using the fit as a constraint, use a multi-objective constraint clustering algorithm to divide the order to be scheduled into multiple order clusters and generate a pre-scheduled production sequence for each order cluster. In step two, the adaptability includes switching cost and resource constraint tightness, and the calculation process includes: For any two orders: By statistically analyzing the actual data of production changeover within the historical cycle, the average time spent on temperature switching, mold replacement, and equipment parameter resetting is calculated as a percentage of the total changeover time, thus obtaining the weights of temperature switching, mold replacement, and equipment parameter resetting. The switching cost is obtained by weighted summation of temperature switching, mold replacement, and equipment parameter resetting. The resource constraint tightness is calculated by proportionally combining the intersection of the resource occupancy durations of the two orders with the total occupancy duration. In step two, the process of dividing the orders to be scheduled into multiple order clusters includes: The clustering objective is set as minimizing the maximum switching cost within a cluster and minimizing the constraint tightness. First, set the value range of K based on the capacity of key resources / order batch. For example, if there are 3 key smelting furnaces, then K∈[2,5] to avoid the failure of dimensionality reduction due to too many / too few clusters. Randomly select K orders from the order set as initial cluster centers, traverse each unassigned order, and calculate its comprehensive distance to each cluster center, including switching cost and resource constraint tightness; For each cluster, recalculate the cluster center and select the order that minimizes the sum of the overall distances of all orders within the cluster as the new center; Repeatedly assign and update until the termination condition is met to determine the final K order clusters, and output the core information of each cluster: the order list within the cluster, the baseline process route of the cluster (the common process path of the orders within the cluster), the maximum switching cost within the cluster, and the average constraint tightness within the cluster. In step two, the generation process of the intra-cluster pre-arranged production sequence includes: The first point to clarify is the baseline process route for generating clusters, which is as follows: Traverse the structured feature vectors of all orders within the order cluster, extract the process path of each order, decompose the continuous process into atomic process nodes (such as melting, pressing, epoxy encapsulation, temperature full inspection according to the US UL60691 standard with 0.1% of the operation temperature sampling inspection, marking, and packaging), and retain the sequence of each order's process. Perform an intersection operation on the atomic process chains of all orders within the cluster, filter out the mandatory processes common to all orders, arrange them in a rigid order of production process, and form the cluster baseline process route. Secondly, it should be noted that the generation process of the pre-sorted production sequence within the cluster includes: The priority rules are as follows: First, sort by urgency from high to low. If the urgency is the same, arrange the orders in ascending order based on the comprehensive process switching cost of each pair of orders. Combined with the short-term load forecast of key resources, the order processes are arranged to the low-load period of resources to avoid resource competition within the cluster. Based on priority rules, orders within the cluster are sorted to form an initial production sequence. And based on the standard working hours of the cluster's baseline process route, the planned start time and planned end time of each process for each order are deduced. The initial sequence of process resource occupancy plans is superimposed and verified with the resource status time sequence and the critical resource load prediction to identify three types of conflicts: critical resources being occupied by multiple processes at the same time, processes being arranged during resource maintenance / failure periods, and resource load rates exceeding the safety threshold. Adjust the initial sequence to address the conflict: High-urgency orders will be prioritized for resource allocation, while low-urgency orders will be deferred to periods when resources are available. Adjust the order adjacency relationship and prioritize combinations with lower overall process switching costs to reduce the total changeover cost within the cluster; Ensure continuous operation of heat-sensitive processes such as melting and packaging to meet process steady-state requirements; After adjustment, the final pre-arranged production sequence within the cluster is formed; It should be noted that the intelligent order aggregation and intra-cluster pre-scheduling based on constraint tightness serve the following purposes: by aggregating orders using process similarity and resource constraints, the frequent changeovers of small-batch orders are reduced. At the same time, resource competition within the cluster is resolved, cross-cluster resource conflicts are avoided, and the problems of high changeover costs and intense competition for key resources caused by the mixed scheduling of small batches of temperature fuses are solved. The complex n-order scheduling problem is reduced to K-cluster scheduling, thereby improving the efficiency of subsequent optimization. Step 3: Based on the pre-arranged production sequence within each order cluster and the time sequence of equipment resource status, calculate the conflict probability of each key resource in the future time period using the attention LSTM prediction model, identify potential conflict periods, and generate conflict early warning information. Please see Figure 2 As shown, in step three, the prediction process for the conflict probability of each key resource in the future time period includes: Organize historical production data, including order cluster scheduling records, resource load data, and conflict event records, and encode input features, including: Process complexity vector: [Number of orders within the cluster, average number of processes, percentage of critical processes, number of melting point types]; Resource demand intensity vector: [occupancy duration of each key resource / duration of time unit, resource type weight]; Equipment historical status sequence: [load rate, number of failures, and number of maintenance visits for the same unit in the past 7 days]; Encode the scheduled order cluster data for the next L time periods into vectors, and then concatenate them with the historical status sequence of the equipment (such as the load rate of the same time period in the last 7 days) to form the model input; Set the prediction period, and the model outputs the probability distribution of the load rate. The conflict probability = P(load rate > safety threshold). The safety threshold is set according to the equipment type, such as 85% for temperature test bench and 90% for smelting furnace. Iterate through all (key resource, time period) pairs. If the conflict probability is greater than or equal to the warning threshold, then trigger a warning. Mark the (key resource, time period) pairs that trigger the warning threshold as potential conflicts and generate warning information, including resource name, conflict time period, conflict probability, and the order clusters involved. It should be noted that the role of dynamic prediction and early warning of production conflicts is to identify load conflicts of key resources in advance by using time series prediction models, and to solve the problem that rescheduling of the thermal fuse process after the fact can easily disrupt the constant temperature steady state and lead to a decrease in yield. Step 4: Using order clusters as scheduling units and combining conflict warning information, construct a multi-objective optimization scheduling model, and generate a look-ahead window scheduling scheme through a rolling optimization algorithm; In step four, the construction process of the multi-objective optimization scheduling model includes: Using order clusters as scheduling units, an optimization model is established with the objectives of minimizing total process time, minimizing the total overload of critical conflicting resources, and minimizing the delay time of high-priority orders. A hybrid algorithm combining standard genetic algorithm and conflict-oriented mutation is used to optimize the solution space for early warning conflicts. Specific steps are as follows: Using real-number encoding, each chromosome corresponds to a scheduling scheme. The gene positions are, in order, the m-th process of the k-th order cluster, the key / general resources allocated to the m-th process of the k-th order cluster, and the timing adjustment coefficient of the pre-arranged production sequence within the k-th order cluster. An initial population is generated based on the pre-arranged production sequence within the cluster; The fitness value is the reciprocal of the multi-objective function F. A higher fitness value indicates a better scheduling scheme. Genetic operations Selection: Use roulette wheel selection to retain high-fitness, high-quality scheduling solutions; Crossover: Single-point crossover is performed on chromosome processing time and resource allocation gene loci to retain superior traits from the parent generation; Conflict-oriented targeted mutation: For gene loci involving potential conflict periods and conflict resources, the mutation probability is increased to 2-3 times the base probability. The mutation direction is to adjust the process of conflict resources to non-conflict periods or replace them with backup resources to actively resolve conflicts. When the number of iterations reaches the preset limit, or when the fitness value does not improve significantly for several consecutive generations, the iteration is terminated and the optimal scheduling feasible solution is output. In step four, the process of generating a look-ahead window scheduling scheme using the rolling optimization algorithm includes: Define the look-ahead window length as the planning period that covers a fixed future duration, the rolling step size as the duration for each window to advance forward, and the fixed interval as the planning segment within the look-ahead window that is adjacent to the current time and needs to be locked for execution; Starting from the current time, the latest state of the production scheduling dynamic environment model and conflict warning information are input into the hybrid evolutionary algorithm to obtain the complete scheduling solution within the look-ahead window W; Only the process allocation, time, and resource configuration within the solidified interval are locked as the planning baseline and issued to the production unit for execution; this part will not participate in subsequent optimization. The variable segment plan within the forward window (length of the forward window minus the fixed interval) is only for pre-scheduling reference and reserves room for optimization. Each time a rolling step is completed, the production scheduling dynamic environment model and conflict warning information are updated synchronously, and the forward window is advanced by one rolling step. The optimization-solidification process is repeated to achieve dynamic iteration of the plan. The optimal schedule is broken down into execution plans for each order cluster, and the cluster's baseline process route and resource status time sequence are associated to form a detailed schedule by resource, time period, and order cluster. Output two types of scheduling results: Fixed planning baseline: a locked-in process plan that has been issued for execution; Forward-looking window pre-scheduling scheme: pre-scheduling timing, resource allocation, and load forecasting for variable segments, for subsequent rolling optimization; It should be noted that the role of the multi-objective rolling optimization scheduling that integrates conflict prediction is as follows: taking order clusters as units, it integrates conflict early warning information for targeted optimization, and balances long-term optimization and real-time adaptation through a rolling window to avoid plan rigidity caused by one-time optimization. Step 5: Collect production data in real time to update the dynamic environment model of production scheduling, and trigger dynamic priority resource preemption by analyzing the deviation between the actual progress and the forward window scheduling plan, and perform local rescheduling updates in combination with the rolling optimization algorithm; In step five, the process of triggering dynamic priority resource preemption through deviation analysis between the actual progress and the forward window scheduling scheme includes: Collect data on process execution status (start / end time), equipment status changes (fault / repair), and urgent order insertion events; Compare the actual progress with the planned progress and calculate the deviation: Schedule Deviation = |Actual Time - Planned Time| / Planned Time; The schedule deviation is compared with the preset deviation. If the schedule deviation is greater than or equal to the preset deviation, dynamic priority resource preemption is triggered, and the affected time window is defined. Resource acquisition rules: Only preempt resources that are pre-allocated but not actually started (such as resources that have been planned to occupy the test bench but have not yet started testing); After a low-priority process is preempted, spare resources (such as spare smelting furnaces) are allocated first. If there are no spares, the process is adjusted to off-peak hours. Local rescheduling: Only the affected time windows are optimized, rather than a full rescheduling, reducing the impact on the overall plan; After the acquisition, a new constraint on local resource allocation is created; Starting from the current time, the system calls the multi-objective rolling optimization schedule that incorporates conflict prediction to perform local rescheduling for the affected time window containing this new constraint, and issues the adjusted process instructions. It should be noted that the priority-based dynamic resource preemption and feedback adjustment function is as follows: by using dynamic priority, resources are flexibly allocated, and only local rescheduling is performed on the scope of abnormal impact, avoiding full rescheduling that would disrupt the steady state of the thermally sensitive process, and balancing flexible response and process stability.

[0021] The technical solution and advantages of this application are as follows: Extracting the inherent and dynamic features of orders to be scheduled, generating order feature vectors, and combining them with the synchronously collected time-series sequences of equipment resource status, constructing a dynamic environment model for production scheduling, and predicting the critical resource load in a future time unit; calculating the fit between any two orders in the order to be scheduled, using the fit as a constraint, employing a multi-objective constraint clustering algorithm to divide the orders into multiple order clusters, and generating intra-cluster pre-scheduled production sequences for each order cluster; based on the intra-cluster pre-scheduled production sequences and equipment resource status time-series sequences of each order cluster, using an attention LSTM prediction model, calculating the conflict probability of each critical resource in future time periods, identifying potential conflict periods, and generating conflict warning information; using order clusters as scheduling units, combining conflict warning information, constructing a multi-objective optimization scheduling model, and generating a look-ahead window scheduling scheme through a rolling optimization algorithm; real-time collection of production data to update the dynamic environment model for production scheduling, and triggering dynamic priority resource preemption through deviation analysis between actual progress and the look-ahead window scheduling scheme, combined with a rolling optimization algorithm for local rescheduling updates. This invention generates structured vectors by extracting inherent and dynamic features of orders, and integrates the time-series sequence of equipment resource status to construct a dynamic production scheduling environment model. It calculates the comprehensive process switching cost and resource constraint tightness of orders, divides order clusters through multi-objective constraint clustering, and generates pre-scheduled sequences. It provides early warning of key resource conflicts through time-series prediction. A multi-objective scheduling model is constructed using order clusters as units, combining rolling optimization to solidify the planning baseline and generate forward-looking window schemes. Finally, based on real-time execution data, the model is updated, and in case of anomalies, dynamic priority preemption and local rescheduling adjustments are used to solve the problems of high cost of small-batch mixed scheduling, delayed conflict identification, easy disruption of steady-state conditions in thermally sensitive processes, and poor adaptation to compliance constraints. This achieves synergistic optimization of scheduling efficiency, production yield, and compliance.

[0022] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the scope of the present invention.

Claims

1. An order scheduling resource allocation optimization method for flexible production of temperature fuses, characterized by: Includes the following steps: Extract the inherent and dynamic features of orders to be scheduled, generate order feature vectors, and combine them with the time-series sequence of resource status of each device collected synchronously to construct a dynamic environment model for production scheduling and predict the key resource load in the next time unit. Calculate the fit between any two orders in the order to be scheduled. Using the fit as a constraint, use a multi-objective constrained clustering algorithm to divide the order to be scheduled into multiple order clusters and generate a pre-scheduled production sequence within each order cluster. Based on the pre-arranged production sequence within each order cluster and the time sequence of equipment resource status, the conflict probability of each key resource in the future period is calculated using the attention LSTM prediction model, potential conflict periods are identified, and conflict early warning information is generated. Using order clusters as scheduling units and combining conflict warning information, a multi-objective optimization scheduling model is constructed, and a forward-looking window scheduling scheme is generated through a rolling optimization algorithm. Real-time production data is collected to update the dynamic environment model of production scheduling. Dynamic priority resource preemption is triggered by the deviation analysis between the actual progress and the forward window scheduling plan, and local rescheduling updates are performed in combination with the rolling optimization algorithm.

2. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 1, characterized in that: The order feature vector is generated as follows: Inherent characteristics include rated operating temperature, rated voltage, rated current, size specifications, material code, process path, batch size, and delivery time; The dynamic characteristic is urgency, and the calculation process is as follows: To determine the maximum and minimum remaining delivery dates within a batch of all pending orders, a linear normalization mapping is used to normalize the remaining delivery date of the current order to its urgency level. By integrating inherent features with dynamic features based on urgency, a structured order feature vector is generated for each order.

3. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 2, characterized in that: The process of constructing the dynamic environment model for production scheduling is as follows: Based on a unified timestamp dimension, the operating status, current load rate, mold online status, estimated maintenance time, and estimated next available time of each device are sorted out and serialized into a single device-level resource status time sequence in chronological order; Establish the association constraints between the structured feature vector of orders and the time sequence of resource status, and clarify the matching rules between orders with different characteristics and corresponding compatible devices; The three types of data—order feature set, resource status time series set, and order-resource matching constraint—are integrated and encapsulated into a dynamic production scheduling environment model.

4. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 3, characterized in that: The method for predicting the critical resource load in the next time unit is as follows: Key resources refer to production equipment in a production line that has limited capacity and is irreplaceable; Set a fixed-duration time unit as the prediction unit; Iterate through the structured feature vectors of all pending orders and extract the demand parameters related to key resources within the prediction unit; Based on the process path, match the key resources required for the order; By combining order urgency and remaining delivery time, the planned production time of the process within the forecast unit is deduced, and then the actual occupation time of the process within the forecast unit is determined. Extract the resource unavailability duration within the prediction unit from the resource status time series of key resources, including the fault duration marked in the sequence, the planned maintenance duration, and the pre-occupancy duration of solidified orders; Effective availability of resources = Total duration of prediction units - Duration of unavailable resources; The total demand duration for the critical resource is aggregated from all orders within the future forecast unit. The ratio of the total demand duration within the future forecast unit to the effective usable duration of the resource is then calculated to obtain the critical resource load factor.

5. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 1, characterized in that: The fitness level is calculated as follows: Adaptability includes switching costs and resource constraints; Calculate the proportion of the average time spent on switching thermal fuse specifications, changing molds, and resetting equipment parameters to the total time spent on changing models within the historical period, and obtain the weights of switching thermal fuse specifications, changing molds, and resetting equipment parameters. The switching cost is obtained by weighted summation of the thermal fuse specification switching, mold replacement, and equipment parameter resetting. The resource constraint tightness is calculated by proportionally dividing the intersection of the resource occupancy time of the two orders by the total occupancy time.

6. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 5, characterized in that: The method for generating pre-arranged production sequences within each order cluster is as follows: Traverse the structured feature vectors of all orders within the order cluster, extract the process path of each order, decompose the continuous process into atomic process nodes, and retain the order of the processes of each order; Perform an intersection operation on the atomic process chains of all orders within the order cluster, filter out the mandatory processes common to all orders, arrange them in the order of production processes, and form the cluster baseline process route; The priority rules are as follows: First, sort by urgency from high to low. If the urgency is the same, arrange the orders in ascending order based on the comprehensive process switching cost of each pair of orders. Combined with the short-term load forecast of key resources, the order processes are arranged to the low-load period of resources to avoid resource competition within the cluster. Orders within a cluster are sorted according to priority rules to form a pre-arranged production sequence within the cluster.

7. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 1, characterized in that: The probability of conflict for each key resource in the future time period is calculated as follows: Organize historical production data, including order cluster scheduling records, resource load data, and conflict event records, and encode input features, including: Process complexity vector: [Number of orders within the cluster, average number of processes, percentage of critical processes, number of melting point types]; Resource demand intensity vector: [occupancy duration of each key resource / duration of time unit, resource type weight]; Equipment historical status sequence: [load rate, number of failures, and number of maintenance visits for the same unit in the past 7 days]; Encode the scheduled order cluster data for the next L time periods into vectors, and then concatenate them with the historical status sequence of the equipment to form the model input; Set the prediction period, and the model outputs the probability distribution of the load rate. The conflict probability = P(load rate > safety threshold), and the safety threshold is set according to the equipment type. Iterate through all key resource and time period pairs. If the conflict probability meets the requirements, trigger an alert. Mark the key resource and time period pairs that trigger the alert as potential conflicts and generate alert information, including resource name, conflict time period, conflict probability, and the order cluster involved.

8. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 7, characterized in that: The process of generating the look-ahead window scheduling scheme is as follows: Define the look-ahead window length as the planning period that covers a fixed future duration, the rolling step size as the duration for each look-ahead window advance, and the fixed interval as the planning segment within the look-ahead window that is adjacent to the current time and needs to be locked for execution; Starting from the current time, the latest state of the production scheduling dynamic environment model and conflict warning information are input into the hybrid evolutionary algorithm, and the algorithm solves the complete scheduling solution within the look-ahead window. Only the process allocation, time arrangement, and resource allocation within the fixed interval of the forward window are locked as the planning baseline and issued to the production unit for execution. The planning baseline of the fixed interval within the forward window will no longer participate in subsequent optimization and adjustment. Within the look-ahead window, the remaining portion after subtracting the fixed interval from the look-ahead window length is the variable segment plan. The variable segment plan is only used as a pre-scheduling reference, reserving room for subsequent optimization and adjustment. Each time a rolling step is completed, the equipment status, order information and conflict warning information of the production scheduling dynamic environment model are updated synchronously, and the forward window is pushed forward by one rolling step, repeating the optimization-solidification process. Output the forward-looking window scheduling plan, including: the fixed planning baseline, which is the locked process plan issued to the production unit for execution, and the forward-looking window pre-scheduling plan, which is the pre-scheduling sequence of variable segments, resource allocation plan, and load forecast results.

9. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 1, characterized in that: The triggering method for the dynamic priority resource preemption is as follows: Collect data on process execution status and equipment status changes; Calculate the schedule deviation between actual progress and planned progress: Schedule Deviation = |Actual Time - Planned Time| / Planned Time; If the schedule deviation meets the requirements, dynamic priority resource preemption is triggered, and the affected time window is marked.

10. The order scheduling resource allocation optimization method for flexible production of temperature fuses according to claim 1, characterized in that: The process of local rescheduling update is as follows: Based on the results of dynamic priority resource preemption, high-priority and low-priority processes are determined. After the resource preemption operation is executed, a new local resource allocation constraint is formed. Starting from the current time, for the affected time window that includes new constraints on local resource allocation, the rolling optimization algorithm is invoked to perform local rescheduling, and the adjusted process instructions are issued.

Images