Intelligent packing material recommendation method and system based on hybrid architecture

Abstract

The invention discloses an intelligent packaging material recommendation method and system based on a hybrid architecture, and relates to the technical field of green supply chain management, and the method comprises the steps: collecting multi-modal data to construct a knowledge graph, and generating a green packaging feature vector and a type label through a Transform attention mechanism; constructing a dynamic graph structure based on the green packaging feature vector, obtaining a node feature matrix by using GNN graph convolution based on the dynamic graph structure, and generating a green packaging material candidate list based on the node feature matrix and the type label; based on the green packing material candidate list, performing multi-dimensional weighted scoring through 3D modeling and visual analysis to obtain a selected green packing material list; and generating an optimal green packing material scheme by selecting the green packing material list, and generating a 3D visual view based on the optimal green packing material scheme. The overall level of the green package recommendation system in the aspects of intelligence, visualization and safety is improved.

Description

Technical Field

[0001] This invention relates to the field of green supply chain management technology, and in particular to a smart packaging material recommendation method and system based on a hybrid architecture. Background Technology

[0002] With the development of the logistics industry, the scale of commodity transportation and distribution continues to expand. Simply relying on static parameters such as size and weight often fails to simultaneously address transportation safety, space utilization, and environmental requirements. The concept of green environmental protection is gradually becoming an industry consensus, requiring packaging materials to not only fulfill their protective function but also meet green environmental indicators such as recyclability, waste reduction, and low carbon emissions. Existing methods have significant limitations in data utilization, often only considering basic attributes such as commodity size and weight, as well as some historical order data, ignoring transportation environment data, logistics link information, and users' environmental preferences and personalized needs. This lack of data dimensions makes recommendation systems unsuitable for complex and ever-changing transportation scenarios, failing to meet the dynamic adaptation and personalized recommendation needs of green supply chains. Most existing recommendation algorithms rely on similarity-based heuristics, lacking deep semantic reasoning capabilities and knowledge bases. The system lacks the ability to express information effectively, failing to construct a systematic knowledge representation covering product characteristics, material properties, and transportation conditions. Furthermore, it struggles to form a queryable and reasonable knowledge graph, resulting in significant shortcomings in green constraints and complex scenario reasoning. Currently, most green packaging recommendation systems still employ static matching methods, lacking the ability to adaptively model order evolution trends and dynamic changes in the transportation environment. They cannot fully capture historical patterns and potential trends in time-series data, thus exhibiting insufficient flexibility and accuracy when addressing diverse and dynamic practical needs in complex supply chains. In evaluating and selecting recommendation results, existing methods often only consider material costs or simple environmental attributes, lacking a multi-dimensional weighted scoring system that combines consumer behavior preferences, resource constraints, and intuitive visualization verification. This makes it difficult to balance green compliance and user experience in recommendation results, limiting their persuasiveness and implementation value in practical applications. Summary of the Invention

[0003] In view of the aforementioned existing problems, the present invention is proposed.

[0004] Therefore, this invention provides a method and system for intelligent packaging material recommendation based on a hybrid architecture, which solves the problems of insufficient data fusion, insufficient algorithm intelligence, and lack of dynamism that are common in existing green packaging recommendation technologies.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a smart packaging material recommendation method based on a hybrid architecture, comprising, Collect multimodal data to construct a knowledge graph, and use the Transformer attention mechanism to generate green-packaged feature vectors and type labels; A dynamic graph structure is constructed based on the green packaging feature vector. A node feature matrix is ​​obtained by using GNN graph convolution based on the dynamic graph structure. A candidate list of green packaging materials is generated based on the node feature matrix and type labels. Based on the candidate list of green packaging materials, a multi-dimensional weighted scoring was performed through 3D modeling and visual analysis to obtain a selected list of green packaging materials. The optimal green packaging solution is generated by carefully selecting a list of green packaging materials. A 3D visualization view is then generated based on the optimal green packaging solution. The optimal green packaging solution is recorded using a blockchain platform. Based on the 3D visualization view and blockchain traceability data, a user access mechanism is constructed.

[0006] As a preferred embodiment of the intelligent packaging material recommendation method based on a hybrid architecture described in this invention, the method involves: collecting multimodal data to construct a knowledge graph, using the Transformer attention mechanism to generate green packaging feature vectors and type labels, collecting product and packaging material data, extracting user preference weights, and generating a multimodal dataset after data cleaning and alignment. A knowledge graph matrix (KG) for the packaging domain is constructed using a multimodal dataset. Based on the knowledge graph matrix KG and the multimodal dataset, a variable-category matrix is ​​constructed. Consumer behavior weights are obtained through improved multiple correspondence analysis (MCA). and resource constraint weights Using weights and Initialize attention weight matrix ; The multimodal dataset is concatenated into an input feature matrix. The knowledge graph matrix (KG) is transformed into a semantic embedding matrix through linear projection, which serves as the key and value of the attention mechanism. A multi-head attention mechanism is then applied to generate single-head feature vectors. The generated single-head feature vector Concatenate the components sequentially to generate a comprehensive green packaging feature vector. , to feature vector Segment by node to obtain the first... The feature vector corresponding to each node The segmented feature vectors As input, the K-means algorithm is applied to cluster the data to obtain different clusters, each cluster representing a type of green packaging. A green packaging type label is defined for each cluster. .

[0007] As a preferred embodiment of the intelligent packaging material recommendation method based on a hybrid architecture described in this invention, the following steps are described: constructing a dynamic graph structure based on green packaging feature vectors; obtaining a node feature matrix using GNN graph convolution based on the dynamic graph structure; and generating a green packaging material candidate list based on the node feature matrix and type labels. Separate into time-series feature sequences based on time steps LSTM is used to extract the long-term dependencies between historical order trends and changes in the transportation environment from time-series data. and Update via LSTM and Output the hidden state sequence ; Constructing dynamic graph structures Using GNN graph convolution to compute the first Layer Node Feature Matrix The node feature matrix is ​​processed through a fully connected layer. Applicability score for conversion to green packaging All candidate green packaging materials are initially screened using rule matching technology to check whether each candidate packaging material meets the requirements of the green packaging type label. Based on the attribute requirements, filter out those that match the type tag. Consistent candidate packaging materials, within the selected candidate set, are ranked according to their suitability score. Sort the packaging materials from highest to lowest score and select the top five to generate a Top-5 green packaging material candidate list.

[0008] As a preferred embodiment of the intelligent packaging material recommendation method based on a hybrid architecture described in this invention, the following steps are taken: Based on a candidate list of green packaging materials, a multi-dimensional weighted scoring is performed using 3D modeling and visual analysis to obtain a selected green packaging material list. This involves 3D modeling of the Top-5 candidate packaging materials, generating three-dimensional images through 3D scanning, analyzing the three-dimensional images using image processing techniques to generate visual feature vectors, extracting predefined template features from a standard green template library, performing a dot product operation between the visual feature vectors and the template feature vectors, and normalizing the results to obtain a similarity value between 0 and 1. This similarity value is then combined with weights. and We formulate green rules with the knowledge graph matrix KG, generate a comprehensive green score for each packaging material through weighted fusion, and select the top-3 packaging materials based on the green scores to form a green packaging material list.

[0009] As a preferred embodiment of the intelligent packaging material recommendation method based on a hybrid architecture described in this invention, the step of generating the optimal green packaging material solution by selecting a list of green packaging materials refers to performing packing simulation based on the output list of selected green packaging materials, calculating the minimum number of boxes, and estimating packaging material costs, freight costs, environmental protection costs, and potential damage costs by combining packaging material prices, transportation distances and methods, carbon emission data of green packaging materials, and historical damage rates. The total cost estimate of each packaging material solution is generated by weighted summation, and the cost and space utilization of the top-3 packaging material solutions are evaluated in parallel using multi-threading technology. The solution with the lowest total cost is selected, a detailed operation manual is generated, and the optimal green packaging material solution is output.

[0010] As a preferred embodiment of the intelligent packaging material recommendation method based on a hybrid architecture described in this invention, the step of generating a 3D visualization view based on the optimal green packaging material solution and using a blockchain platform to record the optimal green packaging material solution refers to performing 3D modeling and scanning on the Top-3 candidate packaging materials based on the output optimal green packaging material solution to generate a three-dimensional packaging material image. The three-dimensional packaging material image is then converted into an interactive 3D visualization view using 3D modeling software. The optimal green packaging material solution, transportation environment data from the modal data acquisition and knowledge graph fusion steps, and environmental protection indicators are recorded using a blockchain platform to generate a publicly searchable traceability record.

[0011] As a preferred embodiment of the intelligent packaging material recommendation method based on a hybrid architecture described in this invention, the step of constructing a user access mechanism based on 3D visualization and blockchain traceability data refers to building a login system based on the output 3D visualization operation manual and blockchain traceability data through the OAuth2.0 protocol, supporting role-based access control, restricting user access permissions to the operation manual and traceability data, storing user operations and system decisions in a database to support subsequent auditing, conducting satisfaction surveys based on feedback data, adjusting the interaction model by comparing user feedback with system predictions, iteratively optimizing the interface design, and outputting user access permissions, operation logs, and the optimized interaction interface.

[0012] Secondly, the present invention provides an intelligent packaging material recommendation system based on a hybrid architecture, comprising: The multimodal data acquisition and knowledge graph construction module is used to generate triples using NER, relation extraction and RDF techniques, and combine them with user preference weights to form an enhanced knowledge graph matrix; The feature extraction and green packaging vector generation module is used to transform multimodal data into a comprehensive green packaging feature vector by using an improved MCA method combined with the Transformer attention mechanism, and obtain green packaging type labels through K-means clustering. The dynamic graph construction and candidate packaging material generation module is used to transform feature vectors into time-series feature sequences, extract historical trends and environmental changes through LSTM, and then combine them with knowledge graphs to construct a dynamic graph structure. The 3D modeling and visual analysis screening module is used to perform 3D modeling and visual feature analysis on candidate packaging materials, and to perform multi-dimensional weighted scoring by combining MCA weights and knowledge graph green rules. The optimal green packaging solution generation module is used to generate the optimal green packaging solution with the lowest total cost through packing simulation and cost evaluation.

[0013] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the intelligent packaging material recommendation method based on a hybrid architecture as described in the first aspect of the present invention.

[0014] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent packaging material recommendation method based on a hybrid architecture as described in the first aspect of the present invention.

[0015] The beneficial effects of this invention are as follows: By introducing multimodal data fusion, deep semantic reasoning, and dynamic graph adaptive modeling, it overcomes the shortcomings of existing technologies, such as single data dimension, insufficient semantic reasoning ability, and lack of dynamic adaptability. By combining product attributes, transportation environment, logistics link information, and user environmental preferences, a comprehensive green packaging recommendation system is constructed, achieving accurate matching and adaptive adjustment in complex transportation scenarios. Utilizing knowledge graphs and improved multiple correspondence analysis methods, semantic reasoning ability is enhanced, enabling a systematic expression of the relationship between product characteristics, green material attributes, and transportation conditions, thereby improving the accuracy and green compliance of recommendations. A multi-dimensional weighted scoring system is established by combining 3D modeling and visual analysis to optimize the selection process of green packaging materials. Furthermore, blockchain technology enables transparent traceability of recommendation results, enhancing trust and traceability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a smart packaging material recommendation method based on a hybrid architecture in Example 1.

[0018] Figure 2This is a schematic diagram of the structure of a smart packaging material recommendation system based on a hybrid architecture in Example 1.

[0019] Figure 3 This is a flowchart of the 3D modeling and scoring process in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, referring to Figures 1 to 3 This is the first embodiment of the present invention, which provides a smart packaging material recommendation method based on a hybrid architecture, including the following steps: S1. Collect multimodal data to construct a knowledge graph, and use the Transformer attention mechanism to generate green packaging feature vectors and type labels; A dynamic graph structure is constructed based on the green packaging feature vector. A node feature matrix is ​​obtained by using GNN graph convolution based on the dynamic graph structure. A candidate list of green packaging materials is generated based on the node feature matrix and type labels. Specifically, multimodal data is collected to construct a knowledge graph, and the Transformer attention mechanism is used to generate green packaging feature vectors and type labels. Data on collected goods and packaging materials is collected, user preference weights are extracted, and after data cleaning and alignment, a multimodal dataset is generated. Product data includes: dimensions (e.g., length, width, height), weight, material (e.g., plastic, glass), fragility (Boolean value), and liquid properties (Boolean value) of the product obtained from the order system; and transportation distance (km), transportation mode (land transport, air transport), and number of transshipments obtained from the logistics system. User preference weights include: parsing user preference text (such as "prioritize recyclable materials") through natural language processing (NLP) to extract keywords and user preference weights (such as environmental weight = 0.8, cost weight = 0.2). Packaging material data includes: material properties such as paper, plastic, foam, and specific characteristics of different metal materials, including density, strength, pressure resistance, impact resistance, temperature resistance, recyclability, biodegradability, as well as procurement costs, processing costs, transportation costs, and the size specifications (length, width, height), shape, and design structure (such as cushioning layer, sealing method) of the packaging materials. Based on a multimodal dataset, a knowledge graph matrix (KG) for the packaging domain is constructed, containing nodes (including product data nodes and packaging material nodes) and edges (each edge connects a product node and a packaging material node, representing the semantic relationship between the product data node and the packaging material node, such as "fragile items - need matching → cushioning material"). Specifically, the Named Entity Recognition (NER) algorithm is used to automatically extract key entities such as fragile glass and recyclable cardboard from the text data. Relationship extraction methods are applied to determine the relationships between entities, such as "need matching" and "substitutable". The identified entities and their corresponding data are then analyzed using RDF (Resource Description Framework). Relationships are transformed into triples (subject-predicate-object). For example, "fragile glass - requires matching - cushioning layer thickness > 2cm" and "recyclable cardboard - environmentally friendly attributes - recyclability > 90%" are formalized and expressed. User preference weights extracted by NLP are integrated as edge confidence. Triple data are stored in a graph database (such as Neo4j) to construct a queryable and reasonable knowledge graph for the packaging domain. A corresponding knowledge graph matrix KG is generated, and green material reasoning rules (such as "fragile items recommend recyclable foam + cardboard") are extracted from it. The edge attributes and MCA weights of the graph reflect the semantic constraints and priorities of green packaging. Based on the knowledge graph matrix KG and a multimodal dataset, a variable-category matrix is ​​constructed. The specific operation is as follows: the product attributes, user preferences, and transportation environment information in the original data are transformed into discrete category forms (for example, the size of the product can be divided into three categories: small, medium, and large). The frequency of the sample (e.g., each product) in each category is counted (if the size of the product is medium, then the frequency in this category will increase by 1). The frequency is used as the elements in the matrix to form a two-dimensional matrix with products as rows and categories as columns. Each row represents a product and each column represents a category. Consumer behavior weights are obtained using an improved multiple correspondence analysis (MCA) process (which introduces additional mechanisms into the standard MCA procedure, such as assigning weights to categories based on domain knowledge (e.g., incorporating user preference weights into the frequency matrix), adjusting the chi-square distance calculation method, and integrating information from external continuous variables). (e.g., environmental preferences, purchasing decisions) and resource constraints weighting (e.g., material availability, transportation environment), the formula is: , , , in, For variable-category matrix The observations in the i-th row (sample, such as a product) and j-th column (category, such as "recyclable") represent the actual frequency, derived from the discretization of KG and multimodal data. For the expected frequency, The total frequency refers to the sum of all observations in the variable-category matrix. To standardize the chi-square distance, which measures the difference between observed and expected values, Point to Perform singular value decomposition, retaining the first two principal components. The number of principal components is used to generate a consumer behavior weight vector. and resource constraint weight vector , To specify constraints in the optimization problem; Based on weight Initialize attention weight matrix The formula is: , in, For balance coefficients, such as T represents the matrix transpose operation; The multimodal dataset is concatenated into an input feature matrix. The knowledge graph matrix (KG) is transformed into a semantic embedding matrix through linear projection (such as a fully connected layer), which serves as the key and value of the attention mechanism. A multi-head attention mechanism is then applied to generate a single-head feature vector. The formula is: , Where Q is the query matrix, derived from the input feature matrix. Generated through linear projection, where K is the key matrix, and derived from... Projection generation, including MCA weight information, where V is the value matrix, generated by... Generated by fusion with X The dimension of the keys and values ​​(e.g., 64) is used to scale the attention weights; Single-head feature vector The multi-head feature matrix is ​​obtained by concatenating the features sequentially. This multi-head feature matrix is ​​then multiplied by a learnable weight matrix (where the learnable weight matrix refers to the parameter matrix in the Transformer linear transformation layer, optimized from multimodal data and the green packaging target through model training (e.g., gradient descent)) through a linear transformation layer (usually a fully connected layer). This maps the concatenated features to the target dimension (e.g., 128 dimensions), generating a comprehensive green packaging feature vector. (The process of generating feature vectors for integrated green packaging incorporates the semantic embedding matrix in the knowledge graph (KG) as the key and value of the attention mechanism, while preserving the original node information of the knowledge graph matrix KG.) Integrate green packaging feature vectors Segment by node to obtain the first... The comprehensive green packaging feature vector corresponding to each node , to feature vector As input, the K-means algorithm is applied to cluster the data, resulting in different clusters. Each cluster represents a type of green packaging (e.g., recyclable and lightweight). Based on the characteristics of each cluster and actual business needs (common characteristics of each group of green packaging derived from cluster analysis (e.g., materials, cost, protective performance) and specific market requirements for packaging characteristics (e.g., environmental standards, consumer preferences, logistical convenience)), a green packaging type label is defined for each cluster. (e.g., biodegradable materials, minimalist packaging), and store the tags in a knowledge graph matrix KG. The K-means algorithm formula is: , , Where A is the sum of squares within the cluster, representing the optimal cluster partitioning result, where each data point is assigned to a cluster. Let j be the j-th feature of sample B (i.e., the j-th dimension component of vector B). For the first Each cluster center represents a green packaging type, and m is the feature dimension (e.g., 128). For the number of clusters, For sample B and the first Cluster center The Euclidean distance between them.

[0024] By systematically cleaning, aligning, and fusing multimodal heterogeneous data such as product attributes, transportation environment, logistics links, and user preferences, and constructing a knowledge graph (KG) for the packaging domain using named entity recognition and relation extraction, the constraints of materials, scenarios, and rules are semantically characterized, fundamentally overcoming the shortcomings of traditional methods in terms of single data dimension and insufficient semantic reasoning ability. An improved multiple correspondence analysis (MCA) is introduced to extract consumer behavior and resource constraint weights and write them back into the KG, enhancing the model's interpretability and achieving a quantitative expression of preference and availability constraints. Furthermore, MCA weights are used as attention initialization, combined with Transformer... Multi-head attention, temporal LSTM, and graph neural networks (GNN) collaboratively model static features, temporal evolution, and graph structure relationships, improving the accuracy and adaptability of recommendations in dynamic orders and complex transportation scenarios. By constructing a multi-dimensional weighted scoring system through 3D modeling and visual similarity evaluation and introducing blockchain traceability, the solution's intuitive visualization and green compliance verification capabilities are enhanced, while also improving the transparency and trustworthiness of business implementation. Systematic improvements are provided in data fusion depth, algorithm intelligence, dynamic adaptation, and traceability, effectively improving the accuracy, interpretability, and engineering feasibility of green packaging material recommendations, and demonstrating significant industrial application value.

[0025] Furthermore, a dynamic graph structure is constructed based on the green packaging feature vector. A GNN graph convolution is then used to obtain the node feature matrix based on this dynamic graph structure. Finally, a candidate list of green packaging materials is generated based on the node feature matrix and type labels, which refers to the green packaging feature vector... Separate into time-series feature sequences based on time steps ,in Let N represent the features at time step t (such as the product and environment data of the current order), and let N represent the total number of time steps in the time series feature sequence. LSTM is used to extract the dependencies between historical order trends and changes in the transportation environment from the time series data. and ,formula: , in, The hidden state at the current time step represents the LSTM's response to the input. Extraction of short-term temporal features from historical information. and These are the hidden states and memory units from the previous time step. The memory unit is the memory unit for the current time step. The memory unit refers to the component in the LSTM model used to store and transmit long-term dependency information, capturing the long-term trend of time series data. For each time step t, update via LSTM and Output the hidden state sequence ; Constructing dynamic graph structures The specific operation is as follows: A graph neural network (such as GCN) is used to encode the knowledge graph matrix KG, generating an embedding vector for each node. Nodes are arranged in a predefined order (e.g., product type ID). If the number of nodes is inconsistent with the number of samples in the hidden state, padding is performed (e.g., copying the hidden state from the last time step or performing mean pooling on the nodes) to align the two numbers, ensuring a one-to-one correspondence (the i-th hidden state corresponds to the i-th node). At each time step, a fully connected layer transforms the hidden state to the same dimension as the node embedding vector, and then the two are concatenated sequentially to generate a comprehensive feature vector. The comprehensive feature vectors from all time steps are combined to generate a comprehensive node feature matrix. Node feature matrix As a feature representation of node V, based on knowledge The green material reasoning rules determine whether two nodes should be connected. For example, if a product node contains "fragile = True" and the material node is "EPE foam", the corresponding position in the adjacency matrix is ​​set to 1 due to rule matching. The semantic similarity method is used to determine whether there is an edge connection, and an adjacency matrix A is generated as edge E. Using GNN graph convolution to compute the first Layer Node Feature Matrix The formula is: , Wherein, the initial is , The adjacency matrix is ​​obtained by normalizing the adjacency matrix A of the dynamic graph structure G=(V,E), and σ is the activation function (such as ReLU). Let L be the learnable weight matrix of the l-th layer; The fully connected layer will convert the node feature matrix Applicability score for conversion to green packaging The formula is: , in, It is a learnable nonlinear transformation function used to map node feature vectors to scalar scores; All candidate green packaging materials are initially screened using rule matching technology to check whether each candidate material meets the requirements of the green packaging type label. Based on the attribute requirements, filter out those that match the type tag. Consistent candidate packaging materials (if) For the "high-protection, recyclable" category, only packaging materials with high cushioning and recyclability properties will be retained. Within the selected candidate set, materials will be ranked according to their suitability score. Sort the packaging materials from highest to lowest score and select the top five to generate the final Top-5 green packaging material candidate list.

[0026] This paper introduces a mechanism combining temporal modeling and dynamic graph structure in green packaging material recommendation to overcome the shortcomings of existing technologies in handling dynamic orders and complex transportation environments, which are static and lack adaptability. LSTM is used to capture the long-term dependencies between historical orders and environmental changes. Temporal features are then concatenated with enhanced knowledge graph node features to construct a dynamically evolving graph structure. Semantic rules and similarity judgments are combined to generate an adjacency matrix, enabling the model to reflect the matching relationships between goods, materials, and transportation conditions in real time across multiple time steps. High-order association features in the graph structure are extracted through GNN graph convolution, and a green packaging material suitability score is generated using fully connected layers, improving the refinement of recommendations and enabling automatic quantification of material and environmental suitability in complex scenarios. By organically combining temporal dependencies, dynamic graph reasoning, and green constraint screening, the dynamic adaptability and green compliance verification capabilities of the recommendation system are enhanced. This provides an improvement to address the shortcomings of existing green packaging recommendations, such as lack of dynamism, lack of high-order semantic reasoning, and difficulty in accurately matching user needs.

[0027] S2. Based on the candidate list of green packaging materials, multi-dimensional weighted scoring is performed through 3D modeling and visual analysis to obtain a selected list of green packaging materials. The optimal green packaging material solution is generated by carefully selecting a list of green packaging materials. Specifically, based on the green packaging material candidate list, a multi-dimensional weighted scoring is performed through 3D modeling and visual analysis to obtain a selected green packaging material list. This involves 3D modeling the top-5 candidate packaging materials, generating 3D images through 3D scanning to represent the spatial structure of the packaging materials (such as the edges of cardboard boxes and the curvature of plastic films). Image processing techniques (such as edge detection and texture analysis algorithms) are used to analyze the 3D images, identifying the packaging material's contours (such as the edges of cardboard boxes), textures (such as surface roughness or smoothness), and colors (such as the natural hue of environmentally friendly materials). Visual feature vectors are generated. Predefined template features (such as standard cardboard texture and green hue) are extracted from a standard green template library (developed independently by enterprises based on their own green concepts, design standards, and practical cases, optimized and expanded in conjunction with industry standards). The visual feature vectors are then multiplied by the template feature vectors, and after normalization, a similarity value of 0 to 1 is obtained, reflecting the degree of matching between the candidate packaging material and the green template (such as the closeness to a minimalist and environmentally friendly design). This is combined with MCA weights (Consumer Behavior Weights). and resource constraint weights The system uses a knowledge graph matrix (KG) to formulate green rules, including: prioritizing packaging materials with a recyclability rate >90%, excluding unsuitable packaging materials (such as cardboard that cannot be used under high humidity) based on transportation environment data, and generating a comprehensive green score for each packaging material through weighted fusion (template similarity accounts for 40%, environmental preference accounts for 30%, and resource constraints account for 30%). The top-3 packaging materials are selected based on the green scores to form a list of carefully selected green packaging materials that meet green concepts and user needs.

[0028] In the selection process for green packaging materials, a multi-dimensional weighted scoring mechanism combining 3D modeling and visual analysis is introduced. This overcomes the limitations of existing technologies that rely solely on two-dimensional attribute data or static rules for coarse evaluation. Three-dimensional modeling and image processing are performed on candidate packaging materials to visually reflect their alignment with green design concepts. Combined with improved MCA-extracted consumer behavior and resource constraint weights, as well as green rule constraints from the knowledge graph, a dynamic fusion of environmental attributes and transportation suitability is achieved. A multi-factor weighted strategy is employed to generate a comprehensive green score, ensuring that the recommendation results achieve global optimization across appearance design, environmental standards, and application scenarios. This enhances the visualization and intuitiveness of the recommendation results, compensates for the lack of multi-dimensional scoring and green compliance verification in traditional methods, and achieves closed-loop optimization from candidate screening to final selection, thereby improving the accuracy of the recommendation scheme.

[0029] Furthermore, generating the optimal green packaging solution through a carefully selected list of green packaging materials refers to conducting packing simulations based on the carefully selected list of green packaging materials. Specifically, the products are sorted by volume from largest to smallest, and the First-Fit-Decreasing algorithm is used to place the products into the selected packaging materials. Green rules (such as reducing filler and using recyclable cushioning materials) are prioritized to optimize the placement method, so that the space utilization rate reaches the target of 80%, and the minimum number of boxes required is calculated. By combining packaging material prices, transportation distances and methods, carbon emission data for green packaging materials, and historical breakage rates, we estimate packaging material costs, freight costs, environmental costs, and potential breakage costs. We then generate a total cost estimate for each packaging option through weighted summation. Specifically, we obtain the unit cost of each packaging material from the price data, calculate the required quantity of packaging material based on product size data, and arrive at the packaging material cost for each option. Based on the transportation distance and method (e.g., land or air transport), and referring to standard logistics rate tables, we calculate the transportation cost for each packaging option. Using carbon emission data for green packaging materials (e.g., carbon emissions per unit of packaging material), and considering packaging material usage, we calculate the carbon emission cost for each option. Finally, we base our calculations on historical breakage rate data (from logistics feedback). The process involves obtaining data and considering the fragility of the goods (e.g., fragile items require stronger protection). It estimates the potential damage cost of each packaging option during transportation, weighting the packaging cost (40%), freight (30%), environmental costs (20%), and potential damage costs (10%) together to calculate the comprehensive cost score for each option. This yields a total cost estimate. Multi-threading technology is used to evaluate the cost and space utilization of the top-3 packaging options in parallel, selecting the option with the lowest total cost. A detailed operation manual is generated, including recommended box types, product placement methods, and filler types and quantities. The optimal green packaging solution (including box type, total cost, and operation manual) is output, and the results are stored and traced via blockchain.

[0030] In the final decision-making stage of green packaging material recommendation, a packing simulation and multi-dimensional cost assessment mechanism is introduced to overcome the shortcomings of existing technologies, which rely on static material selection and lack practical implementation guidance. Packing simulation is conducted using a First-Fit-Decreasing algorithm combined with green rules to ensure that space utilization reaches the target level, optimize the environmental properties of fillers, and solve the problems of space waste and unreasonable filler use in traditional solutions. Packaging material costs, transportation costs, carbon emission costs, and potential damage losses are incorporated into a unified weighted evaluation framework to form a comprehensive cost score. This achieves a holistic balance of economic efficiency, environmental friendliness, and safety for green solutions, avoiding biases caused by single-cost or single-dimensional indicator evaluations. Multi-threaded parallel evaluation accelerates solution selection efficiency and outputs detailed operation manuals including box type, placement method, and filler selection to ensure direct feasibility. Finally, blockchain traceability ensures transparency and traceability of results, improving the accuracy and practicality of recommendations. This achieves an organic unity between green concepts, economic benefits, and engineering execution, enhancing the sustainable development capability of the green packaging recommendation system.

[0031] S3. Generate a 3D visualization view based on the optimal green packaging material solution, record the optimal green packaging material solution using a blockchain platform, and build a user access mechanism based on the 3D visualization view and blockchain traceability data. Specifically, generating a 3D visualization view based on the optimal green packaging solution and using a blockchain platform to record the optimal green packaging solution refers to performing 3D modeling and scanning of the top-3 candidate packaging materials based on the output optimal green packaging solution to generate 3D packaging images. These 3D images are then converted into interactive 3D visualization views using 3D modeling software (such as Blender and SolidWorks) to show how the product is placed inside the box (e.g., stacking order) and the distribution of fillers (e.g., the location of recyclable foam), enhancing the visualization effect and user understanding of the solution. A blockchain platform (such as Hy...) is then used to record the optimal green packaging solution. Perledger records optimal green packaging solutions (i.e., recommended box type selection), transportation environment data (temperature, humidity, vibration) and environmental indicators (recyclability, carbon emissions) from the modal data collection and knowledge graph fusion steps, generating publicly searchable traceability records. This allows customers and supply chain partners to verify the compliance of green packaging (such as environmental standard compliance) through a query interface, collect logistics feedback data (such as damage rate, customer satisfaction), provide data support for optimization, and output 3D visual operation manuals (including views of product placement and filler distribution) and traceability data recorded on the blockchain.

[0032] By introducing 3D visualization and blockchain recording mechanisms, this approach overcomes the shortcomings of existing technologies, such as abstract and unintuitive result presentation and a lack of transparency and traceability in the recommendation process. Through 3D modeling and interactive visualization, it intuitively displays the placement of goods and the distribution of fillers within the packaging, addressing the issue that users can only rely on textual or data descriptions and find it difficult to intuitively understand the recommended solutions. This improves the interpretability of the solutions and the user experience. By leveraging the blockchain platform to record and publicly query the recommendation results and key process data, it achieves full-chain transparency and reliable traceability, avoiding the problems of difficult verification of green compliance and insufficient data credibility in traditional solutions. Combined with logistics feedback data, it enables dynamic optimization, constructs a closed-loop improvement mechanism, enhances the visualization and usability of green packaging material recommendation solutions, and ensures compliance and credibility.

[0033] Furthermore, based on 3D visualization and blockchain traceability data, a user access mechanism is constructed. This mechanism involves building a login system using the OAuth2.0 protocol based on the output 3D visualization operation manual and blockchain traceability data. It supports role-based access control (such as administrators and operators), restricting user access to the operation manual and traceability data to ensure data security. User operations (such as viewing the operation manual and confirming packaging solutions) and system decisions (such as packaging material selection) are stored in the database to support subsequent auditing. Based on feedback data (such as customer satisfaction), a satisfaction survey is conducted to analyze the ease of operation and the accuracy of recommendations. By comparing user feedback with system predictions, the interaction model is adjusted, and the interface design is iteratively optimized (such as simplifying the operation process). The system outputs secure user access permissions, operation logs, and an optimized interactive interface, completing the user interaction closed loop of the green packaging recommendation system.

[0034] The OAuth 2.0 protocol, combined with role-based access control, ensures that users with different roles can only access operation manuals and traceability data related to their responsibilities. This overcomes the shortcomings of loose access management and the easy leakage of sensitive data. User operations and system decisions are recorded in the database, supporting subsequent auditing and compliance traceability, improving the transparency and regulatory oversight of the system. User feedback data is used to conduct satisfaction analysis and compare it with system predictions to iteratively optimize the interaction model and interface design. This solves the problems of insufficient user experience and lack of dynamic improvement of recommendation results in traditional solutions, ensures data security and access compliance, and establishes a user-participatory driven closed-loop optimization mechanism.

[0035] This embodiment also provides a smart packaging material recommendation system based on a hybrid architecture, including: The multimodal data acquisition and knowledge graph construction module is used to generate triples using NER, relation extraction and RDF techniques, and combine them with user preference weights to form an enhanced knowledge graph matrix; The feature extraction and green packaging vector generation module is used to transform multimodal data into a comprehensive green packaging feature vector by using an improved MCA method combined with the Transformer attention mechanism, and obtain green packaging type labels through K-means clustering. The dynamic graph construction and candidate packaging material generation module is used to transform feature vectors into time-series feature sequences, extract historical trends and environmental changes through LSTM, and then combine them with knowledge graphs to construct a dynamic graph structure. The 3D modeling and visual analysis screening module is used to perform 3D modeling and visual feature analysis on candidate packaging materials, and to perform multi-dimensional weighted scoring by combining MCA weights and knowledge graph green rules. The optimal green packaging solution generation module is used to generate the optimal green packaging solution with the lowest total cost through packing simulation and cost evaluation.

[0036] This embodiment also provides a computer device applicable to a smart packaging material recommendation method based on a hybrid architecture, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the smart packaging material recommendation method based on a hybrid architecture as proposed in the above embodiment.

[0037] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a smart packaging material recommendation method and system based on a hybrid architecture as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0039] In summary, this invention overcomes the shortcomings of existing technologies, such as single data dimensions, insufficient semantic reasoning capabilities, and lack of dynamic adaptability, by introducing multimodal data fusion, deep semantic reasoning, and dynamic graph adaptive modeling. By combining product attributes, transportation environment, logistics link information, and user environmental preferences, it constructs a comprehensive green packaging recommendation system, achieving accurate matching and adaptive adjustment in complex transportation scenarios. Utilizing knowledge graphs and improved multiple correspondence analysis methods, it enhances semantic reasoning capabilities, systematically expressing the relationship between product characteristics, green material attributes, and transportation conditions, thus improving the accuracy and green compliance of recommendations. Furthermore, by combining 3D modeling and visual analysis to establish a multi-dimensional weighted scoring system, it optimizes the selection process of green packaging materials. Finally, it uses blockchain technology to achieve transparent traceability of recommendation results, enhancing trust and traceability.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A hybrid architecture based intelligent packaging recommendation method, characterized in that: Comprising, Collecting multi-modal data to construct a knowledge graph, using a Transformer attention mechanism to generate a green packaging feature vector and a type label; Based on the green packaging feature vector, a dynamic graph structure is constructed, and a node feature matrix is obtained based on the dynamic graph structure using GNN graph convolution, and a green packaging material candidate list is generated based on the node feature matrix and the type label; Based on the green packaging material candidate list, multi-dimensional weighted scoring is performed through 3D modeling and visual analysis to obtain a selected green packaging material list; Through the selected green packaging material list, an optimal green packaging material scheme is generated, a 3D visualization view is generated based on the optimal green packaging material scheme, the optimal green packaging material scheme is recorded using a blockchain platform, and a user access mechanism is constructed based on the 3D visualization view and the blockchain traceability data. 2.The hybrid architecture based intelligent packaging recommendation method of claim 1, wherein: The multi-modal data collection and knowledge graph construction module is used to generate triples using NER, relation extraction and RDF technology, and form an enhanced knowledge graph matrix combined with user preference weights; A packaging field knowledge graph matrix KG is constructed using a multi-modal data set, and a variable-category matrix is constructed based on the knowledge graph matrix KG and the multi-modal data set , consumer behavior weights and resource constraint weights are obtained through improved multiple correspondence analysis (MCA), and the weight and are used to initialize an attention weight matrix ; Concatenating multi-modal dataset into input feature matrix , the knowledge graph matrix KG is converted into a semantic embedding matrix through linear projection as the key and value of the attention mechanism, and a single-head feature vector is generated by applying a multi-head attention mechanism , the generated single-head feature vector is concatenated in order to generate a comprehensive green packaging feature vector , the feature vector is divided according to the nodes to obtain the feature vector corresponding to the first node , the segmented feature vector is taken as input, and the K-means algorithm is applied for clustering to obtain different clusters, each cluster representing a green packaging type, and a green packaging type label is defined for each cluster . 3.The hybrid architecture based intelligent packaging recommendation method of claim 2, wherein: The green packaging feature vector is used to construct a dynamic graph structure, a node feature matrix is obtained based on the dynamic graph structure using GNN graph convolution, and a green packaging material candidate list is generated based on the node feature matrix and the type label The time step is split into a time sequence The long-term dependence of historical order trends and changes in transportation environment in time series data is extracted by LSTM And Update And Output the hidden state sequence ; Constructing dynamic graph structures Using GNN graph convolution to compute the first Layer Node Feature Matrix The node feature matrix is ​​processed through a fully connected layer. Applicability score for conversion to green packaging All candidate green packaging materials are initially screened using rule matching technology to check whether each candidate packaging material meets the requirements of the green packaging type label. Based on the attribute requirements, filter out those matching the type tag. Consistent candidate packaging materials, within the selected candidate set, are ranked according to their suitability score. Sort the packaging materials from highest to lowest score and select the top five to generate a Top-5 green packaging material candidate list. 4.The hybrid architecture based intelligent packaging recommendation method of claim 3, wherein: The green packaging material candidate list is subjected to multi-dimensional weighted scoring through 3D modeling and visual analysis to obtain a selected green packaging material list. Top-5 candidate packaging materials are subjected to 3D modeling, three-dimensional images are generated through 3D scanning, visual feature vectors are generated by using image processing technology to analyze the three-dimensional images, pre-defined template features are extracted from a standard green template library, dot product operation is performed on the visual feature vectors and the template feature vectors, and after normalization, a similarity value of 0 to 1 is obtained. A green rule is formulated in combination with a knowledge graph matrix KG, a comprehensive green score of each packaging material is generated through weighted fusion, and Top-3 packaging materials are screened out according to the green score to form a green packaging material list. and and a knowledge graph matrix KG, a comprehensive green score of each packaging material is generated through weighted fusion, and Top-3 packaging materials are screened out according to the green score to form a green packaging material list. 5.The hybrid architecture based intelligent packaging recommendation method of claim 4, wherein: The selected green packaging material list is used to generate an optimal green packaging material scheme, which is based on the output of the selected green packaging material list, performs a simulation of the packing, calculates the minimum number of boxes, combines the packaging material price, transportation distance and mode, carbon emission data of green packaging materials and historical damage rate, estimates the packaging cost, transportation cost, environmental protection cost and potential damage cost, and generates a total cost estimate value of each packaging scheme through weighted summation, uses multi-threading technology to evaluate the cost and space utilization rate of the Top-3 packaging scheme in parallel, selects the scheme with the lowest total cost, generates a detailed operation manual, and outputs the optimal green packaging material scheme. 6.The hybrid architecture based intelligent packaging recommendation method of claim 5, wherein: The 3D visualization view is generated based on the output of the optimal green packaging material scheme, the Top-3 candidate packaging materials are modeled and scanned in 3D, three-dimensional packaging images are generated, the three-dimensional packaging images are converted into interactive 3D visualization views through three-dimensional modeling software, the optimal green packaging material scheme is recorded using a blockchain platform, the transportation environment data of the modal data collection and knowledge graph fusion steps, and the environmental protection indicators are recorded, and traceability records that can be publicly queried are generated. 7.The hybrid architecture based intelligent packaging recommendation method of claim 6, wherein: The 3D visualization view is generated based on the output of the optimal green packaging material scheme, the Top-3 candidate packaging materials are modeled and scanned in 3D, three-dimensional packaging images are generated, the three-dimensional packaging images are converted into interactive 3D visualization views through three-dimensional modeling software, the optimal green packaging material scheme is recorded using a blockchain platform, the transportation environment data of the modal data collection and knowledge graph fusion steps, and the environmental protection indicators are recorded, and traceability records that can be publicly queried are generated.

8. A hybrid architecture based intelligent packaging recommendation system based on the hybrid architecture based intelligent packaging recommendation method of any one of claims 1-7. The 3D visualization view is generated based on the output of the optimal green packaging material scheme, the Top-3 candidate packaging materials are modeled and scanned in 3D, three-dimensional packaging images are generated, the three-dimensional packaging images are converted into interactive 3D visualization views through three-dimensional modeling software, the optimal green packaging material scheme is recorded using a blockchain platform, the transportation environment data of the modal data collection and knowledge graph fusion steps, and the environmental protection indicators are recorded, and traceability records that can be publicly queried are generated. Comprising, A multi-modal data collection and knowledge graph construction module is used to generate triples using NER, relation extraction and RDF technology, and form an enhanced knowledge graph matrix combined with user preference weights; The feature extraction and green packaging vector generation module is configured to convert the multi-modal data into comprehensive green packaging feature vectors by using an improved MCA method combined with a Transformer attention mechanism, and obtain green packaging type labels through K-means clustering. The dynamic graph construction and candidate packaging material generation module is configured to convert the feature vectors into time series feature sequences, extract historical trends and environmental changes through an LSTM, and then construct a dynamic graph structure in combination with a knowledge graph. The 3D modeling and visual analysis screening module is configured to perform 3D modeling and visual feature analysis on the candidate packaging materials, and perform multi-dimensional weighted scoring in combination with MCA weights and green rules of the knowledge graph. The optimal green packaging material scheme generation module is configured to generate an optimal green packaging material scheme with the lowest total cost through container loading simulation and cost evaluation. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The computer program is executed by the processor to implement the steps of the intelligent packaging material recommendation method based on the hybrid architecture according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the intelligent packaging material recommendation method based on the hybrid architecture according to any one of claims 1-7.

Images