Target system for APS and unified construction method and extension system thereof

Abstract

The invention discloses a target system for advanced planning and production scheduling (APS) and a unified construction method and an extension system thereof. By establishing a mapping relationship between business constraints and decision elements, standardizing a data structure and a management process of decision factors, and combining an operator target construction link with a template scene library, automatic construction and dynamic reconstruction of an optimization target are realized. In order to solve the problem that the business target cannot be satisfied, the invention provides a decision-making target switching mechanism and a constraint softening mechanism so as to generate an executable scheme which can be satisfied to the greatest extent. And meanwhile, weight management methods such as weight cutting, normalization, sensitive scanning and automatic optimization are provided, so that the numerical stability and interpretability of the target construction process are ensured. The system supports cloud, local and hybrid deployment, can improve solution robustness, business adaptability and decision efficiency, and promotes engineering landing and popularization of the APS in a manufacturing industry scene.

Description

Technical Field

[0001] This invention relates to the fields of Advanced Planning and Scheduling (APS) and intelligent manufacturing and supply chain optimization, specifically to a target system and its unified construction method for the automated construction and dynamic reconstruction of decision objectives in APS scenarios. It covers technologies such as data structure and management of decision factors, operator-based target construction links, templated scenario library, weight processing and sensitivity analysis, hard constraint softening mechanism and engineering deployment. Background Technology

[0002] Advanced Planning and Scheduling (APS) systems are widely used in manufacturing, supply chain management, and resource optimization to generate executable production plans or scheduling schemes under various business constraints. A typical APS model needs to simultaneously handle multiple complex constraints such as capacity constraints, material availability constraints, process routing constraints, equipment changeover constraints, shift and calendar constraints, and build multi-objective optimization functions based on these constraints, such as minimizing delays, minimizing changeover times, maximizing equipment utilization, and minimizing work-in-process inventory.

[0003] Existing APS systems typically rely on pre-defined objective functions and their weight combinations to drive the solution. However, in real-world business environments, production loads, order structures, resource status, and operational strategies change frequently, causing the original objective functions or weight configurations to gradually become ineffective. When decision-making objectives cannot be adjusted in a timely manner to adapt to business changes, the system may still be driven by outdated objective functions, resulting in planning outcomes inconsistent with current business priorities, constraints, or risk appetite. Such outcomes are often unexecutable, deviate from business objectives, or require significant manual intervention, severely weakening the adaptability and stability of the APS system.

[0004] In addition, existing APS systems generally have the following problems in target construction:

[0005] 1. Fixed target construction method: The target function is usually set once during the deployment phase, lacking an extensible target construction mechanism and making it difficult to support dynamic adjustment;

[0006] 2. Lack of a unified decision-making factor management system: the sources of target items are scattered, the structure is inconsistent, and they are difficult to maintain, reuse, or interpret;

[0007] 3. Lack of ability to handle infeasible scenarios: When constraint conflicts make the solution infeasible, the system lacks an automated target switching or constraint softening mechanism;

[0008] 4. Insufficient weight management capabilities: The lack of numerical stability methods such as weight trimming, normalization, and sensitivity analysis leads to unstable or uninterpretable solution results;

[0009] 5. Lack of engineering-based scalability: It is difficult to support template-based scene switching, operator-based expansion, or multiple deployment modes.

[0010] The aforementioned problems make it difficult for APS systems to adapt to complex and ever-changing manufacturing scenarios, thus limiting their promotion and widespread adoption in the industrial sector. Summary of the Invention

[0011] To overcome the aforementioned problems, this invention proposes a target system for APS and its unified construction method and extension architecture for the automated construction and dynamic reconstruction of decision targets in APS scenarios. By defining the mapping relationship between constraints and decision elements, unifying the data structure and processing flow of decision factors, operator-based target construction links, and template-based scenario libraries, this invention modularizes, extensibly expands, and interoperates the constraint, decision factor, and objective function construction processes. This improves the convenience and interpretability of decision target construction and reconstruction, achieves effective mapping from business intent to optimization targets, and ensures the consistency between business objectives and decision objectives.

[0012] To achieve the above objectives, this invention proposes a target system for APS, the main modules of which and their functions are as follows:

[0013] The decision factor management module maintains the metadata and related information of all decision factors, serving as the core data source for objective function construction / reconstruction. The maintained information includes decision factor name, status, weight, and detailed decision factor information. It supports template import / export and interaction with the objective function module, registration module, settings module, decision factor clearing module, and query module.

[0014] Decision Element Construction Module: Based on the constraint interface provided by the optimization model, it automatically or semi-automatically generates a set of decision elements for constructing decision factors.

[0015] Registration Module: This module registers decision factors to the decision factor management module and supports registration from multiple sources, including preset decision factors and decision factors generated by the first operator. During registration, the name, status, and weight of the decision factor must be specified. Registration information also includes the corresponding set of decision elements and constraint conditions. Default values ​​can be set for status and weight for faster registration.

[0016] The query module provides the ability to retrieve information on decision factors, supporting queries by name, by status, and all queries, and returns metadata of decision factors that meet the criteria, as well as their associated decision elements and constraint information.

[0017] Decision Factor Information Display Module: This module displays decision factors and their related information in a structured view, supporting export, filtering, and historical record viewing for easy user review and decision-making.

[0018] The configuration module provides configuration capabilities for individual or batch decision factors, including status settings, name modifications, weight settings, and records change logs for retrospective purposes.

[0019] Decision factor clearing module: Supports two clearing strategies, shallow clearing and deep clearing, to meet the needs of different scenarios.

[0020] Decision Target Template Module: Used to export, store, and quickly import preset or historical decision target schemes, such as delivery date priority, cost priority, capacity utilization priority, and various user-defined templates.

[0021] Objective function module: Used to construct / reconstruct the decision objective through the second operator before each solution, output an optimization objective expression acceptable to the solver, and support setting the optimization direction and settlement parameters.

[0022] Solution Result Evaluation Module: Used to perform multi-dimensional quantitative analysis, visualization, and evaluation of the results after the solution is completed.

[0023] The present invention also provides a unified construction method for the system, which is performed according to the following steps:

[0024] Step S1: Initialize the decision factor set: Import or define the initial decision factor set through the decision factor management module. Initialization includes: decision factor name, default enabled status, default weight, associated decision element set, and associated constraint set. The system generates corresponding metadata records and writes them to the initialization log to support auditing and backtracking.

[0025] Step S2 Decision Element Generation: In the decision element construction module, the corresponding set of decision elements is generated synchronously by establishing constraints, which serves as the basic data for the subsequent construction of decision factors.

[0026] Step S3 Decision Element Generation: By using the default first operator, specifying the built-in first operator, or a custom first operator, the set of decision elements generated in step S2 is automatically or semi-automatically constructed into the corresponding decision factors.

[0027] Step S4: Decision Factor Registration: Register the pre-set decision factors or the decision factors generated by the first operator to the decision factor management module. Registration information includes: decision factor name, activation status, weight (default value can be used), corresponding set of decision elements, and corresponding set of constraints. Metadata is written during registration, and a change log is generated, recording the operator, timestamp, registration source, and related parameters to support auditing and version management.

[0028] Step S5: Decision Objective Template Selection: Preferably, users can select a preset template or a custom template through the decision objective template module. The template content includes: a set of decision factors, weight configuration, optimization direction (minimization / maximization), and solution parameters (solution accuracy, time constraints, heuristic strategies, etc.). After template selection, the system imports the template content into the decision factor management module.

[0029] Step S6 Objective Function Construction / Reconstruction: Before each solution, the objective function module calls the second operator to generate an objective expression acceptable to the solver based on the currently enabled decision factors and their weights.

[0030] Step S7 Solving and Result Evaluation: After the solver completes the optimization solution, it performs the following through the solution result evaluation module: decision factor score analysis, multi-objective trade-off analysis, visualization, and evaluation report generation.

[0031] Step S8 Iterative Optimization or Factor Cleaning: Based on the evaluation results, users can adjust the status, name, or weight of decision factors through the settings module, and the decision factor cleanup module can perform shallow or deep cleanup. After the adjustments are completed, the system can proceed to the next round of solution verification or be directly used for production deployment.

[0032] This invention also discloses a decision objective switching mechanism based on the system, which softens infeasible hard constraints to allow the optimization problem to continue solving and return the optimal deviation solution, thereby ensuring that the system can still output an executable plan / scheduling scheme even in infeasible situations. This process can be executed automatically by the software system or semi-automatically with user interaction. The specific steps are as follows:

[0033] Step S1: Load the optimization problem model;

[0034] Step S2: Construct / reconstruct the objective function and perform optimization. After the solution is completed, proceed to the next step.

[0035] Step S3: The system performs a feasibility check on the solution returned by the solver, or determines feasibility based on the solver returning no solution or a feasible solution.

[0036] 1. If the process is feasible, proceed to step S4 to conduct result analysis and quantitative evaluation;

[0037] 2. If the process is not feasible, proceed to step S7 to perform constraint transformation and decision element generation;

[0038] Step S4: Perform results analysis and quantitative evaluation, and then proceed to the next step;

[0039] Step S5: Determine satisfaction level.

[0040] 1. If the satisfactory process enters the termination state, output the final optimization result;

[0041] 2. If you are not satisfied with the process, proceed to step S6;

[0042] Step S6: Set the weights of the decision factors. After the adjustment is completed, the process returns to step S2, reconstructs the objective function and executes the solution to achieve iterative optimization.

[0043] Step S7: Constraint Transformation and Decision Element Generation in Infeasible Scenarios:

[0044] 1. Transform infeasible constraints;

[0045] 2. Generate a new set of decision elements;

[0046] 3. Register the generated decision elements as new decision factor candidates;

[0047] After completion, the process proceeds to step S8.

[0048] Step S8: Construct and register decision factors. After completion, the process returns to step S2, reconstructs the objective function, and executes the solution.

[0049] The present invention also discloses a hard constraint softening mechanism based on the system, the specific steps of which are as follows:

[0050] Step S1: Remove the original infeasible hard constraints;

[0051] Step S2: Convert to the corresponding soft constraints and generate a set of decision elements;

[0052] Step S3: Construct the corresponding decision factors based on the decision element set using the first operator;

[0053] Step S4: Register decision factors;

[0054] Step S5: Reconstruct the objective function using the second operator;

[0055] Step S6: Optimize the solution to obtain a feasible solution with the minimum deviation / maximum satisfiability.

[0056] This invention also discloses a specific rule for processing decision factor weights, which ensures the numerical stability of the reconstructed objective function, supports weight sensitivity analysis and automatic tuning, and provides auditable and interpretable records. The specific method is as follows:

[0057] 1. Input and Symbol Definition

[0058] 1) Activate the set of state decision factors ;

[0059] 2) Original weight vector: Provided by the user or template;

[0060] 3) Optional upper and lower bounds for each factor ;

[0061] 4) Generate normalized weight vector This is used to reconstruct the objective function by the second operator.

[0062] 2. Weight Normalization and Boundary Control Rules

[0063] 1) Rule 1 is weight sum normalization: When using a strategy where the weight sum is 1:

[0064]

[0065] 2) Rule 2 is normalization after single-factor boundary clipping: First, the original weights are clipped:

[0066]

[0067] Renormalization:

[0068]

[0069] 3) Numerical stability handling:

[0070] when A uniform distribution strategy is adopted. ;

[0071] 3. Weight Sensitivity Scan: Used to quantify the impact of changes in single or multiple factor weights on Key Performance Indicators (KPIs), identifying inflection points and thresholds. The scanning process is as follows:

[0072] 1) Target Factor In the interval With step size Perform a scan;

[0073] 2) Fix other factors and trim and normalize them according to the current template or baseline weights;

[0074] 3) Reconstruct the objective and call the solver to solve it;

[0075] 4) Record the KPI obtained in each solution;

[0076] 5) Compile the results into a trade-off curve or heatmap and a sensitivity report;

[0077] 4. Automatic tuning and auditability

[0078] Within the weight space, Bayesian optimization, grid search, or other global optimization methods are used to find weight combinations that satisfy the constraints and are optimized. The process is as follows:

[0079] 1) Generate candidate weight vectors;

[0080] 2) The objective is reconstructed by the second operator and submitted to the solver;

[0081] 3) The evaluator returns the target value or uses KPIs as optimization feedback;

[0082] 4) The optimizer generates the next candidate based on the feedback until the termination condition is met;

[0083] Auditability:

[0084] 1) Save the candidate weights and corresponding KPIs for each time to form a weight-KPI mapping table;

[0085] 2) Record optimization process logs, including candidate sequences, evaluation values, timestamps, and operators, for auditing and backtracking purposes;

[0086] 3) After optimization, generate a set of recommended weights with interpretable explanations, such as why the combination outperforms the baseline on a specific KPI.

[0087] The system described in this invention can be deployed using one of the following three methods or a combination thereof, depending on the enterprise's IT architecture, data security requirements, and business scenario needs:

[0088] 1. Cloud Sandbox: This mode is mainly used in the proof-of-concept (PoC) or solution evaluation phase. The system runs in a cloud environment, supports the rapid import of customer orders, processes, capacity, and other data, can quickly solve and display results based on pre-set templates, and supports report export, facilitating solution comparison and decision verification.

[0089] 2. On-premises deployment: This mode is primarily used in production environments. The system is deployed on an internal enterprise server or in a private cloud environment to meet data isolation, information security, and compliance requirements. All data processing, solving, and evaluation processes are completed locally, making it suitable for manufacturing enterprises with high data security requirements.

[0090] 3. Hybrid Deployment: This model deploys system functions in layers based on sensitivity. Non-sensitive modules such as constraint resolution and template management run in the cloud, while the optimization solver and related data processing modules are deployed locally within the enterprise to ensure that sensitive production data does not leave the enterprise's internal network. This model combines the flexibility of the cloud with the security of local deployment.

[0091] As can be seen from the technical solutions provided by the present invention above, the target system for APS and its unified construction method and extended system provided by the present invention have the following beneficial effects:

[0092] Enhance the flexibility and adaptability of APS objective construction: By establishing a mapping relationship between business constraints and decision elements, and adopting an operator-based decision factor construction method, the objective function can be automatically reconstructed as business scenarios, constraints change or strategies are adjusted, avoiding the problems of fixed and difficult-to-maintain objective functions in traditional APS.

[0093] Improve solution feasibility and robustness: When the solution is detected to be infeasible, constraint softening and target switching are automatically triggered, and slack variables or penalty terms are incorporated into the reconstruction target to generate the most feasible solution.

[0094] Improve the numerical stability of objective function construction: By employing strategies such as weight pruning, normalization, boundary control, and uniform distribution, prevent problems such as objective function imbalance, unstable solutions, or uninterpretable results caused by improper weight settings.

[0095] Enhanced interpretability and auditability: Through sensitivity scanning, weight-KPI mapping records, and automatic tuning logs, the system can quantify the impact of weight changes on key performance indicators and provide a traceable record of the optimization process.

[0096] Enhance scalability and engineering capabilities: Through pluggable operators, a templated scenario library, and a standardized solver interface, achieve multi-scenario reuse, solver adaptation, and support for multiple deployment modes.

[0097] Improve overall decision-making efficiency and planning quality: Through a unified goal-setting system, automated refactoring mechanism and intelligent optimization capabilities, APS can continuously output plans that meet business objectives, are executable and of high quality. Attached Figure Description

[0098] Figure 1 System architecture diagram.

[0099] Figure 2 Flowchart of the decision-making objective switching mechanism.

[0100] Figure 3 Schematic diagram of the infeasible hard constraint softening process. Detailed Implementation

[0101] The technical solution of the present invention will be described in detail below with reference to specific embodiments, so that those skilled in the art can more clearly understand the concept, technical features and beneficial effects of the present invention. It should be understood that these embodiments are only used to illustrate the basic principles of the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0102] The technical solutions of this invention can be modified and extended in various ways according to actual application needs. As long as they do not depart from the core idea of ​​this invention and the scope defined by the claims, they should be regarded as the objects of protection of this invention.

[0103] Without causing technical conflicts, the technical features or sub-modules in the following embodiments can be combined arbitrarily to adapt to different application scenarios and system architectures.

[0104] The illustrations are for illustrative purposes only. The processes, module divisions, and interaction relationships shown can be adjusted and optimized according to actual application scenarios and should not be construed as the sole limitation of the collaborative decision-making optimization system and method of this invention. Those skilled in the art can adapt and extend the method steps, system structure, or information interaction methods according to specific needs without departing from the core ideas of this invention.

[0105] To enhance the clarity of the technical expression, terms such as "first" and "second" used in the specification are only used to distinguish similar components or steps and do not indicate their execution order or priority; terms such as "comprising" and "having" are open-ended expressions intended to cover other technical features not explicitly listed; terms such as "and / or" and " / " indicate a parallel or selective relationship, and their specific meaning should be understood in conjunction with the context.

[0106] Example 1

[0107] This embodiment uses Figure 1 For reference, the structure and functional implementation of each module in the target system used for APS in this invention are described in detail, wherein:

[0108] The decision factor management module maintains the metadata and related information of all decision factors in the system. It serves as the core data source for objective function construction / reconstruction and management, and supports querying, template export and import. It also interacts with the objective function module, registration module, settings module, and decision factor clearing module. Its maintenance information includes:

[0109] 1. Decision factor name: Used to clarify semantic meaning and identification. The name reflects the focus of the decision and enhances interpretability.

[0110] 2. Decision Factor Status: Used to identify whether a factor is enabled or disabled. Factors in the enabled state participate in the construction of the objective function; factors in the disabled state do not participate in the objective function construction, but their corresponding variables and constraints are still in effect and are satisfied during optimization.

[0111] 3. Decision Factor Weights: Used to reflect the importance of factors in the objective function. Users can adjust the decision factor weights to achieve the following:

[0112] 1) Target preference and priority control

[0113] Users can adjust the weights to make the solution more focused on meeting specific business objectives, thereby clarifying priorities among multiple objectives and guiding the plan results in the desired direction.

[0114] 2) Quick switching between scenario-based and template-based approaches

[0115] Weights can be saved as decision target templates, and users can switch between preset or historical templates with one click to quickly reproduce target preferences in different business scenarios.

[0116] 3) Constraints and weights work together

[0117] During the construction / reconstruction of the objective function, the system considers the weights and the constraints corresponding to the decision factors together. The weights determine the relative contribution of the decision factors, and the constraints guarantee the feasibility of the solution.

[0118] 4) Normalization and boundary control:

[0119] To ensure the numerical stability of the objective function, the weights are normalized within the system and constrained by upper and lower bounds, for example, the sum of the weights is 1 or the weights of a single factor are limited to a certain interval.

[0120] 5) Traceability and Auditability:

[0121] Each weight adjustment is logged, including the operator, time, previous and next values, and associated templates. Rollback and version management are supported to verify the calculation results under different weight configurations.

[0122] 4. Detailed information on decision factors: including the corresponding set of decision elements and the corresponding set of constraints.

[0123] The decision element building module is used to automatically or semi-automatically generate a set of decision elements for constructing decision factors based on the constraint interface provided by the optimization model.

[0124] The query module provides the ability to retrieve decision factor information, supporting queries by name, by status, and all queries, and returns metadata of decision factors that meet the criteria, as well as their associated decision elements and constraint information.

[0125] The registration module is used to register decision factors to the decision factor management module, and supports registration from multiple sources, including source types:

[0126] 1. Predefined decision factors: Decision factors that are predefined and fixed by the optimization model or system.

[0127] 2. Operator-generated decision factors: The set of decision elements generated by the first operator is used to construct new decision factors. The first operator is the summation operator by default, but operators such as absolute value summation can also be used, and the operator supports custom extensions.

[0128] During registration, you need to specify the name, status, and weight of the decision factors. The registration information also includes the corresponding set of decision elements and the set of constraints. The status and weight can be set to default values ​​for quick registration.

[0129] The decision factor information display module is used to display decision factors and their related information in a structured view. It supports exporting, filtering, and viewing historical records, making it convenient for users to review and make decisions.

[0130] The configuration module provides the ability to configure single or batch decision factors, including status settings, name modifications, weight settings, and other operations, and records change logs for retrospective purposes.

[0131] The decision factor clearing module supports two clearing strategies to meet the needs of different scenarios.

[0132] 1. Shallow shaving: The corresponding decision factors are excluded from the current decision objective, but their related decision elements and constraints are retained. These elements and constraints are still valid and are satisfied during optimization.

[0133] 2. Deep Cleanup: Completely deletes the decision factor and all its associated decision elements and constraints, and simultaneously cleans up related registration records and historical configurations. Deep cleanup is a transactional operation and supports rolling back snapshots for short-term recovery.

[0134] The objective function module allows the objective function to be constructed / reconstructed before each solution iteration using a second operator to reflect the currently enabled decision factors and their weights. The second operator defaults to a linear weighted sum and supports custom operators, such as piecewise objectives or vectorized multi-objectives, in solver-acceptable forms. The objective function reconstruction code is as follows:

[0135]

[0136] The objective function module also supports:

[0137] 1) Optimization direction settings: Minimize or maximize settings.

[0138] 2) Solver parameter settings: Provides an interface for configuring solver parameters, including solution accuracy, time limits, heuristic strategies, etc.

[0139] The decision objective template module is used to export, store, and quickly import preset or historical decision objective schemes, such as delivery date priority, cost priority, capacity utilization priority, and various user-defined templates. Each template includes a set of decision factors, weight configurations, optimization directions, and solution parameters.

[0140] The solution result evaluation module is used to perform multi-dimensional quantitative analysis, visualization, and evaluation of the results after the solution is completed. The corresponding pseudocode is as follows:

[0141]

[0142] 1. The analysis includes:

[0143] 1) Decision Factor Result Analysis: Calculate and output the score or contribution value of each decision factor to quantify the impact of each factor on the solution result.

[0144] 2) The multi-objective trade-off and visualized weight sensitivity analysis system supports scanning or sensitivity analysis of the weight space, quantifies the impact of weight changes on key performance indicators such as delivery rate, work-in-process inventory, and equipment utilization, helps users understand the benefits and costs of weight adjustments, and generates visualized trade-off curves or heat maps.

[0145] 2. Visualization: The solution results are displayed in the form of charts, heatmaps, Gantt charts, radar charts, etc., to compare with key indicators, making it easy for users to understand intuitively.

[0146] 3. Quantitative evaluation: Generates evaluation documents containing key performance indicators, sensitivity analysis and recommendations, and supports export and archiving.

[0147] Example 2

[0148] This embodiment provides a detailed description of the overall operation flow of the target system used for APS in this invention. The steps described in this embodiment can be implemented by software, hardware, or a combination of both, and the execution order between the steps can be adjusted according to the actual business scenario without affecting the core technical solution of this invention.

[0149] When performing advanced planning and scheduling optimization tasks, the system operates according to the following steps:

[0150] Step S1: Initialize the decision factor set:

[0151] System administrators import or define an initial set of decision factors through the decision factor management module. Initialization includes:

[0152] 1. Names of decision factors;

[0153] 2. Enabled by default;

[0154] 3. Default weight;

[0155] 4. A set of related decision elements;

[0156] 5. The set of associated constraints;

[0157] After initialization, the system generates corresponding metadata records and writes them to the initialization log to support subsequent auditing and backtracking.

[0158] Step S2: Decision element generation:

[0159] The decision element construction module synchronously generates a set of corresponding decision elements by establishing constraints. Examples of constraints include, but are not limited to:

[0160] 1. Processing time of each step;

[0161] 2. Equipment capacity and production capacity limits;

[0162] 3. Material availability and supply constraints;

[0163] 4. Process route and priority constraints;

[0164] The generated set of decision elements will serve as the basis for constructing decision factors in the future.

[0165] Step S3: Construct decision factors:

[0166] By using the default first operator, specifying the built-in first operator, or a custom first operator, the set of decision elements generated in step S2 can be automatically or semi-automatically constructed into the corresponding decision factors.

[0167] Step S4: Decision Factor Registration

[0168] The registration module registers pre-set decision factors or decision factors generated by the first operator to the decision factor management module. Registration content includes:

[0169] 1. Names of decision factors;

[0170] 2. Enabled status;

[0171] 3. Weighting;

[0172] 4. The corresponding set of decision elements;

[0173] 5. The corresponding set of constraints;

[0174] The registration module generates a change log while writing metadata, recording the operator, timestamp, registration source, and related parameters to support subsequent auditing and version management.

[0175] Step S5: Selecting a Decision Objective Template:

[0176] Preferably, users can select a preset template or a custom template through the decision-making goal template module. The template content includes:

[0177] 1. Set of decision factors;

[0178] 2. Weighting;

[0179] 3. Optimization direction;

[0180] 4. Solve for parameters;

[0181] After the template is selected, the system will import the template content into the decision factor management module.

[0182] Step S6: Objective function construction / refactoring:

[0183] Before each solution iteration, the objective function module calls the second operator to generate a solver-acceptable objective expression based on the currently enabled decision factors and their weights. The second operator can be:

[0184] 1. Linear weighted sum (default);

[0185] 2. Piecewise objective function;

[0186] 3. Vectorized multi-objective representation;

[0187] 4. Other custom operators.

[0188] Step S7: Solving and Result Evaluation:

[0189] After the solver completes the optimization solution, the solution result evaluation module performs the following operations:

[0190] 1. Decision factor score analysis: Calculate the contribution or score of each decision factor in the final solution;

[0191] 2. Multi-objective trade-off analysis: Scan the weight space or perform sensitivity analysis to quantify the impact of weight changes on key performance indicators (such as on-time delivery rate, equipment utilization rate, and work-in-process inventory);

[0192] 3. Visualization: Results are displayed in the form of charts, heatmaps, Gantt charts, radar charts, etc.

[0193] 4. Evaluation Report Generation: Outputs an evaluation document containing key indicators, sensitivity analysis, and recommendations, and supports export and archiving.

[0194] Step S8: Iterative optimization or factor cleanup:

[0195] Based on the evaluation results, users can:

[0196] 1. Settings module: Adjust the status, name, or weight of decision factors;

[0197] 2. Decision Factor Removal Module: Performs shallow or deep removal;

[0198] After the adjustments are completed, the system can proceed to the next round of solution verification or be directly used for production deployment.

[0199] Example 3

[0200] This embodiment discloses the specific rules and implementation examples for the weighting of decision factors in this invention, in order to ensure the numerical stability of the reconstructed objective function, support weight sensitivity analysis and automatic tuning, and provide auditable and interpretable records.

[0201] 1. Input and Symbol Definition:

[0202] 1) Input set: Set of activation state decision factors ;

[0203] 2) Original weight vector: ,in The original weight values ​​provided to the user or template;

[0204] 3) Single-factor boundaries: Optional upper and lower bounds for each factor. ;

[0205] 4) Normalization objective: Generate a normalized weight vector , used to reconstruct the objective function in the second operator.

[0206] 2. Weight normalization and boundary control rules:

[0207] 1) Rule 1 is weight sum normalization: If a weight sum of 1 strategy is adopted, it is processed as follows:

[0208]

[0209] 2) Rule 2 is normalization after single-factor boundary clipping: If the single-factor weights need to be limited to... First, trim the original weights:

[0210]

[0211] Renormalization:

[0212]

[0213] 3) Numerical stability handling:

[0214] when (For example, when all original weights are pruned to 0), a uniform distribution strategy is adopted. The following is a pseudocode example:

[0215]

[0216] 3. Example of weighted sensitivity scan:

[0217] 1) Used to quantify the impact of single-factor or multi-factor weight changes on key performance indicators (KPIs) (such as on-time delivery rate, work-in-process inventory, equipment utilization rate), identify inflection points and thresholds, and assist in weight setting and decision-making. The scanning strategy is as follows:

[0218] a) Target Factor In the interval With step size Perform a scan;

[0219] b) Fix other factors and trim and normalize them according to the current template or baseline weights;

[0220] c) Reconstruct the objective and call the solver to solve it;

[0221] d) Record the KPIs obtained from each solution and generate a trade-off curve or heatmap;

[0222] e) Output trade-off curves, heatmaps, and sensitivity reports, including inflection points, threshold recommendations, and qualitative descriptions of the impact on business.

[0223] 2) Scan the example pseudocode

[0224]

[0225] 4. Example of automatic tuning:

[0226] Within the weight space, Bayesian optimization, grid search, or other global optimization methods are used to find weight combinations that satisfy the constraints and are optimized. The process is as follows:

[0227] 1) Generate candidate weight vectors;

[0228] 2) The objective is reconstructed by the second operator and submitted to the solver;

[0229] 3) The evaluator returns the target value or KPI as optimization feedback;

[0230] 4) The optimizer generates the next candidate based on the feedback until the termination condition is met, such as the number of iterations or the target threshold.

[0231] Explainability and Auditability:

[0232] 1) Save the candidate weights and corresponding KPIs for each time to form a weight-KPI mapping table;

[0233] 2) Record optimization process logs, including candidate sequences, evaluation values, timestamps, and operators, for auditing and backtracking purposes;

[0234] 3) After optimization, generate a set of recommended weights with interpretable explanations, such as why the combination outperforms the baseline on a specific KPI.

[0235] 5. Implementation Precautions:

[0236] 1) Boundary setting principle: upper and lower bounds It can be determined by business rules, regulations, or historical experience; a default value can be used when there is no prior knowledge. ;

[0237] 2) Step size selection: Sensitivity scan step size A trade-off should be struck between accuracy and computational cost. It is recommended to use a finer step size for critical factors and a coarser step size for minor factors.

[0238] 3) Parallelization and caching: To improve scanning and automatic tuning efficiency, solution tasks can be submitted in parallel and the evaluated weight combination results can be cached to avoid duplicate calculations;

[0239] 4) Robustness extension: When uncertainty exists, sensitivity analysis can be extended to robust optimization or two-stage stochastic optimization to evaluate the robustness of weights under uncertainty conditions;

[0240] 5) Compliance and Audit: All weight changes, scans and auto-tuning operations should be written to the audit log and support rollback by version to meet compliance and traceability requirements.

[0241] Example 4

[0242] This embodiment discloses the definition of a decision factor data structure for decision factor management in this invention, which serves as an example to uniformly represent the metadata and associated information of decision factors, facilitating registration, querying, importing, exporting, version management, and target reconstruction.

[0243] Field description:

[0244] 1.id: A unique identifier for the decision factor, used for indexing and referencing within the system.

[0245] 2. name: The name of the decision factor, used to clarify its semantic meaning and presentation.

[0246] 3. status: Enabled or disabled status. The enumerated values ​​are used to control whether the factor participates in the objective function reconstruction.

[0247] 4. weight: Raw weight, used for weight values ​​input by the user or loaded from the template.

[0248] 5. normalized_weight: The normalized weights, used for the actual weight contribution during target reconstruction.

[0249] 6.elements: A list of decision element IDs, representing the set of decision elements associated with this factor.

[0250] 7. constraints: A list of associated constraint IDs, representing the set of constraints associated with this factor.

[0251] 8. created_by: Creator identifier, used for auditing and access control.

[0252] 9. created_at: Creation timestamp, used for version control and auditing.

[0253] 10. version: Configures the version number, used for concurrency control, rollback, and historical snapshot management.

[0254] Example JSON representation:

[0255]

[0256] Log and audit format: Record all configuration changes to support traceability, auditing, and version rollback. Change logs should be searchable by time, operator, or target object. Each change log contains: {action, target_id, field, old_value, new_value, operator, timestamp, version}.

[0257] Example log entry:

[0258]

[0259] Example 5

[0260] In APS scenarios, hard constraints may not be fully satisfied due to insufficient capacity, material shortages, or process conflicts. Directly submitting a solution will result in the solver finding no solution or returning an infeasible solution. This embodiment, in conjunction with Figure 2, provides a decision objective switching mechanism. By softening infeasible hard constraints, the optimization problem can continue to be solved and return the optimal deviation solution, thereby ensuring that the system can still output an executable plan / scheduling scheme even in infeasible situations. This process can be executed automatically by the software system or semi-automatically through user interaction. The specific steps are as follows:

[0261] Step S1: Load the optimization problem model, including:

[0262] 1. Definition of decision variables;

[0263] 2. Set of constraints;

[0264] 3. Initialize the decision factor management module to form the initial structure of decision factors;

[0265] Once the model is built, the process moves to the next step.

[0266] Step S2: Construct / reconstruct the objective function based on the decision factor registration information and perform optimization. After the solution is completed, proceed to the next step.

[0267] Step S3: Feasibility Assessment

[0268] The system performs a feasibility check on the solutions returned by the solver, including:

[0269] 1. Does it satisfy all hard constraints?

[0270] 2. Does it meet the key business rules?

[0271] 3. Are there any infeasible variable assignments?

[0272] Alternatively, feasibility can be determined by the solver returning either no solution or a feasible solution.

[0273] 1. If the process is feasible, proceed to step S4 to conduct result analysis and quantitative evaluation;

[0274] 2. If the process is not feasible, proceed to step S7 to perform constraint transformation and decision element generation.

[0275] Step S4: Results Analysis and Quantitative Evaluation

[0276] When a solution is feasible, the system performs the following analysis:

[0277] 1. Calculation of key performance indicators;

[0278] 2. Analysis of the contribution of decision factors;

[0279] 3. Visual presentation.

[0280] After the analysis is complete, proceed to the next step.

[0281] Step S5: Satisfaction Assessment

[0282] The system determines whether the current solution meets business requirements based on preset thresholds or user feedback.

[0283] 1. If the satisfactory process enters the termination state, output the final optimization result;

[0284] 2. If you are not satisfied with the process, proceed to step S6.

[0285] Step S6: Setting the weights of decision factors

[0286] The system allows users or automated strategies to adjust the weights of decision factors, including:

[0287] 1. Increase or decrease the weight of a factor;

[0288] 2. Enable or disable certain factors;

[0289] 3. Apply preset templates or historical templates;

[0290] After the weight adjustment is completed, the process returns to step S2, where the objective function is reconstructed / rebuilt and the solution is executed to achieve iterative optimization.

[0291] Step S7: Constraint Transformation and Decision Element Generation in Infeasible Scenarios

[0292] When a solution is not feasible, the system performs the following operations:

[0293] 1. Transform infeasible constraints;

[0294] 2. Generate a new set of decision elements;

[0295] 3. Register the generated decision elements as new decision factor candidates;

[0296] After completion, the process proceeds to step S8.

[0297] Step S8: Construct and register decision factors

[0298] The system generates decision elements based on step S7:

[0299] 1. Construct new decision factors;

[0300] 2. Register to the decision factor management module;

[0301] 3. Update metadata and log records;

[0302] After registration, the process returns to step S2, where the objective function is reconstructed and the solution is executed.

[0303] When the optimization result is deemed satisfactory, the system enters a termination state, outputs the final optimization result, and ends the process. This embodiment demonstrates a complete APS optimization solution closed loop, including:

[0304] 1. A cycle of "Solve → Feasibility assessment → Result analysis → Satisfaction assessment → Weight adjustment → Solve again";

[0305] 2. Compensation mechanism of "infeasibility → constraint transformation → decision element generation → decision factor registration → re-solution";

[0306] 3. Supports a hybrid mode of automation and manual intervention;

[0307] 4. Supports dynamic objective function construction and multi-round iterative optimization.

[0308] This process ensures that the system can continuously approach the optimal solution and meet business needs in complex manufacturing scenarios.

[0309] Example 6

[0310] This embodiment, in conjunction with the flowchart shown in Figure 3, specifically illustrates the implementation method of the present invention for softening infeasible hard constraints and incorporating them into the objective function, thereby demonstrating the engineering feasibility of the present invention.

[0311] 1. The mathematical expression for converting hard constraints to soft constraints is as follows:

[0312] Assume the original hard constraints are:

[0313]

[0314] When this constraint cannot be satisfied, slack variables are introduced. And construct soft constraints based on the direction of deviation:

[0315] 1) The deviation is in the positive direction (insufficient), that is, when When this is done, a nonnegative slack variable is introduced:

[0316]

[0317] And Add it to the objective function as a deviation term:

[0318]

[0319] Used to minimize the degree of deviation.

[0320] 2) The deviation is in the negative direction (excessive), that is, when the constraint is an upper bound constraint or there is a reverse deviation requirement:

[0321]

[0322] And Add it to the objective function as a deviation term:

[0323]

[0324] Used to maximize satisfiability.

[0325] 2. Generating decision elements and constructing decision factors using soft constraints:

[0326] After the constraints are relaxed, this embodiment uses the relaxed variables as decision elements and further constructs decision factors.

[0327] 1) Generate a set of decision elements:

[0328] For each relaxed constraint, the system generates a decision element. This forms a set of decision elements:

[0329]

[0330] 2) Constructing decision factors:

[0331] The system uses the first operator (default is sum) to accumulate all decision elements to construct a new decision factor:

[0332]

[0333] After being registered through the registration module, this decision factor has the following attributes:

[0334] a) Name: Delivery Deviation;

[0335] b) Status: Enabled;

[0336] c) Weight: Can be set by the template or the user;

[0337] d) Related decision factors: ;

[0338] e) Association constraints: .

[0339] 3. After registering the decision factors, the system calls the second operator to reconstruct the objective function:

[0340] 1) Assume the original objective function is:

[0341]

[0342] in:

[0343] a) These are the original decision factors;

[0344] b) For the corresponding weights.

[0345] 2) Objective function after adding soft constraint factors:

[0346]

[0347] in The weights for soft constraint factors (usually high, used to force a reduction in bias).

[0348] Example 7

[0349] The production planning model constructed in this embodiment includes: a set of equipment, a set of products, the processing time for each product on each equipment, and a decision on the quantity of products processed on each equipment. It aims to provide a complete and directly solvable example to demonstrate the feasibility of the technical solution of this invention in engineering implementation. The variable settings, constraint construction, and objective function of this model are illustrative and should not be construed as limiting the scope of protection of this invention.

[0350] 1. Symbols and Data Definitions

[0351] 1) Set:

[0352] a) Product Collection ,index ;

[0353] b) Equipment collection ,index ;

[0354] c) Discrete-time set ,index ;

[0355] d) Set of effective decision factors ,index .

[0356] 2) Parameters:

[0357] a) : product In the equipment The unit processing time (hours / piece);

[0358] b) : product Demand quantity (pieces);

[0359] c) : equipment The total available processing time (man-hours) during the planning period, if modeled by time period, is the processing capacity per time period. ;

[0360] d) If the product Available on device If the processing is performed, the value is 1; otherwise, it is 0.

[0361] e) : product The unit processing cost;

[0362] f) Large numbers (used for linearization constraints).

[0363] 3) Decision variables:

[0364] a) In the equipment Products processed Quantity (pieces);

[0365] b) Slack variables, representing the product The number of undelivered items (pieces);

[0366] c) In the equipment time period Products processed quantity;

[0367] d) If in the equipment Process any number of products Then it is 1 (used for fixed startup costs or allocation constraints);

[0368] e) : Decision factor weights.

[0369] 2. Constraints:

[0370] 1) Demand is met (total output of each product at least meets demand), one of the following scenarios can be selected:

[0371] a) Non-time series model case:

[0372]

[0373] b) Time series model case:

[0374]

[0375] 2) Soft constraint demand constraints in cases where demand / capacity is insufficient (this does not take effect simultaneously with hard constraint demand satisfaction constraints; one must be chosen):

[0376] a) Non-time series model case:

[0377]

[0378] b) Time series model case:

[0379]

[0380] c) The set of decision elements generated by the above constraints:

[0381]

[0382] d) In this embodiment, the first operator uses the sum operator to construct the decision factor, and the registration name is specified as "DeliveryShortfall":

[0383]

[0384] 3) Equipment processing capacity (total processing time per machine shall not exceed available time):

[0385] a) Non-time series model case:

[0386]

[0387] b) Time series model case:

[0388]

[0389] 4) Processability:

[0390]

[0391] 5) Integer property:

[0392]

[0393] 3. Decision factors in this embodiment

[0394] 1) Built-in decision factors in the model

[0395]

[0396] 2) Selectively created non-built-in decision factors

[0397]

[0398] 3) After these decision factors are created, they are registered to the decision factor management module through the registration module for management and use in constructing / reconstructing the objective function.

[0399] 4. Examples of Optional Objective Functions

[0400] 1) Minimize total processing time or equipment usage:

[0401]

[0402] 2) Meet the demand while minimizing the total cost:

[0403]

[0404] 3) In this embodiment, the second operator uses a linear weighted sum operator to construct / reconstruct the objective function of the multi-objective decision-making process, and sets the optimization direction to minimize:

[0405]

[0406] Minimizing total processing time or equipment occupancy and meeting demand while minimizing total cost correspond to respectively and The situation.

[0407] 4) Calculate the minimum outstanding quantity when demand cannot be met / capacity is insufficient.

[0408]

[0409] They can all be constructed using the advanced planning and scheduling target system and its construction method proposed in this invention.

[0410] 5. Small-scale numerical examples

[0411] 1) Products ,need ;

[0412] 2) Equipment Available time ;

[0413] 3) Processing time (h / piece): ;

[0414] 4) Machinability All are machinable;

[0415] 5) The objective function is chosen to minimize the total processing time;

[0416] 6) Verification can be performed using any MILP solver, such as the following solvers:

[0417] a) CPLEX;

[0418] b)Gurobi;

[0419] c) SCIP;

[0420] d)GLPK;

[0421] 7) After solving, we can obtain To meet demand and capacity constraints.

[0422] Example 8: Deployment and Operation

[0423] This embodiment illustrates the operation mode of the target system for APS proposed in this invention under different deployment modes and its application process in typical business scenarios.

[0424] The system of this invention can be deployed in one of the following three ways or a combination thereof, depending on the enterprise's IT architecture, data security requirements, and business scenario needs:

[0425] 1. Cloud Sandbox: This mode is mainly used in the proof-of-concept (PoC) or solution evaluation phase. The system runs in a cloud environment, supports the rapid import of customer orders, processes, capacity, and other data, can quickly solve and display results based on pre-set templates, and supports report export, facilitating solution comparison and decision verification.

[0426] 2. On-premises deployment: This mode is primarily used in production environments. The system is deployed on an internal enterprise server or in a private cloud environment to meet data isolation, information security, and compliance requirements. All data processing, solving, and evaluation processes are completed locally, making it suitable for manufacturing enterprises with high data security requirements.

[0427] 3. Hybrid Deployment: This model deploys system functions in layers based on sensitivity. Non-sensitive modules such as constraint resolution and template management run in the cloud, while the optimization solver and related data processing modules are deployed locally within the enterprise to ensure that sensitive production data does not leave the enterprise's internal network. This model combines the flexibility of the cloud with the security of local deployment.

[0428] The following example, using a Proof-of-Concept (PoC) scenario with manufacturing company A prioritizing delivery time, illustrates the operation of the system of this invention:

[0429] 1. Import order and process data to the cloud sandbox;

[0430] 2. Generate a set of decision elements, specify the first operator, and construct decision factors;

[0431] 3. The registration module registers the "Delivery Priority" factor and sets its weight to 0.8, while the combined weight of other factors is 0.2;

[0432] 4. Specify the second operator to construct / reconstruct the objective function and submit the solver. The solution takes 45 seconds.

[0433] 5. The evaluation module improved the delivery achievement rate from the baseline of 72% to 91%, generated a sensitivity analysis report, and suggested adjusting the delivery factor weight to 0.75 to balance inventory;

[0434] 6. Users can save the current template and export the evaluation document as the basis for project approval.

[0435] The following examples demonstrate typical interface call methods in actual operation of the system, illustrating the feasibility and engineering feasibility of the system of the present invention.

[0436] 1. Query decision factors

[0437] 1) Request

[0438]

[0439] 2) Response

[0440]

[0441] 2. Register new factors

[0442] 1) Request

[0443]

[0444] 2) Response

[0445]

[0446] 3. Update weights

[0447] 1) Request

[0448]

[0449] 2) Response

[0450]

[0451] 4. Trigger the first operator

[0452] 1) Request

[0453]

[0454] 2) Response

[0455]

[0456] 5. Trigger the second operator and solve.

[0457] 1) Request

[0458]

[0459] 2) Response

[0460]

[0461] 6. Obtaining and evaluating the solution results

[0462] 1) Request

[0463]

[0464] 2) Response

[0465] Returns the solution results, scores for each decision factor, key performance indicators (KPIs), and a visual data structure or link.

Claims

1. A target system for advanced planning and scheduling (APS), characterized by The system includes: means for generating decision information for optimization solutions from business constraints and business data; means for constructing and managing decision factors for target expression based on the decision information; means for generating or reconstructing optimization target expressions based on the decision factors; and means for performing optimization solutions and adjusting the decision factors or target expressions based on the solution results.

2. The system as claimed in claim 1, wherein, The decision information includes decision elements generated from deviation values, slack values, or feasibility indicators of business constraints, and may further include performance indicators or resource utilization indicators calculated from business data.

3. The system as described in claim 1, wherein, The device for constructing decision factors processes decision information through accumulation, absolute value summation, segmented aggregation, or user-defined aggregation, and supports the construction of composite decision factors.

4. The system as claimed in claim 1, wherein, The weight governance performed on decision factors includes weight boundary pruning, normalization, and numerical stability processing. When the sum of the pruned weights is zero, a uniform distribution strategy is adopted.

5. The system as claimed in claim 1, wherein, The optimization objective expression is constructed through linear weighting, piecewise weighting, or vectorized combination, and supports setting the direction of minimization or maximization.

6. The system as claimed in claim 1, wherein, When the problem is deemed infeasible, the system generates new decision information by converting infeasible constraints into soft constraints and incorporates slack variables as objective or penalty terms into the reconstructed objective expression.

7. The system as claimed in claim 1, wherein, The system performs sensitivity scans on the weights of decision factors to record the impact of weight changes on key performance indicators and generates trade-off curves or sensitivity heatmaps.

8. The system as claimed in claim 1, wherein, The system uses Bayesian optimization, grid search, or genetic algorithms to search for the optimal weight combination that satisfies the constraints in the weight space.

9. The system of claim 1, wherein, The system saves candidate weights, corresponding key performance indicators and timestamps to form audit records, and performs automatic optimization or rollback decisions based on the records.

10. The system of claim 1, wherein, The first and second operators are pluggable operators, which support user-defined operators and can be registered and invoked through extension points or plugin mechanisms.

11. The system of claim 1, wherein, The decision objective template includes a set of decision factors, weight configuration, optimization direction and solution parameters, and supports template versioning and one-click switching.

12. The system of claim 1, wherein, The registration device records the operator, timestamp, and registration source when registering decision factors, and writes metadata into the change log to support version management.

13. A method for constructing targets for Advanced Planning and Scheduling (APS), characterized in that, include: Extract decision-making information for goal setting from business constraints and business data; Based on the aforementioned decision information, decision factors for target representation are generated and managed; Construct or reconstruct the optimization objective expression based on the decision factors and submit the solution; The decision factors or objective expression are adjusted based on the solution results to achieve iterative optimization.

14. The method of claim 13, wherein, The decision information includes decision elements generated from deviation values, slack values, or feasibility indicators of business constraints, and may further include performance indicators or resource utilization indicators calculated from business data.

15. The method of claim 13, wherein, The generation of decision factors involves processing decision information through accumulation, absolute value summation, segmented aggregation, or user-defined aggregation, and performing weight pruning, normalization, and numerical stability processing on the generated decision factors.

16. The method of claim 13, wherein, The construction or reconstruction of the optimization objective expression includes generating objective terms by linear weighting, piecewise weighting, or vectorized combination, and supports setting the minimization or maximization direction.

17. The method of claim 13, wherein, When the solution is infeasible, the infeasible constraint is converted into a soft constraint and new decision information is generated. The slack variables are incorporated into the reconstructed objective expression as objective terms or penalty terms. The weights or activation states of the decision factors are adjusted based on the solution results to achieve iterative optimization.

Images