Intelligent packing material recommendation method and system based on AI machine learning

Abstract

The invention discloses an intelligent packing material recommendation method and system based on AI machine learning, and relates to the technical field of logistics packaging, and the method comprises the steps: carrying out the fuzzy clustering according to the multi-dimensional data of a commodity, and constructing a knowledge database; performing known packaging material and packaging mode screening for commodity types, training an intelligent packaging material recommendation multi-modal model, and generating a packaging material-packaging mode selectable for the to-be-packaged commodity type; obtaining a transportation destination of a to-be-packaged commodity, and quantifying a transportation demand of the to-be-packaged commodity; judging whether the transportation requirement of the to-be-packaged commodity is met or not, if yes, executing normally, and if not, submitting a strengthening instruction; and if a strengthening instruction is submitted, packaging and substituting the transportation multi-dimensional characteristic parameters, attribute preferences and packaging material-packaging modes with selectable commodity types into an intelligent packaging material recommendation multi-modal model, and generating an optimal packaging material-packaging mode compensation scheme of the to-be-packaged commodity type. The method has the beneficial effects that the loss in the packaging process is reduced, and the transportation efficiency and economic benefits are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics packaging, in particular to an intelligent packaging material recommendation method and system based on AI machine learning. BACKGROUND

[0002] Most of the existing intelligent packaging material recommendation technologies at home and abroad rely on traditional data analysis and rule engines, and fail to fully utilize advanced artificial intelligence and machine learning algorithms for deep learning and self-optimization. The existing methods usually ignore the multi-dimensional features of goods and the dynamic changes of transportation environment, lack personalized recommendation ability based on real data driving, and are difficult to make accurate recommendations when facing complex and diversified packaging requirements. As a result, the adaptability and optimization degree of the recommended results are low, and the system cannot respond to changes in packaging requirements in real time, affecting the effectiveness and economy of the packaging scheme. SUMMARY

[0003] To solve the above technical problems, an intelligent packaging material recommendation method and system based on AI machine learning are provided. The technical solution solves the problem that most of the existing intelligent packaging material recommendation technologies at home and abroad rely on traditional data analysis and rule engines, and fail to fully utilize advanced artificial intelligence and machine learning algorithms for deep learning and self-optimization. The existing methods usually ignore the multi-dimensional features of goods and the dynamic changes of transportation environment, lack personalized recommendation ability based on real data driving, and are difficult to make accurate recommendations when facing complex and diversified packaging requirements. As a result, the adaptability and optimization degree of the recommended results are low, and the system cannot respond to changes in packaging requirements in real time, affecting the effectiveness and economy of the packaging scheme.

[0004] To achieve the above purposes, the technical solution adopted by the present application is as follows: An intelligent packaging material recommendation method based on AI machine learning, comprising: Performing fuzzy clustering according to the multi-dimensional data of the goods to construct a same goods type knowledge database; Based on the same goods type knowledge database, according to the attribute preferences of the goods type, known packaging materials and known packaging methods are screened for the goods type, which is recorded as a packaging material-packaging method classification set of the same goods type, an intelligent packaging material recommendation multi-modal model is trained, and the packaging material-packaging method of the goods type to be packaged is generated; Based on the logistics background, the transportation destination of the goods to be packaged is obtained, the transportation multi-dimensional feature parameters of the goods to be packaged are determined, and the transportation demand of the goods to be packaged is quantified; Determine whether the packaging material-packaging method of the goods type to be packaged meets the transportation demand of the goods to be packaged. If yes, execute the packaging scheme normally, if not, submit a packaging reinforcement instruction; If a packaging enhancement instruction is submitted, the multi-dimensional transportation feature parameters of the goods to be packaged will be used as external interference variables. These parameters, along with the attribute preferences of the goods to be packaged and the available packaging materials and packaging methods for the type of goods to be packaged, will be substituted into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material and packaging method compensation scheme for the type of goods to be packaged.

[0005] Preferably, multi-dimensional data of the product is obtained, product feature attributes are extracted, and a product feature attribute dataset is generated; The optimal number of clusters is determined by the silhouette coefficient, and the cluster fuzziness is adjusted by setting the fuzziness index. An initial membership matrix satisfying the constraints is randomly generated, with the sum of the membership degrees of each product feature attribute to all clusters equal to 1, and the membership degree values ​​ranging from 0 to 1. The weighted Mahalanobis distance formula is used to measure the similarity between the product feature attributes and the cluster centers. The membership matrix and cluster center positions are dynamically updated until the dual convergence conditions of membership matrix change and cluster center offset are met. The membership probability distribution of product feature attributes to clusters is used as the output, as shown in the following formula: ; in, To calculate the maximum value between all product data points i and all clusters j, where i is the index of the product data point and j is the index of the product feature attribute cluster, Let be the membership degree of the i-th product data point to the j-th product feature attribute cluster at the t-th iteration. In the next iteration, the membership degree of the i-th product data point to the j-th product feature attribute cluster is... Let be the center vector of the j-th product feature attribute cluster at the t-th iteration. For the Euclidean norm, For all cluster centers The median of the norm, This is the convergence threshold; The index function is used to comprehensively evaluate the quality of clustering, where n is the total number of products and c is the number of product characteristic attribute clusters. The membership degree of the i-th product data point to the j-th product feature attribute cluster (the value is within the range of...). between), The natural logarithm of membership is used to calculate information entropy. For adjustment coefficients, Let j be the total membership degree of the j-th product feature attribute cluster. Ideally, this represents the average number of items that should be allocated to each category. According to the maximum membership principle, the commodity type set is obtained, the commodity type-feature mapping relationship is generated, and the same commodity type knowledge database is constructed.

[0006] Preferably, the known packaging materials and the known packaging methods of the same commodity type are determined. According to the greedy forward tree, an initial packaging material is assigned to each same commodity type as an initial node according to the known packaging materials of the same commodity type, a search tree is constructed, the optimal child node is selected as the target same commodity type with the minimum cost, the known packaging materials are prioritized using a weighted scoring function, and a local search strategy is introduced to recursively generate the selectable packaging materials of the same commodity type. Using a three-dimensional packing algorithm, combining the selectable packaging materials of the same commodity type, generating commodity type attribute preference ranking, using an adaptive weight allocation method, dynamically adjusting the attribute influence factor of the commodity type, following the heuristic rules and mixed integer programming of the packing process, maintaining the three-dimensional topology structure of the remaining available space in real time through the space partition tree, introducing a rotation feasibility detection mechanism, calculating the optimal placement coordinates and orientation of the goods, substituting into the double-layer iterative optimization architecture, dynamically adjusting the search strategy in the upper layer, verifying the feasibility of the scheme using mixed integer programming to solve the optimal space layout in the lower layer, and generating the selectable packaging methods of the same commodity type.

[0007] Preferably, the greedy forward tree is used as the first layer branch of the known packaging materials of the same commodity type, and the three-dimensional packing algorithm is used as the second layer branch of the known packaging methods of the same commodity type, the intelligent packaging material recommendation multi-modal model is trained, the known packaging materials and the known packaging methods of the same commodity type are used as inputs, and the attribute preferences of the commodity type are used as influence factors to generate the selectable packaging materials and packaging methods of the to-be-packaged commodity type.

[0008] Preferably, based on the logistics background, the transportation destination of the to-be-packaged goods is obtained, the transportation distance data, temperature and humidity range data, expected disassembly times data, and transportation tool type data of the to-be-packaged goods are extracted, the transportation multi-dimensional feature parameters of the to-be-packaged goods are determined, and data normalization processing is performed to map them to a unified interval ranging from 0 to 1. Based on the pre-processed transportation multi-dimensional feature parameters of the to-be-packaged goods, a transportation multi-dimensional feature matrix of the to-be-packaged goods is established. According to the transportation multi-dimensional feature matrix of the to-be-packaged goods, Min-Max range normalization processing is performed, the probability distribution proportion of each transportation parameter of the to-be-packaged goods in the transportation multi-dimensional parameters is calculated, a transportation parameter contribution degree matrix of the to-be-packaged goods is established, the discrete degree of each transportation parameter of the to-be-packaged goods is quantified, the transportation multi-dimensional feature parameter weight of the to-be-packaged goods is calculated, and a transportation risk scoring model of the to-be-packaged goods is constructed, with the formula as follows: ; in, Let be the fusion weight of the transportation parameters for the 'a'-th item to be packaged, where 'a' is the transportation parameter number of the item to be packaged. Let be the contribution vector of the transportation parameters for the a-th item to be packaged. The variance of contribution. It is the arctangent function. This is the historical accident rate vector. The covariance between contribution rate and accident rate, For maximum absolute covariance, b is a temporary index variable for the transportation parameters of the goods to be packaged; Let d be the overall transportation risk score for the d-th item to be packaged, where d is the item number, a is the transportation parameter number, and m is the total number of transportation parameters for the item to be packaged. Let be the normalized principal characteristic value of the a-th product to be packaged on the transportation parameters of the d-th product to be packaged. To amplify the effect of the sensitivity gradient, As a stability regulator, Let be the sensitivity gradient of the a-th product to be packaged with respect to the transportation parameters of the d-th product to be packaged. This is a stability index of the a-th product to be packaged on the transportation parameters of the d-th product to be packaged; Using a linear weighted formula, the comprehensive transportation risk score of the goods to be packaged is calculated. Based on industry safety standards, the transportation risk level of the goods to be packaged is classified, and the transportation demand of the goods to be packaged is quantified.

[0009] Preferably, based on the available packaging materials and packaging methods for the type of goods to be packaged, a unique packaging material and packaging method for the type of goods to be packaged is randomly assigned. In view of the transportation needs of the goods to be packaged, a dual constraint evaluation system based on the maximum load-bearing capacity and the minimum cushioning thickness is established to determine whether the available packaging materials and packaging methods for the type of goods to be packaged meet the transportation needs of the goods to be packaged. If the maximum load-bearing capacity of the packaging materials and packaging methods available for the type of goods to be packaged is greater than the weight of the goods to be packaged, and the minimum cushioning thickness is greater than the cushioning requirement corresponding to the maximum impact force that the goods to be packaged may suffer during transportation, then it is determined that the packaging materials and packaging methods available for the type of goods to be packaged meet the transportation requirements of the goods to be packaged, and the packaging plan is executed normally. If the maximum load-bearing capacity of the packaging materials and packaging methods available for the product type to be packaged is less than the weight of the product to be packaged, or the minimum cushioning thickness is less than the cushioning requirement corresponding to the maximum impact force that the product to be packaged may suffer during transportation, then it is determined that the packaging materials and packaging methods available for the product type to be packaged do not meet the transportation requirements of the product to be packaged, and a packaging reinforcement instruction is submitted.

[0010] Preferably, a multimodal neural network architecture is constructed by using the multidimensional transportation feature parameters of the goods to be packaged as external disturbance variables and the attribute preferences of the goods to be packaged and the packaging materials and packaging methods that can be selected for the type of goods to be packaged as inputs. Using the available packaging materials and packaging methods for the types of goods to be packaged as the state space, the selection of packaging materials and packaging methods for the types of goods to be packaged as the action space, and transportation cost and product protection degree as the reward function, a deep neural network is established as the Q function approximator for transportation cost and product protection degree, and the Q value table for transportation cost and product protection degree is initialized. Using a greedy strategy, a compensation scheme for packaging materials and packaging methods is randomly selected for the type of goods to be packaged. The optimal compensation scheme for packaging materials and packaging methods for the type of goods to be packaged is then predicted based on a deep neural network. Based on the Bellman equation, the Q-values ​​of transportation cost and commodity protection are updated. By utilizing experience replay, the state space, action space, and reward function are stored. Random sampling is used for batch training to optimize the parameters of the deep neural network. The parameters are continuously iterated and updated until convergence. The package is then incorporated into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material-packaging method compensation scheme for the type of commodity to be packaged.

[0011] Furthermore, an AI machine learning-based intelligent packaging material recommendation system, used to implement the aforementioned AI machine learning-based intelligent packaging material recommendation method, includes: The module includes a knowledge database module, a packaging method module for the types of goods to be packaged, a transportation demand module, a module to determine whether the transportation demand of the goods to be packaged is met, and a compensation scheme module for the packaging method of the types of goods to be packaged. The knowledge database module is used to perform fuzzy clustering based on the multi-dimensional data of products to build a knowledge database of the same product type. The packaging method module for the product type to be packaged is electrically connected to the knowledge database module. Based on the knowledge database of the same product type, according to the attribute preferences of the product type, the known packaging materials and known packaging methods are screened for the product type. This is recorded as the packaging material-packaging method classification set of the same product type. The intelligent packaging material recommendation multimodal model is trained to generate the packaging materials-packaging methods that can be selected for the product type to be packaged. The transportation demand module is used to obtain the transportation destination of the goods to be packaged based on the logistics backend, determine the multi-dimensional characteristic parameters of the goods to be packaged, and quantify the transportation demand of the goods to be packaged. The module for determining whether the transportation requirements of the goods to be packaged are met is electrically connected to the packaging method module and the transportation requirements module for the type of goods to be packaged. It determines whether the packaging materials and packaging methods available for the type of goods to be packaged meet the transportation requirements of the goods to be packaged. If yes, the packaging plan is executed normally; otherwise, a packaging reinforcement instruction is submitted. The packaging method compensation scheme module for the type of goods to be packaged is electrically connected to the module that determines whether the transportation requirements of the goods to be packaged are met. If a packaging reinforcement instruction is submitted, the multi-dimensional transportation feature parameters of the goods to be packaged are used as external interference variables, along with the attribute preferences of the goods to be packaged and the available packaging materials and packaging methods for the type of goods to be packaged, and then substituted into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material and packaging method compensation scheme for the type of goods to be packaged.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes an intelligent packaging material recommendation method and system based on AI machine learning. By introducing AI machine learning and multimodal models, this invention significantly improves the accuracy and adaptability of the intelligent packaging material recommendation system. Through fuzzy clustering and product attribute preference analysis, it can dynamically generate personalized packaging solutions based on product type and multi-dimensional transportation characteristics. Unlike traditional systems based on static rules and simple matching, this method can respond to changes in transportation demand in real time, providing more optimized packaging solutions, reducing losses during the packaging process, and improving transportation efficiency and economic benefits. Furthermore, by combining finite element analysis technology to evaluate structural strength and cushioning performance, it ensures that the packaging design can effectively withstand external forces in complex transportation environments, enhancing the safety and reliability of the packaging solution. Attached Figure Description

[0013] Figure 1 This is a flowchart of an AI machine learning-based intelligent packaging material recommendation method. Figure 2 This is a framework diagram of an AI machine learning-based intelligent packaging material recommendation system. Detailed Implementation

[0014] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0015] Reference Figure 1 As shown, an intelligent packaging material recommendation method based on AI machine learning includes: S1. Perform fuzzy clustering based on the multi-dimensional data of the products to construct a knowledge database of the same product type; the multi-dimensional data includes: physical attributes, chemical attributes and historical packaging data; Step S1 includes the following: Acquire multi-dimensional data of products, extract product feature attributes, and generate a product feature attribute dataset; The optimal number of clusters is determined by the silhouette coefficient, and the cluster fuzziness is adjusted by setting the fuzziness index. An initial membership matrix satisfying the constraints is randomly generated, with the sum of the membership degrees of each product feature attribute to all clusters equal to 1, and the membership degree values ​​ranging from 0 to 1. The weighted Mahalanobis distance formula is used to measure the similarity between the product feature attributes and the cluster centers. The membership matrix and cluster center positions are dynamically updated until the dual convergence conditions of membership matrix change and cluster center offset are met. The membership probability distribution of product feature attributes to clusters is used as the output, as shown in the following formula: ; in, To calculate the maximum value between all product data points i and all clusters j, where i is the index of the product data point and j is the index of the product feature attribute cluster, Let be the membership degree of the i-th product data point to the j-th product feature attribute cluster at the t-th iteration. In the next iteration, the membership degree of the i-th product data point to the j-th product feature attribute cluster is... Let be the center vector of the j-th product feature attribute cluster at the t-th iteration. For the Euclidean norm, For all cluster centers The median of the norm, This is the convergence threshold; The index function is used to comprehensively evaluate the quality of clustering, where n is the total number of products and c is the number of product characteristic attribute clusters. The membership degree of the i-th product data point to the j-th product feature attribute cluster (the value is within the range of...). between), The natural logarithm of membership is used to calculate information entropy. For adjustment coefficients, Let j be the total membership degree of the j-th product feature attribute cluster. Ideally, this represents the average number of items that should be allocated to each category. Based on the principle of maximum membership, commodity types are divided into sets, commodity type-feature mapping relationships are generated, and a knowledge database of the same commodity type is constructed.

[0016] When using it, refer to the content of step S1 above: Existing product data clustering technologies mostly rely on hard clustering algorithms, ignoring the fuzziness and uncertainty of product features. This leads to unsatisfactory results when processing multidimensional data, especially when data is abnormal or missing, making it susceptible to interference. Furthermore, the number of clusters is usually manually set, lacking flexibility. This step addresses the uncertainty in product features through fuzzy clustering, dynamically optimizing the number of clusters and the fuzziness factor, avoiding the shortcomings of manual setting and improving clustering accuracy and stability. Combined with the maximum membership principle, it can accurately classify product types and establish a product type-feature mapping relationship, building a more accurate product knowledge base and improving the effectiveness of product recommendation and management.

[0017] S2. Based on a knowledge database of the same product type, according to the attribute preferences of the product type, filter known packaging materials and known packaging methods for the product type, and denot this as the packaging material-packaging method classification set of the same product type. Train an intelligent packaging material recommendation multimodal model to generate the packaging materials-packaging methods that can be selected for the product type to be packaged; the known packaging materials include: cardboard, plastic, metal, glass and wood; the known packaging methods include: box packaging, bag packaging and can packaging; Step S2 includes the following: Identify known packaging materials and packaging methods for the same product type; Based on the greedy forward tree, according to the known packaging materials of the same product type, an initial packaging material is assigned to each product type as the initial node. A search tree is constructed with the goal of minimizing cost. The optimal child node is selected from the known packaging materials of the same product type. The known packaging materials are prioritized using a weighted scoring function. A local search strategy is introduced to recursively generate optional packaging materials of the same product type. Using a 3D packing algorithm, combined with optional packaging materials for the same product type, and sorting by product type attribute preferences, an adaptive weight allocation method is adopted to dynamically adjust the influencing factors of each attribute of the product type. According to the heuristic rules of the packing process and mixed integer programming, the 3D topology of the remaining available space is maintained in real time through a spatial partitioning tree. A rotation feasibility detection mechanism is introduced to calculate the optimal placement coordinates and orientation of the product. Substituted into a two-layer iterative optimization architecture, the upper layer dynamically adjusts the search strategy, and the lower layer uses mixed integer programming to verify the feasibility of the scheme and solve for the optimal spatial layout, generating optional packaging methods for the same product type. Step S2 also includes: A greedy forward tree is used as the first-level branch of known packaging materials for the same product type, and a 3D packing algorithm is used as the second-level branch of known packaging methods for the same product type. A multimodal model for intelligent packaging material recommendation is trained. Known packaging materials and known packaging methods for the same product type are used as inputs, and the attribute preferences of the product type are used as influencing factors to generate optional packaging materials and packaging methods for the product type to be packaged.

[0018] When using it, refer to the content of step S2 above: Most existing intelligent packaging material recommendation systems, both domestically and internationally, only process single-modal data, lacking precise filtering of product type attribute preferences. Furthermore, the separation of packaging method recommendation from cost prediction leads to low system efficiency. In contrast, this step, which comprehensively considers product attributes, packaging materials, and methods, can generate packaging solutions that better meet actual needs. It introduces local search and rotation detection mechanisms to avoid local optima and improve global optimization. The adaptive weight allocation method can dynamically adjust influencing factors, accurately matching products and packaging materials, thus improving efficiency and cost optimization. The trained intelligent model optimizes packaging strategies through real-time feedback, enhancing adaptability and practicality.

[0019] S3. Based on the logistics backend, obtain the transportation destination of the goods to be packaged, determine the multi-dimensional characteristic parameters of the goods to be packaged, and quantify the transportation needs of the goods to be packaged. Step S3 includes the following: Based on the logistics backend, the transportation destination of the goods to be packaged is obtained, and the transportation distance data, temperature and humidity range data, expected number of disassembly data and transportation vehicle type data of the goods to be packaged are extracted. The multi-dimensional characteristic parameters of the transportation of the goods to be packaged are determined, and the data is normalized and mapped to a unified range of 0 to 1. Based on the preprocessed multi-dimensional transportation feature parameters of the goods to be packaged, a multi-dimensional transportation feature matrix of the goods to be packaged is established. Based on the multi-dimensional feature matrix of the goods to be packaged, Min-Max range standardization is performed. The probability distribution weight of each transportation parameter of the goods to be packaged in the multi-dimensional transportation parameters is calculated, and a transportation parameter contribution matrix of the goods to be packaged is established. The dispersion of each transportation parameter of the goods to be packaged is quantified, the weights of the multi-dimensional transportation feature parameters of the goods to be packaged are calculated, and a transportation risk scoring model of the goods to be packaged is constructed. The formula is as follows: ; in, Let be the fusion weight of the transportation parameters for the 'a'-th item to be packaged, where 'a' is the transportation parameter number of the item to be packaged. Let be the contribution vector of the transportation parameters for the a-th item to be packaged. The variance of contribution. It is the arctangent function. This is the historical accident rate vector. The covariance between contribution rate and accident rate, For maximum absolute covariance, b is a temporary index variable for the transportation parameters of the goods to be packaged; Let d be the overall transportation risk score for the d-th item to be packaged, where d is the item number, a is the transportation parameter number, and m is the total number of transportation parameters for the item to be packaged. Let be the normalized principal characteristic value of the a-th product to be packaged on the transportation parameters of the d-th product to be packaged. To amplify the effect of the sensitivity gradient, As a stability regulator, Let be the sensitivity gradient of the a-th product to be packaged with respect to the transportation parameters of the d-th product to be packaged. This is a stability index of the a-th product to be packaged on the transportation parameters of the d-th product to be packaged; Using a linear weighted formula, the comprehensive transportation risk score of the goods to be packaged is calculated. Based on industry safety standards, the transportation risk level of the goods to be packaged is divided, and the transportation demand of the goods to be packaged is quantified. The transportation risk level of the goods to be packaged includes: less than 0.3 is low risk, between 0.3 and 0.6 is medium risk, and greater than 0.6 is high risk.

[0020] When using it, refer to the content of step S3 above: Existing domestic and international technologies for transportation risk assessment typically rely on static data and manual input, lacking real-time performance and accuracy. Furthermore, feature weight allocation is overly dependent on experience, neglecting the complex relationships between features. This step improves the real-time performance and accuracy of data by acquiring transportation destination and multi-dimensional feature data in real time from the logistics backend and employing data normalization processing. Simultaneously, it calculates a comprehensive risk score based on a weighted linear model and classifies risk levels, enabling a more scientific assessment of transportation risks and quantification of needs. This provides precise support for transportation decisions and improves transportation efficiency and safety.

[0021] S4. Determine whether the available packaging materials and packaging methods for the type of goods to be packaged meet the transportation requirements of the goods to be packaged. If yes, execute the packaging plan normally; otherwise, submit a packaging reinforcement instruction. Step S4 includes the following: Based on the available packaging materials and packaging methods for the types of goods to be packaged, a unique packaging material and packaging method is randomly assigned for each type of goods to be packaged. To determine whether the available packaging materials and packaging methods for each type of goods meet the transportation needs of the goods to be packaged, a dual-constraint evaluation system based on maximum load-bearing capacity and minimum cushioning thickness is established. If the maximum load-bearing capacity of the packaging materials and packaging methods available for the type of goods to be packaged is greater than the weight of the goods to be packaged, and the minimum cushioning thickness is greater than the cushioning requirement corresponding to the maximum impact force that the goods to be packaged may suffer during transportation, then it is determined that the packaging materials and packaging methods available for the type of goods to be packaged meet the transportation requirements of the goods to be packaged, and the packaging plan is executed normally. If the maximum load-bearing capacity of the packaging materials and packaging methods available for the product type to be packaged is less than the weight of the product to be packaged, or the minimum cushioning thickness is less than the cushioning requirement corresponding to the maximum impact force that the product to be packaged may suffer during transportation, then it is determined that the packaging materials and packaging methods available for the product type to be packaged do not meet the transportation requirements of the product to be packaged, and a packaging reinforcement instruction is submitted.

[0022] When using it, refer to the content of step S4 above: Existing domestic and international technologies for assessing packaging requirement fulfillment typically rely on static physical parameters, such as maximum load-bearing capacity and cushioning thickness, lacking a comprehensive consideration of dynamic changes during transportation, particularly failing to accurately assess the impact of potential shocks and other environmental factors. Furthermore, traditional methods often lack flexible adjustment mechanisms, unable to respond in real-time to the adaptability of different packaging solutions. In contrast, this step effectively improves the real-time adaptability and safety of packaging solutions by quantifying transportation requirements and defining constraints based on product type. When a packaging solution fails to meet transportation requirements, it automatically submits reinforcement instructions and adjusts the solution, thereby ensuring the rationality of the packaging solution and the safety of transportation, providing more precise and flexible support for packaging design in the logistics process.

[0023] S5. If a packaging enhancement instruction is submitted, the multi-dimensional transportation feature parameters of the goods to be packaged are used as external interference variables, along with the attribute preferences of the goods to be packaged and the available packaging materials and packaging methods for the type of goods to be packaged, and then substituted into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material and packaging method compensation scheme for the type of goods to be packaged; the multi-dimensional transportation feature parameters of the goods to be packaged include: goods weight, transportation method and transportation cost; Step S5 includes the following: A multimodal neural network architecture is constructed by using the multidimensional transportation feature parameters of the goods to be packaged as external disturbance variables and the attribute preferences of the goods to be packaged and the packaging materials and packaging methods that can be selected for the type of goods to be packaged as inputs. Using the available packaging materials and packaging methods for the types of goods to be packaged as the state space, the selection of packaging materials and packaging methods for the types of goods to be packaged as the action space, and transportation cost and product protection degree as the reward function, a deep neural network is established as the Q function approximator for transportation cost and product protection degree, and the Q value table for transportation cost and product protection degree is initialized. Using a greedy strategy, a compensation scheme for packaging materials and packaging methods is randomly selected for the type of goods to be packaged. The optimal compensation scheme for packaging materials and packaging methods for the type of goods to be packaged is then predicted based on a deep neural network. Based on the Bellman equation, the Q-values ​​of transportation cost and commodity protection are updated. By utilizing experience replay, the state space, action space, and reward function are stored. Random sampling is used for batch training to optimize the parameters of the deep neural network. The parameters are continuously iterated and updated until convergence. The package is then incorporated into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material-packaging method compensation scheme for the type of commodity to be packaged.

[0024] When using it, refer to the content of step S5 above: Currently, most domestic and international technologies in the field of intelligent packaging focus on single-factor optimization, such as considering only product attributes or transportation characteristics. However, they lack intelligent decision-making models that comprehensively consider multi-dimensional factors. Especially when considering the complexity of multi-dimensional transportation characteristics, product attribute preferences, and packaging material selection, existing technologies struggle to balance these factors, resulting in insufficient performance in multi-objective optimization. Existing reinforcement learning methods are mostly concentrated in traditional single-task environments, failing to fully consider the real-time dynamics and external interference variables in packaging tasks, and face significant challenges in large-scale data processing and model training efficiency. In contrast, the reinforcement learning method based on a multimodal neural network architecture proposed in this paper can comprehensively consider the transportation characteristics, product attribute preferences, and packaging material-packaging method selection of the product to be packaged. By optimizing the Q-value table through deep neural networks, the packaging scheme is updated in real time, thereby achieving a balance between transportation costs and product protection. This method, through multi-dimensional input and reward function optimization, can generate more accurate and personalized packaging recommendations, significantly improving the efficiency and accuracy of packaging decisions, reducing transportation costs, and enhancing product protection.

[0025] Reference Figure 2 As shown, an intelligent packaging material recommendation system based on AI machine learning is characterized by comprising: The module includes a knowledge database module, a packaging method module for the types of goods to be packaged, a transportation demand module, a module to determine whether the transportation demand of the goods to be packaged is met, and a compensation scheme module for the packaging method of the types of goods to be packaged. The knowledge database module is used to perform fuzzy clustering based on the multi-dimensional data of products to build a knowledge database of the same product type. The packaging method module for the product type to be packaged is electrically connected to the knowledge database module. Based on the knowledge database of the same product type, according to the attribute preferences of the product type, the known packaging materials and known packaging methods are screened for the product type. This is recorded as the packaging material-packaging method classification set of the same product type. The intelligent packaging material recommendation multimodal model is trained to generate the packaging materials-packaging methods that can be selected for the product type to be packaged. The transportation demand module is used to obtain the transportation destination of the goods to be packaged based on the logistics backend, determine the multi-dimensional characteristic parameters of the goods to be packaged, and quantify the transportation demand of the goods to be packaged. The module for determining whether the transportation requirements of the goods to be packaged are met is electrically connected to the packaging method module and the transportation requirements module for the type of goods to be packaged. It determines whether the packaging materials and packaging methods available for the type of goods to be packaged meet the transportation requirements of the goods to be packaged. If yes, the packaging plan is executed normally; otherwise, a packaging reinforcement instruction is submitted. The packaging method compensation scheme module for the type of goods to be packaged is electrically connected to the module that determines whether the transportation requirements of the goods to be packaged are met. If a packaging reinforcement instruction is submitted, the multi-dimensional transportation feature parameters of the goods to be packaged are used as external interference variables, along with the attribute preferences of the goods to be packaged and the available packaging materials and packaging methods for the type of goods to be packaged, and then substituted into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material and packaging method compensation scheme for the type of goods to be packaged.

[0026] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A smart packaging material recommendation method based on AI machine learning, characterized in that, include: S1. Perform fuzzy clustering based on the multi-dimensional data of the products to build a knowledge database of the same product type; S2. Based on the knowledge database of the same product type, according to the attribute preferences of the product type, filter the known packaging materials and known packaging methods for the product type, and denot them as the packaging material-packaging method classification set of the same product type. Train the intelligent packaging material recommendation multimodal model to generate the packaging materials-packaging methods that can be selected for the product type to be packaged. S3. Based on the logistics backend, obtain the transportation destination of the goods to be packaged, determine the multi-dimensional characteristic parameters of the goods to be packaged, and quantify the transportation needs of the goods to be packaged. S4. Determine whether the available packaging materials and packaging methods for the type of goods to be packaged meet the transportation requirements of the goods to be packaged. If yes, execute the packaging plan normally; otherwise, submit a packaging reinforcement instruction. S5. If a packaging enhancement instruction is submitted, the multi-dimensional transportation feature parameters of the goods to be packaged will be used as external interference variables. These parameters, along with the attribute preferences of the goods to be packaged and the available packaging materials and packaging methods for the type of goods to be packaged, will be substituted into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material and packaging method compensation scheme for the type of goods to be packaged.

2. The intelligent packaging material recommendation method based on AI machine learning according to claim 1, characterized in that, S1 includes the following: Acquire multi-dimensional data of products, extract product feature attributes, and generate a product feature attribute dataset; The optimal number of clusters is determined by the silhouette coefficient, and the cluster fuzziness is adjusted by setting the fuzziness index. An initial membership matrix satisfying the constraints is randomly generated, with the sum of the membership degrees of each product feature attribute to all clusters equal to 1, and the membership degree values ​​ranging from 0 to 1. The weighted Mahalanobis distance formula is used to measure the similarity between the product feature attributes and the cluster centers. The membership matrix and cluster center positions are dynamically updated until the dual convergence conditions of membership matrix change and cluster center offset are met. The membership probability distribution of product feature attributes to clusters is used as the output, as shown in the following formula: ； in, To calculate the maximum value between all product data points i and all clusters j, where i is the index of the product data point and j is the index of the product feature attribute cluster, Let be the membership degree of the i-th product data point to the j-th product feature attribute cluster at the t-th iteration. For the first In the next iteration, the membership degree of the i-th product data point to the j-th product feature attribute cluster is... Let be the center vector of the j-th product feature attribute cluster at the t-th iteration. For the Euclidean norm, For all cluster centers The median of the norm, This is the convergence threshold; The index function is used to comprehensively evaluate the quality of clustering, where n is the total number of products and c is the number of product characteristic attribute clusters. The membership degree of the i-th product data point to the j-th product feature attribute cluster (the value is within the range of...). between), The natural logarithm of membership is used to calculate information entropy. For adjustment coefficients, Let j be the total membership degree of the j-th product feature attribute cluster. Ideally, this represents the average number of items that should be allocated to each category. Based on the principle of maximum membership, commodity types are divided into sets, commodity type-feature mapping relationships are generated, and a knowledge database of the same commodity type is constructed.

3. The intelligent packaging material recommendation method based on AI machine learning according to claim 2, characterized in that, S2 includes the following: Identify known packaging materials and packaging methods for the same product type; Based on the greedy forward tree, according to the known packaging materials of the same product type, an initial packaging material is assigned to each product type as the initial node. A search tree is constructed with the goal of minimizing cost. The optimal child node is selected from the known packaging materials of the same product type. The known packaging materials are prioritized using a weighted scoring function. A local search strategy is introduced to recursively generate optional packaging materials of the same product type. Using a 3D packing algorithm and combining optional packaging materials for the same product type, the algorithm sorts the generated product type attributes according to their preferences and uses an adaptive weight allocation method to dynamically adjust the influencing factors of each attribute of the product type. Following the heuristic rules of the packing process and mixed integer programming, the algorithm maintains the 3D topology of the remaining available space in real time through a spatial partitioning tree. A rotation feasibility detection mechanism is introduced to calculate the optimal placement coordinates and orientation of the product. The algorithm is then substituted into a two-layer iterative optimization architecture. The upper layer dynamically adjusts the search strategy, while the lower layer uses mixed integer programming to verify the feasibility of the solution and solve for the optimal spatial layout, generating optional packaging methods for the same product type.

4. The intelligent packaging material recommendation method based on AI machine learning according to claim 3, characterized in that, S2 further includes: A greedy forward tree is used as the first-level branch of known packaging materials for the same product type, and a 3D packing algorithm is used as the second-level branch of known packaging methods for the same product type. A multimodal model for intelligent packaging material recommendation is trained. Known packaging materials and known packaging methods for the same product type are used as inputs, and the attribute preferences of the product type are used as influencing factors to generate optional packaging materials and packaging methods for the product type to be packaged.

5. The intelligent packaging material recommendation method based on AI machine learning according to claim 1, characterized in that, S3 includes the following: Based on the logistics backend, the transportation destination of the goods to be packaged is obtained, and the transportation distance data, temperature and humidity range data, expected number of disassembly times data and transportation vehicle type data of the goods to be packaged are extracted. The multi-dimensional characteristic parameters of the transportation of the goods to be packaged are determined, and the data is normalized and mapped to a unified range of 0 to 1. Based on the preprocessed multi-dimensional transportation feature parameters of the goods to be packaged, a multi-dimensional transportation feature matrix of the goods to be packaged is established. Based on the multi-dimensional feature matrix of the goods to be packaged, Min-Max range standardization is performed. The probability distribution weight of each transportation parameter of the goods to be packaged in the multi-dimensional transportation parameters is calculated, and a transportation parameter contribution matrix of the goods to be packaged is established. The dispersion of each transportation parameter of the goods to be packaged is quantified, the weights of the multi-dimensional transportation feature parameters of the goods to be packaged are calculated, and a transportation risk scoring model of the goods to be packaged is constructed. The formula is as follows: ； in, Let be the fusion weight of the transportation parameters for the 'a'-th item to be packaged, where 'a' is the transportation parameter number of the item to be packaged. Let be the contribution vector of the transportation parameters for the a-th item to be packaged. The variance of contribution. It is the arctangent function. This is the historical accident rate vector. The covariance between contribution rate and accident rate, For maximum absolute covariance, b is a temporary index variable for the transportation parameters of the goods to be packaged; Let d be the overall transportation risk score for the d-th item to be packaged, where d is the item number, a is the transportation parameter number, and m is the total number of transportation parameters for the item to be packaged. Let be the normalized principal characteristic value of the a-th product to be packaged on the transportation parameters of the d-th product to be packaged. To amplify the effect of the sensitivity gradient, As a stability regulator, Let be the sensitivity gradient of the a-th product to be packaged with respect to the transportation parameters of the d-th product to be packaged. This is a stability index of the a-th product to be packaged on the transportation parameters of the d-th product to be packaged; Using a linear weighted formula, the comprehensive transportation risk score of the goods to be packaged is calculated. Based on industry safety standards, the transportation risk level of the goods to be packaged is classified, and the transportation demand of the goods to be packaged is quantified.

6. The intelligent packaging material recommendation method based on AI machine learning according to claim 5, characterized in that, S4 includes the following: Based on the available packaging materials and packaging methods for the types of goods to be packaged, a unique packaging material and packaging method is randomly assigned for each type of goods to be packaged. To determine whether the available packaging materials and packaging methods for each type of goods meet the transportation needs of the goods to be packaged, a dual-constraint evaluation system based on maximum load-bearing capacity and minimum cushioning thickness is established. If the maximum load-bearing capacity of the packaging materials and packaging methods available for the type of goods to be packaged is greater than the weight of the goods to be packaged, and the minimum cushioning thickness is greater than the cushioning requirement corresponding to the maximum impact force that the goods to be packaged may suffer during transportation, then it is determined that the packaging materials and packaging methods available for the type of goods to be packaged meet the transportation requirements of the goods to be packaged, and the packaging plan is executed normally. If the maximum load-bearing capacity of the packaging materials and packaging methods available for the product type to be packaged is less than the weight of the product to be packaged, or the minimum cushioning thickness is less than the cushioning requirement corresponding to the maximum impact force that the product to be packaged may suffer during transportation, then it is determined that the packaging materials and packaging methods available for the product type to be packaged do not meet the transportation requirements of the product to be packaged, and a packaging reinforcement instruction is submitted.

7. The intelligent packaging material recommendation method based on AI machine learning according to claim 6, characterized in that, S5 includes the following: A multimodal neural network architecture is constructed by using the multidimensional transportation feature parameters of the goods to be packaged as external disturbance variables and the attribute preferences of the goods to be packaged and the packaging materials and packaging methods that can be selected for the type of goods to be packaged as inputs. Using the available packaging materials and packaging methods for the types of goods to be packaged as the state space, the selection of packaging materials and packaging methods for the types of goods to be packaged as the action space, and transportation cost and product protection degree as the reward function, a deep neural network is established as the Q function approximator for transportation cost and product protection degree, and the Q value table for transportation cost and product protection degree is initialized. Using a greedy strategy, a compensation scheme for packaging materials and packaging methods is randomly selected for the type of goods to be packaged. The optimal compensation scheme for packaging materials and packaging methods for the type of goods to be packaged is then predicted based on a deep neural network. Based on the Bellman equation, the Q-values ​​of transportation cost and commodity protection are updated. By utilizing experience replay, the state space, action space, and reward function are stored. Random sampling is used for batch training to optimize the parameters of the deep neural network. The parameters are continuously iterated and updated until convergence. The package is then incorporated into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material-packaging method compensation scheme for the type of commodity to be packaged.

8. A smart packaging material recommendation system based on AI machine learning, characterized in that, To implement any one of claims 1-7, the intelligent packaging material recommendation method based on AI machine learning includes: The module includes a knowledge database module, a packaging method module for the types of goods to be packaged, a transportation demand module, a module to determine whether the transportation demand of the goods to be packaged is met, and a compensation scheme module for the packaging method of the types of goods to be packaged. The knowledge database module is used to perform fuzzy clustering based on the multi-dimensional data of products to build a knowledge database of the same product type. The module includes a knowledge database module, a packaging method module for the types of goods to be packaged, a transportation demand module, a module to determine whether the transportation demand of the goods to be packaged is met, and a compensation scheme module for the packaging method of the types of goods to be packaged. The knowledge database module is used to perform fuzzy clustering based on the multi-dimensional data of products to build a knowledge database of the same product type. The packaging method module for the product type to be packaged is electrically connected to the knowledge database module. Based on the knowledge database of the same product type, according to the attribute preferences of the product type, the known packaging materials and known packaging methods are screened for the product type. This is recorded as the packaging material-packaging method classification set of the same product type. The intelligent packaging material recommendation multimodal model is trained to generate the packaging materials-packaging methods that can be selected for the product type to be packaged. The transportation demand module is used to obtain the transportation destination of the goods to be packaged based on the logistics backend, determine the multi-dimensional characteristic parameters of the goods to be packaged, and quantify the transportation demand of the goods to be packaged. The module for determining whether the transportation requirements of the goods to be packaged are met is electrically connected to the packaging method module and the transportation requirements module for the type of goods to be packaged. It determines whether the packaging materials and packaging methods available for the type of goods to be packaged meet the transportation requirements of the goods to be packaged. If yes, the packaging plan is executed normally; otherwise, a packaging reinforcement instruction is submitted. The packaging method compensation scheme module for the type of goods to be packaged is electrically connected to the module that determines whether the transportation requirements of the goods to be packaged are met. If a packaging reinforcement instruction is submitted, the multi-dimensional transportation feature parameters of the goods to be packaged are used as external interference variables, along with the attribute preferences of the goods to be packaged and the available packaging materials and packaging methods for the type of goods to be packaged, and then substituted into the intelligent packaging material recommendation multimodal model to generate the optimal packaging material and packaging method compensation scheme for the type of goods to be packaged.

Images