A scheduling optimization method and system for production project management

Abstract

The present invention discloses a scheduling optimization method and system for production project management, which relates to the technical field of production scheduling. A scheduling optimization system for production project management includes: a production project management module and a production project scheduling module. The present invention standardizes and semanticizes complex production project information through the construction of main project triples and sub-project triples, effectively enhancing the parsability of the scheduling model for business logic; through scheduling-aware encoding and clustering processing of sub-project feature vectors, it can automatically identify similar production unit tasks that can be merged, improve resource utilization and reduce redundant scheduling costs; introduce historical progress fusion and idle time window information, and combine the graph attention network model to construct a scheduling constraint graph, which can comprehensively consider multiple resources, personnel and equipment status restrictions to achieve the global optimal solution of the task scheduling path.

Description

Technical Field

[0001] The present invention relates to the technical field of production scheduling, and in particular to a scheduling optimization method and system for production project management. Background Art

[0002] In modern manufacturing enterprises, production projects typically consist of multiple main projects and several sub-projects. Their scheduling involves complex resource coordination, task dependencies, and real-time change management. Traditional scheduling methods often struggle to achieve refined scheduling and globally optimal allocation in highly concurrent, multi-constrained, and dynamically changing production environments. This leads to low production efficiency, significant resource waste, and difficulty adapting to conflicts in shared resources across multiple projects. In particular, in scenarios involving multiple sub-projects, existing methods lack accurate modeling of inter-task topological relationships and scheduling priorities, making it difficult to dynamically optimize scheduling paths and resource allocation.

[0003] Therefore, a method is needed to achieve scheduling optimization at the sub-project granularity and global rescheduling optimization at the main project level to effectively improve production resource utilization and scheduling flexibility. Summary of the Invention

[0004] The present invention aims to provide a scheduling optimization method and system for production project management, which can effectively improve production resource utilization and scheduling flexibility.

[0005] A scheduling optimization method for production project management includes the following steps:

[0006] Get the list of production work items for the day;

[0007] Based on the production operation project list, project information of several production projects is obtained; feature extraction is performed on the project information to obtain a main project triple set and a sub-project triple set; wherein the main project triple contains the project number, production task, and initial project priority; the sub-project triple contains the project number, production task sub-item, and sub-project production information;

[0008] Extract production sub-projects from all main project triple sets and sub-project triple sets to obtain several production sub-project queues; each production sub-project queue represents several production project scheduling units corresponding to a single sub-project; based on any production sub-project queue, obtain the corresponding unfinished production sub-project queue;

[0009] The scheduling constraint model is used to schedule the production sub-project queue and the unfinished production sub-project queue to obtain the optimized production sub-project queue. Based on the main project triple set, several optimized production sub-project queues are rescheduled to obtain the main project scheduling queue.

[0010] Complete the optimized scheduling of the day's production operation project list based on the main project scheduling queue.

[0011] As a preferred technical solution of the present invention, the specific steps of extracting production sub-items from all main item triple sets and sub-item triple sets include:

[0012] generating a sub-project feature vector based on the sub-project triples in the sub-project triple set;

[0013] Based on the main project triples in the main project triple set, a main project guiding factor with the project number as the identifier is generated;

[0014] Perform scheduling-aware coding based on the sub-project feature vector to obtain the sub-project scheduling-aware coding;

[0015] Clustering is performed based on the scheduling perception coding of all sub-projects using a clustering model to obtain several sub-project groups;

[0016] Based on several sub-project groups, several production sub-project queues are generated. Each production sub-project queue contains several production project scheduling units for subsequent single sub-project scheduling; the production project scheduling unit contains the project number, production project scheduling unit number, production information and progress information.

[0017] As a preferred technical solution of the present invention, the specific steps of performing single sub-project scheduling on the production sub-project queue and the unfinished production sub-project queue using the scheduling constraint model include:

[0018] The unfinished production sub-project queue contains several production project scheduling units with non-empty progress information;

[0019] Merge the production sub-project queue and the unfinished production sub-project queue to obtain a comprehensive production sub-project queue; map the comprehensive production sub-project queue using the topological structure of the sub-project production information to obtain a comprehensive production sub-project graph;

[0020] The idle time window of production sub-projects is introduced; the scheduling constraint model is used to optimize the comprehensive production sub-project graph and the idle time window of production sub-projects to obtain the optimized production sub-project queue.

[0021] As a preferred technical solution of the present invention, the specific steps of optimizing the scheduling constraint model to obtain the optimized production sub-project queue include:

[0022] Based on the project number of the production project scheduling unit in the comprehensive production sub-project queue, the main project guiding factor of the production project scheduling unit is identified;

[0023] Based on the production information of the production project scheduling unit in the integrated production sub-project queue, the production scheduling characteristics required by the production project scheduling unit are identified;

[0024] According to the production scheduling characteristics and the scheduling constraint set, production resources are allocated to obtain the production scheduling constraint factor;

[0025] The scheduling constraint set specifically includes: production personnel fatigue constraint, equipment switching cooling constraint, production heat accumulation constraint, equipment production pre-maintenance constraint, and production preparation pre-constraint;

[0026] Based on the production scheduling constraint factors, a virtual production constraint time window is matched to the production scheduling characteristics; the idle time window of the production sub-item is filled according to the virtual production constraint time window of the production item scheduling unit to obtain the production sub-item time window; the production sub-item time window is fused with the comprehensive production sub-item graph to obtain the fused production sub-item graph;

[0027] Based on the fusion of production sub-project graph and main project guiding factors, the path is solved to obtain the optimized production sub-project queue.

[0028] As a preferred technical solution of the present invention, the specific steps of rescheduling a plurality of optimized production sub-project queues based on the main project triple set include:

[0029] Extract features based on the project information of the production project to obtain the production project requirements;

[0030] The main project triplet associated with the optimized production sub-project queue;

[0031] Generate a production scheduling graph for production projects based on all main project triples; where nodes represent production projects; edges represent the degree of sharing of optimized production sub-project queues shared by two different production projects;

[0032] The production schedule of production projects is sorted based on the main project scheduling priority intensity index, and the final production priority is assigned to each optimized production sub-project queue. The main project scheduling priority intensity index includes the urgency of the main project type, the scale of the main project, and the complexity of the sub-project queue.

[0033] Combine several optimized production sub-project queues with production priorities to obtain the main project scheduling queue.

[0034] As a preferred technical solution of the present invention, the basic model of the scheduling constraint model is the graph attention network model.

[0035] A scheduling optimization system for production project management, comprising:

[0036] The production project management module includes a feature description unit for obtaining a list of production operation projects for the day; obtaining project information of several production projects based on the production operation project list; extracting features from the project information to obtain a main project triple set and a sub-project triple set; wherein the main project triple set includes the project number, production task, and initial project priority; the sub-project triple set includes the project number, production task sub-project, and sub-project production information; extracting production sub-projects from all main project triple sets and sub-project triple sets to obtain several production sub-project queues; each production sub-project queue represents several production project scheduling units corresponding to a single sub-project; based on any production sub-project queue, obtaining the corresponding unfinished production sub-project queue;

[0037] The production project scheduling module includes an optimization scheduling unit, which is used to use the scheduling constraint model to perform single sub-project scheduling on the production sub-project queue and the unfinished production sub-project queue to obtain the optimized production sub-project queue; reschedule several optimized production sub-project queues based on the main project triple set to obtain the main project scheduling queue; and complete the optimized scheduling of the production operation project list for the day based on the main project scheduling queue.

[0038] The present invention has the following advantages:

[0039] 1. The present invention standardizes and semanticizes complex production project information through the construction of main project triples and sub-project triples, effectively enhancing the parsability of the scheduling model for business logic; through scheduling-aware encoding and clustering of sub-project feature vectors, it can automatically identify similar production unit tasks that can be merged, improve resource utilization and reduce redundant scheduling costs; introduces historical progress fusion and idle time window information, and combines the graph attention network model to construct a scheduling constraint graph, which can comprehensively consider various resources, personnel and equipment status constraints to achieve the global optimal solution for the task scheduling path.

[0040] 2. The present invention uses a graph sorting mechanism based on the scheduling priority of main projects to uniformly evaluate the scheduling urgency of multiple order projects, accurately determine the execution order between projects, and take into account multiple dimensions such as urgency, resource sharing, and task complexity. By clustering and integrating the same or similar sub-project tasks in different production projects, the batch processing efficiency can be maximized, the frequency of process switching and equipment idle time can be reduced, and a project scheduling method with centralized processing of common tasks and intelligent interspersal of differentiated tasks can be realized, thereby significantly improving the production line throughput and equipment utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a structural diagram of a scheduling optimization system for production project management adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0043] In this embodiment, the production project is carried out based on the manufacturing of power equipment, and will be described in the context of the manufacturing of power equipment.

[0044] Example 1, a scheduling optimization method for production project management, comprising the following steps:

[0045] Get the list of production work items for the day;

[0046] In the power equipment manufacturing scenario, the day's production operation item list is used to indicate the day's production task scheduling range, representing a list of all production items entering the production planning or scheduling system, distinguished by the item number as a unique identifier.

[0047] Based on the production operation project list, project information of several production projects is obtained; feature extraction is performed on the project information to obtain a main project triple set and a sub-project triple set; wherein the main project triple contains the project number, production task, and initial project priority; the sub-project triple contains the project number, production task sub-item, and sub-project production information;

[0048] In this method, the production tasks in the main project mainly refer to the production goals that need to be completed by the entire project, such as high-voltage switchgear assembly, electricity meter box assembly, control cabinet integration, etc., which are usually manifested as the main deliverables in the customer order; the initial project priority includes the scheduling priority set for the project at the initial stage of the system, which is determined based on factors such as contract delivery date, customer level, project amount, customer credit rating, etc., and can be high, medium, low, or numerical priority weights in form; accordingly, the production task sub-project refers to the specific process that can be executed as a scheduling unit, such as sheet metal cutting, busbar installation, electrical commissioning, injection molding processing, etc., which are the process links that constitute the main project; and the sub-project production information includes the detailed execution data of the sub-task, such as the process name, the workstation or workshop to which it belongs, the equipment used, the person in charge, the task status, the planned quantity, the completed quantity, the estimated working hours, resource constraints, etc. This information constitutes the core basis for the scheduling system to allocate personnel, equipment and time sequence relationships.

[0049] After obtaining the project information of several production projects based on the production operation project list, the specific steps of feature extraction include the following: first, read each project number from the production operation project list, and call the front-end or database interface to obtain the corresponding complete project information, including customer information, product type, delivery time, project priority, process flow, etc.; second, divide each production project into main project and sub-project levels according to the modular structure, where the main project corresponds to the overall production task (such as high-voltage switchgear assembly), and the sub-project corresponds to the schedulable process or task unit (such as sheet metal processing, wire harness welding); then, extract the project number, project name, customer affiliation, initialization, etc. from the main project information. The main project triples are constructed based on key fields such as initial priority and delivery date; the subtask number, project to which it belongs, process name, equipment resources, person in charge, planned output, status, etc. are extracted from the subproject information to construct the subproject triples; then, all structured information is standardized, including field unification, coding mapping (such as priority mapping to numerical values), time format normalization, and data normalization operations (such as Min-Max or Z-score normalization of numerical features such as output, estimated working hours, and process duration). Finally, a main project triple set and subproject triple set with clear structure and unified standards are formed, providing a unified data input basis for subsequent scheduling modeling and cluster analysis.

[0050] Extract production sub-projects from all main project triple sets and sub-project triple sets to obtain several production sub-project queues; each production sub-project queue represents several production project scheduling units corresponding to a single sub-project;

[0051] The specific steps of extracting production sub-projects from all main project triple sets and sub-project triple sets include:

[0052] generating a sub-project feature vector based on the sub-project triples in the sub-project triple set;

[0053] The key attribute information of each sub-project is extracted from the sub-project triple set as input. The input data structure is a set of structured triple records. These triples are vectorized through feature engineering. First, categorical variables (such as process name and equipment type) are one-hot encoded or embedded, numerical variables (such as output and working hours) are normalized, and time variables are time-windowed (such as morning and evening shift identifiers). Finally, a high-dimensional, trainable sub-project feature vector (usually a dense floating-point vector) is constructed to characterize scheduling behavior characteristics. The output of this step is a list of sub-project feature vectors, each vector corresponding to a sub-project, which constitutes the original input for subsequent encoding and clustering models. The beneficial effect of this process is that it realizes the conversion from the original triple to a unified vector space, making the structured task data comparable and measurable, and providing a numerical expression basis for subsequent scheduling tendency analysis and clustering processing.

[0054] Based on the main project triples in the main project triple set, main project guiding factors are generated, with the project number as the identifier. Taking the main project triple set as input and the project number as the index, multiple main project triplets under the project are integrated to form a structured main project information block. Through feature screening and semantic aggregation, different fields are standardized: for example, priority (high / medium / low) is mapped to weight values, customer attributes are converted into delivery sensitivity scores, and delivery time is converted into time urgency (such as the number of days remaining). Contextual scoring logic can also be added for factors such as project size and importance. The output is a set of main project guiding factors, with the project number as the primary key, which is used to subsequently disseminate scheduling importance information at the sub-project level. This step has the beneficial effect of establishing global tonality constraints for task scheduling, allowing subsequent sub-project processing to be clustered not only based on the process characteristics of the tasks themselves but also by incorporating business-oriented information from the main project, enhancing the global rationality and responsiveness of scheduling clustering.

[0055] Perform scheduling-aware coding based on the sub-project feature vector to obtain the sub-project scheduling-aware coding;

[0056] Each sub-project feature vector is concatenated (or fused) with the guiding factor under its corresponding project number and further embedded through a neural network encoder (such as MLP or autoencoder) to generate a sub-project scheduling-aware encoding. This encoding retains the process and resource characteristics of the sub-project, while integrating the scheduling priority and delivery intention of the main project. As a result, the final representation vector of each sub-project has the dual characteristics of project structure and project orientation, capturing the relationship between the task's own process complexity and the upper-level scheduling strategy. The beneficial effect of this step is that it enables the scheduling model to identify key processes in high-priority production projects, thereby guiding the formation of clusters of key tasks in subsequent clustering.

[0057] Clustering is performed based on the scheduling perception coding of all sub-projects using a clustering model to obtain several sub-project groups;

[0058] Use clustering algorithms to group sub-projects. For example, DBSCAN can be selected to automatically determine the number of clusters, or the improved KMeans can be used to set the initial number of clusters based on business-side prior knowledge. The input data structure is a set of sub-project vectors, and the output is a number of sub-project groups. The sub-projects in each group are highly similar in terms of scheduling tendencies, process procedures, equipment requirements, etc. Clustering can be combined with the weights of the main project factors and dynamically adjust the distance measurement method (such as using scheduling-aware distance) to avoid sub-projects with high project priority being diluted by low-priority clusters. The beneficial effect is that the sub-projects within the production operation project list are classified according to scheduling-related attributes, providing a grouping basis for subsequent resource collaboration, batch scheduling, and task merging, thereby improving the controllability and production efficiency of the scheduling system.

[0059] Based on several sub-project groups, several production sub-project queues are generated. Each production sub-project queue contains several production project scheduling units for subsequent single sub-project scheduling. The production project scheduling unit contains the project number, production project scheduling unit number, production information and progress information.

[0060] Each group is restructured, and each sub-project within it is parsed into the smallest scheduling unit, namely the production project scheduling unit. This unit data structure includes the project number (associated with the parent project), the production project scheduling unit number (uniquely identifies this scheduling task), production information (such as process, equipment, and material requirements), current progress (completed / not started / paused), and man-hour forecasts. Each sub-project group is constructed as a production sub-project queue, maintaining consistency and commonality in project allocation resources for subsequent targeted scheduling optimization. The output is several structured scheduling queues, each representing a set of tasks that can be independently optimized. The beneficial effect is that it enables the fusion of sub-tasks between main project tasks (normalization of similar processes), breaks down project barriers for resource rearrangement and scheduling, thereby improving capacity utilization and avoiding scheduling fragmentation.

[0061] Based on any production sub-project queue, obtain the corresponding unfinished production sub-project queue; perform feature matching based on the sub-project name to obtain the corresponding unfinished production sub-project queue;

[0062] The scheduling constraint model is used to schedule single sub-projects in the production sub-project queue and the unfinished production sub-project queue to obtain the optimized production sub-project queue; the basic model of the scheduling constraint model is the graph attention network model; the graph attention network model is used to model the dependency relationship and resource constraint relationship between tasks to achieve intelligent optimization of complex scheduling graphs.

[0063] The specific steps for scheduling a single sub-project in the production sub-project queue and the unfinished production sub-project queue using the scheduling constraint model include:

[0064] The unfinished production sub-project queue contains several production project scheduling units with non-empty progress information;

[0065] Extract the list of unfinished production tasks and filter out the task units that are currently in the unstarted, in-progress or suspended states. The input data includes the production project scheduling unit records of all unfinished sub-projects. Each record contains the project number, production project scheduling unit number, completed working hours / quantity, remaining workload, task status, etc.; based on the judgment standard that the progress information is not empty (such as completed working hours > 0 or status ≠ completed), filter out the task units that still need to participate in subsequent scheduling optimization; the output is a queue of unfinished production sub-projects, each of which is a structured production project scheduling unit record; the beneficial effect of this step is to integrate historical unfinished tasks with today's planned scheduling, so that the scheduling model has continuity and traceability, avoid resource idleness and repeated scheduling, and improve the system's ability to respond to actual production status.

[0066] Merge the production sub-project queue and the unfinished production sub-project queue to obtain a comprehensive production sub-project queue; map the comprehensive production sub-project queue using the topological structure of the sub-project production information to obtain a comprehensive production sub-project graph (directed acyclic graph);

[0067] The current production sub-project queue is unified and integrated with the unfinished sub-project production project scheduling units left over from the past to form a complete set of tasks to be scheduled; the data input includes: the newly generated production sub-project queue (from the clustering output), the unfinished production sub-project queue obtained in the previous step; each record is a production project scheduling unit, and the structure contains at least: project number, process name, task status, required equipment, estimated working hours, resource requirements, etc.; the merging method usually adopts the task-level deduplication and status update mechanism, such as overwriting the execution status of units with the same task number and merging the remaining quantities of overlapping tasks. The output is a comprehensive production sub-project queue, a unified set of production project scheduling units containing all sub-projects to be processed; the beneficial effect of this step is to unify and integrate yesterday's unfinished tasks with today's newly issued scheduling tasks, to construct a real and complete task input space, and to lay an accurate data foundation for the subsequent construction of the scheduling diagram; the topological structure of the sub-project production information refers to the sequential dependency, resource flow path and execution logic connection method between each process or production task within the sub-project;

[0068] The comprehensive production sub-project queue is used as input, and the production information of each production project scheduling unit is recorded with the process name, process sequence number (or preceding process field), the equipment and resource occupancy information. By identifying the relationship between the previous and next processes of each production project scheduling unit, directed edges are established between scheduling tasks, with nodes representing production project scheduling units and edges representing process dependencies. A DAG (directed acyclic graph) is generated using a process flow template, product BOM structure, or a predefined process sequence library. The output is a comprehensive production sub-project graph, in which each node has task execution information and each edge represents the scheduling sequence logic. This graph structure is the basis for subsequent path planning and resource scheduling, and is used to clarify the task execution order, construct a process constraint map, prevent parallel conflicts and scheduling reversals, and ensure the physical feasibility and security of the scheduling plan.

[0069] The idle time window of production sub-items is introduced; the scheduling constraint model is used to optimize the comprehensive production sub-item graph and the idle time window of production sub-items to obtain the optimized production sub-item queue;

[0070] The specific steps for optimizing the production sub-project queue in the scheduling constraint model include:

[0071] Based on the project number of the production project scheduling unit in the comprehensive production sub-project queue, the main project guiding factor of the production project scheduling unit is identified;

[0072] Based on the production information of the production project scheduling unit in the integrated production sub-project queue, the production scheduling characteristics required by the production project scheduling unit are identified;

[0073] For each production project scheduling unit, its structured production information fields are extracted, including process type, resource requirements, equipment model, required process time, bill of materials, etc. The input data is the production project scheduling unit record; the output is the scheduling feature vector of each unit; the data structure is a dense feature vector, which encodes the task load, resource preference, execution characteristics, etc. The feature extraction method uses rule mapping (such as mapping injection molding assembly to equipment type = injection molding machine, process load = high) or embedded coding model to convert task information into a standard expression that can be used for scheduling calculations, providing an input feature basis for subsequent time window matching and path optimization.

[0074] According to the production scheduling characteristics and the scheduling constraint set, production resources are allocated to obtain the production scheduling constraint factor;

[0075] Use the pre-trained production resource allocation model to identify production scheduling features and output production scheduling constraint factors;

[0076] The pre-trained production resource allocation model can be constructed and trained by the BP neural network model. The specific steps include: extracting the production project scheduling unit records of the past year from the historical scheduling log as the training set. These records contain input features such as task process, equipment type, product specification, batch, working hours, main project priority, personnel shift, and the constraint factor labels actually assigned by the manual or expert system at that time; such as the length of the rest interval, equipment cooling buffer, allowable heat load peak, pre-maintenance insertion mark and preparation lead time; all numerical features are normalized, and the category features are encoded by one-hot or embedding; then the samples are The data was randomly divided into an 80% training set and a 20% validation set. The data was input into a BP network with two hidden layers. The multi-output mean squared error was used as the loss function, and the Adam optimizer (initial learning rate 0.001) was used. During training, the loss and R² indicators were monitored on the validation set. If the validation loss did not decrease significantly within multiple training rounds, early stopping was triggered to save the weights and export the model. The model's final output was the production scheduling constraint factor, which corresponds to the five types of constraint parameters that each input production project scheduling unit must comply with, including fatigue, cooling, heat, maintenance, and preparation pre-conditioning. This factor can be directly called by subsequent scheduling algorithms to achieve fast, data-driven resource allocation decisions.

[0077] The scheduling constraint set specifically includes: production personnel fatigue constraint, equipment switching cooling constraint, production heat accumulation constraint, equipment production pre-maintenance constraint, and production preparation pre-constraint;

[0078] Production staff fatigue constraints refer to the need to consider the continuous working hours and rest time of production staff during the production scheduling process to avoid efficiency loss, operational errors or safety hazards caused by excessive work. This is especially critical in key processes (such as high-voltage wiring and precision welding); rules such as a single person cannot work continuously for more than X hours and a daily total working time limit can be set; the scheduling constraint model will automatically insert a mandatory rest window or exclude tasks from being scheduled during this time period; the equipment switching cooling constraint means that when switching equipment between different process tasks (for example, switching from injection molding to molding), physical processing processes such as cleaning, cooling, and preheating may be required to avoid material residue or temperature abnormalities affecting product quality. For example: it is necessary to determine whether the process type has changed for adjacent tasks and automatically insert a cooling time buffer (such as 30 minutes or dynamic prediction by equipment type); otherwise, production scheduling will lead to equipment contamination or process failure;

[0079] The production heat accumulation constraint refers to the fact that when certain hot processing tasks (such as welding, drying, and molding) are scheduled continuously, the heat load in the equipment or workshop continues to increase. Exceeding the threshold will cause equipment overheating and unstable production environment. It is necessary to monitor the density or frequency of the same type of high-heat tasks per unit time; insert low-heat tasks or adjustment segments waiting for cooling into the scheduling path to form a heat load distribution diagram and balance the production rhythm; equipment production pre-maintenance constraint refers to predictive maintenance for key production equipment (such as injection molding machines, CNC machine tools, etc.) based on their usage frequency, historical health data or early warning indicators, and arrange maintenance or calibration before a failure occurs to ensure stable operation of the equipment; predict future maintenance windows based on equipment operation data, proactively block the production route at that time, and allow tasks to be rescheduled only after maintenance is completed; the scheduling engine must have the ability to embed maintenance nodes;

[0080] Production preparation pre-constraints indicate that certain processes require the preparation of specialized tools, fixtures, molds, gauges, and materials before they begin. If these pre-preparation tasks are not completed, the main task cannot start on time. The pre-preparation dependencies of each production project scheduling unit must be analyzed to ensure that these pre-preparation tasks are scheduled before the main task and have sufficient time to complete. Constraints are modeled by defining dependency edges from tool preparation to the main process.

[0081] Based on the production scheduling constraint factors, a virtual production constraint time window is matched to the production scheduling characteristics; the idle time window of the production sub-item is filled according to the virtual production constraint time window of the production item scheduling unit to obtain the production sub-item time window; the production sub-item time window is fused with the comprehensive production sub-item graph to obtain the fused production sub-item graph;

[0082] Based on the fusion of production sub-project graph and main project guiding factors, the path is solved to obtain the optimized production sub-project queue;

[0083] After extracting the production scheduling constraints, the scheduling characteristics of each production project scheduling unit are combined with its corresponding scheduling constraints to predict the time period in which the task can theoretically be scheduled, thereby generating a virtual production constraint time window. This process combines multiple methods: for example, hard rule reasoning can directly exclude explicitly unavailable time periods such as maintenance windows and equipment cooldown periods; a prediction model based on resource load history can assess the likelihood of future resource idleness; and a resource schedulability weight model is used to identify the matching priority between tasks and resources. Ultimately, each production project scheduling unit will obtain a set of candidate virtual time periods, providing constraint boundaries for subsequent path planning.

[0084] Based on the existing idle time windows of production sub-projects, the above virtual time windows are intersected and matched with actual resource gaps. Input includes the available time periods for each piece of equipment, workstation, and worker currently obtained from the MES or forecasting system, as well as the task time windows inferred in the previous step. Using a sliding window algorithm or interval tree structure, we can quickly identify which tasks are truly executable in which time periods, thereby forming a precise production sub-project time window, which is recorded as the legally available time range for each production project scheduling unit on the scheduling resources.

[0085] The time window information of production sub-projects is integrated with the comprehensive production sub-project graph to form a composite scheduling graph with task dependencies and resource time constraints. The integration process includes embedding the adjustable time window information of each graph node (i.e., production project scheduling unit) and adjusting the edge weights according to resource characteristics and time characteristics, such as setting penalties for task connections across shifts or increasing edge costs for paths that require equipment replacement. This structured time-constrained scheduling graph not only retains the process sequence of tasks, but also superimposes the schedulable time dimension, thus laying the foundation for multi-objective path optimization. Path optimization is carried out based on the integrated scheduling graph and the main project guiding factors of each production project scheduling unit. Path solving; the path optimization method can choose heuristic search based on graph structure (such as topological sorting with priority, A*), or introduce deep learning scheduling strategy based on graph neural network, or combine constraint satisfaction and greedy algorithm for multi-objective optimization; the main project guidance factor plays a strategic role in this stage, which is used to dynamically adjust the scheduling priority of task nodes, such as giving priority to tasks with approaching delivery dates and giving priority to resource matching for tasks with high customer levels; the final output is an optimized scheduling sequence that meets multiple conditions such as scheduling order, resource conflict, time window and policy guidance, that is, an optimized production sub-project queue, in which each task has a clear execution order, scheduling time period and resources used.

[0086] Reschedule several optimized production sub-project queues based on the main project triple set to obtain the main project scheduling queue;

[0087] Complete the optimized scheduling of the day's production work project list based on the main project scheduling queue;

[0088] The specific steps for rescheduling several optimized production sub-project queues based on the main project triple set include:

[0089] Extract features based on the project information of the production project to obtain the production project requirements;

[0090] Taking each production project in the production operation project list as input, the system accesses project information (such as customer type, product model, planned delivery time, order type, and urgency) to extract data features representing project scheduling preferences and priorities. These features are mapped into production project demand vectors using structured fields. For example, vector dimensions represent delivery urgency, project volume (measured by the number of processes and man-hours), customer level, and whether it is a special order. The data structure is a key-value mapping table, with a project information dictionary as input and a demand vector corresponding to each project as output. This step has the beneficial effect of converting unstructured project attributes into data that can be used for scheduling modeling, providing a decision-making basis for subsequent sorting and mapping.

[0091] Associate the main project triplet of the optimized production sub-project queue; associate each optimized production sub-project queue with the main project triplet from which it originated, mainly through the main project number to achieve mapping;

[0092] A production scheduling graph for production projects is generated based on all main project triplets. Nodes represent production projects, and edges represent the degree of sharing between optimized production sub-project queues shared by two different production projects. By analyzing whether multiple projects in the optimized sub-project queue share the same sub-project (i.e., tasks in the same process can be scheduled together), edges are constructed to represent the sharing intensity (e.g., the number of shared tasks, sharing weight) between the two projects. This graph can be represented in the form of an adjacency matrix or an edge-weighted graph, with the edge weight being the number of shared sub-project queues or the normalized weight.

[0093] The production schedule of production projects is sorted based on the main project scheduling priority intensity index, and the final production priority is assigned to each optimized production sub-project queue. The main project scheduling priority intensity index includes the urgency of the main project type, the scale of the main project, and the complexity of the sub-project queue.

[0094] In the key step of production scheduling based on the main project scheduling priority intensity index, it is first necessary to comprehensively evaluate the scheduling urgency of each main project; this evaluation consists of three core dimensions: the first is the type of urgency of the main project, which mainly reflects whether the project delivery deadline is approaching, whether the customer level is high, or whether the project is an urgent order; the second is the scale of the main project, which is usually measured by the number of sub-projects included in the project, the overall estimated working hours or the total resource consumption; the third is the complexity of the sub-project queue, which specifically includes the topological depth of the process path, the dependency density between tasks, the frequency of resource conflicts, etc. These three dimensions are quantified and normalized and then integrated into a scoring vector for the main project, which is used to express the priority intensity of the project in the scheduling system. Use a sorting algorithm for scheduling priority reasoning: For example, a weight-aware topological sorting algorithm can be used to comprehensively consider the urgency of individual projects and the coupling effects in the overall task network. Each main project will receive a scheduling priority value, and based on this, a global execution order will be assigned to its associated optimized sub-project queues. This process not only ensures the rationality of task scheduling, but also effectively introduces resource competition relationships and scheduling coordination between projects, making the entire scheduling system have a global optimization tendency.

[0095] Combine several optimized production sub-project queues with production priorities to obtain the main project scheduling queue. According to the scheduling priority of the main project, combine and sort its subordinate optimized sub-project queues to generate the main project scheduling queue. Each sub-project queue of the main project is classified and sorted according to the priority block to which it belongs, forming a hierarchical and orderly task scheduling structure.

[0096] This example, based on a power equipment production scenario, demonstrates how to apply the scheduling optimization method to collaboratively and intelligently schedule orders from multiple customers. For example, an electrical manufacturing company receives the following production project list on a certain day: a total of five production projects from three customers, primarily electricity meter boxes and meter verification boxes.

[0097] Each project is parsed into a main project triple, extracting information such as project number, customer affiliation, delivery date, product type, and priority. Each project is then split into multiple sub-project triples, each representing an independently schedulable process task, such as wiring assembly, housing injection molding, and final inspection and packaging. For all sub-projects, the process type, resource requirements, task volume, and equipment type are extracted as feature vectors. Clustering algorithms (such as dynamic K-means with time window constraints) are used to merge and categorize tasks, resulting in multiple production sub-project queues. For example, injection molding assembly tasks from multiple projects are merged and scheduled to form a unified, cross-project production sub-project queue to maximize batch processing and resource utilization.

[0098] Unfinished tasks of the same type are retrieved from the previous day's production log, merged with new tasks, and constructed into a comprehensive task queue. A directed acyclic graph is constructed based on the process flow to represent the execution order and dependency relationship between sub-projects. For each sub-project queue, a scheduling constraint model is introduced, including: worker fatigue constraint (continuous working hours < 6 hours); equipment switching cooling interval (such as 30 minutes when switching an injection molding machine); heat load accumulation control (drying / welding tasks are dispersed); equipment pre-maintenance window (task arrangement is blocked according to the prediction model); tooling preparation lead time (assembly fixtures must be prepared 1 hour in advance).

[0099] Combined with the idle time window of actual resources, the schedulable window of each task unit is inferred, and path optimization is performed based on the scheduling diagram and resource constraints to output a production scheduling sequence that meets all physical and organizational conditions; the scheduling intensity index of each main project is calculated, including: type urgency (distinguishing between urgent orders and non-urgent orders); project scale (such as the number of sub-projects and the total working hours); sub-project queue complexity (such as dependency graph depth and resource conflict density); the main project production scheduling diagram is constructed based on the degree of sharing of the same sub-project queue among multiple projects; the global scheduling order of the main project is determined through topological sorting and weight scoring, and finally, the main project scheduling queue is generated, and the corresponding optimized production sub-project queues are issued one by one according to priority, and imported into the MES system for actual task allocation on the day. As a result, high-priority tasks will occupy resources first, while low-priority tasks will be delayed or waited to ensure the on-time delivery of key projects.

[0100] Example 2, a scheduling optimization system for production project management, see Figure 1 Shown, including:

[0101] The production project management module includes a feature description unit for obtaining a list of production operation projects for the day; obtaining project information of several production projects based on the production operation project list; extracting features from the project information to obtain a main project triple set and a sub-project triple set; wherein the main project triple set includes the project number, production task, and initial project priority; the sub-project triple set includes the project number, production task sub-project, and sub-project production information; extracting production sub-projects from all main project triple sets and sub-project triple sets to obtain several production sub-project queues; each production sub-project queue represents several production project scheduling units corresponding to a single sub-project; based on any production sub-project queue, obtaining the corresponding unfinished production sub-project queue;

[0102] The production project scheduling module includes an optimization scheduling unit, which is used to use the scheduling constraint model to perform single sub-project scheduling on the production sub-project queue and the unfinished production sub-project queue to obtain the optimized production sub-project queue; reschedule several optimized production sub-project queues based on the main project triple set to obtain the main project scheduling queue; and complete the optimized scheduling of the production operation project list for the day based on the main project scheduling queue.

[0103] It should be understood that those skilled in the art may make improvements or modifications based on the above description, and all such improvements and modifications shall fall within the scope of protection of the appended claims. Any portion of this specification not described in detail is prior art known to those skilled in the art.

Claims

1. A scheduling optimization method for production project management, characterized in that: The following steps are involved: Get the list of production work items for the day; Based on the production operation project list, project information of several production projects is obtained; feature extraction is performed on the project information to obtain a main project triple set and a sub-project triple set; wherein the main project triple contains the project number, production task, and initial project priority; the sub-project triple contains the project number, production task sub-item, and sub-project production information; Extract production sub-projects from all main project triple sets and sub-project triple sets to obtain several production sub-project queues; each production sub-project queue represents several production project scheduling units corresponding to a single sub-project; based on any production sub-project queue, obtain the corresponding unfinished production sub-project queue; The scheduling constraint model is used to schedule the production sub-project queue and the unfinished production sub-project queue to obtain the optimized production sub-project queue. Based on the main project triple set, several optimized production sub-project queues are rescheduled to obtain the main project scheduling queue. Complete the optimized scheduling of the day's production work project list based on the main project scheduling queue; The specific steps of extracting production sub-projects from all main project triple sets and sub-project triple sets include: generating a sub-project feature vector based on the sub-project triples in the sub-project triple set; Based on the main project triples in the main project triple set, a main project guiding factor with the project number as the identifier is generated; Each sub-project feature vector is concatenated with the main project guiding factor under its corresponding project number, and further embedded through a neural network encoder to generate a sub-project scheduling perception code; Clustering is performed based on the scheduling perception coding of all sub-projects using a clustering model to obtain several sub-project groups; Based on several sub-project groups, several production sub-project queues are generated. Each production sub-project queue contains several production project scheduling units for subsequent single sub-project scheduling. The production project scheduling unit contains the project number, production project scheduling unit number, production information and progress information. The specific steps for scheduling a single sub-project in the production sub-project queue and the unfinished production sub-project queue using the scheduling constraint model include: The unfinished production sub-project queue contains several production project scheduling units with non-empty progress information; Merge the production sub-project queue and the unfinished production sub-project queue to obtain a comprehensive production sub-project queue; map the comprehensive production sub-project queue using the topological structure of the sub-project production information to obtain a comprehensive production sub-project graph; The idle time window of production sub-items is introduced; the scheduling constraint model is used to optimize the comprehensive production sub-item graph and the idle time window of production sub-items to obtain the optimized production sub-item queue; The specific steps for rescheduling several optimized production sub-project queues based on the main project triple set include: Extract features based on the project information of the production project to obtain the production project requirements; The main project triplet associated with the optimized production sub-project queue; Generate a production scheduling graph for production projects based on all main project triples; where nodes represent production projects; edges represent the degree of sharing of optimized production sub-project queues shared by two different production projects; The production schedule of production projects is sorted based on the main project scheduling priority intensity index, and the final production priority is assigned to each optimized production sub-project queue. The main project scheduling priority intensity index includes the urgency of the main project type, the scale of the main project, and the complexity of the sub-project queue. Combine several optimized production sub-project queues with production priorities to obtain the main project scheduling queue; The basic model of the scheduling constraint model is the graph attention network model.

2. The scheduling optimization method for production project management according to claim 1, characterized in that: The specific steps for optimizing the production sub-project queue in the scheduling constraint model include: Based on the project number of the production project scheduling unit in the comprehensive production sub-project queue, the main project guiding factor of the production project scheduling unit is identified; Based on the production information of the production project scheduling unit in the integrated production sub-project queue, the production scheduling characteristics required by the production project scheduling unit are identified; According to the production scheduling characteristics and the scheduling constraint set, production resources are allocated to obtain the production scheduling constraint factor; The scheduling constraint set specifically includes: production personnel fatigue constraint, equipment switching cooling constraint, production heat accumulation constraint, equipment production pre-maintenance constraint, and production preparation pre-constraint; Based on the production scheduling constraint factors, a virtual production constraint time window is matched to the production scheduling characteristics; the idle time window of the production sub-item is filled according to the virtual production constraint time window of the production item scheduling unit to obtain the production sub-item time window; the production sub-item time window is fused with the comprehensive production sub-item graph to obtain the fused production sub-item graph; Based on the fusion of production sub-project graph and main project guiding factors, the path is solved to obtain the optimized production sub-project queue.

3. A scheduling optimization system for production project management, characterized in that: The system is a scheduling optimization method for production project management as described in any one of claims 1-2, comprising: The production project management module includes a feature description unit for obtaining a list of production operation projects for the day; obtaining project information of several production projects based on the production operation project list; extracting features from the project information to obtain a main project triple set and a sub-project triple set; wherein the main project triple set includes the project number, production task, and initial project priority; the sub-project triple set includes the project number, production task sub-project, and sub-project production information; extracting production sub-projects from all main project triple sets and sub-project triple sets to obtain several production sub-project queues; each production sub-project queue represents several production project scheduling units corresponding to a single sub-project; based on any production sub-project queue, obtaining the corresponding unfinished production sub-project queue; The production project scheduling module includes an optimization scheduling unit, which is used to use the scheduling constraint model to perform single sub-project scheduling on the production sub-project queue and the unfinished production sub-project queue to obtain the optimized production sub-project queue; reschedule several optimized production sub-project queues based on the main project triple set to obtain the main project scheduling queue; and complete the optimized scheduling of the production operation project list for the day based on the main project scheduling queue.

Images