Intelligent updating method for master-slave edition linkage database

Abstract

The application provides an intelligent updating method for a master and sub-plate linkage database, and belongs to the technical field of database updating, comprising the following steps: based on a multi-dimensional parameterized association model, different key elements of a garment plate type are converted into quantitative association parameters, and a real-time binding relationship between a master plate and a sub-plate is established based on a mapping rule of a database layer; a multi-modal garment process knowledge graph is constructed, multi-modal data are fused, garment plate type data and key elements are deeply associated to obtain a plate type adjustment vector set, a multi-dimensional change type intelligent recognition and priority determination engine is used to automatically recognize the master plate change type and determine the adjustment priority based on the real-time binding relationship, and the preset AI self-optimization garment structure mathematical formula library and the plate type adjustment vector set are combined to drive the sub-plate to complete the adaptive adjustment of the corresponding parts and realize intelligent updating. Without manual adjustment of each plate, the plate type updating efficiency is greatly improved, and the size deviation and process defects are effectively reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of database updating, in particular to an intelligent updating method for a master-sub version linkage database. BACKGROUND

[0002] In the context of fast iteration of clothing styles, multiple sizes and multiple versions in parallel in the clothing industry, traditional pattern management often requires manual adjustment of each sub-pattern after the master version is modified, resulting in low efficiency, size deviation and inconsistent process parameters, etc. Not only time-consuming and laborious, but also prone to production rework due to human error, which seriously affects the production cycle and product quality. The current market lacks mature tools that can realize intelligent linkage of version parameters, and enterprises have an urgent need for standardized and automated pattern management.

[0003] Therefore, the present application provides an intelligent updating method for a master-sub version linkage database. SUMMARY

[0004] The present application provides an intelligent updating method for a master-sub version linkage database to solve the above technical problems.

[0005] The present application provides an intelligent updating method for a master-sub version linkage database, comprising:

[0006] Step 1: Based on a multi-dimensional parameterized correlation model, different key elements of a clothing version are converted into quantitative correlation parameters, and a real-time binding relationship between the master version and the sub-pattern is established based on the mapping rules of the database layer;

[0007] Step 2: Construct a multi-modal clothing process knowledge graph, and fuse multi-modal data to deeply correlate clothing version data and key elements to obtain a version adjustment vector set, wherein the version adjustment vector set includes: a basic adjustment vector , a scene adaptation vector and a process compensation vector ;

[0008] Step 3: Use a multi-dimensional change type intelligent recognition and priority determination engine to automatically identify the master version change type and determine the adjustment priority based on the real-time binding relationship, combine the pre-set AI self-optimization clothing structure mathematical formula library and the version adjustment vector set, and drive the sub-pattern to complete the adaptive adjustment of the corresponding parts to realize intelligent updating, wherein the change type includes: explicit change type and implicit change type.

[0009] Preferably, the key elements include: basic size, structure line, process standard, fabric attribute, version style, adaptation scene and equipment parameter, and the real-time binding relationship includes: rigid binding, flexible binding, predictive binding, time sequence correlation binding, fabric-process coordination binding, user portrait binding and equipment adaptation binding.

[0010] Preferably, the multi-dimensional parameterized association model is constructed, comprising:

[0011] The mutual information entropy is used to quantify the association strength of any two key elements;

[0012] The rigid binding weight wr is determined based on the time effect coefficient, the flexible binding weight wf is determined according to the association strength, the predictive binding weight wp is determined according to the LSTM neural network, the time sequence association binding weight gq is determined according to the historical change time sequence data of the master version and the sub-sample version, the fabric-process collaborative binding weight mb is determined according to the deep association of the master version fabric attribute change and the sub-sample process parameter adjustment, the user portrait binding weight hb is obtained by establishing the association relationship between the master version and the sub-sample version of the target user group according to the portrait features of the target user group, the equipment adaptation binding weight sb is determined by associating the master version change and the parameter constraint of the sub-sample production equipment, and the real-time binding relationship between the master version and the sub-sample version is quantitatively controlled through Normalization, wherein, is a diagonal matrix, and the diagonal elements are the reciprocal of the maximum threshold value of each weight; is a normalization function.

[0013] Preferably, the mapping rule comprises a cross-level mapping logic, and the cross-level mapping logic comprises:

[0014] A double mapping function is constructed , wherein M1 is a master version belonging to a basic version library corresponding to a level L0, and S1 is a sub-sample version belonging to a derived version library corresponding to a level L1-L3; is the i1th key parameter of the master version; is the sub-sample version associated with the i2th derived parameter; is the association degree of and ; is a reverse constraint factor; is the number of key parameters of the master version; is the number of derived parameters of the sub-sample version based on ;

[0015] A level attenuation coefficient is determined , wherein, is the level attenuation coefficient; is the level difference, i.e., the level interval between the master version and the sub-sample version;

[0016] When , the sub-sample version is updated, wherein, is the minimum response threshold; is the change amount of the i1th key parameter of the master version.

[0017] Preferably, the garment pattern data is deeply associated with key elements to obtain a pattern adjustment vector set, including:

[0018] The pre-trained BERT model is used to analyze the process standard document to output a semantic feature vector T1, the modified YOLOv8 model is used to analyze the pattern CAD drawing to output a visual feature vector I1, the PointNey model is used to extract the spatial feature vector V1 of the human-garment fitting model, and the wavelet transform is used to extract the time-frequency feature vector A1;

[0019] The attention weights based on T1, I1, V1 and A1 are determined respectively, and a fusion feature vector F is generated;

[0020] The multi-modal garment process knowledge graph is divided into a core layer, an associated layer and an extended layer, wherein the relationship between nodes is quantified based on the cross-modal association strength R0(a, b): Wherein, is the mutual information of node a and node b in the multi-modal garment process knowledge graph; is the cosine similarity of the fusion feature ; is the set weight;

[0021] When , an explicit association edge between node a and node b is established;

[0022] When , an implicit association edge is established and the association confidence is marked, wherein, , is the strength threshold;

[0023] Based on the association strength of the core layer elements and the target pattern, a basic adjustment vector is determined, wherein, is the key element of the core layer; is the association strength of the core layer and the target pattern; is the current parameter value of ; is the standard parameter value of ;

[0024] Based on the scene-parameter mapping matrix, a scene adaptation vector is determined, wherein, is the feature matrix of the scene ; is the feature matrix of the current pattern; is the scene-parameter mapping matrix based on the scene ; and

[0025] Determine a process compensation vector based on the correlation confidence of the implicit correlation edge wherein, is a compensation coefficient; is an adjustment amount of the implicit correlation parameter; is the correlation confidence of the implicit correlation edge

[0026] Obtain a comprehensive adjustment vector set based on the vector fusion matrix .

[0027] Preferably, automatically identify the master change type and determine the adjustment priority based on the real-time binding relationship, including:

[0028] Match the range influence index consistent with the master change type from the type-range correspondence table;

[0029] Determine the dynamic binding weight based on the real-time binding relationship, and combine the range influence index and the delivery deadline urgency factor to obtain the adjustment priority.

[0030] Preferably, combine the preset AI self-optimization garment structure mathematical formula library and the pattern adjustment vector set to drive the sub-sample pattern to complete the adaptive adjustment of the corresponding parts to realize intelligent updating, including:

[0031] Call the size adaptive formula wherein, , is an AI self-optimization part coefficient; is the adjusted size of the j3th part of the sub-sample pattern; is the size of the j3th part of the sub-sample pattern before adjustment; is the size change rate of the j3th part of the sub-sample pattern; is a master-sub-sample correlation matrix; is the adjustment priority of the j3th part of the sub-sample pattern;

[0032] Call the priority weighting scenario formula wherein, is a scenario sensitivity coefficient; is a style coefficient; is the original length of the structural line of the j3th part of the sub-sample pattern; is the adjusted length of the structural line of the j3th part of the sub-sample pattern;

[0033] Call the conflict resolution enhancement formula wherein, , is a priority compensation coefficient; is the original parameter of the process-related structure of the j3th part of the sub-sample pattern; ​​an adjusted parameter of a process-related structure of a j3th component of a sub-plate;

[0034] generating a virtual prototype of the sub-plate according to the adjusted parameter obtained according to the calling formula and determining an overall deviation, and if the overall deviation is greater than a preset deviation, triggering formula library parameter self-optimization.

[0035] Preferably, the dynamic binding weight is determined based on the real-time binding relationship, and the adjustment priority is obtained in combination with a range influence index and a delivery deadline urgency factor, and comprises:

[0036] performing same-element cluster analysis on the historical vectors in the historical database to obtain an average value under a maximum cluster set, and constructing a reference vector, and simultaneously performing vector cluster analysis on all the historical vectors to obtain a first center cluster vector under a maximum cluster set;

[0037] Taking the maximum cluster set to which the first average value of each element belongs as a first set, performing intra-set cluster analysis on all the historical vectors in the first set to obtain a second center cluster vector of an intra-set maximum cluster;

[0038] determining difference vectors of the reference vector, the first center cluster vector and the second center cluster vector from the normalized vector respectively to obtain a difference matrix, and performing linear fitting on each column vector in the difference matrix to obtain a dynamic binding coefficient;

[0039] ;

[0040] wherein, is a dynamic binding weight of a corresponding column vector; is a weight allocation coefficient; is a corresponding linear fitting slope; is a corresponding linear fitting intercept; is a maximum value of a recent three-month historical fitting slope and a historical fitting intercept of a corresponding binding dimension; is an adjustment coefficient of a discrete point in a linear fitting process under a corresponding binding dimension; is an entropy weight basic weight of a cluster to which a corresponding binding dimension belongs; is a process adjustment coefficient under a corresponding binding dimension;

[0041] obtaining a dynamic binding weight based on the dynamic binding coefficient and the adjustment coefficient of each column vector wherein, is an average value of all satisfying a normal distribution; is a standard deviation of all s; N0 is the number of column vectors; is a factorial symbol; is a maximum value of all The minimum absolute value of the difference;

[0042] The dynamic binding weight, the scope influence index, and the delivery urgency factor are combined into an input vector, which is then input into the vector analysis model to obtain the adjustment priority.

[0043] Compared with the prior art, the beneficial effects of this application are as follows:

[0044] The system achieves automated and intelligent linkage updates between the master pattern and the sub-patterns in three steps, eliminating the need for manual adjustments to each pattern. This significantly improves the efficiency of pattern updates, effectively reduces size deviations and process defects, and meets the diverse and rapidly iterating production needs of the apparel industry.

[0045] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0046] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0048] Figure 1 This is a flowchart of an intelligent update method for a parent-child linked database according to an embodiment of the present invention. Detailed Implementation

[0049] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0050] This invention proposes an intelligent update method for a parent-child version linked database, such as... Figure 1 As shown, it includes:

[0051] Step 1: Based on the multi-dimensional parametric association model, the different key elements of the garment pattern are transformed into quantitative association parameters, and a real-time binding relationship between the master pattern and the sub-pattern is established based on the mapping rules of the database layer.

[0052] Step 2: Construct a multimodal garment technology knowledge graph, integrate multimodal data to deeply correlate garment pattern data with key elements to obtain a pattern adjustment vector set, wherein the pattern adjustment vector set includes: basic adjustment vectors. Scene adaptation vectors and process compensation vector ;

[0053] Step 3: Adopting multi-dimensional change type intelligent recognition and priority judgment engine, automatically identifying the change type of the master plate and judging the adjustment priority based on real-time binding relationship, combining the preset AI self-optimization clothing structure mathematical formula library and the pattern adjustment vector set, driving the sub-sample plate to complete the adaptive adjustment of the corresponding parts to realize intelligent updating, wherein the change type includes: explicit change type and implicit change type.

[0054] Preferably, the key elements include: basic size, structure line, process standard, fabric attribute, pattern style, adaptation scenario and equipment parameter, and the real-time binding relationship includes: rigid binding, flexible binding, predictive binding, time sequence associated binding, fabric-process collaborative binding, user portrait binding and equipment adaptation binding.

[0055] In this embodiment, the multi-dimensional parameterized association model is a model that converts key elements of a clothing pattern into quantifiable association parameters. For example, converting key elements such as basic size (e.g., a dress S size chest circumference of 82 cm), structure line (e.g., the collar contour shape of a shirt), process standard (e.g., the thickness of the adhesive lining of a suit), fabric attribute (e.g., the weight of wool fabric , pattern style (e.g., the style characteristics of a workwear jacket), adaptation scenario (e.g., the scenario attribute of a formal dress), and equipment parameter (e.g., the printing resolution of a digital printing machine of 300 dpi) into specific numerical values or characteristic parameters.

[0056] The mapping rule of the database layer is the corresponding relationship rule between the parameters of the master plate and the sub-sample plate. For example, the length of a master plate sweatshirt is 65 cm, and the length of a sub-sample plate M size is 67 cm and the length of a sub-sample plate L size is 70 cm. The mapping is established according to the rule of master plate length + size difference x 1.5 cm. Taking a men's shirt master plate as an example, the basic size is 40 size chest circumference 90 cm, the structure line is square collar, the process standard is needle distance 3 cm / 12 needles, the fabric is cotton blend, the pattern style is business, the adaptation scenario is commuting, and the production equipment is a cutting bed with cutting accuracy ±0.2 cm. After quantizing these elements, real-time binding relationship is established with sub-sample plate 42 size (chest circumference 94 cm) and sub-sample plate 44 size (chest circumference 98 cm). When the master plate chest circumference is adjusted, the sub-sample plate is adjusted synchronously according to the mapping rule.

[0057] In this embodiment, the multi-modal clothing process knowledge graph is a knowledge graph constructed by fusing multi-modal data such as semantics, vision, space, and time-frequency.

[0058] Pattern adjustment vector set: basic adjustment vector is the adjustment vector of direct change of key elements such as basic size. Scene adaptation vector is the adjustment vector of adaptation scenario requirements. Process compensation vector is a compensation adjustment vector caused by fabric or process factors. Taking a sports shirt as an example, the semantic data is the BERT encoding of the sports shirt process standard, the visual data is the hood contour feature of the CAD drawing of the sports shirt, the spatial data is the fitting space feature of the shirt when the human body moves, and the time-frequency data is the time-frequency feature of the elasticity of the fabric. After fusing these data, the basic adjustment vector (the adjustment of increasing the mother's chest circumference by 2 cm), the scene adaptation vector (the adjustment of increasing the length of the shirt by 1 cm in the sports scene), and the process compensation vector (the adjustment of the elastic fabric needle distance by 0.1 cm) are obtained.

[0059] In this embodiment, the multi-dimensional change type intelligent recognition and priority determination engine can automatically recognize the explicit change and implicit change of the master version, and determine the adjustment priority. The explicit change is, for example, the waistline of the master dress is directly modified from 66 cm to 68 cm; the implicit change is, for example, the terylene content of the master fabric is changed from 30% to 35%, which indirectly affects the sewing process. The priority determination can be combined with the binding relationship, delivery period, etc., such as the priority of the child sample of the urgent order is higher than that of the regular order.

[0060] The AI self-optimizing clothing structure mathematical formula library includes a library of various clothing pattern adjustment formulas, and the formulas can be automatically optimized according to the adjustment effect.

[0061] Explicit change type: direct and easily recognizable changes of the master version, such as the collar of the master shirt is changed from a square collar to a round collar.

[0062] Implicit change type: indirect and not easily recognizable changes of the master version, such as the color fastness of the master fabric changes slightly, which indirectly affects the washing process. Taking the master jacket as an example, the explicit change is that the length of the jacket is changed from 75 cm to 76 cm, and the implicit change is that the cotton content of the fabric is changed from 80% to 75% (which affects the sewing needle distance). After the engine recognizes, it combines the AI formula library and the pattern adjustment vector set to drive the child sample M code (length 77 cm) and L code (length 79 cm) to adjust the length and needle distance at the same time, and according to the urgency of the delivery period, the child sample of the urgent order is updated in priority.

[0063] The beneficial effects of the above technical solutions are: through the three steps, the automatic and intelligent linkage update of the clothing master version and the child sample is realized, without manual adjustment of each version, the efficiency of pattern update is greatly improved, the size deviation and process defects are effectively reduced, and the production demands of the clothing industry in multiple scenes and fast iteration can be met.

[0064] The present application proposes an intelligent updating method of a child-mother version linkage database, and constructs a multi-dimensional parameterized correlation model, which comprises:

[0065] The mutual information entropy is used to quantify the correlation strength of any two key elements;

[0066] The rigid binding weight wr is determined based on the aging influence coefficient, the flexible binding weight wf is determined according to the correlation strength, the predictive binding weight wp is determined according to the LSTM neural network, the time sequence correlation binding weight gq is determined according to the historical change time sequence data of the master version and the sub-sample version, the fabric-process collaborative binding weight mb is determined according to the deep correlation between the master version fabric attribute change and the sub-sample process parameter adjustment, the user portrait binding weight hb is obtained by establishing the association relationship between the master version and the sub-sample version of the target user group, the equipment adaptation binding weight sb is determined by associating the master version change and the parameter constraint of the sub-sample production equipment, and the real-time binding relationship between the master version and the sub-sample version is quantitatively regulated through normalization. is a diagonal matrix, and the diagonal elements are the reciprocal of the maximum threshold value of each weight; is a normalization function.

[0067] In this embodiment, the calculation of the correlation strength is specifically: wherein, the joint information entropy is the single-element information entropy, and , is the parameter probability distribution of the key element ; is the key element; is the parameter probability distribution of the key element ; wherein, is obtained based on the historical version database with a sample size of not less than 1000 groups.

[0068] In this embodiment, wherein, is the aging influence coefficient, the value is 0.3, which is obtained based on the statistical update cycle of the garment industry version, is the decay factor, that is, the decay rate of the weight with time, the value is 0.05, which is set in advance, is the parameter update interval.

[0069] wherein, is the element importance coefficient, which is determined based on the analytic hierarchy process, the basic size =0.3, the structure line =0.25, the process standard =0.2, the sum of the rest elements =0.25, and n is the number of elements. is the correlation strength of the key element .

[0070] ​The wp is output by an LSTM neural network, whose input is the change frequency and adjustment amplitude data of the past 3 months, the network contains 2 hidden layers with 64 and 32 nodes respectively, and the activation function uses ReLU.

[0071] In this embodiment, the weight of the time sequence correlation binding is: wherein, is the change amount of the master version at time ; is the correlation change amount of the sub version at lag time ; is the time sequence correlation coefficient of the sliding window; is the current time, used to weaken the influence of long-term historical data.

[0072] In this embodiment, the weight of the fabric-process coordination binding is: wherein, is the core attribute vector of the master fabric, and is the elastic modulus, is the thickness, is the drape coefficient; is the adaptation parameter vector of the sub process, and is the sewing tension, is the cutting allowance, is the ironing temperature; is the weight, is 0.4, is 0.3, is 0.3, and the weight thereof is obtained based on fabric-process conflict historical data training.

[0073] In this embodiment, the weight of the user portrait binding is: wherein, is the version requirement distribution of the segmented user group; is the parameter set of the master version change, is the intersection-union ratio of demand and change, is the user segmentation weight, which is determined based on the sales proportion, and different user groups correspond to different values, for example, the core user group is 0.8, the ordinary user group is 0.6, and the like.

[0074] The weight of the equipment adaptation degree binding is: wherein, is the upper limit of the process parameter of the equipment, is the maximum fabric thickness of the cutting machine; is the minimum stitch spacing of the sewing machine; is the corresponding parameter of the sub sample version after adjustment; is the equipment weight, and is 0.6, is 0.4, which is set based on the influence of the device on production efficiency, and the higher the value, the better the device compatibility.

[0075] In this embodiment, it is assumed that the reciprocal of each maximum threshold is 1.

[0076] The beneficial effects of the above technical solutions are: by quantifying the element association strength through mutual information entropy, combining the accurate calculation and normalized regulation of multi-dimensional binding weight, the quantification and dynamic optimization of the real-time binding relationship between the master version and the sub-sample version are realized, the influence of time efficiency, prediction, process, user, device and other multi-dimensional factors on the binding relationship can be accurately captured, and the quantifiable and controllable binding relationship is ensured, so that the driving of the master version change on the sub-sample version is more accurate and efficient, and the intelligent upgrading of the garment pattern management is facilitated.

[0077] The present application provides an intelligent updating method for a master-sub version linkage database, wherein the mapping rule comprises a cross-level mapping logic, and the cross-level mapping logic comprises:

[0078] Constructing a double mapping function , wherein M1 is a master version belonging to a basic pattern library corresponding level L0, and S1 is a sub-sample version belonging to a derived pattern library corresponding level L1-L3. is the i1th key parameter of the master version; is the sub-sample version associated with the i2th derived parameter; is the association degree of ; is the association degree of ; is a reverse constraint factor; is the number of key parameters of the master version; is the number of derived parameters of the sub-sample version based on ; determine the level attenuation coefficient , wherein is the level attenuation coefficient; is the level difference, i.e., the level interval between the master version and the sub-sample version;

[0079] when , trigger the sub-sample version update, wherein is the minimum response threshold; is the change amount of the i1th key parameter of the master version.

[0080] In this embodiment, for example, when L0 to L1, d1=1, when L0 to L2, d1=2.

[0081] In this embodiment, the minimum response threshold is 1 / 3 of the master parameter standard deviation, which is calculated based on historical data, such as 1 / 3 of the master chest circumference standard deviation, i.e., 3 cm, .

[0082] In this embodiment, , which is determined based on the process manual.

[0083] In this embodiment, , which is obtained by calibrating 100 master-subsample version change data.

[0084] In this embodiment, It can be adjusted as needed, such as 0.5.

[0085] The beneficial effects of the above technical solutions are: through the double mapping function, the bidirectional optimization of the master and the parameter mapping of the subsample is realized, the hierarchical attenuation coefficient accurately controls the affected degree of different hierarchical subsamples, and combined with reasonable update triggering conditions, the accuracy of the master change driving the subsample is ensured, and invalid update of high-level subsamples is avoided, effectively improving the efficiency and precision of garment pattern cross-level linkage, making the subsample update more in line with production and market demand, and helping enterprises to efficiently manage multi-level pattern system.

[0086] The present application proposes an intelligent updating method for a master-subsample linkage database, which deeply associates garment pattern data with key elements to obtain a pattern adjustment vector set, including:

[0087] The pre-trained BERT model is used to analyze the process standard document to output a semantic feature vector T1, the modified YOLOv8 model is used to analyze the pattern CAD drawing to output a visual feature vector I1, the PointNey model is used to extract the spatial feature vector V1 of the human-clothing fitting model, and the wavelet transform is used to extract the time-frequency feature vector A1;

[0088] The attention weights based on T1, I1, V1 and A1 are determined respectively, and a fusion feature vector F is generated;

[0089] The multi-modal garment process knowledge graph is divided into a core layer, an associated layer and an extended layer, wherein the relationship between nodes is quantified based on the cross-modal association strength R0(a, b): , wherein, is the mutual information of node a and node b in the multi-modal garment process knowledge graph; is the cosine similarity of the fusion feature ; is the set weight;

[0090] When , an explicit association edge between node a and node b is established;

[0091] When , an implicit association edge is established and an association confidence is marked, wherein, , is a strength threshold value;

[0092] Based on the association strength of the core layer element and the target version, a basic adjustment vector is determined , wherein, is a key element of the core layer; is a of the core layer; is the association strength of ; is the current parameter value of ;

[0093] Based on the scene-parameter mapping matrix, a scene adaptation vector is determined , wherein, is a feature matrix of the scene ; is a feature matrix of the current version; is a scene-parameter mapping matrix based on the scene ;

[0094] Based on the association confidence of the implicit association edge, a process compensation vector is determined , wherein, is a compensation coefficient; is the adjustment amount of the implicit association parameter; is the association confidence of the implicit association edge ;

[0095] Based on the vector fusion matrix , a comprehensive adjustment vector set is obtained .

[0096] In this embodiment, the pre-trained BERT model is used to analyze the garment process standard document to extract the semantic feature vector T1. For example, for the set-in process specification document of a knitted dress, the BERT model can capture semantic information such as set-in pitch 3.5 cm / 12 needles in it, output the corresponding semantic feature vector, use the pre-trained model BERT-base-uncased, use 2000 ISO811 garment process standard documents + 5000 enterprise internal process manuals (format: DOCX) for fine-tuning, the word segmentation method is WordPiece, the learning rate is 5e-5, the iteration is 10 rounds, and the dropout probability is 0.1.

[0097] The modified YOLOv8 model is used to analyze the CAD drawings of garment patterns and extract the visual feature vector I1. For example, for a shirt CAD drawing, the modified YOLOv8 can recognize visual elements such as lapel shape and sleeve contour, and generate a visual feature vector describing these elements. In the Neck layer of YOLOv8, a pattern feature attention module is added, which inputs a 300x300 pixel garment CAD vector graph (format: DXF). The labeled data includes 12 types of pattern elements such as collar type, sleeve type, and pocket position. The training set: validation set = 8:2, IOU threshold 0.5.

[0098] PointNet model is used to extract the spatial feature vector V1 of the body-clothing fitting model. For example, for a body model wearing a coat, PointNet can capture the coordinates of the fitting points of the coat on the shoulder and the spatial distribution of the waist wrinkles, forming a spatial feature vector.

[0099] Wavelet transform is used to extract the time-frequency feature vector A1 of the fabric physical performance signal. For example, for the stress change signal of spandex fabric during stretching, wavelet transform is used to analyze its characteristics in the time and frequency domains, and the time-frequency feature vector is obtained.

[0100] Attention weight is the importance allocation of T1, I1, V1, and A1 modal features. For example, when analyzing the process of a business suit, the semantic features (T1) of the process document have a higher weight; when analyzing the design of a fashion dress, the visual features (I1) of the CAD drawing have a higher weight. The fusion feature vector F is generated through the attention mechanism.

[0101] is the weight, based on 1000 sets of pattern association data, through 5-fold cross-validation, when the association prediction accuracy ≥92% =0.6.

[0102] Explicit association edge : node edge with high association strength in the knowledge graph. For example, the (y01=0.7) between the basic size and the process standard, an explicit association edge is established, where y01 is the 90th percentile of the association strength of the historical process conflict-free data, i.e. 0.7.

[0103] Implicit association edge : node edge with medium association strength in the knowledge graph, which needs to be labeled with confidence. For example, the R0 between fabric color fastness and washing process is between 0.5 (y02=0.5) and 0.7, an implicit association edge is established, with a confidence of 0.6, y02 is the 50th percentile of the association strength of the historical process small conflict data, such as 0.5.

[0104] Basic adjustment vector This is an adjustment vector calculated based on the correlation strength between core layer elements and the target pattern. For example, if the current core layer chest circumference is 96cm, the standard value is 94cm, and the correlation strength R0=0.8, then... The adjustment range for the mid-chest circumference is 1.6cm.

[0105] Scene adaptation vector It is an adjustment vector based on the scene-parameter mapping matrix. For example, in a commuting scenario... Feature matrix The difference between the current feature matrix and the current version, multiplied by the scene coefficient. (For commuting scenarios, take 1.1) and mapping matrix This allows for a 1cm adjustment in garment length for commuting scenarios, incorporating... .

[0106] Process compensation vector Adjusted vector based on the confidence of implicitly related edges. For example, the confidence of the relationship between the fabric elasticity and stitch pitch of implicitly related edges. =0.6, adjustment amount =0.2cm, compensation coefficient =0.9, then The needle spacing adjustment is 0.108cm. The value is based on 50 sets of implicitly correlated process adjustment data to ensure that the compensated process defect rate is ≤2%.

[0107] Vector fusion matrix Combined adjustment of vector set, It is fusion The weight matrix is ​​as follows: [[0.5,0.3,0.2],[0.4,0.4,0.2],[0.3,0.3,0.4]]. By merging these, a comprehensive adjustment vector is obtained, which guides the overall adjustment of the pattern. For example, after merging, a comprehensive instruction is obtained to adjust the bust by 1.6cm, the garment length by 1cm, and the stitch length by 0.108cm.

[0108] The beneficial effects of the above technical solution are as follows: by extracting multimodal features, hierarchical association of knowledge graphs and multi-vector fusion, multi-dimensional and precise control of garment pattern adjustment can be achieved. By integrating multi-source information such as semantics, vision, space and time frequency, and capturing process details through explicit and implicit associations, the basic adjustment, scene adaptation and process compensation work together to greatly improve the accuracy of pattern adjustment and scene adaptability, and provide comprehensive and efficient technical support for intelligent garment pattern updates.

[0109] This invention proposes an intelligent update method for a master-child version linked database, which automatically identifies the master version change type and determines the adjustment priority based on the real-time binding relationship, including:

[0110] matching a range influence index consistent with the master change type from a type-range table;

[0111] The dynamic binding weight is determined based on the real-time binding relationship between the master and the sub-sample, and the delivery deadline urgency factor is combined with the range influence index to obtain the adjustment priority.

[0112] In this embodiment, the type-range table is prepared in advance, and the table corresponding to the master change type and the change influence sub-sample range. For example, the master change type is divided into major changes (such as changing the style of the version from casual to business), larger changes (such as adding a size series to one code), and minor changes (such as modifying the pocket decoration details), and the corresponding range influence indexes are 10, 5, and 3, respectively. If the version of the master dress is changed from A-line dress to straight dress (major change), the range influence index 10 is matched from the table.

[0113] The master change type refers to the change category of the master in version, size, style, process, etc. For example, version structure change (such as changing from round neck T-shirt to V-neck T-shirt), size change (such as increasing the waist circumference of the master by 3 cm), style change (such as changing from office style to rural style), process change (such as changing the washing process of the jeans from light washing to heavy washing), etc., which are divided into two categories of explicit and implicit.

[0114] The range influence index is a quantitative index for measuring how many sub-samples are affected by the master change, and the higher the index, the wider the range of affected sub-samples. For example, the master brand logo position change (major change) affects all sub-samples, and the range influence index is 1; the master sleeve button color change (minor change) only affects 2 sub-samples, and the range influence index is 0.2, the value range is 0 to 1.

[0115] The dynamic binding weight is a weight dynamically calculated according to the real-time binding relationship between the master and the sub-sample, which reflects the close degree of the association between the sub-sample and the master.

[0116] The delivery deadline urgency factor is a coefficient set according to the delivery urgency of the sub-sample order. The coefficient of the urgent order is high, the coefficient of the regular order is moderate, and the coefficient of the delayable order is low. For example, the factor of the urgent order that needs to be delivered within 2 days is 1.3; the factor of the regular order that needs to be delivered within 5 days is 1.0; the factor of the order that can be delayed to be delivered within 10 days is 0.8, which is set in advance and can be directly used.

[0117] The beneficial effects of the above technical solutions are: through the type-range table, the influence range of the master change is accurately matched, the dynamic weight of the real-time binding relationship and the delivery deadline urgency factor are combined, and the adjustment priority of the sub-sample can be dynamically determined.

[0118] The application provides an intelligent updating method for a master-sub sample linkage database, which combines a preset AI self-optimization garment structure mathematical formula library and a pattern adjustment vector set to drive a sub sample to complete adaptive adjustment of corresponding parts to realize intelligent updating, including:

[0119] Calling a size self-adaption formula , wherein, , is an AI self-optimization part coefficient; is an adjusted size of a j3th part of a sub sample; is an unadjusted size of the j3th part of the sub sample; is a size change rate of the j3th part of the sub sample; is a master-sub sample correlation matrix; is an adjustment priority of the j3th part of the sub sample;

[0120] Calling a priority weighting scenario formula , wherein, is a scenario sensitive coefficient; is a style coefficient; is an original length of a structure line of the j3th part of the sub sample; is an adjusted length of the structure line of the j3th part of the sub sample;

[0121] Calling a conflict resolution enhancement formula , wherein, , is a priority compensation coefficient; is an original parameter of a process-related structure of a j3th component of a sub sample; is an adjusted parameter of the process-related structure of the j3th component of the sub sample;

[0122] A virtual prototype of the sub sample is generated according to the adjusted parameter obtained by calling the formula, and an overall deviation is determined, if the overall deviation is greater than a preset deviation, triggering formula library parameter self-optimization.

[0123] In this embodiment, the AI self-optimization part coefficient is obtained based on historical adjustment effects, for example, 1.05 for a bust part.

[0124] In this embodiment, the master-sub sample correlation matrix is a matrix describing the correlation degree of each part of a master sample and a sub sample, and the matrix elements reflect the correlation tightness. For example, the master sample bust and the sub sample bust are directly correlated, and the corresponding element in the matrix is 1; the master sample sleeve length and the sub sample shoulder width are indirectly correlated, and the element is 0.5.

[0125] In this embodiment, is a scene-sensitive coefficient, such as 0.8 for a sports scene and 0.6 for a commuting scene, which is set in advance based on historical analysis of scene types.

[0126] In this embodiment, is a style coefficient, which can be directly obtained from a pattern style-value table, such as 1.1 for a retro style.

[0127] In this embodiment, the virtual prototype and the formula library self-optimization are achieved by generating a sub-sample virtual prototype through 3D modeling, simulating the adjusted garment shape, and calculating the overall deviation (such as size deviation, process conflict, etc.). If the deviation is greater than a preset value (such as a size deviation of more than 0.5 cm), the self-optimization of the parameters (such as ) in the formula library is triggered. For example, the value of is adjusted from 1.05 to 1.08, and the size is recalculated to reduce the deviation.

[0128] The beneficial effects of the above technical solutions are: through the precise calling of three types of formulas of size self-adaptation, scene weighting, and conflict resolution, combined with virtual prototype verification and formula library self-optimization, the full-dimensional precise adjustment of the sub-sample from size to process and from basic adjustment to scene adaptation is realized. The accuracy and scene adaptability of the adjustment are ensured, the adjustment effect is continuously improved through self-optimization, the problems of size deviation, scene inadaptation, process conflict, etc. in pattern adjustment are effectively solved, and efficient and precise technical support is provided for the intelligent updating of the garment sub-sample.

[0129] The present application provides an intelligent updating method for a sub-mother version linkage database, determines a dynamic binding weight based on a real-time binding relationship, and obtains an adjustment priority by combining a range influence index and a delivery period urgency factor, comprising:

[0130] The average value under the maximum clustering set is obtained by performing clustering analysis on the same elements of the historical vectors in the historical database, and a reference vector is constructed, and at the same time, vector clustering analysis is performed on all historical vectors to obtain a first center cluster vector under the maximum clustering set;

[0131] Taking the maximum clustering set to which the first average value of each element belongs as the first set, performing intra-set clustering analysis on all historical vectors in the first set to obtain a second center cluster vector of the intra-set maximum clustering cluster;

[0132] The difference matrix is obtained by sequentially determining the difference vectors of the reference vector, the first center cluster vector, and the second center cluster vector with the normalized vector, and the linear fitting is performed on each column vector in the difference matrix to obtain a dynamic binding coefficient;

[0133] ;

[0134] wherein, is the dynamic binding weight of the corresponding column vector. is a weight distribution coefficient; is a corresponding linear fitting slope; is a corresponding linear fitting intercept; is a maximum value of a fitting slope and a fitting intercept of a corresponding binding dimension in a recent 3-month history; is an adjustment coefficient of a discrete point in a linear fitting process under a corresponding binding dimension; is an entropy weight basic weight of a cluster to which a corresponding binding dimension belongs; is a process adjustment coefficient under a corresponding binding dimension;

[0135] a dynamic binding weight is obtained based on a dynamic binding coefficient and an adjustment coefficient of each column vector wherein, is an average value of all is a standard deviation of all N0 is a number of column vectors; is a factorial symbol; is a minimum value of an absolute value of a difference between any two

[0136] The dynamic binding weight, the range influence index, and the delivery deadline urgency factor are combined into an input vector, which is input into a vector analysis model to obtain an adjustment priority.

[0137] In this embodiment, the historical vectors are stored in a historical database, and reflect numerical combinations of the binding relationship between the master version and the sample version. For example, a vector of fabric-process collaborative binding is recorded, which contains parameters such as fabric elasticity coefficient and sewing needle spacing, such as [0.8, 3.2, 1.1], which respectively represent fabric elasticity, needle spacing, and process matching degree. The sample is derived from the production data of the version of the enterprise in the past 5 years, and the abnormal data with a size deviation > 2 cm is removed. The basic size is normalized according to the height / 100, and the fabric attribute is standardized according to the weight / 500 and the elasticity coefficient / 20. The sample size is 1200 groups, covering suits, sweaters, and dresses.

[0138] The cluster analysis of the same element and the reference vector are clustering the same type of elements (such as all rigid binding vectors) in the historical vector, and the average value of the maximum cluster set is taken to construct the reference vector. For example, all the historical vectors of rigid binding are collected, and the average value of the maximum set after clustering is [0.9, 1.0, 0.8], which is taken as the reference vector of rigid binding, representing its typical characteristics.

[0139] ​​In this embodiment, the first central cluster vector is the center vector of the largest cluster set after clustering all historical vectors (including different binding types). For example, when clustering all binding vectors such as rigid and flexible, the center vector of the largest set is [0.7, 0.6, 0.9], which reflects the common trend of various binding types.

[0140] The second central cluster vector is formed by taking the largest cluster set to which the average value of the reference vector belongs as the first set, and then clustering the vectors within it again, taking the center vector of the largest cluster. For example, if the first set is a set of vectors that are mainly rigidly bound, and then clustering is performed within it, the center vector of the largest cluster [0.95, 1.0, 0.85] is the second central cluster vector, which refines the rigidly bound features.

[0141] Normalized vectors, difference vectors, and difference matrices: A normalized vector is a vector whose elements are standardized according to rules (such as dividing by the maximum value of the dimension) so that the values ​​are in the range of 0-1; a difference matrix is ​​composed of multiple difference vectors.

[0142] For example, the reference vector [0.9,1.0,0.8] is normalized to [0.9,1.0,0.8] (assuming the maximum dimension is 1), and the difference vector between it and the normalized vector [0.5,0.5,0.5] is [0.4,0.5,0.3]. Multiple such difference vectors form a difference matrix.

[0143] In this embodiment, the dynamic binding coefficient is obtained by performing a linear regression on each column vector of the difference matrix to obtain the slope. and intercept For example, a column vector represents the difference values ​​of a time series. After fitting, the slope ku = 0.02 (reflecting a monthly increase of 0.02 in the difference) and the intercept b0 = 0.1 (the baseline value of the difference).

[0144] Dynamic binding weights It is a weighted average of the slope, intercept, discrete point adjustment, entropy weight, and process coefficient. For example, =0.6 (slope weight percentage). =0.05 (maximum slope in the past 3 months). =0.2 (absolute value of the largest intercept in the past 3 months). =0.1 (discrete point adjustment) =0.8 (entropy weight basic weight), =1.05 (process adjustment coefficient), then wg=0.524, where, The value is based on 3 months of historical data, ensuring that the correlation between wg and the actual binding effect is ≥0.85.

[0145] In this embodiment, if the proportion of discrete points (residuals > 0.2) in the linear fitting is ≤ 3%, =1; accounting for 3%-5%, =0.9; percentage > 5%, =0.8.

[0146] In this embodiment, Obtaining the information entropy: Calculate the information entropy of the bound dimension, and obtain the corresponding weight based on the ratio of the information entropy of this dimension to the sum of the information entropies of all dimensions.

[0147] Dynamic weight binding combines the normal distribution characteristics, standard deviation, number of column vectors, and other parameters of the weighted average (wg). Weights of the differences.

[0148] In this embodiment, a vector analysis model is trained on a neural network model using 1000 sets of historical samples, consisting of inputs composed of dynamically bound weights, range influence index, and delivery urgency factor normalized, and the input priority evaluation results as outputs. Therefore, the priority adjustment can be directly obtained. For example, if the input is [1,1,0.4], the output is 0.88.

[0149] In this embodiment, the number of column vectors N0 is equal to the number of types of real-time binding relationships.

[0150] In this embodiment, Used for normalizing weighted products.

[0151] The beneficial effects of the above technical solution are: by deeply mining historical binding features through multiple rounds of clustering, and combining multi-dimensional factors such as linear fitting, entropy weight analysis, and process coefficients, the dynamic binding weight can be accurately quantified. Furthermore, by integrating range influence and delivery time factors, the determination of the priority of sub-sample adjustment can not only conform to historical patterns, but also take into account process constraints and delivery requirements, effectively improving the efficiency and accuracy of the linkage update of parent and child versions.

[0152] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An intelligent updating method for a master-slave edition linkage database, characterized in that, Comprise: Step 1: based on multi-dimensional parameterization correlation model, different key elements of garment pattern are converted into quantitative correlation parameters, and real-time binding relationship between master pattern and sub-pattern is established based on mapping rules of database layer; Step 2: Constructing a multi-modal garment process knowledge graph, fusing multi-modal data to deeply associate garment pattern data with key elements to obtain a pattern adjustment vector set, wherein the pattern adjustment vector set includes: a basic adjustment vector , a scene adaptation vector , and a process compensation vector ; Step 3: multi-dimensional change type intelligent identification and priority determination engine is adopted to automatically identify master pattern change type and determine adjustment priority based on real-time binding relationship, and combined with preset AI self-optimization garment structure mathematical formula library and pattern adjustment vector set, sub-pattern is driven to complete adaptive adjustment of corresponding parts to realize intelligent update, wherein, the change type includes: explicit change type and implicit change type; Wherein, garment pattern data is deeply associated with key elements to obtain pattern adjustment vector set, comprising: Pre-training BERT model is used to analyze process standard document to output semantic feature vector T1, YOLOv8 model is used to analyze pattern CAD drawing to output visual feature vector I1, PointNey model is used to extract spatial feature vector V1 of human-garment fitting model, and wavelet transform is used to extract time-frequency feature vector A1; Attention weights based on T1, I1, V1 and A1 are determined respectively, and fusion feature vector F is generated; The multi-modal garment process knowledge graph is divided into a core layer, an association layer and an extension layer, wherein the relationship between nodes is quantified based on cross-modal association strength R0(a,b): wherein, is mutual information of node a and node b in the multi-modal garment process knowledge graph; is a cosine similarity of the fusion feature ; and is a set weight. When an explicit association edge between node a and node b is established; When , an implicit association edge is established and the association confidence is marked, wherein, , is a strength threshold value; determine the basic adjustment vector based on the association strength of the core layer element and the target version wherein, is a key element of the core layer; is a key element of the core layer is the association strength with the target version; is the current parameter value of is the standard parameter value of is the current parameter value of is the standard parameter value of determining a scene adaptation vector based on the scene-parameter mapping matrix wherein, is a feature matrix of the scene ; is a feature matrix of the current pattern; is a scene-parameter mapping matrix based on the scene ; Based on the association confidence of implicitly associated edges, the process compensation vector is determined. ,in, This is the compensation coefficient; This represents the adjustment amount for the implicit correlation parameter; Hidden related edges The association confidence level; Vector fusion matrix Obtaining a set of comprehensive adjustment vectors .

2. The intelligent updating method of the parent-child edition linkage database according to claim 1, characterized in that, The key elements include: basic size, structure line, process standard, fabric attribute, pattern style, adaptation scenario and equipment parameter, and the real-time binding relationship includes: rigid binding, flexible binding, predictive binding, time sequence correlation binding, fabric-process coordination binding, user portrait binding and equipment adaptation binding.

3. The intelligent updating method of the master-slave edition linkage database according to claim 2, characterized in that, A multi-dimensional parameterization correlation model is constructed, comprising: The correlation strength of any two key elements is quantified by mutual information entropy; The following weighting methods are used to determine the binding weights: rigid binding weight wr (based on the time-effect coefficient), flexible binding weight wf (based on the correlation strength), predictive binding weight wp (based on the LSTM neural network), time-series correlation binding weight gq (based on the historical change time series data of the master pattern and the sub-pattern), fabric-process synergy binding weight mb (based on the deep correlation between changes in master pattern fabric attributes and adjustments in sub-pattern process parameters), user profile binding weight hb (based on the profile characteristics of the target user group to establish the correlation between the master pattern and sub-patterns for specific users), and equipment adaptation binding weight sb (based on the correlation between changes in master pattern and parameter constraints of sub-pattern production equipment). Normalization enables quantitative control of the real-time binding relationship between the master template and child templates. It is a diagonal matrix, and the diagonal elements are the reciprocals of the maximum threshold of each weight; This is the normalization function.

4. The intelligent updating method of the parent-child edition linkage database according to claim 1, characterized in that, The mapping rules include cross-level mapping logic, and the cross-level mapping logic includes: Constructing a double mapping function Wherein, M1 is a master, belonging to the base library corresponding to the level L0, S1 is a sub-sample, belonging to the derived library corresponding to the level L1-L3; The i1th key parameter of the master; The sub-sample The i2th derived parameter associated with The master; The degree of association of The master; The reverse constraint factor; The number of key parameters of the master; The number of derived parameters of the sub-sample based on The master; The master; determining a hierarchical attenuation coefficient wherein, is a hierarchical attenuation coefficient; is a hierarchical difference, i.e. the hierarchical separation between the master and the child master. When a sub-master update is triggered, wherein, is a minimum response threshold; is a change in the i1th key parameter of the master.

5. The intelligent updating method of the master-slave edition linkage database according to claim 1, characterized in that, Automatic identification of master pattern change type and determination of adjustment priority based on real-time binding relationship, comprising: Matching the range influence index consistent with the master pattern change type from the type-range comparison table; Based on real-time binding relationship, dynamic binding weight is determined, and combined with range influence index and delivery period urgency factor, adjustment priority is obtained.

6. The intelligent updating method of the master-slave edition linkage database according to claim 1, characterized in that, Combined with preset AI self-optimization garment structure mathematical formula library and pattern adjustment vector set, sub-pattern is driven to complete adaptive adjustment of corresponding parts to realize intelligent update, comprising: Calling a size adaptation formula wherein, , is an AI self-optimization site coefficient; is the adjusted size of the j3th site of the sub-plate; is the size before adjustment of the j3th site of the sub-plate; is the size change rate of the j3th site of the sub-plate; is a master-plate-sub-plate correlation matrix; is the adjustment priority of the j3th site of the sub-plate; Call priority weighting scenario formula wherein, is a scenario sensitivity coefficient; is a style coefficient; is the original length of the structural line of the j3th part of the sub-plate; is the adjusted length of the structural line of the j3th part of the sub-plate; Call conflict resolution enhancement formula wherein, , is a priority compensation coefficient; is an original parameter of the process-related structure of the j3rd component of the sub-reticle; is an adjusted parameter of the process-related structure of the j3rd component of the sub-reticle; The virtual prototype of the sub-pattern is generated according to the adjusted parameters obtained by calling the formula, and the overall deviation is determined, if the overall deviation is greater than the preset deviation, the formula library parameter self-optimization is triggered.

7. The intelligent updating method of the master-slave edition linkage database according to claim 5, characterized in that, Based on real-time binding relationship, dynamic binding weight is determined, and combined with range influence index and delivery period urgency factor, adjustment priority is obtained, comprising: The average value under the maximum clustering set is obtained by clustering analysis of the same elements in the historical vectors in the historical database, and a reference vector is constructed, and at the same time, vector clustering analysis is performed on all historical vectors to obtain the first center cluster vector under the maximum clustering set; Taking the maximum clustering set to which the first average value of each element belongs as the first set, the set-in clustering analysis is performed on all historical vectors in the first set to obtain the second center cluster vector of the maximum clustering cluster in the set; The difference matrix is obtained by sequentially determining the difference vectors between the reference vector, the first center cluster vector, the second center cluster vector and the normalized vector respectively, and performing linear fitting on each column vector in the difference matrix to obtain a dynamic binding coefficient; ； wherein, is a dynamic binding weight for the corresponding column vector; is a weight distribution coefficient; is a corresponding linear fitting slope; is a corresponding linear fitting intercept; , is a maximum value of the recent 3-month history fitting slope, history fitting intercept of the corresponding binding dimension; is an adjustment coefficient of the existence of discrete points in the linear fitting process under the corresponding binding dimension; is an entropy weight basic weight of the cluster to which the corresponding binding dimension belongs; is a process adjustment coefficient under the corresponding binding dimension; The dynamic binding weight is obtained based on a dynamic binding coefficient and an adjustment coefficient of each column vector wherein, is the mean of all is the standard deviation of all N0 is the number of column vectors; is the factorial symbol; is the minimum value of the absolute value of the difference between all arbitrary two ​​ The dynamic binding weight, the range influence index and the delivery deadline urgency factor are combined into an input vector, which is input into a vector analysis model to obtain an adjusted priority.

Images