Formula process extrapolation method, device and equipment, storage medium and product
By generating and optimizing causal directed graphs, and combining self-supervised tasks and causal pruning optimization, the problem of insufficient causal explanation in existing technologies is solved, and efficient and interpretable formulation and process extrapolation is achieved, which is suitable for automated recommendation of new processes and new materials.
Patent Information
- Application Number
- CN202511027441.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-24
- Publication Date
- 2025-11-21
AI Technical Summary
Existing formulation and process extrapolation schemes rely on black-box prediction models, lacking causal explanations and scientific traceability, resulting in low extrapolation efficiency, failure to fully explore the deep logic between columns and discover new correlations, and uncontrollable confidence of the results.
By generating a preliminary causal directed graph, correcting causal edges, optimizing variable node representations, and combining self-supervised tasks and causal pruning optimization, new formula process data is generated, providing traceable recommended paths.
It achieves efficient and accurate automatic extrapolation of unknown process variables, adapts to new processes and materials, provides interpretable recommended paths, and improves extrapolation efficiency and confidence.
Smart Images
Figure CN120996219A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of industrial big data, and in particular to a formula process extrapolation method, device, equipment, storage medium and product. BACKGROUND
[0002] Existing formula process optimization schemes mostly rely on black-box prediction models, lack causal explanation and scientific traceability, and experts cannot intervene. Table fields are only mapped and cannot fully mine deep inter-column logic, especially potential synergy or inhibition. Knowledge graphs usually only navigate static relationships and cannot automatically discover new associations important to final performance. The extrapolation process has poor generalization for new formulas and new processes, and the result confidence is uncontrollable. The above problems result in low extrapolation efficiency. SUMMARY
[0003] The main purpose of the present application is to provide a formula process extrapolation method, device, equipment, storage medium and product, which aims to solve the technical problems of low extrapolation efficiency of existing formula process extrapolation schemes which mostly rely on black-box prediction models.
[0004] To achieve the above purpose, the present application provides a formula process extrapolation method, which comprises:
[0005] Generating a preliminary causal directed graph according to historical table data, the historical table data being table data formed after historical formula process data is structured, and the preliminary causal directed graph comprising preliminary variable nodes and preliminary causal edges;
[0006] Correcting the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph;
[0007] Optimizing the preliminary variable nodes in the corrected causal graph to obtain an optimized causal graph;
[0008] Extrapolating a formula process from the optimized causal graph to generate new formula process data.
[0009] In an embodiment, the step of correcting the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph comprises:
[0010] Physically constraining the preliminary causal edges in the preliminary causal directed graph according to preset causal chain and formula process corresponding physical edge constraints to obtain constrained causal edges;
[0011] Evaluating the constrained causal edges according to a preset scoring function to obtain an evaluation score;
[0012] Correcting the constrained causal edges according to a preset search strategy and the evaluation score to obtain a corrected causal graph.
[0013] In an embodiment, the step of obtaining the optimized causal graph by optimizing the preliminary variable nodes in the revised causal graph comprises:
[0014] obtaining a fused variable node representation by mixing embedding the historical tabular data and the preliminary variable nodes in the revised causal graph;
[0015] constructing a self-supervised task according to the historical tabular data and the fused variable node representation, and optimizing the fused variable node representation according to the self-supervised task to obtain an optimized causal graph.
[0016] In an embodiment, the self-supervised task comprises a consistency task. Figure 1
[0017] The step of obtaining the optimized causal graph by optimizing the fused variable node representation according to the historical tabular data and the fused variable node representation comprises:
[0018] constructing a consistency task according to the historical tabular data and the fused variable node representation; Figure 1
[0019] generating a data feature vector according to the historical tabular data through the consistency task; Figure 1
[0020] generating a structure feature vector according to the fused variable node representation through the consistency task; Figure 1
[0021] optimizing the data feature vector and the structure feature vector according to a preset consistency loss function to obtain an optimized causal graph.
[0022] In an embodiment, the self-supervised task comprises a mask field prediction task.
[0023] The step of obtaining the optimized causal graph by optimizing the fused variable node representation according to the historical tabular data and the fused variable node representation comprises:
[0024] constructing a mask field prediction task according to the historical tabular data and the fused variable node representation;
[0025] obtaining masked tabular data by randomly selecting part of the data in the historical tabular data for masking through the mask field prediction task;
[0026] According to the fused variable node, data prediction is performed on the masked table data, and a prediction result is obtained.
[0027] According to a preset prediction loss function and the prediction result, the fused variable node is optimized, and an optimized causal graph is obtained.
[0028] In an embodiment, the step of formulating process extrapolation on the optimized causal graph to generate new formulation process data comprises:
[0029] Formulating process extrapolation is performed on the optimized causal graph to generate a new decision path.
[0030] According to historical formulation process samples and unlabeled new samples, the new decision path is optimized by causal pruning, and an optimized decision path is obtained.
[0031] According to the optimized decision path, a decision path trace tree is generated.
[0032] The decision path trace tree is visualized and displayed, and when a user's change operation is detected, the decision path trace tree is changed to obtain a changed decision path.
[0033] According to the changed decision path, new formulation process data is generated.
[0034] In addition, to achieve the above-mentioned purpose, the present application also proposes a formulation process extrapolation device, which comprises:
[0035] A causal graph generation module is configured to generate a preliminary causal directed graph according to historical table data, wherein the historical table data is table data formed after historical formulation process data is structured, and the preliminary causal directed graph comprises preliminary variable nodes and preliminary causal edges.
[0036] A causal edge correction module is configured to correct the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph.
[0037] A node optimization module is configured to optimize the preliminary variable nodes in the corrected causal graph to obtain an optimized causal graph.
[0038] A formulation process extrapolation module is configured to perform formulation process extrapolation on the optimized causal graph to generate new formulation process data.
[0039] In addition, to achieve the above-mentioned purpose, the present application also proposes a formulation process extrapolation device, which comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program is configured to implement the steps of the formulation process extrapolation method as described above.
[0040] In addition, to achieve the above object, the present application also provides a storage medium, which is a computer readable storage medium, and a computer program is stored on the storage medium, and the computer program realizes the steps of the formula process extrapolation method when executed by a processor.
[0041] In addition, to achieve the above object, the present application also provides a computer program product, which comprises a computer program, and the computer program realizes the steps of the formula process extrapolation method when executed by a processor.
[0042] The present application provides a formula process extrapolation method, which generates a preliminary causal directed graph from historical formula process table data, including preliminary variable nodes and preliminary causal edges; modifies the preliminary causal edges to obtain modified causal graphs; optimally represents the preliminary variable nodes to obtain optimized causal graphs; and extrapolates the optimized causal graphs to generate new formula process data. The present application generates a preliminary causal directed graph from historical formula process table data, and performs graph-table coupling. The preliminary causal edges are modified, and the preliminary variable nodes are optimally represented, so that a new recommended path can be traced and explained, and unknown process variables can be automatically extrapolated in a non-black-box manner with high efficiency and high accuracy, which is suitable for various scenarios such as new processes and new materials. BRIEF DESCRIPTION OF DRAWINGS
[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and serve to explain the principles of the present application together with the specification.
[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without creative labor.
[0045] Figure 1 A flowchart is provided for the formula process extrapolation method embodiment one of the present application;
[0046] Figure 2 A schematic diagram of the decision path trace tree generated in the present application is provided;
[0047] Figure 3 A flowchart is provided for the formula process extrapolation method embodiment two of the present application;
[0048] Figure 4 A flowchart is provided for the formula process extrapolation method embodiment three of the present application;
[0049] Figure 5An example diagram of a new decision path generated by extrapolation of the present application;
[0050] Figure 6 An example diagram of the present application after pruning of the new decision path;
[0051] Figure 7 An overall flowchart of the recipe process extrapolation method of the present application;
[0052] Figure 8 A module structure diagram of the recipe process extrapolation device of the embodiment of the present application;
[0053] Figure 9 A device structure diagram of the hardware operating environment involved in the recipe process extrapolation method of the embodiment of the present application.
[0054] The object implementation, functional features and advantages of the present application will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION
[0055] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and do not limit the present application.
[0056] In order to better understand the technical solutions of the present application, the following will be described in detail in conjunction with the drawings and specific embodiments of the specification.
[0057] The main solution of the embodiment of the present application is: generating a preliminary causal directed graph according to historical table data, the historical table data is table data formed after historical recipe process data is structured, and the preliminary causal directed graph includes preliminary variable nodes and preliminary causal edges; correcting the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph; optimizing the preliminary variable nodes in the corrected causal graph to obtain an optimized causal graph; and performing recipe process extrapolation on the optimized causal graph to generate new recipe process data.
[0058] Since the existing recipe process optimization scheme depends on a black box prediction model, lacks causal explanation and scientific traceability, and experts cannot intervene, table fields are only mapped and cannot fully mine deep inter-column logic, especially potential synergy or inhibition. Knowledge graph usually only makes static relationship navigation and cannot automatically discover new associations important to the final performance. The extrapolation process has poor generalization for new recipes and new processes, and the result confidence is uncontrollable. The above problems result in low extrapolation efficiency.
[0059] The application provides a solution, which generates a preliminary causal directed graph including preliminary variable nodes and preliminary causal edges according to historical recipe process table data, corrects the preliminary causal edges to obtain a corrected causal graph, optimally represents the preliminary variable nodes to obtain an optimized causal graph, and performs recipe process extrapolation on the optimized causal graph to generate new recipe process data. The application generates a preliminary causal directed graph according to historical recipe process table data, performs chart coupling. The preliminary causal edges are corrected, and the preliminary variable nodes are optimally represented, so that a new recommended path can be traced and explained, and unknown process variables can be automatically extrapolated in a non-black-box manner, thereby achieving high efficiency and high accuracy, and adapting to various scenes such as new processes and new materials.
[0060] It should be noted that the execution subject of the method of the embodiment can be a computing service device with recipe process extrapolation, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone and the like; or a recipe process extrapolation device with the same or similar functions. The embodiment and each of the following embodiments will be described by taking a recipe process extrapolation device as an example.
[0061] Based on this, the application provides a recipe process extrapolation method, which refers to Figure 1 , Figure 1 The flowchart of the first embodiment of the recipe process extrapolation method of the application is shown in FIG. 1.
[0062] In the embodiment, the recipe process extrapolation method includes steps S10-S40.
[0063] In step S10, a preliminary causal directed graph is generated according to historical table data, the historical table data is table data formed after historical recipe process data is structured, and the preliminary causal directed graph includes preliminary variable nodes and preliminary causal edges.
[0064] It should be noted that the historical table data can be table data formed after historical recipe process data (for example, historical recipe data and historical process data) is structured. The historical recipe data can be existing data such as drug ingredients and chemical parameters, and the historical process data can be parameters such as baking temperature and object solidification rate.
[0065] Understandably, historical table data is used as input to the directed causal graph parsing module. A causal directed acyclic graph (C-DAG) is a graphical model used to represent causal relationships between variables. It describes the causal structure between variables using a directed acyclic graph (DAG), where each node represents a variable and each directed edge represents a causal relationship from one variable to another. Based on the table structure, preliminary variable nodes (such as drug components, chemical parameters, etc.) and preliminary causal edges (such as A→B indicating that A is the cause of B) are generated, thus constructing a preliminary causal directed graph.
[0066] Step S20: Correct the initial causal edges in the initial causal directed graph to obtain the corrected causal graph.
[0067] It should be understood that the preliminary causal edges in the preliminary causal directed graph can be corrected by introducing a constrained Bayesian causal structure learning algorithm, which introduces preset physical rules to constrain the direction of the edge connections, thereby correcting the preliminary causal edges and realizing causal inference and expert co-construction.
[0068] In one feasible implementation, step S20 may include steps S201 to S203:
[0069] Step S201: Physically constrain the preliminary causal edges in the preliminary causal directed graph according to the preset causal chain and the physical edge constraints corresponding to the formula process to obtain constrained causal edges.
[0070] In practical implementation, the pre-set causal chain can be a causal chain pre-set based on expert knowledge, serving as a fixed edge or a high-confidence prior. An example of a pre-set causal chain could be a chain of "high-shear stirring → improved dispersion → improved strength". Physical edge constraints can be pre-set known ratios and process physical edge constraints (such as rules like the chemical conservatism law), used to physically constrain the initial causal edges in the initial causal directed graph, obtaining constrained causal edges.
[0071] Step S202: Evaluate the constrained causal edges according to a preset scoring function to obtain an evaluation score.
[0072] Understandably, the Bayesian score-based search algorithm can also be used to refine edges. The Bayesian score-based search algorithm is a core method in Bayesian network structure learning, aiming to automatically discover dependencies between variables (i.e., determine the optimal causal edge) through a scoring function and search strategy. First, a scoring function can be set to evaluate the constrained causal edges and obtain evaluation scores.
[0073] Step S203, according to the preset search strategy and the evaluation score, the constrained causal edge is modified to obtain a modified causal graph.
[0074] It should be understood that the search strategy can be specified secondly, the advantages and disadvantages of different preliminary causal edges are evaluated by defining a scoring function, and the constrained causal edge is modified according to the preset search strategy and the evaluation score to find the optimal causal edge and obtain the modified causal graph.
[0075] Step S30, the preliminary variable node in the modified causal graph is optimized to obtain an optimized causal graph.
[0076] It can be understood that the preliminary variable node representation can be optimized by a self-supervised table-graph embedding learning module. The node representation is enabled by a table-graph hybrid embedding through a self-supervised task, which can automatically supplement the knowledge required for reasoning of new data to obtain an optimized causal graph. The automatic learning rule of history and out-of-domain unlabeled data can be realized, and the success rate in the case of few samples can be significantly improved.
[0077] Step S40, the optimized causal graph is used for formula process extrapolation to generate new formula process data.
[0078] It should be understood that the optimized causal graph is used for formula process extrapolation, and a decision path backtracking tree is automatically output to clearly show the physical reasons and table-graph joint basis for each step of reasoning. It can be referred to Figure 2 The generated decision path backtracking tree is described. The ABC three parameters are used as examples, the initial optimization point of the formula parameter A (acetone concentration) is acetone concentration→dispersion↑, the process parameter B (baking temperature) is baking temperature→solidification rate↑, and the new variable C (additive ratio) is additive+temperature→synergistic effect↑. The upper column chart represents the contribution of different variables in the optimization step (length=weight); the lower Gantt chart represents the time span and type of the key causal path; the line connecting the Gantt chart blocks represents the causal dependence relationship. Finally, the new formula process data can be generated according to the decision path backtracking tree. Thus, the efficient extrapolation and intelligent optimization of the structured table data of the formula process are realized, and are widely applicable to the innovation and research automation of the materials, pharmaceutical, food and other industries.
[0079] The embodiment provides a formula process extrapolation method, generates a preliminary causal directed graph from historical formula process table data, includes preliminary variable nodes and preliminary causal edges; the preliminary causal edges are corrected to obtain a corrected causal graph; the preliminary variable nodes are optimized to obtain an optimized causal graph; the optimized causal graph is extrapolated to generate new formula process data. The embodiment generates a preliminary causal directed graph from historical formula process table data, and performs table coupling. The preliminary causal edges are corrected, and the preliminary variable nodes are optimized to obtain an optimized causal graph. The new recommended path is traceable and interpretable, and the unknown process variable automatic extrapolation with high efficiency and high accuracy can be realized in a non-black box manner, and is suitable for various scenes such as new processes and new materials.
[0080] Based on the first embodiment of the application, in the second embodiment of the application, the same or similar contents as the above-mentioned embodiment one can refer to the above introduction, and the subsequent will not be described in detail. On this basis, please refer to Figure 3 , step S30, the formula process extrapolation method further includes steps S301-S302:
[0081] Step S301, the historical table data and the preliminary variable nodes in the corrected causal graph are mixed and embedded to obtain a fused variable node representation.
[0082] It can be understood that the mixed embedding is performed through a table-graph mixed embedding structure. The table part in the structure refers to structured data (such as numerical or category information such as drug dosage and temperature), which is stored in the form of a two-dimensional table. The historical table data can be used as the table part in the structure. The corrected causal graph (including preliminary variable nodes) is used as the graph in the structure to represent the causal relationship between variables. Then the historical table data and the preliminary variable nodes in the corrected causal graph are mixed and embedded through the structure.
[0083] Specifically, the numerical features in the table and the topological relationship in the graph structure are fused to generate a unified fused variable node representation. For example: the numerical features in the table are encoded into vectors through a neural network. The node relationship (such as neighbor nodes, causal paths) in the graph structure is modeled through a graph neural network. Finally, the two vectors are fused to obtain a node embedding that contains both attribute information and relationship information.
[0084] Step S302, constructing a self-supervised task according to the historical table data and the fused variable node representation, optimizing the fused variable node representation according to the self-supervised task, and obtaining an optimized causal graph.
[0085] It can be understood that self-supervised learning constructs auxiliary tasks through unlabeled data to help the model learn useful features. The self-supervised task can be constructed according to the historical table data and the fused variable node representation, and the fused variable node representation is optimized according to the self-supervised task to obtain an optimized causal graph.
[0086] In a possible implementation, the self-supervised task includes a table Figure 1 consistency task; step S302 can include steps S3021-S3024:
[0087] Step S3021, constructing a table Figure 1 consistency task according to the historical table data and the fused variable node representation.
[0088] It should be noted that the table Figure 1 consistency self-supervised task can be constructed according to the historical table data and the fused variable node representation, to ensure that the variables in the table are consistent with their representations in the graph structure.
[0089] Step S3022, generating a data feature vector according to the historical table data through the table Figure 1 consistency task.
[0090] Step S3023, generating a structure feature vector according to the fused variable node representation through the table Figure 1 consistency task.
[0091] In a specific implementation, for the same variable, a data feature vector is generated by extracting features (such as numerical values, categories) from the historical table data. A structure feature vector is generated by extracting structure features (such as neighbor nodes, causal paths) from the fused variable node representation in the optimized causal graph.
[0092] Step S3024, performing consistency optimization on the data feature vector and the structure feature vector according to a preset consistency loss function to obtain an optimized causal graph.
[0093] It can be understood that a consistency loss function (such as a contrastive loss) can be designed to assist optimization, to perform consistency optimization on the data feature vector and the structure feature vector, to force the two vectors to be close in the embedding space, to ensure the information consistency of the table and the graph, and to obtain an optimized causal graph.
[0094] In another possible implementation, the self-supervised task includes a mask field prediction task; step S302 can further include steps S3025-S3028:
[0095] Step S3025, constructing a mask field prediction task according to the historical table data and the fused variable node representation.
[0096] It should be noted that a self-supervised task of mask field prediction can also be constructed according to historical table data and the fused variable node representation, and the dependency between variables is learned.
[0097] In step S3026, part of the data in the historical table data is randomly selected for masking through the mask field prediction task, and the masked table data is obtained.
[0098] In step S3027, the masked table data is predicted according to the fused variable node, and a prediction result is obtained.
[0099] In a specific implementation, some fields (such as drug dosage, temperature) in the historical table data are randomly masked through the mask field prediction task, and the masked values are predicted according to the context (unmasked fields in the historical table data, graph structure and fused variable node representation), and a prediction result is obtained.
[0100] In step S3028, the fused variable node is optimized according to a preset prediction loss function and the prediction result, and an optimized causal graph is obtained.
[0101] It can be understood that a prediction loss function (for example, mean square error (MSE) or cross-entropy loss) is set, only the masked field is calculated, the fused variable node is optimized according to the preset prediction loss function and the prediction result, and an optimized causal graph is obtained. By reconstructing the mask field, the logical association between variables can be learned.
[0102] The process of optimizing the variable node representation of the self-supervised table-graph embedding learning module is described here. If a drug formula needs to be studied, the dosage, temperature, reaction time and other numerical values of each drug are recorded through table data. The causal relationship such as "dosage affects temperature -> temperature affects reaction speed" is represented through the graph structure. The mixed embedding operation is performed to fuse the dosage value, temperature value and causal chain relationship to generate a comprehensive vector. Two self-supervised tasks are generated, and the consistency discrimination loss judges whether the table record of "dosage = 10ml" is consistent with the causal path of "dosage -> temperature" in the graph; the mask field prediction masks the "reaction time" field, and the reaction time is predicted according to the dosage and temperature. Figure 1
[0103] In this embodiment, the historical table data and the preliminary variable nodes in the corrected causal diagram are mixed and embedded to obtain a fused variable node representation; a self-supervised task is constructed according to the historical table data and the fused variable node representation, and the fused variable node representation is optimized according to the self-supervised task to obtain an optimized causal diagram. The formulation variable node representation is optimized through the self-supervised table-graph embedding learning module. The table-graph mixed embedding and the self-supervised task enable the node representation, which can automatically supplement the knowledge required for reasoning of new data.
[0104] Based on the first embodiment of the present application, the same or similar contents as the above-mentioned embodiment one can be referred to the above introduction, and will not be described in detail. On this basis, please refer to Figure 4 , step S40, the formulation process extrapolation method further includes steps S401-S405:
[0105] Step S401, the optimized causal diagram is used for formulation process extrapolation to generate a new decision path.
[0106] It can be understood that the optimized causal diagram is used for formulation process extrapolation, and a plurality of new decision paths corresponding to new formulations and new processes can be generated. Examples of the new decision paths generated by extrapolation can be referred to Figure 5 , the figure contains a complex network corresponding to all the new decision paths generated, which may contain redundant edges and weak associations, and the dashed line edge represents a weakly associated path that needs to be pruned and eliminated.
[0107] Step S402, the new decision path is pruned and optimized according to the historical formulation process sample and the unlabeled new sample to obtain an optimized decision path.
[0108] It should be understood that after the new formulation and new process are extrapolated, the causal influence pruning verification is performed between the historical formulation process sample and the unlabeled new sample, and only the variable combination that has a significant positive influence on the performance in the example is retained. Combination innovation is also supported, for example, a new solvent + new stirring method is mixed and matched, which can detect the change of the "synergistic / antagonistic" edge and automatically recommend a better structure. The pruned and optimized decision path generated is shown in Figure 6 .
[0109] Step S403, generating a decision path backtracking tree according to the optimized decision path.
[0110] Step S404, visualizing the decision path backtracking tree, and changing the decision path backtracking tree when a change operation of a user is detected to obtain a changed decision path.
[0111] Step S405, generating new formulation process data according to the changed decision path.
[0112] It can be understood that the formula and process extrapolation, automatic output decision path trace tree, clear physical reason and table-graph joint basis of each step reasoning. Provide expert visualization interface, manually add / change formula inter-causal relationship, get changed decision path, form human-machine co-construction progressive optimization chain. Optimization results automatic visualization reasoning link, support experts to adjust business and trace back the historical reason chain. Finally, according to the changed decision path, automatically generate the optimization before and after the causal graph, the "high contribution edge", the final recommended new formula process data and the expected performance improvement details.
[0113] In the present embodiment, the optimized causal graph is extrapolated to generate a new decision path; the new decision path is optimized by causal pruning according to historical formula process samples and unlabeled new samples to obtain an optimized decision path; a decision path trace tree is generated according to the optimized decision path; the decision path trace tree is visualized and changed when a user's change operation is detected to obtain a changed decision path; and new formula process data is generated according to the changed decision path. A new decision path is generated by causal tracing and pruning optimization, and links that are not beneficial or negatively correlated with the final performance are automatically pruned. The decision path trace tree is generated and displayed to support experts to adjust the business and trace back the historical reason chain. Finally, new formula process data is generated.
[0114] For example, in order to facilitate the understanding of the implementation process of the formula process extrapolation method obtained after the above-mentioned embodiment one, please refer to Figure 7 , Figure 7 An overall flowchart of a formula process extrapolation method is provided, specifically:
[0115] First, the table data flow modeling is performed, the historical table data is shown in the following table (1), the system automatically builds a causal graph G, and the nodes are each formula parameter / process step / performance. Combined with part of the expert knowledge, the causal chain (such as "high shear stirring-dispersion improvement-strength improvement" chain) and physical constraints (proportion and process physical edge constraints) are preset, and the edges are corrected by Bayesian score-search algorithm.
[0116] Example of historical table data Table (1)
[0117] id Formulation process Product properties 01 AB resin 40%, acetone 10%, baking 180°C 10 min Bonding strength 12 MPa 02 AB resin 42%, acetone 9%, UV curing Bonding strength 14 MPa 03 New C additive 1%, high shear mixing, baking 200°C 8 min Bonding strength 16 MPa
[0118] Secondly, enter the new formula extrapolation process. 1. Knowledge embedding self-generation: train table with unlabeled new cases Figure 1The task is to determine the cause of the problem, such as automatically guessing the missing values of a process mask downstream performance. 2. Causal tracing and pruning optimization: generate new combinations (such as C additives + UV curing + cooling curing), automatically prune links that are not beneficial or negatively correlated with final performance. 3. Expert collaborative optimization: visual trace shows that "the main contribution of adhesion strength improvement is due to the combination of C additives and high shear mixing", experts can confirm or correct the link with one key. 4. Result inference report export: automatically generate optimization causal diagram, optimization after clear "high contribution edge", the final recommended formula and the expected performance details.
[0119] The beneficial effects of the present application relative to the existing solutions are shown in the following beneficial effect table (2).
[0120] Beneficial effect table (2)
[0121]
[0122]
[0123] It should be noted that the above examples are only for understanding the present application and do not constitute a limitation on the formula process extrapolation method of the present application. Based on this technical concept, more forms of simple transformation are within the protection scope of the present application.
[0124] The present application also provides a formula process extrapolation device, please refer to Figure 8 The formula process extrapolation device comprises:
[0125] A causal graph generation module 10 is configured to generate a preliminary causal directed graph according to historical table data, the historical table data being table data formed after historical formula process data is structured, and the preliminary causal directed graph comprising preliminary variable nodes and preliminary causal edges.
[0126] A causal edge correction module 20 is configured to correct the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph.
[0127] A node optimization module 30 is configured to perform an optimized representation on the preliminary variable nodes in the corrected causal graph to obtain an optimized causal graph.
[0128] A formula process extrapolation module 40 is configured to perform formula process extrapolation on the optimized causal graph to generate new formula process data.
[0129] The formula process extrapolation device provided in the application adopts the formula process extrapolation method in the above embodiment, and can solve the technical problem. Compared with the prior art, the formula process extrapolation device provided in the application has the same beneficial effects as the formula process extrapolation method provided in the above embodiment, and other technical features in the formula process extrapolation device are the same as the features disclosed in the above embodiment method, which will not be described here.
[0130] The application provides a formula process extrapolation device, which comprises at least one processor and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the formula process extrapolation method in the above embodiment one.
[0131] Reference will be made to the following description of the embodiments of the application with reference to the drawings. Figure 9 The formula process extrapolation device provided in the application can include but is not limited to mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), and the like, and fixed terminals such as digital TVs, desktop computers, and the like. Figure 9 The formula process extrapolation device shown is only an example, and should not bring any limitation to the function and use range of the application.
[0132] As Figure 9As shown, the recipe process extrapolation device can include a processing apparatus 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to programs stored in a read-only memory 1002 or loaded from a storage apparatus 1003 into a random access memory 1004. Various programs and data required for the recipe process extrapolation device to operate are also stored in the random access memory 1004. The processing apparatus 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other by a bus 1005. An input / output interface 1006 is also connected to the bus. Generally, the following systems can be connected to the input / output interface 1006: input apparatuses 1007 including, for example, a touch panel, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output apparatuses 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage apparatus 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication apparatus 1009. The communication apparatus 1009 can allow the recipe process extrapolation device to communicate with other devices wirelessly or by wire to exchange data. Although the recipe process extrapolation device having various systems is shown in the figure, it should be understood that all the shown systems are not required to be implemented or possessed. More or less systems can be alternatively implemented or possessed.
[0133] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by a communication apparatus, or installed from the storage apparatus 1003, or installed from the ROM 1002. When the computer program is executed by the processing apparatus 1001, the above-mentioned functions defined in the methods of the embodiments of the present disclosure are performed.
[0134] The recipe process extrapolation device provided by the present disclosure adopts the recipe process extrapolation method in the above embodiments, and can solve the technical problem of recipe process extrapolation. Compared with the prior art, the recipe process extrapolation device provided by the present disclosure has the same beneficial effects as the recipe process extrapolation method provided by the above embodiments, and other technical features in the recipe process extrapolation device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.
[0135] It should be understood that parts of the present disclosure can be realized by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.
[0136] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
[0137] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e. computer programs) for performing the recipe process extrapolation method in the above embodiments.
[0138] The computer readable storage medium provided by the present application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system, system or device, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read only memory (ROM), an erasable programmable read only memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), etc., or any suitable combination of the above.
[0139] The above computer readable storage medium can be contained in a recipe process extrapolation device; or can exist separately without being assembled into a recipe process extrapolation device.
[0140] The computer readable storage medium described above carries one or more programs, when the one or more programs are executed by the recipe process extrapolation device, cause the recipe process extrapolation device to: generate a preliminary causal directed graph according to historical table data, the historical table data being table data formed after historical recipe process data is structured, the preliminary causal directed graph including preliminary variable nodes and preliminary causal edges; correct the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph; perform an optimized representation on the preliminary variable nodes in the corrected causal graph to obtain an optimized causal graph; and perform recipe process extrapolation on the optimized causal graph to generate new recipe process data.
[0141] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).
[0142] The computer program code can also be loaded onto a computer or other programmable information processing apparatus to cause a series of operations to be performed on the computer or other programmable information processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable information processing apparatus implement the functions / acts specified in the flowchart and / or block diagram block or blocks.
[0143] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. In some cases, the name of the module does not constitute a limitation on the module itself.
[0144] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e., a computer program) for executing the above-mentioned formula process extrapolation method, and can solve the technical problems. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the formula process extrapolation method provided by the above-mentioned embodiments, which will not be repeated here.
[0145] The present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the formula process extrapolation method as described above.
[0146] The computer program product provided by the present application can solve the technical problems. Compared with the prior art, the computer program product provided by the present application has the same beneficial effects as the formula process extrapolation method provided by the above-mentioned embodiments, which will not be repeated here.
[0147] The above only describes some embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation, direct / indirect application in other related technical fields based on the technical concept of the present application, and the contents of the present application and the accompanying drawings are included in the patent protection scope of the present application.
Claims
1. A recipe extrapolation method, characterized by, The method comprises: generating a preliminary causal directed graph according to historical table data, the historical table data being table data formed after historical recipe process data is structured, the preliminary causal directed graph comprising preliminary variable nodes and preliminary causal edges; correcting the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph; optimizing the preliminary variable nodes in the corrected causal graph to obtain an optimized causal graph; performing recipe process extrapolation on the optimized causal graph to generate new recipe process data.
2. The method of claim 1, wherein, The step of correcting the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph comprises: physically constraining the preliminary causal edges in the preliminary causal directed graph according to preset causal chain and recipe process corresponding physical edge constraints to obtain constrained causal edges; evaluating the constrained causal edges according to a preset scoring function to obtain evaluation scores; correcting the constrained causal edges according to a preset search strategy and the evaluation scores to obtain a corrected causal graph.
3. The method of claim 1, wherein, The step of optimizing the preliminary variable nodes in the corrected causal graph to obtain an optimized causal graph comprises: performing hybrid embedding of the historical table data and the preliminary variable nodes in the corrected causal graph to obtain fused variable node representations; constructing a self-supervised task according to the historical table data and the fused variable node representations, and optimizing the fused variable node representations according to the self-supervised task to obtain an optimized causal graph.
4. The method of claim 3, wherein, The self-supervised task comprises a table-graph consistency task. The step of constructing a self-supervised task according to the historical table data and the fused variable node representations, and optimizing the fused variable node representations according to the self-supervised task to obtain an optimized causal graph comprises: constructing a table-graph consistency task according to the historical table data and the fused variable node representations; generating a data feature vector according to the historical table data through the table-graph consistency task; generating a structure feature vector according to the fused variable node representations through the table-graph consistency task; optimizing the data feature vector and the structure feature vector according to a preset consistency loss function to obtain an optimized causal graph.
5. The method of claim 3, wherein, The self-supervised task comprises a mask field prediction task. The step of constructing a self-supervised task according to the historical table data and the fused variable node representations, and optimizing the fused variable node representations according to the self-supervised task to obtain an optimized causal graph comprises: constructing a mask field prediction task according to the historical table data and the fused variable node representations; randomly selecting part of the data in the historical table data to obtain masked table data through the mask field prediction task; performing data prediction on the masked table data according to the fused variable nodes to obtain a prediction result; optimizing the fused variable nodes according to a preset prediction loss function and the prediction result to obtain an optimized causal graph.
6. The method of claim 1, wherein, The step of performing recipe process extrapolation on the optimized causal graph to generate new recipe process data comprises: performing recipe process extrapolation on the optimized causal graph to generate a new decision path; optimizing the new decision path according to historical recipe process samples and unlabeled new samples to obtain an optimized decision path; generating a decision path trace tree according to the optimized decision path; visually displaying the decision path trace tree and changing the decision path trace tree when a change operation of a user is detected to obtain a changed decision path; generating new recipe process data according to the changed decision path.
7. A formulation extrapolation device, characterized by, The recipe process extrapolation device comprises: a causal graph generation module configured to generate a preliminary causal directed graph according to historical table data, the historical table data being table data formed after historical recipe process data is structured, and the preliminary causal directed graph comprising preliminary variable nodes and preliminary causal edges; a causal edge correction module configured to correct the preliminary causal edges in the preliminary causal directed graph to obtain a corrected causal graph; a node optimization module configured to perform an optimized representation on the preliminary variable nodes in the corrected causal graph to obtain an optimized causal graph; a recipe process extrapolation module configured to perform recipe process extrapolation on the optimized causal graph to generate new recipe process data.
8. A recipe extrapolation device, characterized by, The device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the recipe process extrapolation method according to any one of claims 1 to 6.
9. A storage medium, characterized by The storage medium is a computer-readable storage medium, and the storage medium stores a computer program, which is executed by a processor to implement the steps of the recipe process extrapolation method according to any one of claims 1 to 6.
10. A computer program product, characterised in that, The computer program product comprises a computer program, which is executed by a processor to implement the steps of the recipe process extrapolation method according to any one of claims 1 to 6.